Abstract

In medical applications it is important that aerosols reach the alveolar

zone of the respiratory tract, to be effective. Before this region is reached,
the aerosols have to pass the upper airway (UA), starting with the mouth
and a 90-degree bend leading into the trachea. The UA geometrys irregularity and constrictions (such as the vocal cords) potentially affect the
deposition of inhaled aerosols.
The goal of this dissertation was to develop from the available CT-scans,
a simplified yet realistic human UA geometry. From this computer generated UA geometry, a suitable physical model was created for Particle Image
Velocimetry (PIV) measurements. Via Reynolds similitude, a seeded waterglycerine mixture, matching the refraction index of the transparent model
was measured in a central sagittal plane of the model at four flow rates (corresponding to 10, 15, 30 and 45 L/min air breathing flow rate). These PIV
measurements were compared with Computational Fluid Dynamics (CFD)
simulations of the fluid phase. Of the various available turbulence models that were combined with the Reynolds Averaged Navier-Stokes (RANS)
equations to compute the fluid phase in this UA model, the k ShearStress Transport (SST) turbulence model best reproduced the experimental
results.
For the simulation of the particle phase, particles with diameter ranging
from 1 to 20 micrometer were tracked in a Lagrangian frame of reference
through the obtained converged flow field. Simulations of total deposition
compared well with experimental deposition data, for particles with a value
for the non-dimensional parameter Stk.Re0.37 higher than 0.1
(Stokesnumber.Reynoldsnumber 0.37 > 0.1). Total deposition of particles
with a smaller value was overpredicted, probably due to exaggerated turbulence simulated at low flow rates. Simulations of local particle deposition
patterns in the UA model were much more realistic than local deposition
patters previously reported in simplified geometries. In particular, simulated mouth deposition more closely resembled that obtained experimentally in realistic upper airway geometries. The influences of gravity, of carrier gas, of degree of turbulence at the model entrance, and of considering
non-steady flow at particle injection were discussed.
i

Finally a clinical problem of tracheal stenosis was tackled by introducing

various degrees of constriction in the upper third of the trachea in the UA
model. CFD simulations of pressure drops across the stenosis allowed us
to propose a rule of thumb from which pressure drops over the stenosis
can be estimated, simply on the basis of breathing flow and stenosis cross
section. In addition, the best-fit exponent in the power law that relates
pressure drop to breathing flow was proposed as a diagnostic tool in the
non-invasive monitoring of tracheal stenosis patients.

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.

Mark Brouns, Vilvoorde,Oktober 2007

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

2 Anatomy of the Human Respiratory Tract

2.1 The respiratory system . . . . . . . . . . . . . . . . . . . . .
2.1.1 Function of the respiratory system . . . . . . . . . .
2.1.2 The structure of the respiratory system . . . . . . .

4
4
4
6

3 Theoretical Background of Particle Image Velocimetry (PIV) 12

3.1 Historical background of fluid measurements . . . . . . . . . 12
3.2 The principle of particle image velocimetry . . . . . . . . . . 15
3.3 Mathematical background of PIV evaluation . . . . . . . . . 17
3.3.1 Auto-correlation . . . . . . . . . . . . . . . . . . . . 19
3.3.2 Cross-correlation of a pair of two single exposed recordings . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3.3 Optimization of the correlation . . . . . . . . . . . . 22
3.4 Evaluation of PIV images . . . . . . . . . . . . . . . . . . . 23
3.4.1 Error estimation . . . . . . . . . . . . . . . . . . . . 26
3.4.2 Detection of spurious vectors (outliers) . . . . . . . . 26
3.5 Tracer particles . . . . . . . . . . . . . . . . . . . . . . . . . 29
v

4 Theoretical Background of Computational

(CFD): The Particle Phase
4.1 Introduction . . . . . . . . . . . . . . . . .
4.2 Geometric properties of particles . . . . .
4.2.1 Particle size . . . . . . . . . . . . .
4.3 Dilute and dense flows . . . . . . . . . . .
4.4 Phase coupling . . . . . . . . . . . . . . .
4.5 Modeling two-phase flows . . . . . . . . .
4.5.1 Eulerian continuum approach . . .
4.5.2 Lagrangian trajectory approach . .
4.6 Mass Balance . . . . . . . . . . . . . . . .
4.7 Momentum Balance . . . . . . . . . . . . .
4.7.1 Interphase Force . . . . . . . . . .
4.7.2 Body Force . . . . . . . . . . . . .
4.8 Stochastic trajectory approach . . . . . . .
5 Theoretical background of Computational
(CFD): The Fluid Phase
5.1 Introduction . . . . . . . . . . . . . . . . .
5.2 History of CFD . . . . . . . . . . . . . . .
5.3 The Navier-Stokes equations . . . . . . . .
5.4 The Reynolds Averaging . . . . . . . . . .
5.5 Turbulence modeling . . . . . . . . . . . .
5.5.1 k turbulence models . . . . . .
5.5.2 k turbulence models . . . . . .
5.5.3 The Reynolds Stress Model (RSM)

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

6 State-of-the-Art of the Research in Upper Airway Geometries

58
7 PIV of the Flow in a Model of the Upper Human Airways 69
7.1 Creation of the phantom . . . . . . . . . . . . . . . . . . . . 69
7.1.1 Creation of the computer model of the upper human
respiratory tract . . . . . . . . . . . . . . . . . . . . 69
7.1.2 Creation of the upper respiratory airways phantom . 72
7.2 Flow measurements . . . . . . . . . . . . . . . . . . . . . . . 77
7.2.1 The experimental-set-up . . . . . . . . . . . . . . . . 77
7.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
vi

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

Upper Human Airways 93

. . . . . . . . . . . . . . . 93
. . . . . . . . . . . . . . . 93
. . . . . . . . . . . . . . . 96
. . . . . . . . . . . . . . . 99
. . . . . . . . . . . . . . . 114

9 Numerical Particles Deposition Study in a Model of

Upper Human Airways
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . .
9.2 Method . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3.1 Validation and total deposition analysis . . . . . .
9.3.2 Local deposition analysis . . . . . . . . . . . . . .
9.3.3 Influence of gravity . . . . . . . . . . . . . . . . .
9.3.4 Influence of turbulence (Eddy Interaction Model)
9.3.5 Influence of the carrier gas (Heliox vs Air) . . . .
9.3.6 Influence of unsteady flow rate . . . . . . . . . . .
9.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . .

the
115
. . 115
. . 116
. . 118
. . 118
. . 122
. . 124
. . 126
. . 128
. . 131
. . 135

10 Clinical Application: Tracheal Stenoses

10.1 Introduction . . . . . . . . . . . . . . . .
10.2 Materials and methods . . . . . . . . . .
10.3 Results . . . . . . . . . . . . . . . . . . .
10.4 Discussion . . . . . . . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

138
138
140
143
148

11 Conclusions and Future Challenges

153

Bibliography

170

A List of publications

171

vii

List of Figures
2.1
2.2
2.3
2.4
2.5
2.6

representation of the breathing cycle of a human [4] . . .

representation of an alveolus [4] . . . . . . . . . . . . . .
picture from the cilia in the respiratory tract [5] . . . . .
frontal view of the mouth with the different structures [2]
side view of the mouth and pharynx [6] . . . . . . . . . .
detailed front(left) and top (right) view of the larynx [8]

.
.
.
.
.
.

.
.
.
.
.
.

3.1 Leonardo da Vinci sketched various flow fields over objects

in a flowing stream . . . . . . . . . . . . . . . . . . . . . . .
3.2 Ludwig Prandtl next to his famous water tunnel . . . . . . .
3.3 Experimental arrangement for PIV in a wind tunnel . . . . .
3.4 The three modes of particle image density: (a) low (PTV),
(b) medium (PIV) and high image density (LSV) . . . . . .
3.5 Schematic representation of geometric imaging . . . . . . . .
3.6 Example of an intensity field I . . . . . . . . . . . . . . . . .
3.7 Schematic representation of the auto-correlation of the intensity field I . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.8 The intensity field I recorded at time t and the intensity field

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

3.13 Light scattering by a (from top to bottom) 1 m, 10 m and

30 m glass particle in water . . . . . . . . . . . . . . . . . .

31

4.1 Particle size distribution in gas-solid flows (after Soo, 1990) .

4.2 Map for particle-turbulence modulation (after Elghobashi,
1994) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 Drag coefficient for spheres as a function of particle Reynolds
number (after Schlichting, 1979) . . . . . . . . . . . . . . . .
4.4 Drag coefficient computed with the different formulations for
spheres as a function of particle Reynolds number; right
panel shows a zoom . . . . . . . . . . . . . . . . . . . . . .

40

6.1
6.2
6.3
6.4

.
.
.
.

59
60
63
65

7.1 Creation of realistic geometry . . . . . . . . . . . . . . . . .

7.2 Comparison of flow field in realistic (left) and simplified geometry (right) . . . . . . . . . . . . . . . . . . . . . . . . . .
7.3 side-view and rear-view of the smoothed 3D model of the upper human airways with cross-sections at different locations
of the model . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.4 STL-model suspended in plasticine . . . . . . . . . . . . . .
7.5 metal positive placed in perspex box . . . . . . . . . . . .
7.6 Phantom of upper human airway model with the glycerine/water mixture in the pharynx . . . . . . . . . . . . . . .
7.7 scheme of the experimental set-up . . . . . . . . . . . . . .
7.8 an image pair (a. image 1, b. image 2) with the obtained
correlation coefficient (c) . . . . . . . . . . . . . . . . . . . .
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 . . . .
7.10 streaklines and contour of normalized magnitude of velocity
in the upper human airway model at 15 l/min (a) and 30
l/min (b) air flow rate . . . . . . . . . . . . . . . . . . . . .

70

geometry
geometry
geometry
geometry

developed by Katz and Martonen [68]

used by Corcoran and Chigier [27] . .
developed by Zhang et al. [145] . . . .
developed by Stapleton et al. [115] . .

ix

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

33
35
38

71

75
76
76
79
80
81

82

83

7.11 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 85
7.12 detailed view of the streaklines of velocity in the pharynx at
10 l/min (a), 15 l/min (b), 30 l/min (c), 45 l/min (d) air flow
rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

86

7.13 Cartesian plot of the normalized axial velocity in the pharynx

for 10, 15, 30 and 45 l/min air flow rate . . . . . . . . . . . .

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

7.15 contour plots of the normalized turbulent kinetic energy (k norm)

for 10 l/min (a), 15 l/min (b), 30 l/min (c) and 45 l/min (d) 90
7.16 normalized turbulent kinetic energy at four different locations
in the trachea for all measured flow rates . . . . . . . . . . .

91

8.1 Three dimensional view of the grid in the central sagittal

plane with a detailed view of the boundary layer cells . . .

94

8.2 Comparison of normalized magnitude of velocity (left) and

normalized turbulent kinetic energy (right) for three grid
sized (400 000, 800 000 and 1 500 000 cells) at 5mm above
the epiglottis (up) and one tracheal diameter downstream the
glottis (down) . . . . . . . . . . . . . . . . . . . . . . . . .

95

8.3 Comparison of normalized turbulent kinetic energy (right)

for two different inlet boundary conditions at 5mm above
the epiglottis (left) and one tracheal diameter downstream
the glottis (right) at 45 l/min . . . . . . . . . . . . . . . . .

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

8.7 zoom of the streaklines in the pharynx at 15 L/min: panel

a: k--sst, panel b: k--realizable, panel c: reynolds stress
model and panel b: experiments at 15 l/min . . . . . . . . .
8.8 zoom of the streaklines in the trachea at 15 L/min: panel
a: k--sst, panel b: k--realizable, panel c: reynolds stress
model and panel b: experiments at 15 l/min . . . . . . . . .
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
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 . . . . . . . . . . . . . .
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 . . . . . . . . . . . . . . .
8.12 cross-sectional view of the streamlines in the mouth (a), pharynx (b) and trachea (c) . . . . . . . . . . . . . . . . . . . . .
8.13 a three-dimensional view of the streamlines . . . . . . . . . .
9.1 Inspiratory deposition efficiency: comparison of simulated
deposition with reported experimental data . . . . . . . . . .
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] . . . . . . . . . . . . . . . . . . . . . . .
9.3 sites of deposition: inlet tube (blue), mouth (green), pharynx
(red) and larynx + trachea (yellow) . . . . . . . . . . . . . .
9.4 Simulated deposition values (expressed as % of total number
of particles) in three model subparts for four different flow
rates (10, 15, 30 and 45 L/min) . . . . . . . . . . . . . . . .
9.5 Two-dimensional representation of individual particle deposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.6 Top view of the deposited 10 m particles for a flow rate of
15 L/min . . . . . . . . . . . . . . . . . . . . . . . . . . . .
xi

103

103

107

108

111
112
113
119

120
122

124
125
125

9.7 Total deposition in zero gravity, gravity vector under an angle

of 45 ,gravity vector under an angle of 90 , normal gravity
vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
9.8 Comparison of total deposition between mean flow tracking
and Eddy Interaction Model . . . . . . . . . . . . . . . . . . 128
9.9 Comparison of total deposition between particles suspended
in air and heliox . . . . . . . . . . . . . . . . . . . . . . . . . 131
9.10 Scheme of the inhalation profile and particle injection for
unsteady flow accelerating through 30 L/min for FIR of 2 L/s2 132
9.11 Scheme of the inhalation profile and particle injection . . . . 134
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 model141
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 . . . . . . . . . . . . . . 143
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 . . . . . . 144
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). . . . . . . . . . . . . 146
xii

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

position vector of particle i at given time

the coordinates in the image plane, position

magnification factor

image intensity field

V0

transfer function giving the light energy of the particle image

in the correlation volume
auto-correlation of a single exposure image

R1
ai

interrogation area i

RC

convolution of the mean intensities of I

RF

fluctuating noise component

RC

convolution of the mean intensities of I

RP

self-correlation peak

d
RD

d (t)

separation vector in the correlation plane

constant displacement of all particles inside the interrogation

volume
constant displacement of all particles inside the interrogation
area
cross-correlation of particle images from the first exposure
with identical particle images of the second exposure
distance traveled of particle images within the pulse separation time
xv

pulse separation time

U, u

(instantaneous)fluid velocity (in PIV it is assumed that fluid

velocity is equal to the tracer particle velocity)
particle diameter

dp
Up , up

particle velocity

residual

normalized tracer diameter

Dp

distance between particles

resolution length (size of the computational cell)

characteristic length scale of the flow field

Gb

mass of a particle

FD

CD
s
S

aerodynamic interphase force, inviscid flux

body force, generation of turbulent kinetic energy due to
buoyancy
drag force
exposed frontal area of the particle
drag coefficient
surface area of a sphere, having the same volume as the particle
actual surface of the particle, scalar measure of the deformation tensor, local cross section

gravity vector

rp

particle position

TL

lagrangian integral time

Le

eddy length scale

tcross

particle eddy crossing time

tji

viscous stress tensor

sij

strain rate tensor

pressure
xvi

time

turbulent kinetic energy

fk
Gk , G
Sk

generation of the turbulent kinetic energy due to mean velocity gradients

contribution of the fluctuating dilitation in compressible turbulence
source term in the turbulent kinetic energy equation

source term in the turbulent dissipation rate equation

source term in the specific turbulent dissipation rate equation

generation of the specific turbulent dissipation

Yk

dissipation of the turbulent kinetic energy

dissipation of the specific turbulent dissipation rate

cross diffusion in the specific turbulent dissipation rate equation

YM

Gij

generation of turbulent kinetic energy due to buoyancy

DT,ij

turbulent diffusion in RSM

DL,ij

molecular diffusion in RSM

Cij

convection term in RSM

Pij

stress production term in RSM

Fij

production of system rotation in RSM

Mt

turbulent Mach number

volume flow rate

y+

dimensional wall distance

distance to the nearest wall

friction velocity

DH

hydraulic diameter

damping function

volume, local velocity

central sagittal line of the geometry

xvii

u, v, w
p, P

three velocity components

pressure
viscous flux

total energy

Cp

constant-pressure specific heat capacity

temperature

Greek symbols

state of the ensemble at a given time

point spread function of the imaging lens

tot

absolute measurement errors

resid

residual errors of the measured image displacements

sys

systematic errors

average value of the nearest neighbors of scalar U

relaxation time

particle density

dynamic fluid viscosity

kinematic fluid viscosity

wavelength of the incident light, mean free path of the flow

field
particle shape factor, pressure strain term in the Reynolds
Stress Model
characteristic lifetime of an eddy

ij

specific Reynolds stress tensor

turbulent viscosity

turbulent dissipation rate

specific turbulent dissipation rate

turbulent Prandtl number for the turbulent kinetic energy

xviii

turbulent Prandtl number for the turbulent dissipation rate

turbulent Prandtl number for the specific turbulent dissipation rate

inverse turbulent Prandtl number for the turbulent kinetic
energy
inverse turbulent Prandtl number for the turbulent dissipation
rate
turbulent Prandtl number for the specific dissipation rate

effective diffusivity of the turbulent kinetic energy

effective diffusivity of the specific turbulent dissipation rate

ij

turbulent dissipation rate tensor

deposition efficiency

wall shear stress

preconditioning matrix

Subscripts
air
mixture

of air
of the water/glycerine mixture

inlet

at inlet

norm

normalized

rms

root mean square

ref

reference

loc

local

gauge

Other symbols
.

averaged

e.

favre averaged
xix

fluctuating part

Abbreviations
LDV

Laser Doppler Velocimetry

LDA

Laser Doppler Anemometry

PDI

Phase Doppler Interferometry

PIV

Particle Image Velocimetry

PTV

Particle Tracking Velocimetry

LSV

Laser Speckle Velocimetry

FFT

Fast Fourier Transform

CFD

Computational Fluid Dynamics

DNS

Direct Numerical Simulation

LES

Large Eddy Simulation

DES

Detached Eddy Simulation

RANS

Reynolds Averaged Navier Stokes

EIM

Eddy Interaction Model

RNG

Renormalization group

SST

Shear Stress Transport

RSM

Reynolds Stress Model

MRI

Magnetic Resonance Imaging

CT

Computer Tomography

STL

stereolithography

DPI

Dry Powder Inhaler

pMDI
FIR

pressurized Metered Dose Inhaler

flow increase rate
xx

Dimensionless numbers
Re

Reynolds number

Stk

Stokes number

De

Dean number

xxi

xxii

Chapter 1
Introduction
1.1

Asthma

Asthma is a chronic disease of the respiratory tract in which the airway

occasionally constricts, in response to one or more triggers, by things such
as exposure to an environmental stimulant (or allergen), cold air, warm air,
moist air, exercise or exertion, or emotional stress. The airway often becomes inflamed, and is lined with excessive amounts of mucus. In children,
the most common triggers are viral illnesses such as the ones that cause
the common cold [120]. This narrowing causes symptoms such as wheezing, shortness of breath, chest tightness, and coughing. Between episodes,
most patients feel well but can have mild symptoms and they may remain
short of breath after exercise for longer periods of time than the unaffected
individual.

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

Asthma in the world

According to World Health Organization (WHO) estimates, 300 million

people suffer from asthma and 255 000 people died of asthma in 2005.
Asthma is the most common chronic disease among children. It is not just a
public health problem for high income countries but it occurs in all countries
regardless of the level of development. Over 80% of asthma deaths occurs
in low and lower-middle income countries. Asthma deaths will increase by
almost 20% in the next 10 years if urgent action is not taken [9].
2

A recent study of the European Federation of Allergy and Airway (EFA)

Diseases Patients Associations, presented in Brussels on March 5 2007 involved 1,300 people with severe asthma in five European countries: France,
Spain, Germany, Sweden and the UK and was conducted by NOP healthcare and coordinated by Asthma UK on behalf of EFA. The study revealed
that 90% of the 6 million people in Europe with severe asthma are not receiving optimum care, leaving 1.5 million of them to live in constant fear
that their next attack could be fatal. All these people are missing at least
one of the five treatment goals recommended by the Global Initiative for
Asthma (GINA) [33]. Unfortunately, patients who are inadequately treated
are more likely to suffer frequent attacks and are at greater risk of hospitalization and in some cases, death [122], [124].
Asthma affects 30 million people across Europe [39] and costs healthcare
services approximately 17.7 billion euro a year [11], a cost which could be
significantly reduced if access to effective patient centered care was a rule
not a privilege across Europe. In Western Europe one person dies every
hour as a result of severe asthma [95] , but 90% of these deaths could be
prevented with effective management of the disease [10].
Many respondents of the survey are optimistic that new more effective
drugs will be available in future (71%). Approximately one in three, 29%,
say investing in research is the single most useful thing their government
could do to improve their asthma.
With this in mind, this dissertation aims to make a small but hopefully
a meaningful contribution to the control of this disease.
The dissertation is organized as follows: first an explanation is given of al
essential parts of the upper human airways, followed by a theoretical background of Particle Image Velocimetry, the applied computational methods
of the fluid phase and particle phase. In the next chapter the creation of
the model of the upper human airways is described and the measurements
of the flow in this model. Followed by chapter 8 where the applied fluid
flow simulations are compared with the experiment. The simulation of the
aerosol deposition in the developed model is described in chapter 9. To
conclude, a numerical study, which is carried out on the developed model
with inclusion of a tracheal stenosis, is discussed.

Chapter 2
Anatomy of the Human
Respiratory Tract
This chapter describes the terminology of the respiratory tract

2.1

The respiratory system

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

Function of the respiratory system

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.

Figure 2.1: representation of the breathing cycle of a human [4]

As already been mentioned, the primary function of the respiratory system is to exchange gases. This exchange happens at the level of the alveoli (figure 2.2), where oxygen attaches to the hemoglobin molecules in the
blood. These molecules transport oxygen from regions of supply (the alveoli) to the region of demand (the blood and cells).
Carbon dioxide travels from the metabolically active cells into the capillaries. The concentration of carbon-dioxide in the cells is much greater than
in the capillaries, this process of movement of materials from a higher to a
lower concentration is called diffusion. According to Ficks law of diffusion
diffusion, the rate of gas transfer through a sheet of tissue is proportional
to the tissue area and inversely proportional to the sheet thickness. In the
blood, water is combined with carbon-dioxide to form bicarbonate. This removes the carbon-dioxide from the blood and keeps the concentration levels
5

Figure 2.2: representation of an alveolus [4]

of carbon-dioxide low in the blood, so that the diffusion process can continue. In the alveolar capillaries, bicarbonate combines with a hydrogen ion
to form carbonic acid, which breaks down into water and carbon-dioxide.
This carbon-dioxide diffuses into the alveoli.

2.1.2

The structure of the respiratory system

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

called cilia (figure 2.3).

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

Figure 2.5: side view of the mouth and pharynx [6]

situated in this casting of cartilage. The function of the vestibular folds

is to prevent air from coming from the lungs and prevent material from
entering the larynx, like the epiglottis. Speech is produced by letting the
vocal cords vibrate with moving air. The greater the amplitude of the
vibration, the louder the sound will be. Male adults usually have longer
vocal cords and therefor have lower voices. The opening between the vocal
cords is called the glottis, which varies within the breathing cycle. The
larynx is connected to a membraneous tube of approximately 12 cm long
and 2 cm wide, called trachea or windpipe. It consists of dens regular
connective and smooth muscle reinforced with 15 to 20 C-shape pieces of
9

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

Historical background of fluid measurements

Since the early ages mankind is interested in the observation of nature.

Even now children place obstacles in a flow and observe the most fascinating structures, or by throwing pieces of wood in a river allowing them to
roughly estimate the velocity of the river. The well known artist and scientist Leonardo Da Vinci made very detailed drawings of the flow structures
within a water flow by simple observation (figure 3.1).
Ludwig Prandtl made a great step forward in the observation of flow
structures. In contrast with Da Vinci, he not only observed the flow passively but he tried to extract information out of well planned experiments.
He designed a water tunnel (figure 3.2) and by adding mica particles on the
surface of the water he studied the structures of the flow in steady as well as
unsteady flow behind wings and other objects. By changing the model systematically , flow velocity and other parameters of the experimental set-up
12

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

Figure 3.2: Ludwig Prandtl next to his famous water tunnel

the wire has to be calibrated before each experiment which can be
cumbersome
the disturbance of the flow by the probe
like the pitot tube, only one point can be simultaneously measured
The temporal resolution is a lot higher than the pitot tube and the disturbance is smaller.
In the mid 1960s, the Laser Doppler Velocimetry (LDV), also called
Laser Doppler Anemometry (LDA) was developed . This is an optical technique to measure flow velocity in a desired point without disturbing the
flow. The operating principle of LDV is based on sending a highly coherent
monochromatic light beam (laser beam) toward the target, collecting the
light reflected by small particles in the target area, determining the change
14

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 particle image velocimetry

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 3.3: Experimental arrangement for PIV in a wind tunnel

interrogation area. One image recording contains more than thousand interrogation area. The projection of the vector of the local flow velocity into
the plane of the light sheet is calculated taking into account the magnification at imaging and the time delay between two illuminations. The time
of the illumination should be very short in order to avoid streaks made by
the reflection of the illuminated particles. The time between two pulses
should be long enough to have a displacement of the particles between two
recordings but also not too long to avoid out-of-plane displacement of the
particles. As already mentioned, tracer particles have to be added into the
flow. Particles should faithfully follow the motion of the fluid and have to
scatter the light very effectively. Therefore a high energy light source (laser)
has to be used for generation of the light sheet.
Another important issue is the density of images of tracer particles on
the PIV recording. Qualitatively three different types of image density can
be distinguished ([13]), which is illustrated in figure 3.4.
In the case of low image density (figure 3.4a), the images of individual
tracer particles can be detected. Low image density requires tracking methods for evaluation. This situation is often referred to as Particle Tracking
Velocimetry (PTV). In the case of medium image density (figure 3.4b), it is
16

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

Mathematical background of PIV evaluation

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

Figure 3.5: Schematic representation of geometric imaging

describes the state of the ensemble at a given time. Xi is the position

vector of particle i at that time.
 

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)

Figure 3.6 an example of an intensity field I of a single expose image is

shown.

Figure 3.6: Example of an intensity field I

3.3.1

Auto-correlation

The auto-correlation of a single exposure image can be defined as follows:

R1 (
s , ) = hI(
x , )I(
x +
s , )i

(3.5)

where hi represents the spatial average over interrogation area a1 . Equation

3.5 can be approximated by:

R1 (
s , ) =

Z
N

1 X

V0 (Xi )V0 (Xj ) (

x
xi )(
x
xj +
s )d
x
a1
i6=j

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.

Figure 3.7: Schematic representation of the auto-correlation of the intensity field I

In figure 3.7 the schematic representation of the auto-correlation of the
example intensity field I is given. Correlation peaks (RP and RF ) occur at
locations which are given by the vectorial differences between particle locations. Their strength is proportional to the number of all possible differences
which result in that location.

3.3.2

Cross-correlation of a pair of two single exposed

recordings

Mostly used in PIV, is the so-called cross-correlation technique. Hereby, a

tracer ensemble consists of two single exposure images (figure 3.8).
If an identical light sheet and windowing characteristics are considered,
the cross-correlation function of two interrogation areas can be written as:
20

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

V0 (Xi )V0 (Xj + D )

(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)

D is the constant displacement of all particles inside the interrogation vol

ume,
s represents the separation vector in the correlation plane. The terms
i 6= j are the correlation of different randomly distributed particles, considered as noise in the correlation plane. On the other hand, the terms i = j
contain the displacement information desired. Equation 3.9 can again be
decomposed into three parts:

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

flow, the displacement correlation peak reaches a maximum for

s = d.
Therefore the location of this maximum yields the average in-plane displacement, and thus the U and V components of the velocity. In figure
3.9, the schematic of the cross-correlation of the example intensity fields I

and I are given. Correlations of x2 dont appear on figure 3.9 because this
particle image is located outside the interrogation window.

Figure 3.9: Schematic representation of the cross-correlation of the inten

sity fields I and I
One can also consider the correlation of a double exposed recording,
but this is less important than the previous correlation. This correlation
contains more noise than the correlation of the single exposed recording.
Also an ambiguity in the displacement peaks appears. The reader is referred
to [96] for a mathematical description of this technique.

3.3.3

Optimization of the correlation

The following equation can define the locally measured velocity:

22

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)

As been shown, the accuracy of PIV measurements can be increased

by increasing pulse separation time between the exposures. But for high
values of the separation time the measurement noise increases. For very
large pulse separation time, the particle displacement will exceed the extent
of the interrogation volume.

3.4

Evaluation of PIV images

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

I known, the aim is to estimate the displacement d while excluding the

effects of the noise process N. This can be done by finding locally the
best match between the images in a statistical sense, like explained in the
previous section. One way is to directly calculating the cross-correlation,

where the template I is linearly shifted around in the sample I without

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)

where I and I are the Fourier transforms of the functions I and I . In

practice the Fourier transform is efficiently implemented for discrete data

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

In such cases standard evaluation schemes cannot use small interrogation

areas without losing the correlation signal due to the larger displacements.
Also sub-pixel based image shifting (image deformation) on multiple-passor multi-grid interrogation can be applied. This technique is a second order method and involves a complete deformation of the image data using
the displacement data of the previous interrogation passes. More detailed
information on these methods can be found in the literature ([101], [102],
[103], [131]).

3.4.1

Error estimation

An important aspect of measurements is the estimation of the errors, which

always arise within an experiment. The absolute measurement error, tot ,
can be decomposed into a group of systematic errors, sys , and a group of
residual errors, resid :
tot = sys + resid
(3.16)
Systematic errors are those which arise due to the inadequacy of the statistical method of cross-correlation in the evaluation of a PIV record, such as
the use of an inappropriate sub-pixel peak estimator. These errors follow a
consistent trend which makes them predictable. The second type of errors
remain in a form of a measurement uncertainty, even when all systematic
errors are removed. A very popular approach to estimate these errors is
based on numerical simulation (Monte Carlo simulations) ([96], [129], [70],
[71], [72], [73]). Artificial particle images with known content is generated
and by varying only one parameter at a time the influence of that parameter can be evaluated. In literature detailed information can be found for
optimizing a number of experimental parameters.([96], [101]). An example
of such a result is shown in figure 3.11. It is shown that the optimal particle
image diameter is slightly higher than 2 pixel.

3.4.2

Detection of spurious vectors (outliers)

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

Figure 3.11: Measurement uncertainty in digital cross-correlation PIV

evaluation with respect to varying particle image diameter
the illuminated area). Mostly, they appear as a single incorrect vector. This
can be caused by insufficient particle images in the interrogation area, noise
or artifacts not due to the correlation of matched image pairs. Most of the
spurious vectors can easily be detected by visual inspection, but when dealing with thousands of recorded images it is not obvious to deal with these
outliers in an interactive way. These erroneous results have to be removed,
certainly when wanting to calculate for instance, vorticity. Because of the
great amount of data involved, this can only be done by an automatic algorithm. This algorithm must ensure with a high level of confidence that no
questionable data is stored in the data set. Several algorithms are described
in literature but only the two most used will be explained.
The dynamic mean value operator
This algorithms checks each velocity vector individually by comparing its
magnitude with the average value over its nearest (mostly 8) neighbors.
The velocity vector will be rejected if the absolute difference between its
magnitude and the average over its neighbors is above a certain threshold.
|U U0 | < threshold

(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

Figure 3.12: Arbitrary example of a PIV measurement result containing

spurious displacement vectors
The (normalized) median test
The local median test proposed by Westerweel ([130]) showed superior performance in detection of outliers in case of a homogeneous and orthogonal
random field with an single step evaluation procedure. Median filtering simply speaking means that all neighboring velocity vectors are ordered linearly
with respect to the magnitude of their components and a velocity vector is
valid if the difference between the central value of this order (the median)
is smaller than a threshold value:
|U(median) U0 | < threshold

(3.18)

Westerweel ([132]) proposed an adaptation to this test, that normalizes the

median residual with respect to a robust estimate of the local variation
of the velocity. This method was verified on a large variety of flows and
a single threshold value can be applied to effectively detect the outliers.
Consider a displacement vector denoted by U0 and its eight neighbors by
{U1 , U2 , U8 } and U(median) as the median of this set (without U0 ). The
28

residual ri is defined as:

ri = |Ui U(median)|

(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)

with being a minimum normalization level. Two seemed to be the

optimal threshold value for r0 . Smaller values lead to a more stringent
where larger values lead to less stringent outlier detection.

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)

with is the wavelength of the incident light. If q is greater than one,

approximately q local maxima appear in the angular distribution over the
range from 0 to 180 .
In general, the averaged intensity increases with q 2 and the scattering
efficiency depends on the ratio of the refractive index of the particles to that
of the fluid. Therefore the scattering of particles in air will be at least one
order of magnitude larger than that of the same particles in water, because
the refractive index of water is considerably larger than the index of air. As
can be seen on figure 3.13, the light intensity is not blocked by the particles
but spread in all directions and the highest intensity can be found back in
the forward scattered position. Unfortunately, this direction cannot be used
in PIV in contrast to LDA. In case of heavily seeded flows, a certain amount
of the light scattered by the particles is not only due to direct scattering
but also due to scattered light from other particles. This implies that not
only the diameter of the particles are important but also the density of
the seeding plays an important role in the scattering efficiency. But with
denser seeded flows, the background noise and thus the noise on recordings
will increase.

30

Figure 3.13: Light scattering by a (from top to bottom) 1 m, 10 m and

30 m glass particle in water

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

Geometric properties of particles

The geometric properties of particles such as size and distribution have

direct influence on the dynamic behavior of gas-solid flows.
32

Figure 4.1: Particle size distribution in gas-solid flows (after Soo, 1990)

4.2.1

Particle size

Each particle is implicitly supposed to be made up of large number of

molecules (minimum of 103 104 ) which is a pre-requisite to define their
physical properties. The particle size is assumed to be much higher when
compared to mean free path (average distance a molecule travels in a gas
between collisions) of continuous phase. Quantitatively, a particle size of
1m corresponds to a gas flow with particle size about 10 times more than
the mean free path.
Fig. 4.1 shows the distribution of particle sizes encountered in various
multiphase systems as given by Soo [111]. It can be seen that the typical
range of particle sizes of interest to gas-solid flows is approximately between
1 m - 10 cm.

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

 << dp << Dp << l << L

where,
= mean free path of flow field
dp = particle size
Dp = distance between particles
l = resolution length (size of computational cell)
L = characteristic length scale of flow field

4.3

Dilute and dense flows

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

The phenomenon of mutual mass, momentum and energy transfer between

the phases is termed as coupling. Elghobashi [36] proposed a map of regimes
of interactions between particles and fluid turbulence which is given by
Fig.4.2. For values of dispersed-phase volume fraction less than 106 , particles have negligible effects on turbulence and is termed as one-way coupling.
In the second regime which lies between 106 103 , the existence of particles can augment the turbulence if the ratio of the particle response time
to the turnover time of a large eddy is greater than unity, or can attenuate
turbulence if the ratio is less than unity. This interaction is called two-way
coupling. In the third regime where the volume fractions are greater than
103 , in addition to two-way coupling between particles and turbulence,
34

Figure 4.2: Map for particle-turbulence modulation (after Elghobashi,

1994)
particle collisions take place and hence this regime is termed as four-way
coupling.

4.5

Modeling two-phase flows

The two most widely used approaches for mathematical modeling of twophase flows are Eulerian continuum approach and Lagrangian trajectory
approach.

4.5.1

Eulerian continuum approach

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

Lagrangian trajectory approach

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

The mass transfer of a particle is given by,

dm

=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

The momentum transfer of a particle is given by,

d(m
up )

= F + mGb + (m
up )
dt

(4.3)

Time rate of change of momentum of a particle is equal to the sum of

external forces acting on it. F is the aerodynamic interphase force, mGb

is the body force and m

up is the momentum gained/lost by particle due to
mass transfer.

4.7.1

Interphase Force

For an idealized sample, the aerodynamic interphase force can be considered

as the sum of Drag force, Basset force, Saffman force and Magnus force,
where each force represents some feature of particle fluid interaction. For
most of the practical applications, drag force is the most significant and
often the only considered part of the interface force.
Generally the particle moves with a different velocity than the fluid

at any given point. The difference in fluid velocity (

u ) and the particle

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

Figure 4.3: Drag coefficient for spheres as a function of particle Reynolds

number (after Schlichting, 1979)
incoming flow. For a rigid sphere,
A = d2p /4

(4.5)

CD is the coefficient of drag which is a function of the particle Reynolds

number and the local turbulent intensity of the fluid. The particle Reynolds
number is given by,
Rep = dp

|
u
up |

(4.6)

Laws for Drag Coefficients

Various experimental data on the drag coefficient for a single sphere at
various Rep were compiled into a single standard curve by Schlichting [105]
as shown in Fig. 4.3.
At high Reynolds number of about 3 105 the wake structure behind
the particle changes which changes the surface pressure distribution around
the particle resulting in steep reduction of drag coefficient and transition
38

of flow from laminar to turbulent in the boundary layer of particle surface.

Mathematically, the curve can be expressed as follows,
For Rep << 2 (creeping flow regime) where the flow is governed by
viscous effects,
24
CD =
Rep < 2
(4.7)
Rep
For Rep >> 1, but within the limits of transition regime, the flow is governed by inertia effects.
CD =

18.5
Re0.6
p

CD = 0.44

2 < Rep < 500

500 < Rep < 2 105

(4.8)
(4.9)

Jayaraju et al. [64] used a relatively simple formulation of Schraiber et

al. [106]. This formulation is valid for high Reynolds number up to 6 103
. It is given by,
CD =


p
24 
1 + 0.179 Rep + 0.013 Rep
Rep

0.5 < Rep < 6 103 (4.10)

Substituting 4.5, 4.6 and 4.10 in 4.4 results in the following final form
of drag force,


p

FD = 3 dp ( u up ) 1 + 0.179 Rep + 0.013 Rep

(4.11)
In Fluent several possible laws for the drag coefficient are available. The
drag coefficient can either be taken from
CD = a1 +

a2
a3 2
+
Rep Rep

(4.12)

where a1 , a2 , and a3 are constants that apply to smooth spherical particles

over several ranges of Re given by Morsi and Alexander [89], or
CD =

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

=
=
=
=

exp(2.3288 6.4581 + 2.44862)

0.0964 + 0.5565
exp(4.905 13.8944 + 18.42222 10.25993)
exp(1.4681 + 12.2584 20.73222 + 15.88553)

(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

Stochastic trajectory approach

Stochastic modeling involves direct simulation of particle motion through

a random turbulent flow field. Most stochastic approaches use the technique of generating turbulent-like carrier flow field using the variables of
turbulence model which is used for carrier flow simulation.
The instantaneous motion of particle is governed by,



p

d(m
up )

= 3 dp ( u up ) 1 + 0.179 Rep + 0.013 Rep + mGb (4.20)

dt

The instantaneous fluid velocity

u in the above equation is represented

as the sum of the mean and fluctuating velocity,

u=u+u

(4.21)

The stochastic model in Fluent is based on the Eddy Interaction Model of

Gosman and Ioannides [45]. Assuming isotropic turbulence, the fluctuating
velocity is given by,
r
2

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)

where the coefficient CL is to be determined as it is not well known. For the

k model and its variants a value of 0.15 is chosen for CL . In equation
4.23 = k can be substituted. The characteristic lifetime of an eddy is
defined either as a constant:
e = 2TL
(4.24)
or as a random variation about TL :
e = TL log(r)

(4.25)

where r is a random number between 0 and 1. The eddy length scale Le

can be described as follows:
Le =

k
C3/4

3/2

(4.26)

where C is equal to 0.09 The particle eddy crossing time is defined as

tcross = ln[1

Le
]
|u up |

(4.27)

with the particle relaxation time, given by

4
p dp
3 CD |u up |

(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

In CFD several methods can be applied in the simulation of turbulent fluid

flow. The most well-known methods are: Direct Numerical Simulation
(DNS), Large Eddy Simulation (LES), Detached Eddy Simulation (DES)
and Reynolds-averaged Navier-Stokes (RANS).
DNS captures all of the relevant scales of turbulent motion, so no modeling of the turbulence has to be applied. This approach is extremely expensive, if not impossible, for complex problems on modern computing machines.
LES is a technique in which the smaller eddies are filtered and are modeled using a sub-grid scale model, while the larger energy carrying eddies
are simulated. This method generally requires a relatively fine mesh, but a
far coarser mesh than a DNS solution.
DES, originally formulated for the Spalart-Allmaras model is a modification of the RANS model in which the model switches to a subgrid scale
formulation in regions fine enough for LES calculations. Regions near solid
43

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

of mass, momentum, pressure, species and turbulence. These equations are

coupled and so difficult to solve that it was not until modern digital computers were invented in the 1960s and 1970s that these could be resolved for
real flow problems within reasonable timescales. Other key figures who developed theories related to fluid flow in the 19th century were Jean Le Rond
dAlembert, Simon-Denis Poisson, Joseph Louis Lagrange, Jean Louis Marie
Poiseuille, John William Rayleigh, M. Maurice Couette, Osborne Reynolds,
and Pierre Simon de Laplace.
In the early 20th Century, much work was done on refining theories
of boundary layers and turbulence in fluid flow. Ludwig Prandtl (18751953) proposed a boundary layer theory, the mixing length concept, compressible flows, the Prandtl number, and many more. Theodore von Karman (1881-1963) analyzed what is now known as the von Karman vortex
street. Geoffrey Ingram Taylor (1886-1975) proposed a statistical theory
of turbulence and the Taylor microscale. Andrey Nikolaevich Kolmogorov
(1903-1987) introduced the concept of Kolmogorov scales and the universal
energy spectrum for turbulence, and George Keith Batchelor (1920-2000)
made contributions to the theory of homogeneous turbulence.
The earliest numerical solution for the flow past a cylinder was carried
out in 1933 by Thom and reported in England: A.Thom, The Flow Past
Circular Cylinders at Low Speeds, Proc. Royal Society, A141, pp. 651-666,
London, 1933
Kawaguti in Japan obtained a similar solution for flow around a cylinder
in 1953 by using a mechanical desk calculator, working 20 hours per week
for 18 months! M. Kawaguti, Numerical Solution of the NS Equations for
the Flow Around a Circular Cylinder at Reynolds Number 40, Journal of
Phy. Soc. Japan, vol. 8, pp. 747-757, 1953. [7]

5.3

The Navier-Stokes equations

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)

The vectors ui and xi are velocity and position, t is time, p is pressure,

is density and tji is the viscous stress tensor defined by
tji = 2sij

(5.3)

where is molecular viscosity and sij is the strain rate tensor,

1 ui uj
sij = (
+
)
2 xj
xi

(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)

Because of the mass conservation equation 5.5 leads to,

uj

ui
(uj ui )
=
xj
xj

(5.6)

The Navier-Stokes equation in conservation form is

5.4

(uj ui )
p
(2sij )
ui
+
=
+
t
xj
xi
xj

(5.7)

The Reynolds Averaging

The three forms most pertinent of the averaging concepts introduced by

Reynolds (1895) are the time average, the spatial average and the ensemble
average: the general term used to describe these averaging processes is
mean. Time averaging is appropriate for stationary turbulence, i.e., a
turbulent flow that, on the average, does not vary with time. For such a
46

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

ui(x, t) = ui (x) + ui (x, t)

(5.9)
Time averaging of equations 5.1 and 5.7 yields the Reynolds averaged
equations of motion in conservation form
ui
= 0
xi

p
ui

+
(uj ui + uj ui) =
+
(2sji )
t
xj
xi xj

(5.10)
(5.11)

Aside from replacement of instantaneous variables by mean values, the

only difference between the time-averaged and instantaneous momentum

equations is the appearance of the correlation uj ui. This is a time-averaged

rate of momentum transfer due to the turbulence. A prescription of uj ui is
needed for computing all mean-flow properties. Herein lies the fundamental
problem of turbulence. Equation 5.11 can be written in its most recognizable
form by using equation 5.5 in reverse. The resulting equation is

ui
ui
p

+ uj
=
+
(2sji uj ui )
t
xj
xi xj

(5.12)

Equation 5.12 is usually referred to as the Reynolds-averaged Navier-Stokes

equation (RANS). The quantity uj ui is known as the Reynolds-stress

tensor and it will be denoted here as ij , so that ij is the specific Reynolds
stress tensor given by

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

In order to solve the system of equations, the Reynolds stresses have to be

modeled. One way to solve this problem is the Boussinesq approach. This
method makes use of the Boussinesq hypothesis (equation 5.16) to relate
Reynolds stresses to the mean velocity gradients.

uj ui = t (

ui
ui
2
uk
+
) (k + t
)ij
xj
xj
3
xk

(5.16)

This assumption is used in different turbulence models ranging from

algebraic (e.g. Baldwin-Lomax), one-equation (e.g. Spalart-Allmaras) to
two-equation models (e.g k- and k-). The advantage of this approach is
the relatively low computational cost with the computation of the turbulent
viscosity, t . The disadvantage is the assumption of t being an isotropic
scalar quantity, which is not strictly true. Models based on the Boussinesq
assumption provide good predictions for many flows of engineering interest.
48

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

In Fluent three different turbulence models are implemented: the standard,

renormalization group RNG and the realizable k models. All three
models have similar transport equations for k and . The major differences
in the models are:
1. the method of calculating turbulent viscosity.
2. the turbulent Prandtl numbers governing the turbulent diffusion of k
and .
3. the generation and destruction terms in the equation.
The Standard k Model
The standard k model is a semi-empirical model based on model transport equations for the turbulence kinetic energy (k) and its dissipation rate
(). The transport equation for k is derived from the exact equation, while
49

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)

In equations 5.17 and 5.50, Gk represents the generation of turbulent

kinetic energy due to mean velocity gradients, Gb is the generation of turbulent kinetic energy due to buoyancy and YM represents the contribution of
the fluctuating dilatation in compressible turbulence to the overall dissipation rate. For details on the modeling of those terms, the reader is referred
to the manual of Fluent. The source terms Sk and S can be defined by
the user. The model constants in equations 5.17, 5.50 and 5.19 have the
following values:
C1 = 1.44, C2 = 1.92, C = 0.09, k = 1.0 and = 1.3

(5.20)

The RNG k Model

This turbulence model is derived from the instantaneous Navier-Stokes
equations, using a mathematical technique called renormalization group
methods.
The transport equations are given by:

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 Sk/, 0 = 4.38, = 0.012. S is the scalar measure of the

deformation tensor. The model constants C1 and C2 in equations 5.21 and
5.22 are derived analytically by the RNG theory. These values are:
C1 = 1.42, C2 = 1.68
The scale elimination procedure in RNG theory results in a differential
equation for turbulent viscosity:

where

2 k
b
d( ) = 1.72 3
db

b 1 + C

(5.24)

ef f

100

b =

Equation 5.24 is integrated in order to obtain an accurate description

of how the effective turbulent transport varies with the effective Reynolds
number, allowing the model to handle low-Reynolds-number and near-wall
flows.
The Realizable k Model
The term realizablemeans that the model satisfies certain mathematical
constraints on the normal stresses, consistent with the physics of turbulent
flows. If the strain is large enough it is possible that the normal Reynolds
stress (which is by definition a positive quantity) becomes negative, i.e.,
51

non-realizable. Another weakness of the standard k model or other

traditional k models lies with the modeled equation for the dissipation
rate, e.g. the prediction of the spreading rate for axisymmetric jets is unexpectedly poor. The realizable k model proposed by Shih et al. [110]
was intended to address these deficiencies by adopting the following:
A new eddy-viscosity formula involving a variable C originally proposed by Reynolds [98].
A new model equation for dissipation based on the dynamic equation
of the mean-square vorticity fluctuation.
The modeled transport equations in the realizable k model are:

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)

In case of the realizable k turbulence model, C is not a constant but

is a function of the mean strain rate, rotation strain rates, the turbulent
kinetic energy and the dissipation rate.
52

5.5.2

k turbulence models

In Fluent, two different k turbulence models are implemented: the

standard and the Shear-Stress Transport (SST) k models. Both models
have similar transport equations for k and . The major differences between
the SST model and the standard model are:
the gradual change from the standard k in the inner region of the
boundary to a high-Reynolds-number version of the k model in
the outer part of the boundary layer.
the modified turbulent viscosity formulation to account for the transport effects of the principal shear stress.
The Standard k Model
The standard k model is an empirical model based on model transport
equations for the turbulent kinetic energy (k) and the specific dissipation
rate ().
The turbulent kinetic energy and the specific dissipation rate are obtained from the following transport equations:

and

k
(k) +
(kui ) =
(k ) + Gk Yk + Sk
t
xi
xi
xi

(5.28)

() +
(ui ) =
( ) + G Y + S
t
xi
xi
xi

(5.29)

In equations 5.28 and 5.29, Gk represents the generation of turbulent

kinetic energy due to mean velocity gradients, G is the generation of the
specific dissipation rate. Yk and Y represent the dissipation of k and due
to turbulence. The source terms Sk and S can be defined by the user. For
details on the modeling of these terms, the reader is referred to the manual
of Fluent [1]. k and represent the effective diffusivity of k and and
are given by
t
k
t
= +

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)

The model constants are given by

Ret =
Rk =
0 =
i

=
=
=
=

=6
i
3
0.072
1
2.0
2.0

The SST k Model

The SST k model is a variation of the standard k model, which
accounts for the transport of the principal turbulent shear stress by a modification of the definition of the turbulent viscosity. This feature gives the
SST k model an advantage in terms of performance over both the standard k model and the standard k model. Other modifications include
the addition of a cross-diffusion term in the equation and a blending function to ensure that the model equations behave appropriately in both the
near-wall and far-field zones.
The turbulent kinetic energy and the specific dissipation rate are obtained from the following transport equations:

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)

ek represents the generation of turbulent

In equation 5.34, the term G
kinetic energy due to mean velocity gradients. The term D , in equation
5.35 represents the cross-diffusion. For details on modeling of those terms,
the reader is referred to the manual of Fluent [1]. The effective diffusiveness
for the SST k model are given by
t
k = +
(5.36)
k
t
= +
(5.37)

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 Reynolds Stress Model (RSM)

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

2k (uj um ikm + ui um jkm )

+Suser
|
{z
}
Fij Production by system Rotation

(5.48)

In equation 5.48, DT,ij is the turbulent diffusion, Gij represents the

production of turbulence due to buoyancy, is the pressure strain term and
ij represents the dissipation tensor. All these terms have to be modeled to
close the set of equations. In Fluent, two different models are available for
modeling the pressure-strain term. First is the linear pressure-strain model,
which has a low-Reynolds-number modification available. The second is the
quadratic pressure-strain model, which should be more accurate for a wider
class of flows. The dissipation tensor is modeled as follows
2
ij = ij ( + 2Mt2 )
3

(5.49)

where Mt represents the turbulent Mach number. The scalar dissipation

rate ,, is computed with a model transport equation similar to that used
56

in the standard k model:

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

Figure 6.1: geometry developed by Katz and Martonen [68]

deposition near the larynx. However, no experimental results to validate
the numerical simulations were available for validation.
Katz et al. [67] calculated in the same geometry particle trajectories
based on fluid results obtained in previous work [68],[69] and using the
stochastic model of Gosman and Ioannides [45] for the influence of turbulence on the particle motion. To the authors knowledge, the influence of
gravity was not taken into account. The main conclusion of this numerical study was that turbulence can have a profound effect on the particle
deposition of particles in the larynx and trachea and any calculation for
deposition of inhaled aerosols must consider this effect.
A symmetric three-dimensional computer model of the human upper
respiratory tract , based a silicone rubber cast of the human head, throat
and trachea was created by Martonen et al [83] and CFD-simulations [82]
were performed with a commercial code (CFX-F3D). Martonen et al. [82]
considered the flow, during quiet breathing to be laminar. This was based
on measurements performed by Hahn et al. [51] in a large-scale model of
the nasal cavity and experimental measurements of Schroter and Sudlow
[107] and Zhao et al. [146]in the upper tracheobronchial airways. Air flow
patterns were investigated during inhalation and exhalation for were investigated for flow rates corresponding to resting (17 l/min) and light (34 l/min)
physical activity. A difference in flow patterns of inhalation and exhalation
was observed. The magnitude of the respiratory flow rate had a significant
effect on the different jets and the size of the recirculation zones.
In the same geometry Martonen et al. [84] studied the influence of gravity and particle size on the deposition region. As previously explained, the
influence of turbulence was neglected. It was found that the magnitude of
the inspiratory flow had a significant influence on the deposition region but
less influence on the deposition fractions and this in contrary to the particle
size, which had a major influence in the deposition fractions and almost no
59

effect on the deposition region.

Corcoran and Chigier [27] measured the axial velocity and turbulence
intensity, using Phase Doppler Interferometry (PDI) in a cadaver based
larynx-trachea model (Figure 6.2). The model consisted of a polyurethane
casting of the human larynx, connected to a glass tube with an inside diameter matching the tracheal diameter of the cadaver. Velocities were measured
at three different flow rates representing quiet and normal breathing of an
adult and the breathing conditions of a six year old child. At all three
flow rates a tracheal jet occurred at the posterior side of the trachea and
reverse flows are apparent at the anterior side. According to Corcoran and
Chigier, the posterior location of the tracheal jet can be attributed to the
wider opening in the vocal cords that existed at the posterior side at the
posterior side of the glottal plane. Regions of high axial turbulence intensity
were observed within two tracheal diameters downstream the larynx which
correlated with results from Schlesinger and Lippman [104]. A surprising
observation was that the average turbulence levels of the lower flow rate
were 20 % higher than those of the higher flow rate. This was attributed to
the possible faster expansion of the tracheal jet at higher flow rates.

Figure 6.2: geometry used by Corcoran and Chigier [27]

Gemci et al [40] studied the deposition levels of inhaled sprays in a
simplified model of the human throat(larynx), consisting of two circular
60

tubes which were connected to each other through a semi-triangular shape,

representing the vocal cords. Numerical simulations of the fluid flow were
compared to experimental results obtained with PDI. Numerical simulations
of dispersion and deposition of the inhaled sprays were performed. The
fluid flow was simulated, using a commercially available code (KIVA-3V),
which solves the RANS -equations coupled to a standard k- turbulence
model. The spray equation for the particle phase with a stochastic particle
technique, taking into account the influence of the turbulence in the fluid
flow on the particle movement, was solved. High peaks of turbulence are
noted in both experimental and numerical study were noted but the effect
on the particle deposition had to be determined.
Gemci et al. [42] performed numerical simulations in a cadaver based
throat model [27]. The numerical results were obtained by using RANS
combined with the standard k- turbulence model. The air flow and particle deposition of a poly-dispersed size distribution were studied at an air
flow rate of 18 l/min. The deposition of the spray droplets was strongly
effected by the highly turbulent gas flow in the laryngeal jet and the recirculation regions. The same model connected to a transparent glass tube,
which represented the trachea was used by Corcoran and Chigier [28] to
measure axial velocity and axial turbulence intensity, using Laser Doppler
Velocimetry (LDV) and specific tracheal deposition mechanisms, using fluorescent dye. Measurements were made at two different steady flow rates
(18.1 l/min and 41.1 l/min). Regions of high deposition within the trachea
were found where the laryngeal jet passed near the tracheal wall. In contrary
to the numerical deposition study of Gemci et al. [42], there appeared no
obvious correspondence between deposition levels and turbulence intensity.
A numerical and experimental study of a very simple throat model was
carried out by Gemci et al. [41]. In the numerical particle deposition study,
evaporation, coalescence and turbulent dispersion was taken into account.
The latter was computed by using the model of ORourke [93] which is based
on the model of Gosman and Ioannides [45]. The numerical fluid flow results
(axial contour plots) differ from the experimental results obtained by LDV,
the authors contribute this difference to the decreased in-plane measurement
points in the experimental result versus computational cells in the numerical
simulation. High deposition levels were found at the posterior wall of the
model caused by the laryngeal jet. The inclusion of the coalescence model
resulted in a lower number of droplets and an increase of mean droplet
61

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.

Figure 6.3: geometry developed by Zhang et al. [145]

Zhang et al. [145] simulated air flow and micro-particle transport in
a simplified model of the upper human airways (Figure 6.3). The model
consist of a single circular tube with local diameter variations, based on
data sets provided by Cheng et al [24]. Fluid flow results were obtained
by solving the RANS-equations combined with a low-Reynolds adaptation
of the standard k equations. This model was validated on a locally constricted tube by Zhang and Kleinstreuer [142]. To the authors knowledge
the particle trajectory equation used by Zhang et al. only drag force was
taken into account and gravity was neglected. To account for the influence of turbulence the model Gosman and Ioannides [45] was used. The
63

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

Figure 6.4: geometry developed by Stapleton et al. [115]

Stapleton et al [115] investigated the suitability of using a k- model for
the prediction of aerosol deposition in a simplified mouth-throat model. Total deposition of aerosols and pressure drop over the model were experimentally measured at two different flow rates, representing laminar (2 l/min)
and turbulent (28.3 l/min) flow rate. The RANS equations in combination
with a standard k- turbulence model were solved and the eddy interaction
model of Gosman and Ioannides [45] was used for predicting the influence
of the turbulence on the particle movement. The pressure drop along the
model was quite well predicted by the numerical simulations. Also good
agreement between experimental deposition measurements and the numerically predicted particle deposition levels was found for the laminar case.
But the simulated particle deposition differed quite dramatically for the
turbulent case. The conclusion was that the standard k- turbulence model
was not suitable for predicting the aerosol deposition in the upper human
airway system.
65

Heenan et al [54] used endoscopic Particle Image Velocimetry (PIV) to

measure the flow field at three different flow rates (15, 30 and 90 l/min) in
the idealized human oropharynx described in Stapleton et al [115]. These
results were compared with numerical simulations. The RANS-equations
combined with a standard k- turbulence model with a low-Reynolds number modification were solved. Mainly qualitative results were presented.
Detailed comparison was given at only one location in the geometry (at
the beginning of the pharynx). The conclusion was that RANS-simulations
could capture the basic structures of the flow quite accurately but the increased viscous effects at lower Reynolds numbers were not predicted by the
numerical simulations.
Matida et al. [85] used the Eddy Interaction Model of Schuen et al.
[108] , which is based on the model Gosman and Ioannides [45] to predict
the particle deposition in a simplified geometry of the upper human airways.
The use of this model overpredicted the deposition of small particles in the
upper airway model and therefore the authors presented a near-wall correction of this model based on the model of Wang and James [126] for taken
into account the anisotropy of turbulence. This adaptation in combination
with a standard k- turbulence model showed relatively good agreement
with experimental data.
At seven different locations in the same geometry as Stapleton et al [115]
Johnstone et al. [66] used single and X-hot wire to measure the mean and
axial velocity field at seven different flow rates (10, 15, 30, 45, 60, 90 and 120
l/min). The normalized mean axial velocity profiles at the first five locations
(from the oral inlet to the oropharynx) increased at decreasing inhalation
flow rate. At these locations the normalized axial velocity profiles were
found to converge onto a similar curve for the higher flow rates.
Grgic et al. [49] experimentally studied the deposition of particles (3,
5 and 6.5 micro-meter) in an idealized average human upper airway model
[115] at two different flow rates (30 and 90 l/min). Effects of flow rate
particle size and flow Reynolds number were investigated. It was found
that inertial impaction is the mechanism that governing aerosol deposition
and in addition to flow rate and particle size, a Reynolds number effect
could be found back. An empirical correction were the Stokes number was
multiplied with the Reynolds number to the power 0.37 was derived.
Zhang et al [139] experimentally measured the deposition of aerosols in
two standard United States Pharmacopeia (USP) and a proposed highly
66

idealized mouth-throat for steady inhalation flow rates of 30, 60 and 90

l/min. The deposition curves of the USP-throats differed dramatically with
the in-vivo average curve where the curve of the proposed highly idealized
mouth-throat the in-vivo average curve followed. In addition numerical
simulations with the method proposed by Matida et al. [85]. The total
deposition results showed relatively good agreement when compared with
the measured data.
The relationship between flow field and regional deposition at a inhalation flow rate of 90 l/min was investigated by Heenan et al. [53] using PIV
measurements of the time-averaged flow field in the central sagittal plane of
two realistic extra-thoracic airway geometries of one human subject taken
at different days for the flow field and gamma scintigraphy measurements
to characterize regional deposition of 3 and 5 micro-meter particles. Mainly
two factors influenced the local deposition levels, namely, deposition increased with both increasing local velocity magnitude and increasing local
flow curvature. These results indicated that morphological differences between two different geometries from the same subject create differences in
deposition distribution.
Grgic et al. [47] chose a seven different realistic geometries out a set
of 80 geometries, obtained via MRI scans and gamma scintigraphy and
gravimetry were used to determine local and total deposition. A large intersubject and some intra-subject difference was found. All total deposition
data collapsed on a single curve when the data was set-out to the Stokes
number with inclusion of the Reynolds number effect. An equation was
derived for this curve.
Grgic et al [49] studied numerically and experimentally the effect of
unsteady accelerating flow on the deposition of inhaled boluses in the simplified average upper human airway model of Stapleton et al [115]. In the
numerical study, the effect of turbulence on the flow and deposition of particles was neglected. The main conclusion of the study was that accelerating
flow enhanced mouth-throat deposition. This could be explained that by
the time the particles reached the main deposition area of this geometry the
velocity was higher than in the case of steady or non-accelerating inhalation.
Jin et al. conducted Large Eddy Simulations on the idealized average
geometry of Stapleton et al. [115] connected to a triple bifurcation, which
was based on Weibels model [128]. Flow field results were compared with
the experimental results obtained by Heenan et al [54]. According to Jin et
67

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

Creation of the phantom

Creation of the computer model of the upper
human respiratory tract

Local ethics approval was obtained for a multi-slice CT imaging study in

five otherwise healthy never-smoker male subjects who were referred for
lung imaging due to possible low concentration occupational asbestos exposure. CT images were acquired using a Sensation 16 CT-scanner (Siemens,
Erlangen, Germany) with following settings: acquisition 16x0.75 mm, reconstruction slice thickness 1 mm, pitch 1.25, rotation time 0.5 seconds,
Care dose, reference mAS 100, KV 120, Medium filter B45f. Imaging was
initiated above the nasal cavity and synchronized to start with the subject
slowly inhaling to near total lung capacity where a breath hold was inserted
to complete data acquisition just below the carina. In this way, the glottic
opening was imaged with the subject being in a slow inhalation maneuver.
69

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.

Figure 7.1: Creation of realistic geometry

With the use of preliminary CFD-simulations, a smoothed symmetric
70

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)

Figure 7.3 shows a side-view and rear-view with several characteristic

dimensions of the smoothed 3D model of the upper human respiratory tract
are shown.
71

7.1.2

Creation of the upper respiratory airways phantom

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

Another possibility was to rapid prototype a wax model of the airways.

But the transparent silicone is porous when heated and melting the wax
causes it to penetrate the silicone, leaving a irremovable non-transparent
layer behind.

74

Figure 7.3: side-view and rear-view of the smoothed 3D model of the

upper human airways with cross-sections at different locations of the model
75

Figure 7.4: STL-model suspended in plasticine

Figure 7.5: metal positive placed in perspex box

76

7.2
7.2.1

Flow measurements
The experimental-set-up

Matching index of refraction

To eliminate the refraction and reflection of the laser sheet as it passes
through the convoluted flow passage, and the distortion of the light scattered by the seeding particles, the index of refraction of the working fluid
had to be matched to the refraction index of the model. The fluid which
was finally chosen was a mixture of water and glycerine. By changing the
concentration of water the index of refraction could be tuned. Dow specifies an index of refraction of 1.43 for the selected silicone. However, due to
inherent differences in mixing and curing, the actual index will vary from
model to model. Corino and Brodkey [30] and Johnston et al. [65] used in
their experiments the index of refraction but the use of PIV for measuring a
convolute geometry requires a more accurate match of refraction index. A
grid of lines was placed behind the model and a mixture with excess glycerine, was pumped through the model. The distortion of the grid lines was
clearly visible and water was added to the mixture. By adding water, the
distortion in the grid lines slowly decreased. When the matching of index of
refraction was reached, the distortion completely disappeared. This method
assured accurate match of indices. Figure 7.6 shows a side-view of the transparent phantom. The posterior side of the pharynx is invisible because in
that region water/glycerine mixture with matching index of refraction was
added while other parts are clearly visible.
Reynolds number matching
The flow under investigation was an incompressible, non-oscillatory flow. In
order for two such flows to be similar they must have the same geometry, and
have equal Reynolds numbers. The Reynolds similitude can be expressed
as follows:
Re1 =

U1 D1
U2 D2
=
= Re2
1
2

(7.1)

with U the characteristic velocity, D the characteristic length (in case

of internal flows, the characteristic hydraulic diameter) and the kinematic
77

viscosity. The viscosity of the glycerin/water mixture had to be known,

so that Reynolds number matching could be applied. The viscosity of the
mixture depends on temperature and had to be accurately measured at
operating temperature, which was around 25.2 C. The viscosity of the
mixture was 5.88106m2 /s, measured with the AVS300 viscosimeter (Schott
Gerate GmbH, Ludwigshafen, Germany) at 25.2 C. The model is not scaled
(1:1) and consequently the Reynolds number matching amounts to:
Qair =

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 7.7: scheme of the experimental set-up

32x32 pixels with an overlap of 50%. Spurious vectors were detected by
using the normalized median test, which eliminates a dependence of the
detection criterion on the interrogation domain size [132]. Only 0.1 % of
the vectors had to be interpolated. A least squares 3-point Gauss fit algorithm was used to recover the sub-pixel displacement of the correlation.
Since it uses more data for the location of the peak, the noise in the result
is generally less compared to more simple fits. A combination of image
parameters and numerical effects influence the stochastic and systematic
80

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

with the obtained correlation coefficient in the respective measured section

(7.8c: correlation coefficient).
Normalized magnitude of velocity and normalized turbulent kinetic energy at two different locations in the trachea at flow rate of 45 l/min obtained by evaluating 3000 and 4000 image pairs was compared. As can be
seen on figure 7.9, almost no difference could be found. Hence, 4000 image
pairs were certainly enough to capture all physical phenomena.

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

Figure 7.10: streaklines and contour of normalized magnitude of velocity

in the upper human airway model at 15 l/min (a) and 30 l/min (b) air flow
rate
the tip of the tongue and follows the tongue to the end where it turns into
the pharynx. Heenan et al [54] also report the formation of a separation
bubble at the roof of the mouth in their average simplified model but it is a
lot smaller. This is probably due the major difference in mouth geometry,
compared to the present model. They also reported a concentration of
streaklines at the posterior side of the mouth when entering the pharynx.
This phenomena is caused by the strong curvature which generates a local
counter-rotating secondary motion, where fluid close to the central plane of
symmetry flows from anterior to posterior and then around the walls from
anterior to posterior side. Such behavior is not observed in the present
geometry. However, flow visualization, with air bubbles showed the threedimensionality of the flow. When entering the mouth, a part of the flow
bends downwards and turns then upwards and follows the upper part of
the mouth and finally enters the pharynx. This, of course could not be
visualized by the 2D PIV measurements.
The beginning of the pharynx resembles a backward facing step. This
explains the formation of a recirculation zone at the posterior side of the
oropharynx. At the anterior side the epiglottis valve pushes the flow towards the posterior side. The epiglottis valve is a local constriction and
creates recirculation zone on its roof. The change in cross-sectional area
causes the flow to accelerate. This high-speed fluid flow, the laryngeal jet
bends around the tip of the epiglottis and a very small separation bubble
83

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 7.11: 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
separation region is about hs (where hs is the height of the step at the start
of the region). Le et al. [77] performed DNS calculation of a backward facing
step at a Reynolds number of 5000. They reported a mean reattachment
length of approximately 6hs . Other authors reported reattachment lengths
between 5-7 hs , which is much higher than the measured one in the upper
human airway model. Heenan et al. [54] found a reattachment length of
2hs at the posterior side of the oropharynx of their average simplified upper
airway geometry. Therefore, the differences in reattachment length can be
85

attributed the constriction imposed on the flow by the epiglottis at the

anterior wall. The recirculation zone under the epiglottis is notably smaller
for quiet breathing compared to normal breathing.

Figure 7.12: detailed view of the streaklines of velocity in the pharynx at

10 l/min (a), 15 l/min (b), 30 l/min (c), 45 l/min (d) air flow rate
Figure 7.13 displays the velocity profiles, normalized by the inlet plug
flow velocity, in the pharynx, 5 mm above the epiglottis for all flow rates.
A clear difference can be seen between the profiles for the lower flow rates
(10 and 15 l/min) and the higher flow rates (30 and 45 l/min). At the
two highest inhalation flow rates, the profiles display essentially the same
characteristic shape, with the profile peaks being more moderate than at
the two lower flow rates. The stronger viscous effects at lower Reynolds
numbers cause thicker boundary layers and hence the core of the jet should
relatively move faster to maintain the flow rate. A Similar effect has been
described by Johnstone et al. [66] and Heenan et al. [54]. However a
difference between the profiles of 10 and 15 l/min can be observed. The
profile of the jet is wider for 10 l/min compared the the one of 15 l/min,
86

which is an unexpected result and not observed by other authors. The

reason can be that the flow separation zone at the posterior side of the
pharynx is larger for 10 l/min (axial velocities are positive at that location)
causing the jet to be wider in order to maintain the same flow rate.

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

Figure 7.15: contour plots of the normalized turbulent kinetic energy

(k norm) for 10 l/min (a), 15 l/min (b), 30 l/min (c) and 45 l/min (d)
90

Figure 7.16: normalized turbulent kinetic energy at four different locations

in the trachea for all measured flow rates

91

7.4

Conclusion

A computer-model of the upper human airways was successfully developed

and an suitable model for PIV was created. 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 results show the
complex nature of the flow in the geometry in the upper human airways.
Several regions of recirculation in mouth, pharynx and trachea show the
possible existence of three dimensional flows. Profiles of normalized axial
velocity in the pharynx show distinct differences between the lower flow
rates, being 10 and 15 l/min and the higher flow rates, 30 and 45 l/min. A
surprising result was that the jet for 10 l/min was wider than the jet of 15
l/min inhalation flow rate. A reason for the behavior can be the influence
of the recirculation zones on the jet. The profile of normalized velocity
magnitude of the laryngeal jet in case of 30 l/min air flow is lower than in
the case of 15 l/min air flow rate. The region of high normalized turbulent
kinetic energy in the trachea at an air flow rate of 30 and 45 l/min are very
similar, while in the case of 10 and 15 l/min the region is shifted towards
the end. In the rest of the model the flow seems to be laminar. A clear
amplification in kinetic energy is seen for a flow rate as low as 15 l/min. At
higher flow rates (30 and 45 l/min) the increase in turbulent kinetic energy
starts at the glottis while at the lower flow rates this amplification starts at
approximately two tracheal diameters downstream the glottis.

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

Figure 8.2: Comparison of normalized magnitude of velocity (left) and

normalized turbulent kinetic energy (right) for three grid sized (400 000,
800 000 and 1 500 000 cells) at 5mm above the epiglottis (up) and one
tracheal diameter downstream the glottis (down)

The importance of the numerical uncertainty in CFD-simulations has

been pointed out by several authors (for example, Roache [100] and Stern
et al. [116]). In the present dissertation, the grid error was tested with two
different grid sizes. Simulations were performed on three grids containing
respectively 400 000, 800 000 and 1 500 000 cells for a flow rate of 45
l/min, which is the highest flow rate studied in this dissertation. a value
of y + 1 in the first grid point near the wall was maintained for all grids.
At two locations, 5 mm above the epiglottis and at 1 tracheal diameter
downstream the glottis, the magnitude of velocity and the turbulent kinetic
energy were compared for the different grid sizes. As can be seen on Figure
8.2, a grid size of approximately 400 000 cells is not enough to obtain a
grid independent solution, while a grid size of approximately 800 000 cells
is enough to get an independent solution for both magnitude of velocity and
turbulent kinetic energy at the two locations. Therefore in all simulations
the grid, containing approximately 800 000 cells is chosen.
95

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
=
=
=
=

surface area vector

diffusion coefficient for
gradient of
source of per unit volume

Equation 8.3 is applied to each cell, or control volume, in the computational

domain. Discretization of this equation leads to:
Nf aces

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

= number of faces enclosing cell

= value of convected through face f
= area of face f
= mass flux through the face f
= magnitude of normal to face f
= cell volume

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

turbulent dissipation can be expressed as follows:

3/4

inlet =

c k 3/2
l

(8.6)

where c is a constant which is used in the equation for calculation the

turbulent viscosity,which is equal to 0.09 for most model and l is a turbulent
length scale which can be estimated by:
l = 0.1DH

(8.7)

where DH is the hydraulic diameter of the inlet section.

Figure 8.3: Comparison of normalized turbulent kinetic energy (right) for

two different inlet boundary conditions at 5mm above the epiglottis (left)
and one tracheal diameter downstream the glottis (right) at 45 l/min
The sensitivity of turbulent kinetic energy in the upper airway geometry
to the inlet turbulence intensity Iinlet is tested for a flow rate of 45 l/min. On
Figure 8.3 the influence of the inlet turbulence intensity on the normalized
turbulent kinetic energy at 5mm above the epiglottis and at one tracheal
diameter downstream the glottic opening. Simulations were performed with
an inlet turbulent intensity of 5% and 10% and as can be seen the variation
of turbulence intensity at the inlet has no influence on the turbulent kinetic
energy in the pharynx and trachea. A uniform profile is chosen at inlet.
Different inlet profiles tested by Heenan et al. [54] did not have any influence
on the results obtained by the simulations.
The flow is modeled using the different turbulence models extensively
discussed in Chapter 5. The flow was assumed to be converged when the
dimensionless mass and momentum residuals were less than 105 .
98

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

Results and Discussion

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

Figure 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

Figure 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

Figure 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

Figure 8.7: zoom of the streaklines in the pharynx at 15 L/min: panel a:

k--sst, panel b: k--realizable, panel c: reynolds stress model and panel b:
experiments at 15 l/min

Figure 8.8: zoom of the streaklines in the trachea at 15 L/min: panel a:

k--sst, panel b: k--realizable, panel c: reynolds stress model and panel b:
experiments at 15 l/min
103

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)

where is the density, k is the turbulent kinetic energy, is the laminar

dynamic viscosity and is the specific turbulent dissipation. Numerical
simulations using this turbulence model where compared with experimental
measurements in a straight round tube with a local 75% constriction. Velocity profiles obtained with different turbulence models (RNG k-, Menter k-,
Low Reynolds Number k- and the proposed Low Reynolds Number k-) at
different sections at Reynolds number 1000, 2000 and 5000 were compared
with experimental measurements performed by Ahmed and Giddens [14].
The proposed model gave the best results. No differences of velocity profiles in mouth and pharynx at 10 l/min using all turbulence models in the
simplified geometry of [74] were found. Only in the trachea differences were
found. The low Reynolds number SST k- turbulence model gave also good
results for the velocity profiles in the trachea but comparison of experimental velocity profiles with the simulated data showed important differences
in the pharynx. Therefore only comparing velocity profiles in the trachea is
not enough for choosing one turbulence model over the other.
Figure 8.11 shows the profiles of turbulent kinetic energy at one (a and
c) and two (b and d) tracheal diameters downstream the glottis obtained
106

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)

where D is the diameter of the channel, v is the velocity, is the kinematic

viscosity, L is the length of the channel and R is the radius of the curvature
of bend. The Dean number in the case of 30 l/min is of order 103 , which
is high. Heenan et al. [54] also found a secondary flow pattern in the
mouth but only reported a single vortex pair. The difference between the
secondary motion in the present model and the simplified model of Heenan
et al. [54] can be attributed to the difference in mouth geometry. The upper
109

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

Figure 8.12: cross-sectional view of the streamlines in the mouth (a),

pharynx (b) and trachea (c)

112

Figure 8.13: a three-dimensional view of the streamlines

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

The dispersed phase, in a Lagrangian frame of reference, which is discussed

in detail in chapter 4 is simulated by tracking simultaneously a large number
of particles through the converged flow field. The volume fraction of medical
aerosols is usually in the order of 106 or less (dilute flow) and therefore
one-way coupling is taken into account where the particle motion is treated
as having no effect on the fluid motion, particle-particle interactions can
also be ignored [31]. Only drag force and gravity are considered to be
acting on the particles. Brownian motion, caused by the collision of air
molecules was considered to have no influence on the trajectories of particles
with a diameter larger than 0.1 m [84]. The influence of turbulence on
the particle motion is taken into account by using the Eddy Interaction
Model, which is discussed in chapter 4 Stapleton et al. [115] reported the
inadequacy of the Eddy Interaction Model in combination with the k-
turbulence model to simulate correctly the particle deposition in the human
oral airways. In combination with the standard k- turbulence model, the
numerical prediction of deposition showed improvement but was still far
off the experimental curve [85]. In fact tracking the particles without the
influence of turbulence gave better results, but under predicted the particle
deposition of the larger particles. Matida et al. [85] proposed an improved
Eddy Interaction Model, where anisotropy of the turbulence near the wall
was taken into account. A damping function f , based on the function
normal to the wall, introduced by Wang and James [126] was proposed:
f = 1 e0.02y

(9.1)

For values of y + less than approximately 80 a new turbulent kinetic energy

was proposed:
knew = f 2 k
(9.2)
The introduction of this near-wall correction showed better agreement
with experiment. However Zhang et al. [144] introduced the same correction in combination with their low-Reynolds number k- turbulence model
but almost no difference could be found with the results obtained without
116

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

simulated deposition results on an idealized geometry did not change when

the number of particles injected were changed from 1000 to 10 000.
particle
30 l/min
30 l/min
diameter Deposition (%) Deposition (%)
(m)
4500 particles 15000 particles
1
4.75
4.67
2
5.42
5.74
4
8.59
9.09
6
19.23
18.65
8
42.75
41.59
10
70.19
69.13
12
85.31
85.35
16
97.11
97.43
20
99.51
99.51
Table 9.1: Influence on total injected particles on the total deposition
percentage

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

Figure 9.1: Inspiratory deposition efficiency: comparison of simulated

deposition with reported experimental data
assuming no expiratory deposition in the oral airway:
= 1

1
4.2e6 (d2p Q)1.7

+1

(9.3)

where represents the deposition efficiency. Cheng et al. [24] proposed

function 9.4 obtained by applying nonlinear least-squares regression on experimentally obtained particle deposition in an airway replica.
= 1 exp(0.000276d2p Q)

(9.4)

The numerically obtained total deposition efficiency was computed by

neglecting the deposition in the inlet tube, because this is not a part of
119

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

ability. In order take into account this intersubject geometric variability,

Grgric et al [47] proposed to plot the extrathoracic deposition as a function
of the non-dimensional Stokes number. First, the Stokes number is defined
as:
p d2p Umean
Stk =
(9.5)
18Dmean
where p is the particle density, is the dynamic viscosity of the carrier
fluid, Dmean the mean diameter, which is calculated using the cast volume
V and the path length L of the central sagittal line of the airway geometry
and defined by:
p
(9.6)
D = 2 V /L
The corresponding velocity scale is calculated as
Umean = QL/V

(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

Local deposition analysis

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

Figure 9.4: Simulated deposition values (expressed as % of total number

of particles) in three model subparts for four different flow rates (10, 15, 30
and 45 L/min)
more on the roof of the oral cavity.
Figure 9.6 shows the deposition of 10 m particles in the upper airway
respiratory tract in the case of a flow rate of 15 L/min. The dark grey line
represents the top view of the central sagittal plane. As can be seen the
particles are scattered all over the geometry.

9.3.3

Influence of gravity

By simulation of the particle transport with and without the influence of

gravity, the effects of gravity and inertial impaction on the particle deposition could be separated. Turbulence also has an non negligible influence on
the deposition and will be discussed in the following section.
124

Figure 9.5: Two-dimensional representation of individual particle deposition

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

sitting or standing position of the human subject. As already explained a

simulation in the case of 30 L/min inhalation flow rate was performed using
zero gravity.
The orientation of the gravity vector was also studied. A simulation
was performed with the model in the horizontal position, corresponding to
a lying position of the subject. Finally, the gravity vector was placed under
an angle of 45 , corresponding to a position where the human subject leans
backwards. Both simulations are performed in case of a flow rate of 30
L/min.
All results are presented on Figure 9.7, where the total deposition in zero
gravity, gravity vector under an angle of 45 , gravity vector under an angle
of 90 and normal gravity vector (connected line) are shown. As can be seen
the difference in total deposition of small (1 and 2 m) and bigger (8, 10, 12,
16 and 20 m) between no gravity and normal gravity is very small. Only
for particles with a diameter of 4 and 6 m show smaller deposition in zero
gravity condition. However this difference is less than 2%. The direction
of the gravity vector has the most influence on the deposition of particles
with a diameters ranging from 8 till 12 m. This difference is maximum
10 % and mainly occurs in the pharynx (not shown on the figure). Smaller
particles, which have a less weight and the bigger particles, which already
deposit in the mouth are much less affected by the change in direction of
the gravity vector.

9.3.4

Influence of turbulence (Eddy Interaction Model)

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

Figure 9.8: Comparison of total deposition between mean flow tracking

and Eddy Interaction Model

9.3.5

Influence of the carrier gas (Heliox vs Air)

One of the currently used methods to improve the alveolar deposition of

medical aerosols is the use of a helium/oxygen (Heliox) mixture instead of
air as the carrier gas. Breathing impairment caused by upper airway constrictions such as Croup [20] and airway tumors [121] are often treated by
using Heliox. Jaber et al. [61] experimentally observed a temporary relief
of breathing impairment during Heliox administration with post-intubation
patients. Kress [75] noticed a therapeutic improvement in the treatment of
acute athma with the 2 antagonist albuterol when delivered with 80/20
Heliox as compared to oxygen. After three treatment doses, the change in
Forced Expiratory Volume (F EV1 ) from pre-treatment levels for patients
receiving albuterol in Heliox was 65.1% as compared to 26.6 % for patients
using standard albuterol delivery in oxygen. Anderson et al. [118] have
128

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

Figure 9.9: Comparison of total deposition between particles suspended

in air and heliox
the best experimental fit of Grgic et al. [47]. As can be seen the deposition
values both for air and Heliox are close to this experimental fit, suggesting
once again the adequacy of the applied method.

9.3.6

Influence of unsteady flow rate

Effect of flow increase rate on the deposition

Devices, such as dry powder inhalers (DPIs) release the medication via
rapid inhalation instead of using chemical propellants. The strength of the
inhalation required is stronger than that needed to release the medication of
a metered-dose inhaler (MDI). From literature Grgic et al. [49] reported values for flow rates of 20 to 90 L/min and flow acceleration of 0.5-8.0 L/s2 at
the moment of particle release. Therefore, Grgic et al. [49] developed a procedure for measuring the effects of flow increase rate(FIR) on mouth-throat
deposition. Comparison of deposition of 5 m dioctylphthalate (DOP) particles was made between bolus injection into a steady-state flow of 30 L/min,
and bolus injection into unsteady flow accelerating through 30 L/min at
either 2 or 4 sL2 . In addition, Grgic et al. [49] performed an unsteady numerical simulation of the effects on mouth-throat deposition for 5 and 10
m particles. In their numerical study no turbulent dispersion was taken
into account.
For validation similar numerical procedures as Grgic et al. [49] where
131

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

Figure 9.11: Scheme of the inhalation profile and particle injection

particle
diameter
(m)
1
2
4
6
8
10
12
16
20

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

The transport of particles, with diameter ranging from 1 to 20 m is tracked

in a Lagrangian frame of reference through the converged flow field, obtained
135

with RANS in combination with a low Reynolds number k- turbulence

model, in the developed model of the upper human airways. Newtons
second law was solved for the dispersed phase and drag force and gravity
were the acting forces. The influence of turbulence was taken into account
by using the standard Eddy Interaction model.
Simulations of total deposition compared well with experimental deposition data, for particles with a value for the non-dimensional parameter
Stk.Re0.37 higher than 0.1
(Stokesnumber.Reynoldsnumber 0.37 > 0.1). Total deposition of particles
with a smaller value was overpredicted, probably due to exaggerated turbulence simulated at low flow rates.
The study of local deposition showed that the deposition in the oral
cavity increases with increasing particle diameter, while deposition in the
pharynx first increases and then decreases. Deposition in the larynx and
trachea is negligible. This agrees with experimental observation of particle
deposition in realistic upper airway geometries. This in contrary to other
simplified geometries described in the literature.
The influence of gravity and turbulence on the total deposition was studied and it was found that gravity only has a minor effect on the deposition
of micron sized particles in the upper human airways. Turbulence has a
greater effect on the deposition. The major influence on the extra-thoracic
deposition is the inertial impaction.
The positive influence of Heliox on the deposition is shown. The differences in total deposition values between particles carried by air and
Heliox can go up to 35 % (16 m at 15 L/min)
A transient calculation with flow increase rate was carried out and compared with experimental measurements reported in literature. Similar deposition differences between the steady and unsteady flow rate accelerating
through 30 L/min for FIR of 2 L/s2 were found. The transient numerical simulation had a difference of 2 % for 4 m particles and almost 9 %
for 6 m particles. Difference which were not found back in the reported
numerical study.
The influence of the particle injection time was studied by performing an
unsteady RANS calculation where the particles were released at the moment
when the maximum flow rate was obtained and compared with deposition
results obtained by using a fully developed (steady) flow rate. Deposition of
particles, released immediately after reaching maximum flow rate is consid136

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).

A simple way of characterizing flow fields in a trachea or upper airway

is to study its pressure-flow relationship. Wheatley et al. [133] measured in
situ nasal and oral pressure drops in healthy men and related these to the
corresponding flows, to study the exponents of the power law that related
both. When the power of flow had a value of 1.00, 1.75 and 2, these were assumed to represent laminar, turbulent and orifice flow regimes respectively.
While the exact labeling of such categories may be debatable in strict fluid
dynamic terms, the use of the power exponent to somehow quantify the
flow regime at hand is interesting. Whether it can also provide a diagnostic
tool that can be used in a clinical setting may be more complicated. In its
most invasive experimental setting, this would require in situ pressure measurements beyond the stenosis, as was done by Wasserman et al. [127] in
view of proposing a cut-off for stenotic resistance, beyond which surgery was
indicated. A less invasive possibility is the use of frequency dependence of
resistance, measured by body plethysmography, as was proposed by Fasano
et al [38] in patients with laryngeal hemiplegia.

The exploration of any of these possibilities also requires a detailed study

of the flow fields in a representative airway structure with and without stenosis. The focus of the present study was on the so-called weblike stenosis,
with a stenosis length of only 2mm along the tracheal axis, and typically
occurring in the upper third of the trachea; this is a typical presentation of
type I post-intubation stenosis [92]. For a subset of conditions, an elongated
stenosis (length: 30mm of tracheal axis) was also considered; this is a typical presentation of a complex or type II stenosis [92]. With a particular
attention to the relationship between pressure drop and flow over stenosis
constrictions, we set out to quantify the exponents of the power laws at
hand, but also to determine the pertaining proportionality constants for
easy estimation of actual pressure drops.
139

10.2

Materials and methods

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

pseudo-wave speed c of the same order of magnitude as the fluid speed.

The preconditioned Reynolds-Averaged Navier-Stokes equations including turbulence and species transport equations written in a Cartesian frame
of reference and integrated over a control volume are expressed as:
ZZZ

1 Q

d +

ZZ
S

141

F dS =

ZZ
S

V dS

(10.1)

with
Q = (pg , u, v, w, Eg , k, )

(10.2)

The variable pg = p pref is the gauge pressure, u, v, w are the three

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

One typical simulation of a flow field in the realistic geometry of Figure

10.1 took approximately 12 to 15 hours to reach convergence (HP Integrity
rx5670 4 x 1.3 GHz Itanium2, Hewlett Packard, Palo Alto, CA ).

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

exceeding the velocities in the glottic area. As expected in the geometry

without stenosis, the glottic area results in a so-called laryngeal jet which
impinges on the anterior side of the trachea with a huge flow separation
zone forming at the posterior side of the trachea. The symmetry typical of
a weblike stenosis basically reorients the jet along the tracheal axis, with
flow separation occurring on either side of the jet in the case of a relatively
mild stenosis (e.g. 50% constriction; panel B). In the case of a more severe
constriction (e.g. 90 % constriction; panel C), the laryngeal jet is dwarfed
by the jet developed at the site of stenosis.

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

a flow of 30 L/min. The effect of the glottic narrowing (located at 152mm

from model inlet) is seen to be of the same order as that of the 50% constriction, 34mm further downstream. By contrast, a stenosis narrowing to 75%
constriction (i.e., leaving 25% of the tracheal lumen unobstructed) more or
less doubles the pressure drop across the stenosis. An extra 10% and 15%
constriction produces dramatic increases in pressure drop. For the greatest stenosis constriction (90%), the pressure obtained in a longer (30mm)
stenoses with a central channel of 10, 20 and 30mm length, is also represented. These long stenoses show a modest difference in pressure profile
with a slightly smaller magnitude of total pressure drop than the weblike
stenosis of comparable constriction (90%). One would expect the pressure
drop in the long stenoses to be higher than the pressure drop in the webliken stenosis because of the higher friction in the small channel. However,
the shape of the entrance and exit of the long stenoses is slightly different
to the entrance and exit of the web-like stenosis, which can explain this
observation. The difference in pressure drop between the 10mm and 30mm
stenosis (26Pa) is approximately twice the pressure drop between the 10mm
and 20mm stenosis (12Pa).
When expressing the simulated pressure drops over the stenosis as a
function of constriction (Figure 10.4A), a dramatic P increase with constriction is observed well beyond 70% constriction, in a characteristic pattern which is similar for 15 and 30 L/min inlet flows (closed and open circles
respectively). When considering heliox instead of air at a flow rate of 30
L/min (crosses in Figure 10.4A), this resulted in a reduction of pressure
drop to approximately one third of its value with air at 30 L/min for all
constrictions in Figure 10.4A, consistent with the density of heliox being
one third that of air.
From these CFD simulated curves an attempt was made to derive an
empirical formula which could serve as a rule of thumb, in a clinical situation
where inlet flow and stenosis cross section can be reliably estimated. For
this purpose, we first consider the generalized Bernoulli equation (excluding
the effect of gravity), which takes into account cross sectional changes (as if
they were occurring in a straight tube) and a sum of so-called losses Piloss
which account for frictional losses (e.g., induced by the overall geometry)
and local losses owing to particular geometrical features (e.g. an orifice at
the level of the stenosis)
145

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)

where P is pressure, indices 1, 2 and 3 refer to locations at respectively

model entry, stenosis, and model exit; is gas density (1.2 kg/m3 for air;
0.4 kg/m3 for heliox), V is local velocity and SPis local cross section. If we
consider that in our particular case, the term i Piloss refers to a simple
sum of the pressure loss generated by the overall upper airway geometry
(P[nostenosis] ) and the pressure loss generated by the stenosis (P[stenosis]),
the latter term can be isolated to estimate the pressure drop generated
by the stenosis. In engineering applications, P loss (and applied here to
P[stenosis] ) is usually expressed as
loss
Pstenosis
= K

146

Q2
2S22

(10.7)

where Q is bulk flow and K a constant which is empirically determined

for each particular geometrical feature (in this case, location 2 at the stenosis). K = 0 corresponds to zero loss and a typical K-value for airflow in
a 90 miter degree bend is 1.1, but K can increase to more than 100 for a
partially opened butterfly valve [134].

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)

A more detailed inspection of the flow dependence of pressure drop over

the entire model provides another perspective for clinical decision making with respect to tracheal stenoses. Indeed, the dependence of pressure
drop on flow, with its specific power law exponent depending on whether a
stenosis is present (2.00) or not (1.77) (Figure 10.5), suggests an attractive
non-invasive way to probe the trachea for presence of considerable stenosis.
Wheatley et al [133] have determined the power law exponent to quantify pressure-flow relationships they were measuring in human nasal and
oral passages, based on the rationale that power exponents 1.00, 1.75 and 2
would respectively correspond to laminar, turbulent and orifice flow regimes.
Recording oral and nasal pressure-flow power relationships, they obtained
best-fit exponent values around 1.75, with a consistently greater exponent
during inspiration (1.830.04) versus expiration (1.660.07), which was
thought to signify greater level of turbulence during inspiration. Surely,
the high quality of fit in Wheatley et al [133] was partly due to the fact
that pressure and flow were measured in situ, over a very well-defined oral
149

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

obtained by Brancatisano et al. [17] at mid-inspiration. The difference in

pressure drop obtained here for 30 L/min (15Pa) compared to those obtained on simplified models ([19], [54]) (30Pa across both studies), can
be almost entirely accounted for by the difference in glottic cross section.
Given that in these previous models, cross sections ranged 9-10mm2 , and
considering that simulated pressure drop was seen to decrease with the
square of increasing cross section [19], the squared ratio (0.95/1.25)=0.58
accounts for most of the difference in pressure drop between the present
model without stenosis and previous unobstructed models. The remaining
difference can be probably accounted for by the abruptness of cross sectional
changes in the oropharyngeal structure proximal to the glottis. In any case,
the present results clearly shows that the pressure drop over the glottis
which actually corresponds to a 40% constriction (i.e., 125mm2 /310mm2 ),
is negligible with respect to those induced by the constrictions greater than
70%, which impair breathing. In fact, Jaeger and Matthys [62] used an
equation of the form of Equation 10.7 to study the effect of laryngeal constriction on the relationship between pressure drop and flow. Using the plot
of the experimentally obtained discharge coefficients (corresponding to 1/K
in Equation 10.7) versus Reynolds number (as illustrated in Figure 11 of
Chang [21]), and assuming that the same relationship would hold for stenosis constrictions, we computed the estimated P for the cases considered in
Fig.10.4A. For 50% constriction, P values were similar to those obtained
with CFD in our model, whereas for 80% and 90% constrictions, the CFD
computed P was underestimated by respectively 30% and 50% when using
the discharge coefficient proposed for the larynx.
While the focus of this chapter was on weblike stenosis, the effect that an
elongated stenosis could have on pressure drop in the case of the most severe
constriction, was investigated. In fact, the pressure drop with the three long
stenoses are slightly smaller compared to that obtained for a weblike stenosis
of similar constriction (90%). This is due to the fact that in the weblike
stenosis, the constriction is immediately followed by the expansion, while in
the long stenoses the flow is able to develop in the small channel between
constriction and expansion. The dependence of pressure drop on length of
the long stenosis can be explained on the basis of the frictional losses that
come into play along the channel lengths. In turbulent regime, the pressure
drop due to friction of a fluid in a horizontal pipe of length L and diameter
151

d, can be approximated by ([87],[134]):

P = 0.158L0.75 0.25 d1.25 V 1.75

(10.8)

, where is density of the fluid, is dynamic viscosity and V is velocity

in the tube; the coefficient 0.158 was taken from White et al [134] which
was considered representative of a rough walled tube. In the case of 90
% stenosis, flow rate of 30 l/min and a stenosis length of 10mm, 20mm
and 30mm, respective frictional losses are then estimated at 8.6 Pa, 17.1
Pa and 25.7 Pa (i.e., corresponding to the multiplicative factor on length).
Considering the difference in inlet conditions to the stenosis of our model
versus that of a straight tube of diameter d and length L to which this
equation applies, the CFD computed pressure drops between 10mm and
30mm stenosis (26Pa) are more or less double that between 10mm and
20mm (12Pa), which is consistent with the above predicted frictional losses.
In conclusion, by applying computational fluid dynamics in a realistic
model of tracheal stenosis, pressure drops during normal breathing that
are consistent with the clinically observed sudden appearance of breathing
impairment at a relatively advanced stage of tracheal obliteration, were
simulated. A rule of thumb is derived from which pressure drops over the
stenosis can be estimated on the basis of breathing flow and stenosis cross
section. In addition, from the CFD simulations the best-fit exponent in the
power law that relates upper airway resistance to flow could indeed be used
as a diagnostic tool in the non-invasive monitoring of stenosis patients, was
determined.

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 chapter 9 the transport of particles, with diameters ranging from

1 to 20 m, is simulated in a Lagrangian frame of reference through the
converged flow field in the developed geometry of the human upper airway
model. Simulations of total deposition compared well with experimental
deposition data, for particles with a value for the non-dimensional parameter
Stk.Re0.37 higher than 0.1. Total deposition of particles with a smaller value
was overpredicted, probably due to excessive turbulence in the simulations
at low flow rates.
Also, local deposition was studied and the results confirm that deposition
in the developed model resembles experimental local deposition studies in
realistic geometries. This in contrary to other reported simplified models.
The influence of gravity and turbulence on the total deposition was studied
and it was found that gravity only has a minor effect on the deposition
of micron sized particles in the upper human airways. Turbulence has a
greater effect on the deposition. The major influence on the extra-thoracic
deposition is the inertial impaction.
The positive influence of Heliox, as carrier gas on the deposition is shown.
A transient calculation with flow increase rate was carried out and successfully compared with experiments reported in literature.
A transient calculation with flow increase rate was carried out and compared with experimental measurements reported in literature. Similar deposition differences between the steady and unsteady flow rate accelerating
through 30 L/min for flow increase rate of 2 L/s2 were found.
The influence of the particle injection time was studied by performing
an unsteady RANS calculation where the particles were released at the
moment when the maximum flow rate was obtained and compared with
deposition results obtained by using a fully developed (steady) flow rate.
Deposition of particles, released immediately after reaching maximum flow
rate is considerably lower than particles injected at the moment of fully
developed flow.
In chapter 10 a CFD study in the developed model in combination with
tracheal stenoses was performed. Pressure drops during normal breathing
that are consistent with the clinically observed sudden appearance of breathing impairment at a relatively advanced stage of tracheal obliteration, were
simulated. A rule of thumb was derived from which pressure drops over
the stenosis can be estimated on the basis of breathing flow and stenosis
cross section. In addition, from the CFD simulations the best-fit exponent
154

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

will probably not take place.

The developed model could also be coupled with models of the intrathoracic airways and simulations of particle deposition in the complete human airways could help engineers and physicians to better understand the
complex behavior of the aerosols in the airways and to develop tools to
improve the alveolar deposition of medical aerosols.

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

[15] G. M. Allen, B. P. Shortall, T. Gemci, T. E. Corcoran, and N. A.

Chigier. Computational simulations of airflow in an in vitro model
of the pediatric upper airway. Journal of Biomechanics, 126:604613,
2004.
[16] S. M. Bowes and D. L. Swift. Deposition of inhaled particles in the oral
airway during oronasal breathing. Aerosol Sci. Technol., 11:157167,
1989.
[17] T. Brancatisano, P.W. Collett, and L. A. Engel. Respiratory movements of the vocal cords. J Appl Physiol, 54:12691276, 1983.
[18] E. O. Brigham. The fast fourrier transform. Prentice Hall, Englewood
Cliffs, New Jersey, 1974.
[19] M Brouns, S Verbanck, and C Lacor. Influence of glottic aperture on
the tracheal flow. Journal of Biomechanics, 40:165172, 2007.
[20] J. Brown. The managment of croup.
61:189202, 2002.

Britisch Medical Bulletin,

[21] H. K. Chang. Flow dynamics in the respiratory tract, in: Respiratory

Physiology: An Analytical Approach. Marcel Dekker inc., New York,
1988.
[22] K. H. Cheng, Y. S. Cheng, H. C. Yeh, and D. L. Swift. Measurements
of airway dimensions and calculation of mass transfer characteristics
of the human oral passage. J. Biomed. Eng., 119:475482, 1997.
[23] K. H. Cheng, Y. F. Su, H. C. Yeh, and D. L. Swift. Deposition of
thoron progeny in human head airways. Aerosol Sci. Technol, 18:359
375, 1997.
[24] Y. S. Cheng, Y. Zhou, and B. T. Chen. Particle deposition in a cast
of human oral airways. Aerosol Science Technology, 31:286300, 1999.
[25] D. Choi and C. L. Merkle. Prediction of channel and boundary-layer
flows with a low-reynolds number turbulence model. AIAA Journal,
23:15181524, 1985.
158

[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

[37] G. T. Reid F. P. Chiang. Optics and Lasers in engineering. Elsevier,

New York, 1988.
[38] V. Fasano, L. Raiteri, E. Bucchioni, S. Guerra, G. Cantarella, M. G.
Massari, B. M. Cesana, and L. Allegra. Increased frequency dependence of specific airway resistance in patients with laryngeal hemiplegia. Eur Respir J, 18:10031008, 2001.
[39] GINA Global Initiative for Asthma. The global burden of asthma
report, 2004.
[40] T. Gemci, T. E. Corcoran, and N. Chigier. Dispersion and deposition
of inhalation therapy sprays in the larynx and trachea using experimantal and numerical methods. Proc. eighth int. conf. on liquid
atomization and spray systems, Passadena, 2000.
[41] T. Gemci, T. E. Corcoran, and N. Chigier. Characterization of the
laryngeal jet using phase doppler interferometery. Aerosol Science and
Technology, 36:1838, 2002.
[42] T. Gemci, T. E. Corcoran, K. Yakut, B. Shortall, and N. Chigier.
Spray dynamics and deposition of inhaled medications in the throat.
Proc. ILASS-Europe, Zurich, 2001.
[43] T. Gemci, B. Shortall, G. M. Allen, T. E. Corcoran, and N. Chigier.
A cfd study of the throat during aerosol drug delivery using heliox
and air. Journal of Aerosol Science, 34:11751192, 2003.
[44] M. Gharib and C. E. Willert. Particle tracing revisited. Lecture Notes
in Engineering: Advances in Fluid Mechanics Measurements, 1990.
[45] A. D. Gosman and E. Ioannides. Aspects of computer simulation of
liquid-fueled combustors. Journal of Energy, 7:482490, 1983.
[46] L. P. Goss, M. E. Post, D. D. Trump, and B. Sarka. Two-color particle
velocimetry. Proc. ICALEO 89, 1989.
[47] B. Grgic, W. H. Finlay, P. K. P. Burnell, and A. F. Heenan. In
vitro intersubject and intrasubject deposition measurements in realistic mouth-throat geometries. Journal of Aerosol Science, 35:1025
1040, 2004.
160

[48] B. Grgic, W. H. Finlay, and A. F. Heenan. Regional aerosol deposition

and flow measurements in an idealized mouth and throat. Journal of
Aerosol Science, 35:2132, 2004.
[49] B. Grgic, A.R. Martin, and W. H. Finlay. The effect of unsteady flow
rate increase on in vitro mouth-throat deposition of inhaled boluses.
Experiments in Fluids, 37:673689, 2004.
[50] D. M. Habib, S. S. Garner, and S. Brandeburg. Effect of heliumoxygen on delivery of albuterol in a pediatric, volume cycled, ventilated lung model. Pharmacotherapy, 19(2):143149, 1999.
[51] I. Hahn, P. W. Scherer, and M. M. Mozell. Velocity profiles measured
for airflow through a large scale model of the human nasal cavity.
Journal of Applied Physiology, 75:22732287, 1993.
[52] A. Haider and O. Levenspiel. Drag coefficient and terminal velocity of
spherical and non-spherical particles. Powder Technology, 58:6370,
1989.
[53] A. F. Heenan, W. H. Finlay, B. Grgic, A. Pollard, and P. K. P. Burnell.
An investigation of the relationship between the flow field and regional
deposition in realistic extra-thoracic airways. Journal of Aerosol Science, 35:10131023, 2004.
[54] A. F. Heenan, E. Matida, A. Pollard, and W. H. Finlay. Experimental
measurements and computational modeling of the flow field in an
idealized human oropharynx. Experiments in Fluids, 35:7084, 2003.
[55] L. Hesselink. Digital image processing in flow visualization. Annu.
Rev. Fluid Mech., 20:421485, 1988.
[56] W. C. Hinds. Aerosol Technology: properties, behavior and measurement of airborne particles. John Wiley and Sons, 1982.
[57] J. O. Hinze. Turbulence. McGraw-Hill, New York, 1975.
[58] C. Hirsch, C. Lacor, C. Rizzi, P. Eliasson, I. Lindblad, and J. Hauser.
A multiblock/multigrid code for the efficient solution of complex 3d
navier-stokes flows. European Symposium on Aerodynamics for Space
Vehicles, pages 415420, 1991.
161

[59] V. Hoffstein and J. J. Fredberg. The acoustic reflection technique for

non-invasive assessment of upper airway area. AIAA Journal, 4:602
611, 1991.
[60] L. M. Hopkins, J. T. Kelly, A. S. Wexler, and A. K. Prasad. Particle
image velocimetry measurements in complex geometries. Experiments
in Fluids, 29:9195, 2000.
[61] S. Jaber, A. Carlucci, M. Boussarsar, R. Fodil, J. Pigeot, S. Maggiore,
A. Harf, D. Isabey, and L. Brochard. Helium-oxygen in the postextubation period decreases inspiratory effort. Am J Respir Crit Care
Med, 164:633637, 2001.
[62] M. J. Jaeger and H. Matthys. The pattern of flow in the upper human
airways. Respir Physiol, 6:113127, 1968.
[63] A. Jameson, W. Schmidt, and E.. Turkel. Numerical simulation of
euler equations by finite volume methods using runge-kutta time stepping schemes. AIAA Journal, 81-1259, 1981.
[64] S.T. Jayaraju, M. Brouns, S. Verbanck, and C. Lacor. Fluid flow and
particle deposition analysis in a realistic extrathoracic airway model
using unstructured grids. Journal of Aerosol Science, 38:494508,
2007.
[65] W. Johnston, A. Dybbs, and R. Edwards. Measurement of fluid velocity inside porous media with a laser anemometer. Physics of FLuids,
29:913914, 1975.
[66] A. Johnstone, M. Uddin, A. Pollard, A. Heenan, and W. H. Finlay.
The flow inside an idealised form of the human extra-thoracic airway.
Journal of Aerosol Science, 37:10251040, 2004.
[67] I. M. Katz, B. M. Davis, and T. B. Martonen. A numerical study
of particle motion within the human larynx and trachea. Journal of
Aerosol Science, 30-2:173183, 1999.
[68] I. M. Katz and T. B. Martonen. Flow patterns in three-dimensional
laryngeal models. journal of aerosol medicine, 9-4:501511, 1996.
162

[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

[80] S. G. Marketos and C. N Ballan. Bronchial asthma in the medical

literature of greek antiquity. J Asthma, 19(4):263269, 1982.
[81] T. B. Martonen and J. Lowe. Assessment of aerosoldeposition patterns in human respiratory tract casts. In: Aerosols in the mining
and industrial work environments, Vol 1. Fundamentals and status.
Ann Arbor Science Publishers, 1983.
[82] T. B. Martonen, L. Quan, Z. Zhang, and C. J. Musante. Flow simulation in the human upper respiratory tract. Cell. biochemistry and
biophysics, 37:2736, 2002.
[83] T. B. Martonen, Z. Zhang, G. Yu, and C. J. Musante. Threedimensional computer modeling of the human upper respiratory tract.
Cell. biochemistry and biophysics, 35:255261, 2001.
[84] T. B. Martonen, Z. Zhang, G. Yue, and C. J. Musante. 3-d particle transport within the human upper respiratory tract. Journal of
Aerosol Science, 33:10951110, 2002.
[85] E. A. Matida, W. H. Finlay, C. F. Lange, and B. Grgic. Improved numerical simulation of aerosol deposition in an idealized mouth-throat.
Journal of Aerosol Science, 35:12221233, 2006.
[86] T. Matsuo and H. G. Colt. Evidence against routine scheduling of
surveillance bronchoscopy after stent insertion. Chest, 118:14551459,
2000.
[87] J. Mead, J. M. Turner, P. T. Macklem, and J. B. Little. Significance
of the relationship between lung recoil and maximum expiratory flow.
Respir Physiol, 22:95108, 1967.
[88] W. Merzkirch. Flow visualization. New York Academic, 1987.
[89] S. A. Morsi and A. J. Alexander. An investigation of particle trajectories in two-phase flow systems. J. Fluid Mech, 55(2):193208,
1972.
[90] S. P. Newman. Principles of metered-dose inhaler design. Respiratory
Care, 50(9):11771190, 2005.
164

[91] J. Noguiera, J. Lecuona, and P. A. Rodriguez. Local field correction

piv: on the increase of accuracy of digital piv systems. Experiments
in Fluids, 27:107116, 1999.
[92] M. Noppen. Airway injury and sequelae : conservative view. Eur
Respir Mon, 29:234245, 2004.
[93] P. J. ORourke. Statistical properties and numerical implementation
of a model of droplet dispersion in a turbulent gas. Journal of Computational Physics, 83:345360, 1989.
[94] W. K. Pratt. Digital Image processing. John Wiley and Sons, New
York, 1991.
[95] K. F. Rabe, P. A. Vermeire, and J. B. Soriano. Clinical management
of asthma in 1999: the asthma insights and reality in europe (aire)
study. Eur Respir J, 16:802807, 2000.
[96] M. Raffel, C. E. Willert, and J. Kompenhans. Particle image velocimetry - a practical guide. Springer, Berlin, Heidelberg, 1998.
[97] C. Renotte, V. Bouffioux, and F. Wilquem. Numerical 3d analysis
of oscillatory flow in the time-varying laryngeal channel. Journal of
Biomechanics, 33:16371644, 2000.
[98] W. C. Reynolds. Fundamentals of turbulence for turbulence modeling
and simulation, 1987.
[99] C. M. Rhie and W. L. Chow. Numerical study of the turbulent flow
past an airfoil with trailing edge separation. AIAA Journal, 21:1525
1532, 1983.
[100] P. J. Roache. Quantification of uncertainty in computational fluid
dynamics. Annual Review of Fluid Mechanics, 29:739749, 1997.
[101] F. Scarano. Iterative image deformation methods in piv - review article. Meas. Sci. and Tech., 13:R1R19, 2002.
[102] F. Scarano and M. L. Riethmuler. Iterative multigrid approach in piv
image processing with discrete window offset. Expperiments in Fluids,
26:513523, 1999.
165

[103] F. Scarano and M. L. Riethmuler. Advances in iterative multigrid piv

image processing. Expperiments in Fluids, 26:S51S60, 2000.
[104] R. B. Schlesinger and M. Lipmann. Particle deposition in the trachea:
In vivo and in hollow casts. Thorax, 31:678684, 1976.
[105] H. Schlichting. Boundary-layer theory. MCGraw-Hill, New York,
1968.
[106] A. A. Schraiber, L. B. Gavin, V. A. Naumov, and V. P. Yatsenko. Turbulent flows in gas suspensions. Hemisphere Publishing corporation,
New York, 1990.
[107] R. C. Schroter and M. F. Sudlow. Flow patterns in models of the
human bronchial airways. Respiratory Physiology, 7:341355, 1969.
[108] J. S. Schuen, L. D. Chen, and G. M. Faeth. Evaluation of a stochastic model of particle dispersion in a turbulent round jet. American
Institute of Chemical Engineers Journal, 29:167170, 1983.
[109] M. M. Schuurmans and C. T. Bolliger. Silicone airway stents. In: Interventional pulmonary medicine. Lung Biology in health and Disease,
volume 189. Marcel Dekker, New York.
[110] T. H. Shih, W. W. Liou, A. Shabbir, Z. Yang, and J. Zhu. A new
k- eddy-viscosity model for high reynolds number turbulent flows model development and validation. Computers and Fluids, 24(3):227
238, 1985.
[111] S.L. Soo. Fluid Dynamics of Multiphase systems. Blaisdell, London,
1990.
[112] W. Stahlhofen, J. Gebhard, and J. Heyder. Experimental determination of the regional deposition of aerosol particles in the human
respiratory track. American Industrial Hygiene Association Journal,
41:385398a, 1980.
[113] W. Stahlhofen, J. Gebhart, J. Heyder, and G. Scheuch. New regional deposition data of the human respiratory tract. J. Aerosol
Sci., 14:186188, 1983.
166

[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

[123] E. Turkel. Preconditioning methods for solving the incompressible and

low-speed compressible equations. Journal of Computational Physics,
72:277298, 1987.
[124] M. O. Turner, K. Noetjoj, S. Vedal, T. Bai, S. Crump, and J. M.
Fitzgerald. Risk factors for nearfatal asthma. a casecontrol study
in hospitalized patients with asthma. Am J Respir Crit Care Med,
157:18041809, 1998.
[125] J. M. Vergnon, F. Costes, M. C. Bayon, and A. Emonot. Efficacy of
tracheal and bronchial stent placement on respiratory function tests.
Chest, 107:741746, 1995.
[126] Y. Wang and J. P. James. On the effect of anisotropy on the turbulent
dispersion and deposition of small particles. International Journal of
Multiphase Flow, 25:749763, 1999.
[127] K. Wassermann, A. Koch, A. Warschkow, F. Mathen, J. MullerEhmsen, and H. E. Eckel. Measuring in situ central airway resistance
in patients with laryngotracheal stenosis. Laryngoscope, 109:1516
1520, 1999.
[128] E. Weibel. Morphometry of the human lung. Springer-Verlag, Berlin.
[129] J. Westerweel. Digital particle image velocimetry - Theory and application. PhD thesis, Technical University of Delft, Delft, the Netherlands, 1993.
[130] J. Westerweel. Efficient detection of spurious vectors in particle image
velocimetry data. Meas. Sci Tech., 16:236247, 1994.
[131] J. Westerweel, D. Dabiri, and M. Gharib. The effect of a discrete
window offset on the accuracy of cross-correation analysis of digital
piv recordings. Expiments in Fluids., 23:2028, 1997.
[132] J. Westerweel and F. Scarano. Universal outlier detectionfor piv data.
Expiments in Fluids, 39:10961100, 2005.
[133] J. R. Wheatley, T. C. Amis, and L. A. Engel. Nasal and oral airway
pressure-flow relationships. J Appl Physiol, 71:23172324, 1991.
168

[134] F. M. White. Fluid Mechanics. Marcel Dekker, New York, 2003.

[135] C. Willert, M. Raffel, J Kompenhans, B. Stasicki, and C. Khler. Recent applications of particle image velocimetry in aerodynamic research. Flow Meas. Instrum., 7:247256, 1997.
[136] C. E. Willert and M. Gharib. Digital particle image velocimetry. Exp.
Fluids, 10:181193, 1991.
[137] Z. Yang and T. H. Shih. a k- model for turbulence and transitionary
boundary layer, In: Near-wall turbulent flows. Elsevier-Science Publishers B. V., 1993.
[138] Y. Zhang and W. H. Finlay. Experimental measurements of particle
deposition in three proximal lung bifurcation models with an idealized
mouth-throat. Journal of Aerosol Medicine, 18:460473, 2005.
[139] Y. Zhang, W. H. Finlay, and E. A. Matida. Particle deposition
measurements and numerical simulation in a highly idealized mouththroat. Journal of Aerosol Science, 35:789803, 2004.
[140] Z. Zhang and C. Kleinstreuer. Effect of particle inlet distributions on
deposition in a triple bifuration lung airway model. Journal of Aerosol
Medicine, 14:1329, 2001.
[141] Z. Zhang and C. Kleinstreuer. Transient airflow structures and particle transport in a sequentially branching lung airway model. Physics
of Fluids, 14:862880, 2002.
[142] Z. Zhang and C. Kleinstreuer. Modeling of low reynolds number
turbulent flows in locally constricted conduits: A comparison study.
AIAA journal, 41:831840, 2003.
[143] Z. Zhang and C. Kleinstreuer. Species heat and mass transfer in a
human upper airway model. International Journal of Heat and Mass
Transfer, 46:47554768, 2003.
[144] Z. Zhang, C. Kleinstreuer, J.F. Donahue, and C. S. Kim. Comparison
of micro- and nano-size particle depositions in a human upper airway
model. Journal of Aerosol Science, 36:211233, 2005.
169

[145] Z. Zhang, C. Kleinstreuer, and C. S. Kim. Micro-particle transport

and deposition in a human oral airway model. Journal of Aerosol
Science, 33:16351652, 2002.
[146] Y. Zhao, J. H. Citriniti, and B. B. Lieber. Flow characteristics in a
symmetric bifurcation. Adv. Bioengineering, ASME BED-22:489492,
1992.

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)

articles in scientific journals with an international referee system(2nd author)

Fluid Flow and Particle Deposition Analysis in a realistic Extrathoracic Airway Model Using Unstructured Grids, Journal of
Aerosol Science, Volume 38, Issue 5, pp 494-508 , 2007, Santhosh Tovinakere Jayaraju, Mark Brouns, Sylvia Verbanck, Christian Lacor
Flow characterization using a laser Doppler vibrometer, Optics and
171

Lasers in Engineering, pp 19-26, Vol 45 (1) 2007, 2007, Joris Vanherzeele,

Mark Brouns, Patrick Guillaume, Steve Vanlanduit

communications at international congresses

integrally published in proceedings or only available as an abstract
PIV on the flow of a simplified upper airway model, Proceedings of
the 13th International Symposium on Applications of Laser Techniques to
Fluid Mechanics, 2006, Mark Brouns, Sylvia Verbanck, Jeroen Van Beeck,
Steve Vanlanduit, Joris Vanherzeele, Christian Lacor
Reynolds-averaged Navier-Stokes simulations of the airflow in the
extra thoracic airways, Proceedings ECCOMAS European Congress on
Computational Methods in Applied Sciences and Engineering, 24-28/07/04,
Jyvskyl, Finland, 2004, Mark Brouns, Christian Lacor, Sylvia Verbanck
Effect of Tracheal Stenosis on Local Pressure Drop, American Thoracic Society Conference, San Francisco, , 2007, Mark Brouns, Santhosh
Tovinakere Jayaraju, Christian Lacor, Johan De Mey, Marc Noppen, Walter Vincken, Sylvia Verbanck
Comparison of Navier-Stokes Simulations of the Airflow in the
Upper Airways in an Idealized and a Realistic Geometry, Proceedings EUROMECH Colloquium No. 456, 2004, Mark Brouns, Christian
Lacor, Sylvia Verbanck
Numerical study of aerosol flows in the upper airways, Proceedings ERCOFTAC BPC Seminar 2003, VUB, 2003, Mark Brouns, Christian
Lacor, Sami Hawash

172