Modeling Air and Particle Transport in the Human

Upper and Tracheobronchial Airways using RANS

and LES
by
Vivek Agnihotri
MTech, Mechanical Engineering, IIT Kanpur, 2007

Thesis submitted in fulfillment of the requirements for the award of the

degree of Doctor in de Ingenieurswetenschappen (Doctor in Engineering)
At
Vrije Universiteit Brussel, February 2014

Promoter: Prof. Dr. Ir. Chris Lacor

Prof. Dr. Ir. Ghader Ghorbaniasl
Prof. Dr. Sylvia Verbanck

Print: Silhouet, Maldegem

2014 Vivek Agnihotri

2014 Uitgeverij VUBPRESS Brussels University Press

VUBPRESS is an imprint of ASP nv (Academic and Scientific Publishers nv)
Ravensteingalerij 28
B-1000 Brussels
Tel. +32 (0)2 289 26 50
Fax +32 (0)2 289 26 59
E-mail: info@vubpress.be
www.vubpress.be

ISBN 978 90 5718 030 9

NUR 950
Legal deposit D/2014/11.161/034

All rights reserved. No parts of this book may be reproduced or transmitted in any form
or by any means, electronic, mechanical, photocopying, recording, or otherwise, without
the prior written permission of the author.

Abstract
Attempts are being made from several researchers to understand aerosol transport inside
the human respiratory system. The basic goal of these studies is acquire an in-depth
knowledge of flow-particle transport with the ultimate aim to access local and total
particle deposition. The knowledge of which can then be used to make improved inhaler
designs or evaluate toxicological impact of inhaling toxic matter. The key to this
understanding depends mainly on three factors: human airway geometry, material
characteristics of inhaled aerosols (i.e. shape, size and mass) and knowledge of local flow
structures of carrier gas. Evaluation of aerosol deposition characteristics relies on
experiments and Computational Fluid Dynamics (CFD). The human airway geometry
varies significantly from person to person and as such realization of experiments is a time
consuming, and costly affair. Additionally, the inherent intricacies involved with the
airway geometry may render experiments unfeasible. It is to this end that CFD is a boon
to many biological studies. This work is concerned with the understanding of aerosol
deposition behavior in human airways using CFD.
The main goal of the present PhD research is to evaluate the existing flow-particle
modeling methods and propose more efficient CFD techniques in particulate flows.
The work described in this dissertation considers a sedentary breathing rate of 30 l/min
and 60 l/min corresponding to light activity. The particle sizes range from 2 m to 10
m. Two airway geometries, representing extrathoracic airway (consisting of oral,
pharynx, larynx and throat) and a 5-generation intrathoracic airway, are used for the
current research. The Eulerian or fluid part is solved by employing Reynolds Averaged
Navier-Stokes equations (RANS) and using Large Eddy Simulation (LES). Moreover, the
particle phase is represented in the Lagrangian frame and calculated by numerically
integrating the particle equation of motion.
iii

The main highlights of the current research are:

(i) Introduction of a Helicity based Eddy Interaction Model (HEIM) to account the effect
of turbulence on particles in the framework of RANS.
(ii) Implementation and validation of a new Rotational based Smagorinsky Model
(RoSM) developed at VUB, as a subgrid scale model for LES.
(iii) Introduction of an efficient multiple LES frozen field approach based on the proper
orthogonal decomposition method, for particle simulation.

iv

Acknowledgement
Foremost, I would like to express my deepest acknowledgement to my two supervisors,
Prof. Dr. Ir. Chris Lacor and Prof. Dr. Ir. Ghader Ghorbaniasl. Their patience, wisdom
and constant encouragement were the key driving force behind this PhD research.
I would like to express my deepest gratitude and respect to Prof. Dr. Ir. Ghader
Ghorbaniasl who has been my advisor and the guiding beacon through four years of my
PhD research. I deeply owe to him for my intellectual gains and knowledge I learnt
during these years.
I would also like to thank my co-promoter Dr. Sylvia Verbanck for making me
understand the physiological aspects of the research project and giving vital suggestions
during the meetings.
I owe special gratitude for our system administrator Alain Wery for his everlasting
patience and support from day one of my VUB. Without his support this research
wouldnt have completed.
Special thanks go to our secretary Jenny Dhaes for helping me out in various
administrative tasks.
I would also like to thanks my colleagues, Leonidas Siozos-Rousoulis, Anna Sunol
Jimenez, Joo Duarte Miranda, Dinesh Kumar Verma, Khairy Elsayed for providing me
a good office environment with their good mood and humor. I would also thank my
former colleagues Matteo Parsani, Willem Diconinck, and Patryk Widera for making my
first two years at VUB truly fun and enjoyable. I have shared many laughs with them.
I would like to thank my family, specially mom, dad, brother and sister in law for their
constant love, motivation and unlimited support throughout my education. My gratitude

also goes to my parents in law for their support and their unconditional belief in me.
Special word of affection goes to my nephew Akshat and my brother in law Pranjal.
Finally, I cannot end without thanking my wife Prachi. You came into my life at the most
important time of my PhD, my final year. I thank you for giving me your unending
support, motivation and the belief that success will eventually follow despite having the
rough time of 2013. Without your support and love this work couldnt have completed.
You bring the light to my life.

vi

Jury Members
President

Prof. Johan Deconinck

Vrije Universiteit Brussel

Vice-President

Prof. Rik Pintelon

Vrije Universiteit Brussel

Secretary

Prof. Steve Vanlanduit

Vrije Universiteit Brussel

External Members

Prof. Wolfgang Schroeder

RWTH AAchen University
Prof. Jan Vierendeels
University of Ghent

Advisors

Prof. Chris Lacor

Prof. Ghader Ghorbaniasl
Prof. Sylvia Verbanck

vii

Contents
CHAPTER 1 ........................................................................................................................................... 1
Human respiratory system and its interaction with aerosols
1.1 Introduction............................................................................................................................... 1
1.2 Anatomy of human respiratory system ........................................................................................ 2
1.2.1 Extra-thoracic region.......................................................................................................................... 3
1.2.2 Tracheobronchial region..................................................................................................................... 5
1.2.3 Alveolar region .................................................................................................................................. 6

1.3 Aerosols .................................................................................................................................... 6

1.4 Aerosol particle dynamics .......................................................................................................... 7
1.5 Aerosol deposition mechanism ................................................................................................... 9
1.6 Factors affecting aerosol particle deposition ............................................................................ 12
1.7 Outline of the thesis ................................................................................................................. 13

CHAPTER 2 ......................................................................................................................................... 15
Literature survey
2.1 Airway modelisation ................................................................................................................ 15
2.1.1 Extrathoracic airway geometry ......................................................................................................... 16
2.1.2 Tracheobronchial geometry .............................................................................................................. 19

2.2 Modeling methods for aerosols transport ................................................................................. 24

2.2.1 Modeling fluid phase........................................................................................................................ 24
2.2.2 Modeling particle phase ................................................................................................................... 26

CHAPTER 3 ......................................................................................................................................... 29
Governing equations
3.1 Introduction............................................................................................................................. 29
3.2 Importance of turbulence ......................................................................................................... 30
3.3 Incompressible Navier-Stokes equation .................................................................................... 31
3.4 Modeling turbulence ................................................................................................................ 32
3.4.1 Reynolds averaged Navier-Stokes equation....................................................................................... 33
3.4.2 Two equation SST

EVM ..................................................................................................... 35

viii

3.4.3 Large eddy simulation ...................................................................................................................... 39

3.4.4 Smagorinsky model.......................................................................................................................... 41

3.5 Modeling particle phase........................................................................................................... 45

3.5.1 Modeling assumptions...................................................................................................................... 46
3.5.2 Eulerian approach ............................................................................................................................ 47
3.5.3 Lagrangian approach ........................................................................................................................ 49
3.5.4 Eddy interaction model .................................................................................................................... 52

CHAPTER 4 ......................................................................................................................................... 55
Particle deposition in an extrathoracic airway using RANS
4.1 Introduction............................................................................................................................. 55
4.2 Mathematical background........................................................................................................ 57
4.2.1 EIM and correction functions ........................................................................................................... 59
4.2.2 Isotropic EIM .................................................................................................................................. 59
4.2.3 Wang and James EIM ...................................................................................................................... 59
4.2.4 Helicity EIM .................................................................................................................................... 60

4.3 Results and discussion ............................................................................................................. 63

4.3.1 Test geometry 1, 90 degree bend, Re=10000 ..................................................................................... 63
4.3.2 Test geometry 2, Simplified human upper airway model ................................................................... 66

4.4 Conclusions ............................................................................................................................. 72

CHAPTER 5 ......................................................................................................................................... 75
Performance of helcity eddy interaction model
5.1 Introduction............................................................................................................................. 75
5.2 Simulation condition ................................................................................................................ 76
5.3 Results and discussion ............................................................................................................. 77
5.3.1 Inspiration ....................................................................................................................................... 77
5.3.2 Expiration ........................................................................................................................................ 79
5.3.3 Influence of inlet conditions and flow field data ................................................................................ 85

5.4 Conclusion .............................................................................................................................. 95

ix

CHAPTER 6 ......................................................................................................................................... 99
Rotational based Smagorinsky model
6.1 Introduction............................................................................................................................. 99
6.2 Description of RoSM model.................................................................................................... 101
6.3 Results and discussion ........................................................................................................... 107
6.3.1 Fully developed turbulent channel flow .......................................................................................... 108
6.3.2 Fully developed turbulent flow in a square duct .............................................................................. 113
6.3.3 Flow past a circular cylinder ........................................................................................................... 117
6.3.4 Application to UAM ...................................................................................................................... 122

6.4 Conclusions ........................................................................................................................... 126

CHAPTER 7 ....................................................................................................................................... 129

Particle deposition in an extrathoracic airway using LES
7.1 Introduction........................................................................................................................... 129
7.2 Multiple LES frozen field approach ........................................................................................ 130
7.3 Mathematical background...................................................................................................... 132
7.3.1 Discrete POD method..................................................................................................................... 133
7.3.2 Procedure to derive the optimal set of frozen fields ......................................................................... 136

7.4 Description of model geometry and sample data sets .............................................................. 139
7.5 Results and Condition ............................................................................................................ 140
7.5.1 Step 1, Evaluation of sampling period ............................................................................................. 140
7.5.2 Step 2, Evaluation of time interval .................................................................................................. 144

7.6 Particle deposition results...................................................................................................... 147

7.6.1 Effect of SGS models ..................................................................................................................... 148
7.6.2 Accounting SGS motion................................................................................................................. 150
7.6.3 Particle deposition comparison between RANS and LES ................................................................. 154

7.7 Conclusion ............................................................................................................................ 156

CHAPTER 8 ....................................................................................................................................... 159

Particle deposition in a 5 generation intrathoracic airway using LES
8.1 Introduction........................................................................................................................... 159
8.2 Upper airway geometry.......................................................................................................... 160
8.3 Simulation condition .............................................................................................................. 161
8.4 Results and discussion ........................................................................................................... 163
8.4.1 Mesh convergence study ................................................................................................................ 163

8.4.2 Flow dynamics............................................................................................................................... 165

8.4.3 Optimal set determination .............................................................................................................. 168
8.4.4 Particle deposition results ............................................................................................................... 171

8.5 Conclusions ........................................................................................................................... 175

CHAPTER 9 ....................................................................................................................................... 177

Conclusion and furture work
9.1 RANS..................................................................................................................................... 177
9.2 LES ....................................................................................................................................... 178
9.3 Future work ........................................................................................................................... 179
9.3.1 Boundary condition ........................................................................................................................ 179
9.3.2 Detailed comparison of particle data in present tracheobronchial geometry ...................................... 180
9.3.3 Moving Larynx .............................................................................................................................. 181
9.3.4 Uncertainty quantification .............................................................................................................. 181

LIST OF PUBLICATIONS ................................................................................................................ 183

BIBLIOGRAPHY............................................................................................................................... 185

xi

Nomenclature
N-S

Navier-Stokes

DNS

Direct Numerical Simulation

RANS

Reynolds Averaged Navier-Stokes

LES

Large Eddy Simulation

SGS

Subgrid Scale

DSM

Dynamic Smagorinsky model

RoSM

Rotational based Smagorinsky model

EVM

Eddy Viscosity Model

SST

Shear Stress Transport

EIM

Eddy Interaction Model

HEIM

Helicity Eddy Interaction Model

POD

Proper Orthogonal Decomposition

UAM

Upper airway Model

PDI

Phase Doppler Interferometry

CT

Computer tomography

Symbols

Re

Reynolds number, dimensionless

xii

Pr

Prandtl number, dimensionless

Fr

Froude number, dimensionless

Stk

Stokes number, dimensionless

Re p

Particle Reynolds number

Re t

Turbulent Reynolds number

Sct

Turbulent Schmidt number

Fluid density, kg/m3

Fluid dynamic viscosity, kg/m-s

Fluid kinematic viscosity, m2/s

Turbulent eddy viscosity, m2/s

Particle density, kg/m3

Cd

Particle drag coefficient

ui

Velocity component, m/s

Velocity vector, m/s

Gravity vector, m/s2

up

Particle velocity vector, m/s

position vector, m

xp

Particle position vector, m

xiii

Force vector, N

Turbulent kinetic energy, (m/s)2

Dissipation

Specific dissipation

Particle relaxation time, s

tint

Particle interaction time

tcross

Particle crossing through eddy, s

le

eddy length scale, m

time step, s

filter width

Sij

strain rate tensor, s-1

ij

stress tensor, N/m2

absolute value

time average

Cs

Smagorisnky constant

xiv

Figures
FIGURE 1.1, HUMAN RESPIRATORY TRACT [1] ...................................................................................................... 3
FIGURE 1.2, SCHEMATIC OF EXTRATHORACIC REGION REPRESENTING DIFFERENT REGIONS ................................................ 3
FIGURE 1.3, DIFFERENT TYPES OF INHALERS ......................................................................................................... 7
FIGURE 1.4, DEPOSITION DUE TO INERTIAL IMPACTION ........................................................................................... 9
FIGURE 1.5, DEPOSITION DUE TO SEDIMENTATION .............................................................................................. 10
FIGURE 1.6, DEPOSITION DUE TO BROWNIAN MOTION ......................................................................................... 11
FIGURE 1.7, DEPOSITION PROBABILITY WITH RESPECT TO PARTICLE SIZE, [7]............................................................... 13
FIGURE 2.1, SIMPLIFIED LARYNX GEOMETRY BY KATZ AND MARTONEN [13] .............................................................. 16
FIGURE 2.2, LARYNX GEOMETRY USED BY CORCORAN AND CHIGIER [14] .................................................................. 17
FIGURE 2.3, EXTRATHORACIC AIRWAY DEVELOPED BY KLEINSTREUER AND ZHANG [15] ................................................. 18
FIGURE 2.4, EXTRATHORACIC AIRWAY DEVELOPED BY [20] .................................................................................... 18
FIGURE 2.5, (LEFT) REALISTIC GEOMETRY; (B) SIMPLIFIED EXTRATHORACIC GEOMETRY OF VUB. ..................................... 19
FIGURE 2.6, TRACHEOBRONCHIAL GEOMETRIES BASED ON WEIBEL A MODEL, (LEFT) ZHANG AND KLEINSTREUER [26], (MIDDLE)
LONGEST AND VINCHURKAR [27], (LEFT) VASCONCELOSET ET AL. [28] ............................................................ 21
FIGURE 2.7, TRACHEOBRONCHIAL GEOMETRY GENERATED BY VAN ERTBRUGGEN ET AL. [11] ......................................... 22
FIGURE 2.8, DOUBLE BIFURCATION GENERATION GENERATED BY HOLBROOK ET AL. [29]............................................... 23
FIGURE 2.9, CT-SCAN BASED UPPER RESPIRATORY AIRWAY GEOMETRY USED BY LIN ET AL. [30], (LEFT) EXTRATHORACIC REGION
DEPICTING VARIOUS REGIONS, (RIGHT) TRACHEOBRONCHIAL REGION ............................................................... 23

FIGURE 3.1, FLOW STRUCTURES INSIDE UAM AT 60 L/MIN ................................................................................... 31

FIGURE 3.2, CHARATERISTIC MAP FOR PARTICLE-FLOW COUPLING, [78] ................................................................... 47
FIGURE 4.1, 90 DEGREE TEST BEND. (LEFT) GEOMETRY DETAILS; (RIGHT) MESH CUT PLANE AT MIDDLE OF THE BEND. ........... 64
FIGURE 4.2, PARTICLE PROFILE AT THE INLET ...................................................................................................... 65
FIGURE 4.3, TOTAL DEPOSITION EFFICIENCY. (CONNECTED SYMBOLS) - SIMULATIONS; ( DISCONNECTED SYMBOLS)
EXPERIMENTS; (LEFT) EXPLODED VIEW ..................................................................................................... 65

FIGURE 4.4, SIMPLIFIED UAM MODEL AND RESPECTIVE COMPARTMENTS USED IN UAM .............................................. 66
FIGURE 4.5, MESH AT DIFFERENT SECTION PLANES IN UAM .................................................................................. 67
FIGURE 4.6, COMPARISON OF NORMALIZED VELOCITY COMPONENT FOR DIFFERENT MESH COUNTS. SECTION (A), (B), (C) AND
(D) CORRESPONDS TO FIVE MM ABOVE EPIGLOTTIS AND ONE, TWO AND THREE TRACHEAL DIAMETERS DOWNSTREAM OF
GLOTTIS, RESPECTIVELY ........................................................................................................................ 68

xv

FIGURE 4.7, INSPIRATORY TOTAL DEPOSITION EFFICIENCIES, (LEFT) 30 L/MIN AND (RIGHT) 60 L/MIN; PARTICLE SIZES RANGE
FROM 2 M-10 M; (CONNECTED SYMBOLS)-SIMULATIONS; (DISCONNECTED SYMBOLS)-EXPERIMENTS ................... 70

FIGURE 4.8, INSPIRATORY RELATIVE DEPOSITION EFFICIENCIES, (UPPER) 30 L/MIN AND (LOWER) 60 L/MIN; (LEFT) 3M AND
(RIGHT) 6M; (CONNECTED SYMBOLS) SIMULATIONS; (DISCONNECTED SOLID SQUARES,-EXPERIMENTS) ................. 71
FIGURE 5.1, SCHEMATIC OF AEROSOL DEPOSITION EXPERIMENTS, VERBANCK ET AL. [45].............................................. 77
FIGURE 5.2, INSPIRATORY TOTAL DEPOSITION EFFICIENCIES, (LEFT) 30 L/MIN AND (RIGHT) 60 L/MIN; PARTICLE SIZES RANGE
FROM 2M-10M; (CONNECTED SYMBOLS) SIMULATIONS; (DISCONNECTED SYMBOLS)-EXPERIMENTS ................... 78

FIGURE 5.3, INSPIRATORY RELATIVE DEPOSITION EFFICIENCIES, (UPPER) 30 L/MIN AND (LOWER) 60 L/MIN; (LEFT) 3M AND
(RIGHT) 6M; (CONNECTED SYMBOLS) SIMULATIONS; (DISCONNECTED SOLID SQUARES,-EXPERIMENTS) ................. 79
FIGURE 5.4, INSPIRATORY RELATIVE DEPOSITION EFFICIENCIES, (LEFT) 30 L/MIN AND (RIGHT) 60 L/MIN; ALL PANELS TOP TO
BOTTOM CORRESPOND TO 2, 4, 8AND 10 M ........................................................................................... 81

FIGURE 5.5, EXPIRATORY TOTAL DEPOSITION EFFICIENCIES, (LEFT) 30 L/MIN AND (RIGHT) 60 L/MIN; PARTICLE SIZES RANGE
FROM 2M-10M; (CONNECTED SYMBOLS)SIMULATIONS; (DISCONNECTED SQUARE SYMBOLS)-EXPERIMENTS ......... 82

FIGURE 5.6, EXPIRATORY RELATIVE DEPOSITION EFFICIENCIES, (UPPER) 30 L/MIN AND (LOWER) 60 L/MIN; (LEFT) 3M AND
(RIGHT) 6M; (CONNECTED SYMBOLS)SIMULATIONS; (DISCONNECTED SQUARE SYMBOLS)-EXPERIMENTS ................ 83
FIGURE 5.7, EXPIRATORY RELATIVE DEPOSITION EFFICIENCIES, (LEFT) 30 L/MIN AND (RIGHT) 60 L/MIN; ALL PANELS TOP TO
BOTTOM CORRESPOND TO 2, 4, 8AND 10 M; (CONNECTED SYMBOLS) SIMULATIONS ........................................ 84

FIGURE 5.8, (LEFT) UAM WITH CONNECTOR TUBING (SHOWN IN SHADED AREA) AT THE INLET FOR INSPIRATORY BREATHING
0

PHASE SETUP; (RIGHT) UAM WITH 90 ELBOW BEND CONNECTOR (SHOWN IN SHADED AREA) AT THE INLET FOR
6

EXPIRATORY BREATHING PHASE SETUP. THE ADDITIONAL MESH FOR THE CONNECTOR TUBINGS ARE (LEFT) 0.4 X 10 AND

(RIGHT) 0.8 *106 HEXAHEDRAL CELLS. .................................................................................................... 86

FIGURE 5.9, (TOP) TOTAL DEPOSITION EFFICIENCY FOR PARTICLE SIZES RANGING FROM 2-10M AT 60 L/MIN, (BOT-TOM)
RELATIVE DEPOSITION EFFICIENCY FOR 6M AT 60 L/MIN. (LEFT) INSPIRATION, (RIGHT) EXPIRATION. ....................... 87

FIGURE 5.10, IMPACT OF TOP HAT AND PARABOLIC INSPIRATORY VELOCITY PROFILES IMPOSED EITHER DIRECTLY AT UAM INLET
(UPPER AND MIDDLE PANELS) OR VIA CONNECTOR TUBING (BOTTOM PANEL). VELOCITY MAGNITUDE CONTOURS, FLOW
STREAMLINES IN THE CENTRAL SAGITTAL PLANE, AND PARTICLE DISTRIBUTION PROFILES AT DIFFERENT UAM STATIONS ARE
SHOWN FOR 60 L/MIN

........................................................................................................................ 89

FIGURE 5.11, IMPACT OF EXPIRATORY VELOCITY PROFILES: (LEFT AND MIDDLE PANEL) ARE NAMELY TOP-HAT AND PARABOLIC
WITHOUT CONNECTOR TUBING; (RIGHT PANEL) TOP-HAT WITH CONNECTOR TUBING. VELOCITY MAGNITUDE CONTOURS,
STREAMLINES IN THE CENTRAL SAGITTAL PLANE, AND PARTICLE DISTRIBUTION PROFILES AT DIFFERENT UAM STATIONS ARE
SHOWN FOR 60 L/MIN.

....................................................................................................................... 91

FIGURE 5.12, TOTAL DEPOSITION EFFICIENCY; (LEFT PANEL) INSPIRATION; (RIGHT PANEL) EXPIRATION FOR 60 L/MIN .......... 92
FIGURE 5.13, RELATIVE DEPOSITION EFFICIENCIES,(TOP) 8M, (LOWER) 10M; (LEFT) INSPIRATION, (RIGHT) EXPIRATION .. 94

xvi

FIGURE 5.14, DEPENDENCE OF HELICITY EIM ON THE TURBULENCE MODEL; (LEFT) INSPIRATION, (RIGHT) EXPIRATION FOR
60L/MIN ......................................................................................................................................... 94
FIGURE 6.1, (A) RESOLVED AND UNRESOLVED EDDIES IN TURBULENT FLOW FIELD. (B) TRANSLATIONAL VELOCITY AND ROTATION
RATE COMPONENTS IN X, Y AND Z DIRECTIONS.......................................................................................... 104

FIGURE 6.2, DIMENSIONS OF CHANNEL FLOW TEST CASE ..................................................................................... 109

FIGURE 6.3, (A) THE CALCULATED MEAN STREAMWISE VELOCITY PROFILE COMPARED WITH THE DYNAMIC SMAGORINSKY, AND
DNS. (B) THE RELATED ABSOLUTE ERROR FIELD. (C) THE MODEL COEFFICIENT PROFILES. .................................... 111
FIGURE 6.4, (A) THE CALCULATED XY-COMPONENT OF REYNOLDS STRESS TENSOR COMPARED WITH THE DYNAMIC
SMAGORINSKY, NO MODEL LES AND DNS. (B) THE RELATED ABSOLUTE ERROR FIELD....................................... 111
FIGURE 6.5, (LEFT) THE CALCULATED DEVIATORIC DIAGONAL REYNOLDS STRESSES COMPARED WITH THE DYNAMIC
SMAGORINSKY MODEL AND DNS. (RIGHT) THE RELATED ABSOLUTE ERROR FIELD ............................................. 112
FIGURE 6.6, SCHEMATIC OF THE DUCT GEOMETRY AND THE COORDINATE SYSTEM IS SHOWN ........................................ 113
FIGURE 6.7, THE CALCULATED MEAN STREAMWISE VELOCITY COMPARED WITH THE DYNAMIC SMAGORINSKY MODEL, NO
MODEL , AND DNS. .......................................................................................................................... 114

FIGURE 6.8, THE CALCULATED RMS VALUES OF THE DIAGONAL REYNOLDS STRESSES COMPARED WITH THE DYNAMIC
SMAGORINSKY, NO

MODEL , LES OF MADABHUSHI AND VANKA AND DNS ................................................... 116

FIGURE 6.9, (A) THE CALCULATED XY-COMPONENT OF REYNOLDS STRESS TENSOR COMPARED WITH THE DYNAMIC SMAGORINSKY
MODEL , NO MODEL AND DNS. (B) THE MODEL COEFFICIENT PROFILES.......................................................... 116

FIGURE 6.10, GRID IN THE X, Y-PLANE ............................................................................................................ 117

FIGURE 6.11, THE CALCULATED MEAN STREAMWISE VELOCITY COMPARED THE DYNAMIC SMAGORINSKY, NO MODEL, LES DATA
[111], DNS [114] AND EXPERIMENTS [115]. (A) x / D 1.06 , (B) x / D 1.54 . ............................... 118
FIGURE 6.12, THE CALCULATED MEAN VERTICAL VELOCITY COMPARED WITH THE DYNAMIC SMAGORINSKY, NO MODEL, LES
DATA [111] DNS [114] AND EXPERIMENTS [115]. (A)

x / D 1.06 , (B) x / D 1.54 .......................... 119

FIGURE 6.13, THE CALCULATED STREAMWISE TURBULENT INTENSITY COMPARED WITH THE DYNAMIC SMAGORINSKY MODEL,
NO MODEL, LES DATA [111], DNS [114] AND EXPERIMENTS [115]. (A) x / D 1.06 , (B) x / D 1.54 . 120
FIGURE 6.14, THE CALCULATED CROSSWISE TURBULENT INTENSITY COMPARED WITH THE DYNAMIC SMAGORINSKY MODEL, NO
MODEL , LES DATA [111], DNS [114] AND EXPERIMENTS [115]. (A)

x / D 1.06 , (B) x / D 1.54 ....... 120

FIGURE 6.15, THE CALCULATED XY-COMPONENT OF THE REYNOLDS STRESS TENSOR COMPARED WITH THE DYNAMIC
SMAGORINSKY MODEL, NO MODEL, LES DATA [111], DNS [114] AND EXPERIMENTS [115]. (A) x / D 1.06 , (B)

x / D 1.54 . ............................................................................................................................ 121

FIGURE 6.16, LOCATION OF SECTION LINES CORRESPONDING TO SAGITTAL PLANE...................................................... 124

xvii

FIGURE 6.17, NORMALIZED TWO COMPONENT ( ux

AND

uz ) VELOCITY MAGNITUDE CORRESPONDING TO CENTRAL SAGITTAL

PLANE, (A), (B), AND (C) ARE ONE, TWO AND THREE TRACHEAL DIAMETER DOWNSTREAM OF LARYNX, RESPECTIVELY; ( D)

FIVE MM ABOVE EPIGLOTTIS. ............................................................................................................... 124

FIGURE 6.18, NORMALIZED TIME AVERAGED 2 COMPONENT VELOCITY MAGNITUDE; (A) PIV, (B) RANS; (C) ROSM; (D) DSM
................................................................................................................................................... 125
'2

FIGURE 6.19, NORMALIZED TWO COMPONENT ( ux

AND

uz'2 ) KINETIC ENERGY CORRESPONDING TO CENTRAL SAGITTAL

PLANE, (A), (B), AND (C) ARE ONE, TWO AND THREE TRACHEAL DIAMETER DOWNSTREAM OF LARYNX, RESPECTIVELY; ( D)
FIVE MM ABOVE EPIGLOTTIS. ............................................................................................................... 126

FIGURE 7.1, SCHEMATIC OF PARTICLE CALCULATION PROCEDURE, A) DYNAMIC APPROACH, B) MULTIPLE LES FROZEN FIELD
APPROACH ...................................................................................................................................... 132

FIGURE 7.2, UAM MODEL GEOMETRY SHOWING LOCATION OF PLANES WITH ONE SAGITTAL PLANE AND FIVE PERPENDICULAR
PLANES NUMBERED 1-5 ..................................................................................................................... 140

FIGURE 7.3, RELATIVE INFORMATION CONTENT (RIC) FOR VARIOUS SAMPLE SETS DETAILED IN TABLE 7.1; CONSIDERING (A)
CASE 1 SAGITTAL PLANE; (B) CASE 2 SAGITTAL AND FIVE PERPENDICULAR PLANES (SEE FIGURE 7.2)...................... 141
FIGURE 7.4, CONTOURS AND PROFILES OF AVERAGE VELOCITY MAGNITUDE AT SAGITTAL PLANE FOR SELECTED SETS OF TABLE 1,
(A) VELOCITY MAGNITUDE CONTOUR; (B), (C) AND (D) ARE THE VELOCITY PROFILES AT SECTION 1-1 ..................... 143
FIGURE 7.5, AUTO CORRELATION INDEX; (A, D) X-VELOCITY, (B, E) Y-VELOCITY, (C, F) Z-VELOCITY RESPECTIVELY; (TOP) CASE 1,
(BOTTOM) CASE 2 ............................................................................................................................ 145
FIGURE 7.6, CONTOURS AND PROFILES OF AVERAGE VELOCITY MAGNITUDE AT SAGITTAL PLANE FOR SETS DETAILED IN TABLE 3,
(A) VELOCITY MAGNITUDE CONTOUR; (B), (C) AND (D) ARE THE VELOCITY PROFILES AT SECTION 1-1, 2-2 AND 3-3
RESPECTIVELY. ................................................................................................................................. 146

FIGURE 7.7, PARTICLE DEPOSITION EFFICIENCIES FOR SET I, II AND III; (A)TOTAL DEPOSITION EFFICIENCY, (B)-(C) RELATIVE
DEPOSITION EFFICIENCY FOR 3M AND 6M ............................................................................................ 148

FIGURE 7.8, PARTICLE DEPOSITION EFFICIENCIES COMPARISON ; (A) TOTAL DEPOSITION EFFICIENCY, (B)-(C) RELATIVE
DEPOSITION EFFICIENCY FOR 3M AND 6M ............................................................................................ 149

FIGURE 7.9, CONTOURS AND PROFILES OF AVERAGE VELOCITY MAGNITUDE AT SAGITTAL PLANE, (A) VELOCITY MAGNITUDE
CONTOUR; (B), (C) AND (D) ARE THE VELOCITY PROFILES AT SECTION 1-1, 2-2 AND 3-3 RESPECTIVELY ............... 150
FIGURE 7.10, PARTICLE DEPOSITION EFFICIENCIES COMPARISON ; (A)-(B)TOTAL DEPOSITION EFFICIENCY, (C)-(D) RELATIVE
DEPOSITION EFFICIENCY FOR 3M, (E)-(F) RELATIVE DEPOSITION EFFICIENCY FOR 6M ....................................... 154

FIGURE 7.11, COMPARISON OF k res (AVERAGED KINETIC ENERGY OF THE RESOLVED FIELD),

ksgs (AVERRAGED SGS

KINETIC ENERGY) AND k rms (KINETIC ENERGY CONTAINED IN THE FLUCTUATION), CORRESPONDING TO REFERENCE DATA
SET;

DENOTES ENSEMBLE AVERAGE AND RMS DENOTES ROOM MEAN SQUARE FLUCTUATION ......................... 154

xviii

FIGURE 7.12, COMPARISON OF PARTICLE DEPOSITION BETWEEN RANS AND LES AT INSPIRATORY FLOW RATE OF 60 L/MIN; (A)
TOTAL DEPOSITION EFFICIENCY, (B)-(C) RELATIVE DEPOSITION EFFICIENCY CORRESPONDING TO 3M AND 6 M ........ 155

FIGURE 8.1, UPPER AIRWAY GEOMETRY RECONSTRUCTED BY FUSING SCALED UAM MODEL WITH CT-SCAN DATA OF A FEMALE
ADULT. .......................................................................................................................................... 161

FIGURE 8.2, SAGITTAL PLANE WITH SECTION PLANES USED FOR MESH CONVERGENCE STUDY ........................................ 163
FIGURE 8.3, NON DIMENSIONAL MEAN VELOCITY MAGNITUDE AT SECTIONS A-F ........................................................ 164
FIGURE 8.4, CONTOURS OF MEAN VELOCITY MAGNITUDE (A) AND MEAN TURBULENT KINETIC ENERGY (B) ....................... 165
FIGURE 8.5, CONTOURS OF MEAN VELOCITY MAGNITUDE AND TURBULENT KINETIC ENERGY AT CENTRAL PLANE ................. 166
FIGURE 8.6, COMPARISON OF FLOW DISTRIBUTION AT OUTLETS AGAINST EXPERIMENTALLY MEASURED DATA AT 30 L/MIN .. 167
q

FIGURE 8.7, RELATIVE INFORMATION CONTENT RIC

i
i 1

k , WHERE M=1201 AND Q DENOTES VARIOUS

k 1

REDUCED SETS. THE RED LINE CORRESPONDS TO Q=1041 AND THE BLUE LINE CORRESPONDS TO THE REFERENCE DATA
BASE OF Q=1201 .............................................................................................................................
95%

FIGURE 8.8, N q

/ N q95%
1201

168

FOR VARIOUS SETS ........................................................................................... 169

FIGURE 8.9, AUTO CORRELATION INDEX; (A) X-VELOCITY, (B) Y-VELOCITY, (C) Z-VELOCITY ........................................... 170
FIGURE 8.10, ORAL DEPOSITION EFFICIENCY .................................................................................................... 172
FIGURE 8.11, TOTAL DEPOSITION EFFICIENCY ................................................................................................... 173
FIGURE 8.12, DEPOSITION EFFICIENCY AT GENERATION G1 ................................................................................. 173
FIGURE 8.13, LOBAR DEPOSITION EFFICIENCY: (A) 2 M, (B) 10 M ...................................................................... 174
FIGURE 9.1, LOBAR PARTICLE DEPOSITION VERSUS LOBAR FLOW DISTRIBUTION AT 60 L/MIN, PRELIMINARY EXPERIMENTAL
RESULTS. ........................................................................................................................................ 181

xix

iii

Chapter 1
Human respiratory system and its
interaction with aerosols
Contents
1.1 Introduction.............................................................................................................................................1
1.2 Anatomy of human respiratory system.................................................................................................2
1.2.1 Extra-thoracic region........................................................................................................................3
1.2.2 Tracheobronchial region..................................................................................................................5
1.2.3 Alveolar region..................................................................................................................................6
1.3 Aerosols....................................................................................................................................................6
1.4 Aerosol particle dynamics......................................................................................................................7
1.5 Aerosol deposition mechanism...............................................................................................................9
1.6 Factors affecting deposition mechanism.............................................................................................12

1.1 Introduction
The respiratory system present in mammals and other life forms, is a vital organ whose
primary function is to supply oxygen to the blood (needed by cells to function) and
remove carbon monoxide (by product of cells). This is achieved through the act of
inhaling air (termed as Inhalation) and exhaling air (termed as Exhalation). The whole
act/process is termed as respiration. Some amazing facts regarding human respiratory
system are listed as follows:

 Breathing rate is faster in newborn than compared to woman and comparatively

slower in men. A newborn up to 6 weeks can takes in 40-60 breaths per minute.
At rest an adult respire at a normal rate of 12-20 breaths per minute. This accounts
to approximately 10000 to 20000 liters of inhaled air per day.
The total lung capacity (maximum amount of air that someones lungs are capable
of holding) is between 4 and 6 liters of air in an adult. Males usually have higher
total lung capacities than females.
The total airways when stacked up will amount to about 2400 km.
This chapter briefly discusses lung physiology and discusses how aerosols are connected
to human respiratory system.

1.2 Anatomy of human respiratory system

Topologically, a human respiratory tract consists of a series of connecting pipes, with
pipes becoming progressively smaller and ending at alveoli where the exchange of
oxygen with blood takes place. Giving all the details of the respiratory tract is almost
beyond the scope of current work, therefore a brief schematic detailing important parts is
shown in Figure 1.1. The entire respiratory tract can be classified into three basic regions
(described briefly in below sections) namely, extrathoracic region, tracheo-bronchial
region and the alveolar region.

Figure 1.1, Human respiratory tract [1]

Figure 1.2, Schematic of extrathoracic region representing different regions

1.2.1 Extra-thoracic region

The extra-thoracic region represents those areas that start the beginning of respiratory
system. It consists of nasal cavity (or nose), oral (or mouth) cavity, pharynx, and larynx
(see Figure 1.2). The extra-thoracic region is sometimes also referred to as upper airway.
3

Nasal cavity
The nasal cavity forms the main entrance point for the outside air into the respiratory
tract. The nasal cavity represents a hollow space lined up with hairs and mucus. The
primary function is to condition the air before it is conducted deeper inside the respiratory
system. As the air passes through the nasal cavity it gets warm, moisturized and filtered
for the foreign particle. The hair and mucus helps to trap foreign contaminants present in
the air before it reaches deeper inside lungs. In addition, mucus also helps in moistening
the air.
Oral cavity
Oral cavity is the secondary opening to the respiratory tract. Normally breathing takes
place through nasal cavity, but the oral cavity can be used to supplement the breathing
such during cold or heavy exercises. The mouth however does not condition the air
because of the lack of hairs and mucus. Since the mouth passage is shorter than nasal and
also because of larger cross sectional area, more air reaches quickly in the lungs.
Pharynx
Pharynx also known as throat extends from the posterior end of the nasal cavity to the
superior end of the esophagus and larynx (see Figure 1.1). It consists of three regions,
namely nasopharynx, oropharynx, and laryngopharynx. It shapes like a funnel and
collects air coming from mouth and nose and passes it down towards trachea. The
epiglottis, which is a flap of elastic cartilage present at the opening of the trachea,
prevents swallowed material from getting entered into the trachea.
Larynx
Larynx known as the voice box is a small region in the respiratory tract that connects
laryngopharynx and trachea. It is composed of several cartilage structures. It consists of
special structures known as vocal chords which vibrate when one expires air. The vocal
folds are made up of mucous membrane that vibrates to produce vocal sounds. Humans
4

have the ability to control the tension and vibration speed of the vocal folds to make
sound.

1.2.2 Tracheobronchial region

Distal to extrathoracic region are the tracheobronchial airways. This is referred to as
lower airways. This region is consists of airways that are responsible for conducting air to
the oxygen exchange part of the lungs. The tracheobronchial region is formed of
components starting from trachea and ending with terminal bronchioles.
Trachea
Trachea so called as the wind pipe consists of many C-shaped cartilage rings. It measures
roughly 10-14 cm in length and 16-20 mm in diameter. It is passage which connects
larynx to the bronchi. Similar to nasal cavity its inner surface is covered with mucus to
further trap foreign particles for getting deeper inside lungs. The main function of the
trachea is to provide a clear airway for air to enter and exit the lungs.
Bronchi and Bronchioles
At the inferior side, trachea bifurcates into two cartilage-ringed airways known as main
bronchi. The left and right main bronchi run into each lung bifurcating further into
smaller secondary bronchi. The secondary bronchi carry air into the different lobes of the
lungs (2 in the left lung and 3 in the right lung). Within each lobe the secondary bronchi
further splits into smaller airways called as tertiary bronchi which inturn bifurcates into
even smaller airway tubes. These smaller airway tubes are called as bronchioles and
spread throughout the lungs. These bronchioles further splits into many smaller branches
less than a millimeter in diameter called terminal bronchioles. Finally, the millions of tiny
terminal bronchioles conduct air to the alveoli of the lungs. The main function of the
bronchi and bronchioles is to carry air from the trachea into the lungs.

1.2.3 Alveolar region

Distal to tracheobronchial region is the region whose primary function is to exchange
oxygen and carbon monoxide between blood and air. This region is known as alveolar
region and composed of cup like structures called as alveoli, attached at the end of
respiratory bronchioles (bifurcations of terminal bronchioles, Figure 1.1). A typical lung
contains approximately 30 million alveoli.

1.3 Aerosols
Suspension of fine solid particles or liquid droplets in a carrier fluid (liquid or gas) is
termed as Aerosols. Examples include, haze, dust, smoke, mist etc. In the context of
human respiratory system, the study on aerosols is required for, a) evaluating
toxicological impact of inhaling toxic matter on human lungs and b) making efficient
devices delivering therapeutic drugs to affected sites of the lungs.
I) Toxicological impact, Prolonged exposure to particulate matter present in air can have
severe ill effects on human health. It is now been well recognized that deposition of
particulate matter present in aerosol scan is linked to many lung related ailments such as
asthma, chronic obstructive pulmonary diseases (COPD) and lung cancer. The recent
report from World Health Organization (WHO) [2] concludes that indoor air pollution
resulted in nearly 2.5 million premature deaths. Urban outdoor air pollution is estimated
to have caused nearly 1.3 million deaths worldwide per year. The size of the particulate
matter present in the air is the determining factor where in lungs particles will deposit.
Particle sizes smaller than 2.5 m pose more threat than particles having size greater than
10 m, since former has the higher probability of getting much deeper inside the lungs.
II) Therapeutic consideration, Delivering therapeutic drug via respiratory tract is
considered the preferred method for treating COPD and asthma. The advantages include
delivery of medication directly to the site of action thereby resulting in faster onset of
action. Moreover, one dose of inhaled medication contains less medication than a tablet
6

yet delivers the same effect. This mode of treatment achieved by devices called as
inhalers. Based on the working mechanism, there are three principle types of inhalers
available in the market, i) pressurized metered dose inhalers (pMDI), ii) dry powder
inhalers (DPI), and iii) nebulizers (see Figure 1.3). Nebulizer systems are typically less
portable than pMDI and DPI and are used in case of acute asthma attacks and for patients
unable to use other inhalers. The development of these devices is not a trivial task and
includes many design considerations, importantly, particle properties (density, diameter,
shape, chemical composition etc.), aerosol particle properties (volume fraction loading of
particles (concentration), particle size range, flow rate etc.), respiratory tract properties
(geometry, presence of disease etc.).

Figure 1.3, Different types of inhalers

1.4 Aerosol particle dynamics

The motion of a particle inside human respiratory tract is governed by certain physical
laws and its useful to have a brief look into each of these.
I) Drag Force ( Fd ), Drag force is the retarding force exerted by the fluid to the aerosol
particle and is caused because of the relative motion between particle and fluid. The drag
force in general terms is given as

Cd

Fd
1 2
u A
2 rel

(1.1)

where Cd represents the drag coefficient and depends upon the particle Reynolds number

Re p . The term is the fluid density, urel is the magnitude of relative velocity vector
u up ( u is the fluid velocity vector and up is the particle velocity) and A is the cross
sectional area. The direction of drag force acts in a direction parallel to vector u up .
II) Gravitational force ( Fg ), As with every object on earth, aerosol particle is pulled by
earths gravity

Fg m p g

(1.2)

where m p is the particle mass and g is the acceleration vector due to gravity.
III) Brownian diffusion force, Particles having size less than 1 m are subjected to
random movement as a result of collision with fluid molecules. The accounting Brownian
force becomes increasingly important with decreasing particle size.
IV) Electrostatic force, If an aerosol particle has a net charge, its motion is affected by the
electrostatic forces. The lung airways in general do not have any net charge but they are
electrically conducting. As a result, close to the airway wall, the tissue in the wall gets
oriented because of the charged aerosol particle and creates a net electric field. In this
work, since we are dealing with particles having zero net charge we neglect this force.
As will be discussed later in chapter 3, due to the fact that particle density is 1000 times
greater than carried fluid density, other forces such as Buoyancy force, Magnus force, lift
force, Basset force etc. are neglected.

1.5 Aerosol deposition mechanism

The human respiratory system essentially acts as a filter by removing particles present in
the inhaled air. As soon as the particle gets in contact with walls of the respiratory system
it gets trapped. The mucus lining present on the walls prevent the particle from getting reentrained. There are varying physical mechanism by virtue of which a particle can reach
the airway wall and get deposited. The most important deposition mechanisms are, i)
inertial impaction, ii) sedimentation, iii) brownian diffusion, and iv) interception.
Inertial impaction
Deposition due to inertial impaction is associated with the particle inertia. The deposition
occurs since the particle in transit is unable to follow the streamlines of air negotiating a
curve, bend or airway bifurcations (illustrated in Figure 1.4) i.e. the particle follows its
original trajectory. Inertial impaction becomes important for bigger sized particles or at
locations where high air velocity exists. In respiratory system this generally occurs at first
few airway generations, where the stream lines are highly curved and flow velocities are
high [3].

Figure 1.4, Deposition due to inertial impaction

The importance of deposition due to inertia is represented via dimensionless Stroke

number, and interpreted as the ratio of characteristic particle time and characteristic flow
time,

Stk

rU
D

(1.3)

where r is particle characteristic time, U is the air velocity and D is the airway diameter.
For a particle Stk 1 indicates that it will closely follow flow streamlines and will
quickly adjust to any change in air flow path. Similarly, for a particle with Stk 1 or
Stk 1 , will not follow any sudden change in air flow streamline.

Sedimentation
Deposition due to sedimentation or gravitational settling refers to particle deposition due
to gravity (illustrated in Figure 1.5). The probability of a particle being deposited due to
sedimentation depends upon particle density, size and the time spent by the particle in
airway segment (called as residence time). In the respiratory system this usually occurs at
distal airway segments (smaller bronchi and bronchioles, alveolar region), where the flow
velocity is very low [4].

Figure 1.5, Deposition due to sedimentation

10

Brownian diffusion
At smaller airway segments (such as bronchioles, alveoli), where the particle residence
time is relatively long, smaller diameter particles ( d p 1 m ) can come into contact of
airway walls and hence deposit as a result of Brownian motion (as illustrated in Figure
1.6). Brownian motion is a microscopic three dimensional stochastic random walk of
particle and occurs due to the collision of particle with the air molecules. For times much
longer than the time between molecular collision, Einstein formulated the displacement
of particle due to brownian motion as

xd 2 Dbt

(1.4)

where Db is the particle diffusion coefficient given as

Db

kTCc
3 d p

(1.5)

where k is the Boltzmann constant, T is the temperature in kelvin and Cc is the

Cunningham slip correction factor, d p is the particle diameter and is fluid viscosity.

Figure 1.6, Deposition due to Brownian motion

11

Interception
A particle following the air stream without any deviation can still deposit at the airway
wall because of its physical size greater than the airway. This is termed as interception.
Usually this type of deposition mechanism occurs for long fibers (long in one dimension)
and ignored for pharmaceutical aerosols which mainly consist of spherical particles.

1.6 Factors affecting aerosol particle deposition

As explained in the above section three main deposition mechanisms (Impaction,
Sedimentation and Brownian motion) are usually considered responsible for the
deposition of inhaled particles. However, the above deposition mechanisms inturn
requires a priori knowledge of three important parameters,
I) Respiratory airway geometry, In order to evaluate the deposition characteristic of
aerosol particles, the first and foremost requirement is to have a geometrical model
representing lung. Unfortunately, there is not a single unique geometrical model which
can be said to represent the human population. In literature there exist simplified
mathematical models based on the classical symmetric model of Weibel [5] or the
asymmetric model Horsfield [6]. But such models are based on simplifying assumptions
such as representing the airways as smooth circular cylinders, basing airway dimension
and branch connectivity. Although with the recent advancement in imaging capability
one can now have anatomically accurate physical model of respiratory tract but this is
still restricted to first few generations of the airway and are subject specific. Further, from
the effectiveness point of view of inhaled pharmaceuticals, the extrathoracic airway
possesses some serious barrier. The extrathoracic airway region with is bends and sudden
cross-sectional changes reduce the quantity of the drug reaching alveolar region, thereby
rendering it ineffective.
II) Aerosol particle properties, Apart from the chemical composition particle size is a
critical parameter determining the fate of particle being inhaled. Particles having size
12

between 0.5-5m has the ability to reach much deeper inside the lungs. Particles having
size greater than 5 m tend to deposit mainly in the extrathoracic region, while particles
smaller than 0.5 m exhaled out without depositing. Figure 1.7, illustrates a typical
deposition pattern in various regions of the lungs depending upon the particle size
diameter.

Figure 1.7, Deposition probability with respect to particle size, [7]

III) Inhalation flow rate: The mode of inhalation directly affects the particles deposition.
With increased inhalation flow rate, the probability of particle depositing due to inertial
impaction in the extra-thoracic region increases. Conversely, reducing inhalation flow
rate allows particles to negate sudden changes in flow path as such the probability of
particles being carried much deeper inside the lungs increases. In general, the inhalation
flow rates are broadly classified into slow breathing (15 l/min), tidal breathing (30 l/min)
and heavy breathing (60 l/min).

1.7 Outline of the thesis

While Chapter 1 introduces the reader to various aspects of Human respiratory system
and its interaction with Aerosols in general, Chapter 2 provides an overview of the
13

existing CFD literature on human airways. Chapter 3 gives detailed mathematical models
used to describe fluid and particle phase.
Chapter 4 till Chapter 8 presents results from various numerical experiments conducted in
the present research. In particular within the framework of RANS, Chapter 4 discusses
and compares simulated particle data using classical eddy interaction model and Wang &
James model and describes a new simpler helicity based eddy interaction model (HEIM).
Further Chapter 5 presents various numerical experiments to evaluate the performance of
HEIM.
The last part of the research focuses on Large Eddy simulation (LES) (Chapter 6 till
Chapter 8). In this part, first a new simpler rotational based subgrid scale model for LES
is proposed and evaluated in Chapter 6. In continuation the next Chapter 7 discusses and
evaluates an efficient multiple LES frozen field approach based on Proper Orthogonal
decomposition (POD).

Finally, Chapter 8 presents the preliminary results in a 5

generation intrathoracic airway model using LES and the approach of Chapter 7.

14

Chapter 2
Literature survey
Contents
2.1 Airway modelisation...............................................................................................................................15
2.1.1 Extrathoracic airway geometry.......................................................................................................16
2.1.2 Tracheobronchial airway geometry................................................................................................19
2.2 Modeling methods for aerosol transport..............................................................................................24
2.2.1 Modeling fluid phase.......................................................................................................................24
2.2.2 Particle phase....................................................................................................................................26

The entire spectrum representing aerosol transport in human respiratory tract can be
broadly divided in two main fields, namely i) airway modelisation, and ii) predictive
methods for aerosol transport. While the former concerns with development of airway
geometries for numerical and experimental studies. Latter concerns mainly with the
numerical modeling of aerosol transport i.e. modeling fluid phase and particle phase.

2.1 Airway modelisation

The anatomical description of human respiratory tract given in Chapter 1, shows that the
entire human respiratory tract is composed of many bends, sudden cross-sectional
changes, dichotomous bifurcation (i.e. parent branch divides in two daughter braches)
making the process of airway modelisation complex. Add to this complexity we have in
total around 16,000,000 airway segments, making the morphometric description of
airway structure unfeasible.
In literature there exist two approaches of generating airway structure. The first approach
uses mathematical algorithms which contain inhaled volume based rules for establishing
15

relationships between parent and daughter airways (e.g. Kitaoka et al. [8], Tawhai et al.
[9]). The second approach consists of digitally reconstructing the airway structure based
on the data available from CT-scan or micro CT-scan of patients (e.g. Matida et al. [10],
van Ertbruggen et al. [11] and Nithiarasu et al. [12]).
This work uses the second approach, whereby the airway geometries are modelled based
on the available CT-scan. Traditionally, for computational and experimental studies the
respiratory tract is divided into extrathoracic region or tracheobronchial region (typically
till 5th generation) or alveolar region. In this work we have kept ourselves to the
extrathoracic (or upper airway) and tracheobronchial region.

2.1.1 Extrathoracic airway geometry

Katz and Martonen [13] carried out flow studies in a simplified three dimensional model
of larynx. The simplified larynx model (Figure 2.1) was constructed based on the
morphometric measurements of human casts and Weibel [5] morphology of the tracheal
dimension. The larynx was modeled as a 6 cm long cylinder with a circular entrance and
exit cross-sections. The apertures created by ventricular and vocal cords were modeled as
ellipse.

Figure 2.1, Simplified larynx geometry by Katz and Martonen [13]

Corcoran and Chigier [14] measured the axial velocity and turbulence intensity, using
Phase Doppler Interferometry (PDI) in a cadaver-based simplified larynx-trachea model
(Figure 2.2). 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.
16

Figure 2.2, Larynx geometry used by Corcoran and Chigier [14]

The preliminary flow results obtained in the above mentioned geometries (Figure 2.1 and
Figure 2.2) did provided some useful initial flow physics but these geometries are over
simplified. As a result authors that followed used more realistic geometries such as
Kleinstreuer and Zhang [15], Matida et al. [10], Farkas et al. [16], Jin et al. [17]; Xi and
Longest [18]. For example, Kleinstreuer and Zhang [15] modeled the extrathoracic region
as 1800 curved bend (see Figure 2.3). The diameter variations along the airway model
from oral cavity to trachea where based on the Cheng et al. [19] morphometric
measurements of a human oral cast. Similarly, Stapleton et al. [20] , DeHaan and Finlay
[21] and Grgic, Finlay, and Heenan [22]

performed their studies in an average

geometrical model based on the morphometric information available in the literature.

Further the model was refined by separate measurements using computed tomography
(CT) scans of patients (n=10) having no visible airway abnormalities and by the
observation of living subjects.

17

Figure 2.3, extrathoracic airway developed by Kleinstreuer and Zhang [15]

Figure 2.4, extrathoracic airway developed by [20]

On similar lines, a simplified extrathoracic airway shown in Figure 2.5 was developed at
Vrije Universiteit Brussel (VUB). The simplified model was based on the work of Brouns
et al. [23] of VUB, and was based on the CT-scan of five otherwise healthy never-smoker
male subjects. The simplification of the geometry was done so as to facilitate the
comparison of numerical and experimental studies. The simplification is done by keeping
the critical features such as shape of mouth cavity, position of trachea and the epiglottis
(as shown in Figure 2.5). Due to the availability of experimental, the simplified model is
extensively used in the present work.
18

Figure 2.5, (left) Realistic geometry; (b) simplified extrathoracic geometry of VUB.

2.1.2 Tracheobronchial geometry

While there are numerous successful aerosol studies of the extrathoracic region, the
complexity of obtaining morphometric description of human tracheobronchial tree has
defied detailed aerosol studies in anything more than small sections. In order to
adequately understand the aerosol dynamics, modelisation of the distal airways is
essential.
One of the earliest and well known airway model used for the description of
tracheobronchial airways is the Weibel [5] symmetric model also known as Weibel A
model. The model assumes that each generation of the airway bifurcates symmetrically
19

into two daughter branches. Although this simplify analysis but is not accurate since in
actual human airway diameters and lengths of daughter airways can be quite different
from one another. It is stated that the absence of airway curvature and surface
irregularities make the flow fields in a Weibel based model very different from those in a
real lung. For example, Nowak et al. [24] , who were first to compare Weibel A model
with a CT-scan of a cadaver lung cast (till 4th generation). Unsteady and steady
calculations were compared. These authors concluded there is no consistent pattern of
similarity between symmetric model of Weibel and asymmetric model based on CT-scan.
They also stated that differences between steady and unsteady flow solutions make the
former unreliable. Although physiologically incorrect, Weibel A geometry has been used
in several studies (such as Shi et al. [25], Zhang and Kleinstreuer [26] , Longest and
Vinchurkar [27], Jin et al. [17] and more recently by Vasconceloset al. [28], etc.). Figure
2.6, illustrates the upper airway geometries based on Weibel A model. The left most
panels represent the airway geometry used by Zhang and Kleinstreuer [26]. As shown the
upper airway model consists of two parts, an oral airway model, including oral cavity,
pharynx, larynx and trachea, and a symmetric, planar, triple-bifurcation lung airway
model representing generations G0 (trachea) to G3 based on Weibel A model. The
authors used this geometry to study the differences between micron and nano particle
deposition characteristics. Further, Figure 2.6 (middle panel), illustrates the geometry
used by Jin et al. [17] for conducting large eddy simulation calculations. The geometry
was generated by fusing the mouththroat geometry developed at ARLA (Aerosol
Research Laboratory of Alberta) [7] with the triple bifurcation based on Weibel A model.
Similarly, the left most panel of Figure 2.6 consists of a 4 bifurcation Weibel A model.
This geometry was recently used by Vasconceloset et al. [28] for investigating their
proposed approach of determining micron particle deposition by estimating escape rate at
each airway segment. In most of these studies the simplified model was used with the
sole purpose of proposing or validating a numerical approach, effect of mesh styling,
inlet boundary conditions etc. These studies were not meant to draw any physiological
conclusion.

20

Figure 2.6, Tracheobronchial geometries based on Weibel A model, (left) Zhang and Kleinstreuer
[26], (middle) Longest and vinchurkar [27], (left) Vasconceloset et al. [28]

In order to simulate aerosol deposition in physiologically more accurately, further to

Weibel A model, Horsfield and Cumming [6] in 1968 (subsequently refined in 1971)
made measurements of the tracheobronchial measurements as close as possible and
introduce an asymmetric model. Weibel A model differs from the Horsfield geometry in
terms of branching angles (or bifurcation angles). While the former gave no information
regarding the branching angles, later gave the branching angle down to segmental
bronchi. Unlike Weibel [5] geometry who made measurements till 5th generation and
incomplete till 10th generation, Horsfield [6] accurately measured values of length,
diameter, curvature ratio and branching angle till segmental bronchi. Several aerosol
studies have utilized this information given by Horsfield. For example, Van Ertbruggen et
al. [11] used the morphometric data of Horsfield [6] to construct a realistic 3D model (see
Figure 2.7) of bronchial tree in order to study the flow and particle dynamics. Similarly,
Holbrook et al. [29] recently studied localized aerosol deposition in a double bifurcation
geometry having generations 3 to 5 (see Figure 2.8). While these studies present a more
extensive estimation of airflow behavior in the lungs, they still rely on an idealized
smooth representation of the airway geometry.
21

Because of the advent of new imaging modalities (Magnetic resonance imaging (MRI
scan) and Computer tomography (CT)) and increase computing resources, it is now
possible to recreate digitally realistic human airway geometries. The only shortcoming
with these modalities is that the reconstruction of human geometry is restricted to the
resolution of the camera. Therefore, to date geometries till 5 th generation has been
reported in literature. For instance, in a detailed study Lin et al. [30] conducted direct
numerical simulation on CT imaged derived upper airway (shown in Figure 2.9). The
geometry of the human upper respiratory tract is derived from volumetric scans of a
volunteer imaged via multi detector row computed tomography. In this study Lin et al.
[30] used to geometries one starting with mouth and ending at generation 5, while other
starting at trachea and ending at generation 5. The author reported that neglecting
extrathoracic part and using simple inlet boundary conditions do not adequately represent
the effect of the upper airway structures. This inturn affects the estimate of flow through
several generations of airway and the tracheal wall shear stress. Similar to Lin et al. [30]
few more studies have been reported using CT-scan derived upper airway geometry (such
as Ma and Luchten et al. [31], Ghalati et al. [32] etc).

Figure 2.7, Tracheobronchial geometry generated by Van Ertbruggen et al. [11]

22

Figure 2.8, Double bifurcation generation generated by Holbrook et al. [29]

Figure 2.9, CT-scan based upper respiratory airway geometry used by Lin et al. [30], (left)
extrathoracic region depicting various regions, (right) tracheobronchial region

23

2.2 Modeling methods for aerosols transport

Modeling methods involved for simulating particle transport involve prediction of fluid
phase and particle phase. The fluid phase is represented by numerically solving Navierstokes equation (called as computational fluid dynamics (CFD)). There is large amount of
scientific papers describing the use of CFD for studying the flow behavior in the human
airways. The different numerical methods are classified based on the nature of the flow
i.e. laminar or turbulent.
In past, aerosol studies where predominantly simulated assuming laminar flow (for
example, Hofmann et al. [33], Comer et al. [34], Li et al. [35], Kleinstreuer et al. [36]).
The reason for representing flow as laminar is mainly to avoid the complexity associated
with turbulence modeling. For example, Martonen et al. [37] performed laminar
calculation in larynx, trachea and main bronchi to provide basic flow information.
However, as Stapleton et al. [20] reported that the flows in larynx and trachea are
normally turbulent or at least transitional. Therefore, particle results assuming laminar
flow should be interpreted carefully. Recent aerosol studies have now focused in
applying turbulence modeling for predicting fluid-particle dynamics in airways which is
also the scope of the present work.

2.2.1 Modeling fluid phase

As will be seen in latter chapters, predicting flow simulations accurately is the
prerequisite for analyzing particle deposition characteristics inside human airways. This
in turn depends on the approaches used for modeling turbulence. So far, in literature three
approaches have been generally applied, namely, i) Reynolds averaged Navier-stokes
equation, ii) Large eddy simulation and iii) Direct numerical simulation (DNS).
In the framework of RANS, mostly two equation turbulence models have been applied.
The flow rates that are mostly considered in human airways correspond to an inlet
Reynolds number that lies in the transitional regime (for e.g. the Reynolds number range
24

in the present work ranges from 2500 for 30 l/min and 5000 for 60 l/min). At this
Reynolds number range as reported by Wilcox [38], and Pope [39], the standard
turbulence models (for e.g. k-, RNG k etc.) fails to predict laminar to turbulence
transition. Therefore, under such conditions low-Reynolds-number (LRN) turbulence
models have to be considered. For example, Stapleton et al. [20] reported significant
deviations from experimentally observed pressure drop when comparing against standard
k- simulations. To further investigate this problem Zhang and Kleinstreuer [40],
compared the performance of LRN k- model against k-, RNG k-, LRN k- and found
comparatively good experimental agreement for velocity and kinetic energy levels across
the geometry. This model has also been successfully employed in various other airway
geometries [ [15], [41], [42], [43]].
Even though LRN based turbulence model predicted comparatively better mean flow
field, still there were deviations in particle deposition when compared against
experiments (for e.g. Jayaraju et al. [44], Verbanck et al. [45]). This mainly because in
RANS there is no information of velocity fluctuations and as such it needs to be modelled
from simulated turbulent kinetic energy. The modelling is not straightforward and
involves some adhoc assumptions (more discussed in Chapter 4). Owing to this
ambiguity and the importance of predicting the flow field accurately, authors such as
Matida et al. [46], Jayaraju et al. [44], Lambert et al. [47], Jin et al. [48] etc., have
employed large eddy simulation (LES). For example, Matida et al. [46], employed LES
using constant Smagorisnky model and reported improved particle results over RANS.
Similarly, Jayaraju et al. [44] employed LES using WALE and constant Smagorinsky as
subgrid scale (SGS) model and found good overall agreement of flow and particle results
when compared against experimental data.
However, RANS till date is still preferred over LES because of computational
requirements. In this view, the present work proposes a new SGS model which will be
shown to be computationally faster and gives accurate results (see chapter 6 and 7).
Besides RANS and LES, direct numerical solution (DNS) in human airways has also
been cited. Till date the most comprehensive work remains that of Lin et al. [30], who
25

systematically studied the importance of upper and intrathoracic airways on air flow
patterns and turbulence characteristics. They found that the regions of high turbulence
intensity are associated with TaylorGrtler like vortices.

2.2.2 Modeling particle phase

Invariably, when simulating particle deposition inside human airways, the common
practice is to assume particles as spherical, non-interacting, and monodisperse. These
assumptions allow decoupling of fluid calculation from particle calculation i.e. one-way
coupling is assumed. Above assumptions are reasonable for inhaled aerosols owing to the
particle to air density ratio, volume fraction loading (Finlay [49]), and these assumptions
also apply in the present work. When considering dilute suspension in general there are
two approaches reported in literature, namely Euler-Lagrange and Euler-Euler. While the
Euler-Euler approach has been applied to particles having size less than 1m, EulerLagrange approach is mostly confined to simulating micron particle sizes i.e. particle size
greater than 1m. This work is related to the conducting fluid-particle simulations using
Euler-Lagrange methodology.
In the Euler-Lagrange methodology, the Eulerian or the fluid phase is solved using
aforementioned approaches, whereas the Lagrangian phase is simulated by numerically
integrating individual particle equation of motion. In case of RANS, stochastic modeling
of the particle phase involves direct simulation of particle motion through a random
turbulent flow field. There are several approaches in developing stochastic models. Few
examples in literature include models based on the Langevin dispersion equation [50],
random Fourier modes [51] and pdf models [52]. However, one of the most widely used
models is the Eddy Interaction Model (EIM) first introduced by Hutchinson et al. [53]
and further developed by Gosman and Ioannides [54]. Even though several variants of
EIM model have been proposed in the literature ( [55], [56], [57], [58], [59]), the
originally proposed EIM of Gosman and Ioannides [54] remains the most widely used,
mainly because of its simplicity. Although largely used, the classical EIM when used
have resulted in large deviations in particle deposition values compared to experimental
26

data. For example, Matida et al. [10] studied the deposition of monodisperse particles (126 m) in human mouth-throat geometry, at inhalation flow rates of 30 and 90 l/min. For
the Eulerian phase they employ LRN k turbulence model, while the Lagrangian phase
is simulated based on EIM by Gosman and Ioannides [54]. Contrary to the good
performance of LRN k- model reported by Zhang et al. [60], the simulations of Matida
et al. [10] showed huge deviations in deposition percentages (>50%), even for the lowest
Stokes number particles. Similar observations were reported in Jayaraju et al [44],
Verbanck et al. [45] etc. The main reason according to these authors was the underlying
assumption of isotropy in the EIM model. To this Matida employed near wall correction
function based on Wang and James [61] and obtained better overall deposition prediction.
Similar correction function were used by [62], [63], [64] etc. Inspite of using such adhoc
correction function one can introduce the anisotropy retrieved from the flow itself. This
new methodology will be discussed in chapter 4. As a result of ambiguities present with
EIM several authors applied large eddy simulation (LES) methods for the study of
particle deposition in the human airway (e.g. Matida et al. [46], Jayaraju et al. [44],
Lambert et al. [47]). Still the number of such studies are few because of the involved
computational and memory requirement. To this Agnihotri et al. [65] used a systematic
procedure to reduce computational requirement discussed in detail in chapter 7.

27

28

Chapter 3
Governing equations
Contents
3.1 Introduction............................................................................................................................................29
3.2 Importance of turbulence.....................................................................................................................30
3.3 Incompressible Navier-Stokes equation..............................................................................................31
3.4 Modeling turbulence.............................................................................................................................32
3.4.1 Reynolds averaged Navier-Stokes equation................................................................................33
3.4.2 Two equation SST k- EVM.........................................................................................................35
3.4.3 Large eddy eimulation...................................................................................................................39
3.4.4 Smagorinsky model.........................................................................................................................41
3.5 Modeling particle phase........................................................................................................................45
3.5.1 Modeling assumption......................................................................................................................46
3.5.2 Eulerian approach..........................................................................................................................47
3.5.3 Lagrangian approach.....................................................................................................................49
3.5.4 Eddy interaction model...................................................................................................................52

3.1 Introduction
In this chapter, we first begin by describing fluid dynamics from a purely heuristic point
of view. We then give a brief overview on the mathematical equations used in this work
to describe the transport of the fluid phase and particle phase. Firstly, the Navier-Stokes
equations (N-S) which govern the transport of fluid phase will be presented. Followed by
the representation of Navier-Stokes equation in the frame work of Reynolds Averaged
Navier Stokes equation (RANS) and Large Eddy Simulation (LES) will be presented.
Finally, the chapter concludes with the description of equation of motion for the particle
phase.

29

3.2 Importance of turbulence

Understanding of turbulence in fluid flows is the most important and yet most complex of
the phenomena to be understood in all of the classical physics. It is a fact that turbulence
exists at all scales, spanning from interior of biological cells to geophysical and
astrophysical phenomena. The human respiratory system is not an exception to this.
According to Lumley [66], turbulence is recognized as a distinct fluid behavior,
characterized by randomness,

increased diffusivity, three dimensionality, and

dissipativity.
O. Reynolds [67] systematically investigated the transition from laminar to turbulence
through experiments in a pipe flow. He discovered that the flow instability resulting in
transition from laminar to turbulent flow is dependent on a non-dimensional parameter
known as Reynolds number. This parameter denoted by Re is a ratio of inertial forces to
viscous forces, i.e.

Re

UL

(3.1)

where is the dynamic viscosity, is the fluid density. The terms U and L are the
characteristic velocity and length scales, respectively. At large Reynolds numbers
(inertial forces much larger than viscous forces), flow instabilities grow rapidly resulting
into intense mixing, and the flow is then termed as turbulent. On the other hand, at
moderate or low Reynolds number viscous forces are able to suppress the flow
instabilities and the resulting fluctuations, this behavior is termed as laminar.
Turbulence is generally perceived as the spectrum of eddies having varied sizes, where an
eddy is conceived as a coherent flow structure over a specified region (see Figure 3.1).
The spectrum spans from the largest energy containing eddies known as integral scale to
intermediate Taylor scale down to smallest scales known as Kolmogorov micro scales.
The integral scale corresponds to those flow structures which contain most of the
30

turbulent energy. At these scales, the viscosity effects are negligibly small and are mainly
determined by the geometrical domain and boundary conditions. The Taylor micro scales
are intermediate scales, basically corresponding to Kolmogorovs inertial sub range. The
Kolmogorov (or dissipation) scales are the smallest turbulence scales. These scales are of
universal nature and are strongly affected by viscosity.

Figure 3.1, Flow structures inside UAM at 60 l/min

3.3 Incompressible Navier-Stokes equation

For the flow of constant-property Newtonian fluid, conservation of mass and
conservation of momentum results in the following set of dynamical equations
u j
x j

31

(3.2)

ui ui u j
1 p 1 ij

t
x j
xi x j

(3.3)

where ui represents the i-th component of the fluid velocity, (constant) is the fluid
density, and ij represents the symmetric stress tensor. For a constant property
Newtonian and incompressible fluid, ij is given as

ij 2 Sij

(3.4)

where p is the static pressure, is the dynamic viscosity, and Sij is the strain rate tensor
given by

Sij

1 ui u j

2 x j xi

(3.5)

Using equation (3.4) in equation (3.3) results into the following final form
S
ui ui u j
1 p

2 ij
t
x j
xi
x j

(3.6)

where / is the kinematic viscosity of the fluid.

3.4 Modeling turbulence

The three main paradigms in the order of decreasing computational requirement
employed for modeling turbulence are; Direct Numerical Simulation (DNS), Large Eddy
Simulation (LES), and Reynolds Averaged Navier-Stokes equation (RANS).
DNS is the most accurate approach and consists in solving the Navier-Stokes equation for
one realization of the flow without the need for any turbulence model. In DNS all the
32

length scales and time scales need to be resolved, thereby making the approach
computationally most expensive. Because the number of grid points required to fully
resolve DNS in three dimensions varies with Re9/4 , DNS is restricted to low to moderate
Reynolds number flows. DNS is generally used as a research tool for studying the
mechanics of turbulence, such as identifying dominant flow structures, energy cascading,
energy production etc.
RANS, which is computationally the least expensive method, consists in solving an
equation for a time averaged mean velocity field ui x and all the flow fluctuations
are averaged out. Since only mean quantities are represented, it requires additional
strategies for modeling the effect of turbulence on the mean quantities.
In terms of flow field description, LES lies in between DNS and RANS. In this approach,
equations are solved for an instantaneous filtered velocity ui ( x, t) , where small scale
turbulent structures are filtered out. The small scale motions that are not explicitly
represented are modeled via a subgrid scale model.

The next two sections explain

RANS, and LES approaches in details.

3.4.1 Reynolds averaged Navier-Stokes equation

The derivation of RANS equation for an incompressible flow starts by representing terms
of equation (3.2) and equation (3.6) into their respective mean and fluctuating part. Such
decomposition is referred to as Reynolds decomposition
ui ui ui'
p p p'

(3.7)

where the symbol denotes time average. The time average of a quantity is defined as:

x, t

T
33

t T

x, t dt
t

(3.8)

Finally the RANS equations are derived by time averaging equations (3.2), (3.6) and
using the property

i , i' 0 as follows
uj
x j

(3.9)

ui' u 'j
ui u j
Sij
ui
1 p

t
x j
xi
x j
x j

The last term in equation (3.10), ui' u 'j

(3.10)

is known as Reynolds stress tensor and the

mean strain rate tensor Sij is given as

uj
1 ui
Sij

2 x j
xi

(3.11)

Now for a three dimensional flow, we have in total 10 unknowns in the above equations
namely, 3-mean velocity components, 1-mean pressure, and 6-Reynolds stress tensor
components. And total equations available are four namely, 1-continuity equation (3.9) ,
and 3-momentum equation (3.10). This is an underdetermined system i.e. system of
equations represented in (3.9) and (3.10) are not closed (closure problem).
In order to close the above system of equations we must introduce a turbulence model.
This is achieved either by invoking eddy viscosity hypothesis or more directly by
introducing Reynolds-stress transport equation.
The eddy viscosity (or Boussinesq) hypothesis intrinsically assumes that Reynolds
stresses ui' u 'j scales linearly with the mean strain rate i.e.
2
ui' u 'j 2 t Sij k ij
3

34

(3.12)

where the constant t is known as turbulent eddy viscosity and k is the turbulent kinetic
energy

1 ' '
ui ui
2

(3.13)

The last term in equation(3.12) is required to guarantee that upon contraction the
equations remain correct. Finally one requires a model to describe the turbulent eddy
viscosity t , thereby closing the set of equation represented by equation (3.9) to (3.12).
There are a number of Eddy Viscosity Models (EVM) to describe t , all are classified
based on the number of equations involved in describing t . The simplest of the EVM are
the zero equation or Algebraic models (e.g. Baldwin and Lomax [68]). These models use
algebraic relation i.e. no PDE to compute t . In terms of complexity, the next hierarchal
EVMs are the one equation model. In this a model transport equation is solved for just
one turbulence quantity, namely, turbulent kinetic energy k . Both zero and one equation
EVMs are incomplete in the sense that both require the specification of a turbulence
parameter to describe t (mainly specification of mixing length). The widely used and
most accurate EVM are the two equation EVMs, where t is computed by solving two
transport equations for two turbulence quantities. These are also complete models in the
sense they dont require specification of any turbulence parameter.

3.4.2 Two equation SST k EVM

The SST k eddy viscosity model, developed by Mentor [69], is a popular variant
based on Wilcox [38] k model. It is known from experience that the standard k
turbulence model performs well for free shear flows but often fails miserably for wall
bounded flow, especially at regions close to the walls. In contrast, the k turbulence
model from Wilcox [38] performs quite well close to the wall but as shown by Menter
[69] the model is sensitive to free stream values of . Therefore, in order to make a
model suitable for variety of applications, Menter [69] combined the above two models
35

via a blending function. By doing this the new model called as SST k amalgamates
the advantages of both models, thereby making it more versatile. Following are the
features of SST k turbulence model,
The blending function F1 in the model is defined in such a way so that it is zero at
the walls and gradually become equal to one away from the wall. This gradual
change causes the model to behave as k near to the wall and k away from
the wall.
In addition to the above the SST model also features a modification to eddy
viscosity to account for the transport of turbulent shear stress.
3) Menter [70], formulated the equations by first transforming the standard k model
into a k and formulation. Next the Wilcox k model is multiplied by function
F1 and the transformed equation are multiplied by 1 F1 . Finally both models are
added together to form what is named as SST k . The modeled equations for SST
k read as below,

k k ui
k

k
Pk Dk
t
xi
x j x j

(3.14)

ui

P D C
t
xi
x j x j

(3.15)

In the above equations, the effective diffusivities k and are defined as

t
k

(3.16)

(3.17)

36

where k and are Prandtl turbulence numbers

1
F1 1 F1

k ,1 k ,2

1
F1 1 F1

,1 ,2

(3.18)

(3.19)

where k,1 , k,2 , ,1 and ,2 are constant parameters. The turbulent viscosity t
appearing in equations(3.16) and (3.17) is given by

k
1

1 S F2
max ,

a1

The term S is the strain rate magnitude given as

(3.20)

2 Sij Sij . The coefficient , is a

lowRe correction factor introduced to dampen the eddy viscosity close to wall
0.024 Ret 6

1 Re t 6

(3.21)

where the term Ret k / is the turbulent Reynolds number. Functions F1 and F2
appearing in the above equations (3.19) and (3.20) are the blending functions (for more
information on blending functions reader is referred to Menter [69]). Pk and P represent
respectively the production of turbulent kinetic energy and the specific dissipation ,
given as
Pk min( t S
37

,10 k )

(3.22)

Pk
t

(3.23)

For incompressible flows, the parameters and appearing in the above equations are
defined as below
0.267 Re t 8 4
0.09

1 Re 84
t

(3.24)

1 9 Ret 2.95

1 Ret 2.95

(3.25)

F1 ,1 1 F1 ,2

(3.26)

where

where ,1 and ,2 are constant values. Further, the dissipation of turbulent kinetic
energy Dk and specific dissipation rate Dw are given by
Dk k

(3.27)

D 2

(3.28)

F1 i ,1 1 F1 i ,2

(3.29)

where

where i ,1 and i ,2 are constant values. Finally, the cross-diffusion term C , is a result of
transformation of standard k into k and quantities, required for blending with
Wilcox [38] k model, is represented as
38

C 2 1 F2 ,2

1 k
x j x j

(3.30)

The various constants appearing in the above equations are given as

,1 0.5532, ,2 0.440, ,1 2.0, k ,2 1.0, ,2 1.168, a1 0.31

i ,1 0.075, i ,2 0.0828

3.4.3 Large eddy simulation

In a LES computation, the large eddies are resolved explicitly while small eddies are
modeled. The underlying premise is that large eddies carry much of the turbulence
information, therefore, must be resolved accurately. Smaller eddies, on the contrary are of
universal nature and are more amenable to modeling. Since LES involves modeling of
small eddies, the mesh can be made much coarser than that required in DNS. Hence from
the computational cost it is much cheaper than DNS.
To separate the larger eddies ( , resolved scale) from smaller eddies ( ' , subgrid scale
or SGS), an idea similar to Reynolds decomposition is followed, i.e. a quantity is
decomposed as

'

(3.31)

Such decomposition is possible by using a low pass spatial filter. In finite volume
methods, the filtering operation is implicitly provided by the discretization itself and
there is no need for separate filtering operation. Leonard [71], gave the formal definition
of filter as a convolution integral, defined as

( x, t ) G x ( , t ) d
V

39

(3.32)

where integration is performed over the entire flow domain. G is the filtering kernel and
should satisfies the normalization condition

G( x ) d 1

(3.33)

In terms of filter function, the box filter is given as

1

G( x) 3
0

x / 2

(3.34)

other wise

Based on the suggestion of Deardroff [72], the filter width in anisotropic meshes is
defined based on cell volume
1/3

hx h y hz

(3.35)

Here hx , h y and hz represents grid spacing in x, y and z coordinate directions. Finally, the
governing equation for a constant property incompressible LES is obtain by spatially
filtering equation (3.2) and equation (3.3)
u j
0
x j

(3.36)

Sij ij sgs
ui ui u j
1 p

t
x j
xi
x j
x j

(3.37)

where Sij is the resolved strain rate tensor, determined as

1 u u
Sij i j
2 x j xi
40

(3.38)

In the above equation the term ijsgs represents the subgrid scale tensor, defined as

ijsgs ui u j ui u j

(3.39)

The mathematical form of above equations looks similar to RANS (see equation (3.10)),
but there is a considerable difference between the two formalism, for above we have

i i and i 0 . Similarly to RANS the filtered LES equations (3.36) and (3.37) are
unclosed. The closure is achieved by modeling ijsgs by a suitable subgrid scale model
(SGS). There exists large number of closure models (such as mixed models, similarity
models etc,), but the most common and widely used are the eddy viscosity based SGS
models (Sagut [73], Pope [39])

3.4.4 Smagorinsky model

One of the earliest eddy viscosity based closure model for ijsgs was provided by
Smagorinsky [74]. The model is based on applying Boussinesq hypothesis similar to
RANS

1
ijsgs kksgs ij 2 t Sij
3

(3.40)

where Sij is the strain rate of large scale or resolved field. Equation (3.36)-(3.40) are still
not closed and requires the specification of t . On dimensional grounds one can write

t UL

(3.41)

Equation(3.41), indicates that t can be thought of as the product of a characteristic

velocity U and a characteristic length L. Smagorinsky used the Prandtl mixing length
hypothesis to develop relations for the characteristic velocity and length. In the case of
the Smagorinsky model, the characteristic velocity U is given as
41

ULS

(3.42)

where the term S indicates norm of the resolved strain rate magnitude and is given by
(3.43)

S 2 Sij Sij

The characteristic length L of the unresolved motions is taken proportional to filter width

(3.44)

L Cs

(3.45)

or

1/3

where hx h y hz

is the filter width. The parameter C s is referred to as the

Smagorinsky model coefficient. Substituting relations (3.42) and (3.45) into equation
(3.41) yields
2

t L2 S = Cs S

(3.46)

In the above equations, the model coefficient C s is not a self-adjusting flow dependent
parameter and requires specification in order to close the above system of equations.
With respect to the Smagorinsky constant, several values have been proposed. For the
case of isotropic turbulence, 0.2 has been suggested, however, in a channel flow,
Deardorff [72] found that a smaller value 0.1 is better. Following the work of Lilly [75],
in the case of high Reynolds number, one can estimate the value of the constant as 0.17,
by assuming the filter width in the inertial subrange, where mean subgrid scale
production equals dissipation. Assuming a constant value for the model coefficient is
42

only justified when is in the inertial subrange. This is not valid in the viscous wall
regions as, assuming a constant value would imply a non-zero viscosity at the walls. Also
for laminar flow regions there should be no subgrid contribution, implying that the value
of the constant C s should be zero.
In order to overcome the aforementioned problems associated with the Smagorisnky
constant, Germano [76] proposed a dynamic procedure wherein, the model coefficient is
determined locally i.e. Cs Cs ( x, t ) . Such class of SGS models is known as Dynamic
Smagorinsky model (DSM). The model coefficient is determined locally by employing
an additional filter with a filter width larger than the grid spacing (test filter). In principle,
this permits identification of the fluctuating part of the resolved scale, and this is then
used to obtain an estimate of the unresolved stresses. A scale similarity is then invoked,
leading to the formulae for C s . The starting point of the dynamic model is the Germano
identity, given as
sgs
Lij Tij
ij

(3.47)

Tij u
i u j ui u j

(3.48)

where

and ijsgs is the subgrid stress tensor as defined in equation(3.39). The symbol ^
represents a second filtering operation with respect to a filter width which is usually
taken as 2 . Equation(3.47), can be further simplified to

Lij u
i u j ui u j

43

(3.48)

The term Lij with respect to test filter can be interpreted as the contribution to the
residual stress from the largest unresolved motions. Following equation(3.40) and (3.46),
we have

1
2
ijsgs ,d ijsgs kksgs ij 2 Cs S Sij
3

(3.49)

In the above equation, the superscript d represents the anisotropic part of stress tensor.
Equation(3.49), can also be written for the test filter width

1
Tijd Tij Tkk ij 2 Cs
3

S Sij

(3.50)

where the terms S and Sij are computed based on definition given in equation (3.43)
and (3.38) and by replacing u by u . The superscript d represents the anisotropic part
of stress tensor. Considering C s being uniform, we have quantity M ij defined as
2
M ij 2
S S ij 2

S Sij

(3.51)

Comparing equation(3.49) and (3.50)

sgs , d
Ldij Tijd
C s2 M ij
ij

(3.52)

The above term is nothing but the Smagorinsky model for the anisotropic part of Lij , i.e.

Ldij Lij Lkkij / 3 . Also it can be seen that to arrive at the above form of equation that Cs2
is assumed constant over a subdomain corresponding to the test filter width. As can be
now seen, both M ij and Ldij are known in terms of resolved u . This information can be
utilized to compute Cs2 . As shown by Lilly [75], the mean-square error given as
44

1 d
Lij Cs2 M ij
2

(3.52)

is minimized by specifying Cs2 as

2
s

M ij Ldij
M kl M kl

M ij Lij
M kl M kl

(3.53)

With the above formulation of Cs2 , the dynamic model overcomes many of the
aforementioned limitations of constant Smagorinsky model. Even though there has been
considerable progress by using a dynamic procedure, the model leads to difficulties in
practical simulations; this will be discussed in Chapter 6, where a new model will be
proposed.

3.5 Modeling particle phase

Primarily there are two fundamental approaches for modeling transport of inhaled
pharmaceutical aerosols (or particle phase). In the first approach, one treats the
particulate second phase as single particles and the particle trajectories are as a result of
forces acting on particles. This approach is called as Lagrangian approach. In the second
approach the particle phase is assumed to behave as a second continuum and equations
are developed for averaged properties of the particle. This second approach is referred to
as Eulerian approach. For a complete description of these two methodologies, interested
readers are referred to Fan and Zhu [77]. Herein, we discuss briefly about the two
methodologies, with main focus on the Lagrangian trajectory approach.

45

3.5.1 Modeling assumptions

Before going into the details of each of the approaches mentioned above, it is worthwhile
to discuss about the modeling assumptions involved with particle transport inside the
human lungs.
The major simplifying assumptions are as follows,
1) The particle is assumed to be spherical, This is a reasonable assumption particularly
for liquid inhaled pharmaceutical droplets, since the small liquid droplets are spherical.
Also, in case of dry powder aerosols and evaporated metered dose inhalers, the generated
aerosols consist of closely packed particles. Therefore an assumption of sphericity is
reasonable.
2) The particle density is much larger than the fluid density, The densities of
pharmaceutical compounds are much larger than their carrier fluid, typically the ratio of
particle to fluid density is approximately p / 1000 .
3) Drag force is the dominant force, This assumption is a direct result of assumption 2.
Since particle density is much greater than fluid density forces such as buoyancy force,
Magnus force, Basset force, lift force, pressure force, virtual mass force are considered
negligible.
4) The particle phase is considered dilute, In pharmaceutical aerosols, the particle
relaxation time r , defined as the time taken by the particle to drop to 37% of its initial
value is much smaller than particle collision time c , making the phase dilute.
5) One-way coupling, The volume fraction, defined as the ratio of volume occupied by all
particles combined to that occupied by the fluid i.e.

46

NV p
V

(3.54)

Here, N denotes total number of particles, Vp denotes volume of single particle, and V
total volume occupied by particle and fluid. According to classification map given by
Elghobashi [78] (see Figure 3.2) if p 10 6 , then particles have negligible effect on fluid
turbulence, this is termed as one way coupling. In this work and in most of
pharmaceutical aerosols the volume faction p remains less than 10-6, so one way
coupling is a reasonable assumption.

Figure 3.2, Charateristic Map for particle-flow coupling, [78]

3.5.2 Eulerian approach

In the context of pharmaceutical aerosols this approach is mostly applied for predicting
nanoparticle (particle size less than 1m) transport (for e.g. [79] [80]). The particle
transport are generally calculated by a convection-diffusion mass transfer equation, given
as

Y
Y
Y
uj

t
x j x j x j

47

(3.55)

where Y is the mass fraction, D is the effective particle diffusivity which for a spherical
particle and for relatively long particle time steps can be calculated based on StokesEinstein relation

kTCc
3 d p

(3.56)

Here, k 1.38 1023 JK 1 is the Boltzmann constant, T is the temperature in kelvin, C c is

the Cunningham slip factor, d p is the particle diameter, is dynamic viscosity of the
surrounding fluid and u j represents, jth component of the carrier fluid. When using
RANS, a similar Reynolds decomposition of Y yields a mean and fluctuating part

Y Y Y '

(3.57)

Using above equation (3.57) and time averaging equation (3.55), results in
Y
Y

uj

t
x j x j

Y
D
x j

u 'jY '

x j

(3.58)

The last term u 'jY ' represents the scalar-flux term and is modeled based on a gradient
diffusion hypothesis

u'jY ' Dt

Y
x j

(3.59)

Using equation (3.59) into (3.58), yields time averaged mass transfer equation

Y
Y

Y
uj

D Dt

t
x j x j
x j
48

(3.60)

In the above equation the term Dt represents, turbulent diffusivity given as

Dt

t
Sct

(3.61)

where Sct is the turbulent Schmidt number and relates turbulent momentum diffusivity
with turbulent mass diffusivity. In the case for Euler-Euler approach, assuming the wall
to be a perfect sink the boundary condition for the aerosols is considered to be Yw 0 ,
where Yw is the mass fraction at the wall [81].

3.5.3 Lagrangian approach

In this approach several particles are released in the carrier fluid and then the particle
phase is described by tracking the motion of individual particles in Lagrangian coordinates. The velocity and position of individual particles through the carrier fluid are
calculated by directly integrating their equation of motion. Therefore, the Lagrangian
approach has the advantage that there is a possibility of obtaining detailed statistical
information about the motion of individual particles. Also, accounting of different forces
is easy unlike the Eulerian continuum approach. Since particles are being tracked
individually, this approach is practically viable for very dilute suspension such as in case
of pharmaceutical aerosols.
Based on aforementioned modeling assumptions, the equation for the motion of a single
particle reads
dup
dt

u up

(3.62)

with dx p / dt u p . In this equation, x p is the particle position vector, up is the

instantaneous particle velocity vector, u is the instantaneous fluid phase velocity vector
and g is the gravity vector. The particle re-laxation time is denoted by r
49

pd p
4
3 up u Cd

(3.63)

where Cd is the drag coefficient, up u is the particle absolute relative velocity

magnitude, p and are the particle density and fluid phase density, respectively. The
product of r and the particle relative velocity is interpreted as particle stopping distance
relative to the fluid. At lower values of r , particle responds quickly to the fluid motion
and therefore, follows the fluid velocity fluctuation. On the other hand at higher values of

r , the particle is slow to respond to the fluctuations of the fluid flow and does not follow
the fluid so closely. In general, the particle relaxation time is dependent upon the particle
Reynolds number, Re p f up u d p / . In case of Stokesian drag, Re p 1 , the drag
coefficient Cd is equal to 24 / Re p . In such cases, r is dependent upon the material
properties of particle and fluid

p d p2
r
18

(3.64)

Various experimentally derived empirical relations exist in the literature (such as Schiller
and Neumann [82]), however, the one followed in this work is from Morsi and Alexander
[83], given by as
Cd a1

a2
a
32
Re p Re p

(3.65)

where Re p is the particle Reynolds number and coefficients ai ( i 1,2,3 ) are constants
as given by Morsi and Alexander [83], shown in Table 3.1. One of the important aspects
in Lagrangian approach is the representation of instantaneous fluid phase velocity which
in turn depends upon the methodology used for modeling the fluid phase i.e. DNS, LES
or RANS. In case of DNS is directly represented and there is no need for any modeling.
50

Similarly, in LES u is approximated as an instantaneous resolved velocity. Whereas, in

RANS, to take into account the effect of turbulence one requires certain modeling
approaches. These are known as stochastic dispersion models.
In Euler-Lagrange approach, as soon as the particles comes in contact with wall, this is
determined by first determining whether particle has entered the wall cell, if yes the a
particle distance from the cell wall face is calculated if this is less than half of particle
diameter, particle is considered to have crossed the wall and is removed from the
calculation. This approach is called as trap boundary condition. This is a reasonable
assumption since the airway walls are covered with sticky mucus. However in the present
thesis, the movement of the thin mucus layer is not modelled.

Re p

a1

a2

a3

<0.1

24

0.1<1.0

3.69

22.73

0.0903

1<10

1.222

29.1667

-3.8889

10<100

0.6167

46.5

-116.67

100<1000

0.3644

98.33

-2778

1000<5000

0.357

148.62

-4.75

5000<10000

0.46

-490.546

57.87

10000<50000

0.5191

-1662.5

5.4167

Table 3.1, Coefficients used in Morsi and Alexander [83]

51

3.5.4 Eddy interaction model

Within the framework of RANS the need for a stochastic approach arises due to the
unavailability of statistics relating to the flow fluctuation. There are several stochastic
dispersion models available in literature, few examples include Langevin dispersion
model [84], random Fourier modes [51] and pdf models [52]. However, the most
frequently used model is the eddy interaction model from Gosman and Ioannides [54].
The eddy interaction model (EIM) is a stochastic random walk treatment in which
particles are made to interact with the instantaneous velocity field u , where u is
represented similar to RANS decomposition
u u u

(3.65)

Here, u is the mean fluid phase velocity vector and u is the fluctuating velocity vector.
By computing the trajectories of a large enough number of particles, the effects of the
fluctuating flow field can be taken into account. The heart of EIM lies at reconstructing
the instantaneous field from the local mean values of velocity and turbulent intensity.
Assuming isotropic turbulence, we have

u2 v2 w2

2
k
3

(3.66)

Here, k is the turbulent kinetic energy determined through a turbulence model such as
k , k etc. Furthermore, in EIM at a given particle position x p , a hypothetical

eddy is considered, with the particle being assumed to be located at the center of the
eddy. Following Gosman and Ioannides [54], the velocity fluctuation in Cartesian
components of this hypothetical eddy is expressed as
u' N r

2
2
2
k , v' N r
k , w' N r
k
3
3
3

52

(3.67)

Here, N r is a random number generator drawn from a normal probability distribution

having zero mean and unit standard deviation. The second important aspect of EIM is to
determine the time step t required for numerical integration of equation(3.62). A
typical way of estimating this, is by estimating the particle-eddy interaction time tint ,
given as
(3.68)

tint min(te , tcros )

Here, te and tcross are respectively, the eddy life time and particle crossing time through
the eddy. When using the k model, te and tcross are estimated as

te

le
2k

(3.69)

with le , the characteristic turbulent length scale, defined as

le C3/4

k 3/2

(3.70)

where is the specific dissipation. For a k turbulence model is estimated as k .

The particle crossing time through the eddy is estimated as

le
tcross r ln 1
r u up

(3.71)

(3.72)

Here, tcross is the particle crossing time in the eddy [56]

le
tcross r ln 1
r u up

53

where

u up

is the magnitude of slip velocity. In circumstances where

le / r u up 1 , equation(3.72) has no solution, in such a case tint te . Finally, using

an implicit Euler discretization scheme, one obtains the particle velocity at time level
n 1 as follows,

unp 1

unp un t / r
1 t / r

(3.73)

In order to update the particle position, a trapezoidal discretization scheme is used. The
relation for the particle position is given by
x np1 x np 0.5t unp1 upn

54

(3.74)

Chapter 4
Particle deposition in an extrathoracic
airway using RANS
Contents
4.1 Introduction............................................................................................................................................55
4.2 Mathematical background....................................................................................................................57
4.2.1 EIM and correction function..........................................................................................................59
4.2.2 Isotropic EIM...................................................................................................................................59
4.2.3 Wang and James EIM.....................................................................................................................59
4.2.4 Helicity EIM.....................................................................................................................................60
4.3 Results and discussion..........................................................................................................................63
4.3.1 Test Geometry 1, 90 degree bend, Re=10000...............................................................................63
4.3.2Test Geometry 2, Simplified human upper airway model..........................................................66
4.4 Conclusion..............................................................................................................................................72

4.1 Introduction
Quantifying the local and the total deposition fraction of the inhaled aerosols in a human
extrathoracic airway is essential for the development of effective drug delivery systems.
The complex extrathoracic airway (comprising of oral cavity, pharynx, larynx and
trachea) acts as an effective filter which limits the amount of inhaled aerosols that enter
the intrathoracic airways. The transport and deposition of these aerosols in the oral
airway present a significant health risk considering that these particles carry a large dose,
have a high probability for impaction and may generate large local regions of enhanced
particle deposition, referred to as hot spot. Such quantifications pose some serious
challenges for both experimental and numerical studies. From the numerical studies
point of view, the most important are, a) airway geometry used, and b) mathematical
equations involved with the modeling of air and particle transport.
55

The extensive literature available on the numerical studies in extrathoracic airways

mostly consists of applying RANS which requires: a) an adequate choice of turbulence
model, and b) a model to account for the effect of turbulence on the particles. Based on
their ability to accurately predict both laminar and turbulent regions which exists in the
extrathoracic region, previous studies recommended the use of k , and Menter [69]
SST k turbulence model. In the framework of RANS, however, accounting the effect
of turbulence on the particles remains a problematic area. In general there exist, two
methods to account this effect, namely, a) eddy interaction model (EIM), originally
proposed by Gosman and Ioannides [54] (see chapter 3), b) continuous random walk
(CRW) model based on solving Langevin dispersion equation. While CRW has been
successfully applied to a similar extrathoracic airway by Debhi [85], the classical
isotropic EIM still remains the most widely used model (Graham & James [56], Graham
[58], Zhang et al. [60], Kleinstreuer & Zhang [15], Matida et al. [10], Jayaraju et al. [44],
Sandeau et al. [62], Verbanck et al. [45]). However, the major drawback with the
classical EIM is that it assumes fluctuating velocities to be isotropic, an assumption that
is inappropriate, especially near the walls where strong anisotropic turbulent structures
develop (Kallio and Reeks [86]). When EIM with isotropic velocity fluctuations was
considered, resulting simulations largely overpredicted aerosol deposition especially for
particle sizes in the range 1-5 m, the so-called respirable range (Matida, Nishino, &
Torii, [87], Zhang, Finlay, & Matida, 2004 [64]). Hence, many authors have employed
large eddy simulation (LES) in an attempt to obtain a more accurate prediction of aerosol
deposition in the human upper airway (Matida, Finlay, Breuer, & Lange [46], Jayaraju,
Brouns, Lacor, Belkasem and Verbanck [44]). While LES coupled with Lagrangian
particle tracking indeed greatly improves the simulation of experimental deposition
efficiency, the method is time consuming. Therefore, RANS together with an appropriate
turbulence model and EIM still remains the most widely used method. In order to
overcome the drawbacks of classical EIM assuming isotropy, Wang & James [61] have
proposed correction functions for velocity fluctuations obtained by curve fitting direct
numerical simulation (DNS) duct flow data of Kim et al. [88] and Mansour et al. [89]. As
indicated by Wang and James these correction functions are valid for y 80 . In
56

addition, Wang and James [61] pointed out that, these functions are applicable only for
flows with a low Reynolds number against which they are calibrated. For example,
Matida et al. [10] calibrated the range of y 20 for a flow rate of 90 l/min and y 1
for a flow rate of 30 l/min. While such correction functions do improve the simulation of
particle deposition compared to classical EIM assuming isotropy, the main shortcoming
of this approach is that the correction functions were derived for ducted flow geometry.
Finally, it critically depends on the wall distance which is usually difficult to estimate in
complex geometries such as the human upper airway. The main contribution of this work
was to avoid these limitations by using an EIM methodology (further referred to as
helicity EIM) which incorporates anisotropy but without any a priori estimation of y+.
In the present work, we apply the classical EIM from Gosman and Ioannides [54]
(isotropic EIM), EIM with correction functions from Wang and James [61] (Wang
and James EIM) and helicity EIM to two test geometries, a) 900 bend pipe at
Re=10000, b) human upper air way model at a steady inhalation rate of 30 l/min and 60
l/min. The calculated numerical results were compared against the experimental results of
Pui et al. [90] and Verbanck et al. [45] for both test cases, respectively. The fluid phase
and the particle phase were solved numerically using FLUENT 13.0 solver; the helicity
EIM and Wang and James EIM was implemented in FLUENT 13.0 using userdefined functions.

4.2 Mathematical background

The Reynolds number based on the inlet diameter in the case for UAM ranges from 2500
for 30 l/min and 5000 for 60 l/min) i.e. there exists laminar and turbulence flow regimes
in UAM (explained in detail in S T Jayaraju [91]). At this Reynolds number range as
reported by Wilcox [38], and Pope [39], the standard turbulence models (for e.g. k-,
RNG k etc.) fails to predict laminar to turbulence transition. Therefore, under such
conditions low-Reynolds-number (LRN) turbulence models have to be considered. The
fluid phase is simulated by numerically solving time-averaged RANS equations
57

employing SST k- turbulence model (Menter [69]) with low-Re corrections; the
mathematical formulation for turbulence modeling is not reiterated here (see Chapter 3).
The choice of the turbulence model is based on its ability for better handling transitional
flows as is the case in mouth-throat geometries compared to other widely used k-
turbulence model (Matida et al. [10], Zhang et al. [92], Xi and Longest [93], S T Jayaraju
[91]). Also for the sake of maintaining consistency we have kept the turbulence model
same for the both test cases discussed below.
The particle phase is solved employing Lagrangian particle equation of motion, described
in Chapter 3 and rewritten in non-dimensional form as below
dup
dt

up

Stk

g
Fr

(4.1)

where Stk and Fr are dimensionless Stokes and Froude number. The parameter Fr gives
the importance of inertial deposition versus deposition due to gravity and is given as,

Fr

U 02
Lg

(4.2)

where U 0 and L are the characteristic velocity and length, respectively. For the present
work Fr varies approximately from 50 in case of 30 l/min and 100 in case of 60 l/min,
implying that deposition by inertial impaction will be dominant. Moreover, considering
the high velocities inside human UAM (the velocity of the glottal jet reaches around 10
m/sec for 60 l/min) especially, in the extra-thoracic part, the gravity and Brownian force
in general have a very marginal effect on micron particle deposition. As such, the main
deposition mechanism in the upper airway is the deposition by inertial impaction (See
Finlay [49]). Following the above reasoning many authors like Longest and Xi [94],
Robinson et al. [95], Longest et al. [96] etc. ignore the gravity term. Therefore, in the
present chapter the gravity is ignored in equation 3.61. The volume fraction, defined as

58

the ratio of volume occupied by all particles and that occupied by the fluid, is of the order
of 1e-7 i.e. the heterogeneous system of fluid-particle is a dilute suspension.

4.2.1 EIM and correction functions

As described in chapter 3, in EIM, the effect of turbulence on the particle motion is
introduced by decomposing u in equation (3.51) as u N r u , where u , is the mean flow
velocity vector, N r , a Gaussian random number with zero mean and unit standard
deviation, and u the fluctuation due to turbulence which needs to be modeled. The
turbulence fluctuation u is obtained from the computed turbulent kinetic energy, k of
the fluid-phase.

4.2.2 Isotropic EIM

In terms of Cartesian components, the velocity fluctuations are modeled based on
equation (3.57), reiterated as below
u' N r

2
2
2
k , v' N r
k , w' N r
k
3
3
3

(4.3)

4.2.3 Wang and James EIM

Wang and James [61] introduced three correction functions namely, f u , f v and f w . To
account for the anisotropic nature of the velocity fluctuations close to the wall. The
correction functions, f u , f v and f w are respectively in the streamwise, wall normal and
spanwise direction, and were obtained by curve fitting DNS data of Kim et al. [88] and
Mansour et al. [89] as a function of y . The velocity fluctuations in Cartesian component
form can again be expressed as,

u' fu N r

2
2
2
k , v' f v N r
k , w' f w N r
k
3
3
3

59

(4.4)

Here
f u 1 0.285( y 6)exp( 0.455( y 6) 0.53 )
f v 1 exp( 0.02 y )

(4.5)

f w 3 f u2 f v2

These correction functions are valid for y 80 . As already mentioned by Wang and
James (1999), the functions are applicable for flows with a low Reynolds number against
which they are calibrated. For example, Matida et al. [10] calibrated the range of y 20
for a flow rate of 90 l/min and y 1 for a flow rate of 30 l/min. The applicability of
above equation is restricted to cases where the flow essentially remains 1D for e.g.
channel flow, pipe flow etc. For cases that involve recirculation regions, flow separation
the use of equation (4.5) is questionable. Moreover, equation (4.5) is derived using
turbulent statistics of channel flow will be different from for e.g. turbulent statistics in a
duct flow. Therefore, equation (4.5) should be used with caution since the functions
provide an ad-hoc technique to account anisotropy which doesnt have any physical
meaning.

4.2.4 Helicity EIM

The notion behind this EIM correction methodology has been proposed very recently by
Ghorbaniasl et al. [97], where the correction functions were used to estimate subgridscale stress coefficients. Even though the correction functions were meant for LES
applications, in the present work, we use the same notion to account for the anisotropic
nature of turbulent velocity fluctuations close to the wall via helicity. In fact, it modifies
the velocity fluctuations based on the flow field data obtained from RANS, considering
that the velocity fluctuations should go to zero on the wall and in the region where the
flow is laminar. In particular, the correction functions have the property of giving zero
values, both in the laminar region and on the wall in each direction, as opposed to the
damping functions introduced by Wang and James [61] that give zero values only on the
60

wall and in the wall normal direction. Since the correction functions fu and f w do not
the guarantee zero values on the wall for fluctuation velocity components u and w , this
can affect the particle deposition characteristics close to the wall via equation (4.1). In
general, it is believed that the deposition of particles on the wall is mainly affected by the
wall normal velocity fluctuation; as such there can be no observable differences in the
particles deposition efficiency, but there can be differences in the local variation of
particle deposition. In the helicity EIM, the correction functions do not depend on the
estimation of the wall distance parameter ( y ). The flow field is the only input that is
being used in these functions. Furthermore, the introduced functions are such that, when
applied to the velocity fluctuations, the turbulent kinetic energy k remains unchanged.
We shall model the velocity fluctuations as follows

u ' bGx N r

2
k,
3

v ' bG y N r

2
k,
3

w' bGz N r

2
k
3

(4.6)

In equation, bGx , bG y and bG z are the correction functions, where from Ghorbaniasl et al.
[97], one has

Gx

Gy
Gz

0.5 H x
H x2 H y2 H z2
0.5 H y
H x2 H y2 H z2
0.5 H z
2
x

H H y2 H z2

61

(4.7)

Here, the value of the term b is set to be 2 3 such that the computed turbulent kinetic
energy remains unchanged, the terms H x , H y and Hz are evaluated as,

H x x u,
H y y v,

(4.8)

H z z w

and symbol

denotes the absolute value. In equation(4.8), u , v and w are the mean

flow velocity components in x, y, and z directions, respectively. The terms x , y and z

are the x-, y- and z-component of the vorticity vector, defined as
u

(4.9)

Using these correction functions, zero values for velocity fluctuations are guaranteed on
the walls and in laminar regions. As can be seen from equation (4.6)-(4.9), the accuracy
of thehelicityEIM in predicting near wall fluctuations depends on the accurate
resolution of the velocity field especially in the vicinity of the wall. As such it becomes
important to have adequate cells near the walls so that the first cell adjacent to wall is in
the viscous sub-layer. By using these correction functions, zero values for velocity
fluctuations are guaranteed on the walls and in laminar regions. However, some obvious
limits can now be verified, in the limit when H x , H y and H z approaches zero or small
value, the terms Gx , G y and Gz reach a constant value of 0.28 and we get back the
original form of isotropic EIM. It should however be mentioned that in the case where
H x , H y and H z are all equal to zero equation (4.7) is ill defined such a condition is

overcome by putting a very small number (for e.g. 1e-30) at the denominator.

62

4.3 Results and discussion

In this section, two test geometries are considered and the experimental data of aerosol
deposition efficiency is compared with the results obtained by the classical EIM
(isotropic EIM), the correction method proposed by Wang and James [61] (Wang and
James EIM), and the present methodology (helicity EIM).
First, the flow is simulated by employing RANS with SST k- turbulence model
(described in Chapter 3) and then particle trajectories are calculated as described in
section 4.2. Second-order upwind scheme for momentum and k- equation were
employed. Semi implicit pressure linked equation (SIMPLE) was used for pressurevelocity equation. The particle equation of motion is discretized as mentioned in Chapter
3.

4.3.1 Test geometry 1, 90 degree bend, Re=10000

The computed deposition efficiencies are compared to the experimental results of Pui et
al. [90] for a 90 degree bend (Figure 4.1) with a diameter, 8.51 mm and a radius of
curvature, RO 5.6 mm. The test geometry was discretized into 0.6 million hexahedral
cells with a near wall clustering of elements with a stretching ratio of 1.1. The y value of
the first layer of cells next to the wall is about 1. The inlet section of the test bend has an
entrance length, le 4.4 Re1/6 so as to have a fully developed turbulent velocity
profile at the entrance to the 90 degree bend. At the inlet a steady top-hat/uniform
velocity is imposed and at the outlet an outflow/zero-gradient boundary condition is
imposed. The specification for k and at the inlet is given by using well known rule of
thumb (M Brouns [98], Ertbruggen [99]),
3
( I inlet uinlet ) 2
2
0.25
k 0.5
C
(0.1d h )
k

63

(4.10)

where I inlet is the inlet turbulent intensity which is consider as 5%, uinlet is the inlet
velocity magnitude, constant C 0.09 and d h is the hydraulic diameter. Using I inlet as
either 5% or 10% results in negligible change in the velocity or the turbulent kinetic
energy profile at entrance to the bend.

Figure 4.1, 90 degree test bend. (left) Geometry details; (right) mesh cut plane at middle of the
bend.

In the experiments, it is assumed that Re=10000 and Stokes number (Stk) varies in the
range 0.25-1.35, where Stk cc p d p2U in / 18 r

with cc as a slip correction factor

(considered=1, since in present simulations d p 1 m ), p the particle density and d p the

particle diameter. Here U in is the inlet velocity magnitude, the dynamic viscosity and r
the tube radius. In the experiments Pui et al. [90] used oleic acid aerosols with air, where
p / 755 and p 895 kg/m3 . In the simulations, the same material properties are

adopted and Stokes range of 0.05 Stk 1.5 is considered. The total number of particles
injected at the inlet is around 5000. However, it is to be mentioned that using 5000 or
1000 particles resulted in less than 3% change in the particle deposition efficiency. The
injection file for each of the particle size considered is generated in Matlab by uniformly
distributing injection points at the inlet. Figure 4.2, shows randomly distributed injection
points for a typical particle size. The particles were assumed to be carried by the flow i.e.
initial particle velocities were set equal to the inlet flow velocity.

64

Figure 4.2, Particle profile at the inlet

Figure 4.3, Total deposition efficiency. (connected symbols) - simulations; (disconnected

symbols) experiments; (left) exploded view

Figure 4.3 shows total deposition efficiency defined as the ratio of number of particles
deposited to the total number of particles injected at the inlet. As illustrated the total
deposition efficiency is overpredicted by isotropic EIM, in particular for the Stk range
below 0.5. By contrast, the total deposition efficiencies simulated by Wang and James
EIM and the helicity EIM methods are in good agreement with the experimental curve
in the entire Stokes range considered. Interestingly, the latter two methods also show a
65

good agreement with the LES result of Breuer et al. [100]. With the latter being more
close to the experiments. At first glance, both Wang and James EIM and helicity
EIM looks giving identical results, in fact Wang and James EIM gives slightly better
results but the difference with helicity EIM is very marginal.

4.3.2 Test geometry 2, Simplified human upper airway model

The CT based simplified UAM model is created by preserving all the critical geometrical
features based on the work of Brouns et al. [23] (see Figure 4.4). The simulations were
performed for mono-disperse particle sizes 2m, 3m, 4m, 6m, 8m and 10m having
particle density 912 kg/m3 at breathing rates of 30 and 60 l/min. For the fluid phase, air at
a density of 1.2 kg/m3 and dynamic viscoity 1.884e 5 kg/m-s was considered. The
performance of each of the aforementioned EIM models is based on evaluating total
deposition efficiency defined as the ratio of number of particles deposited to the total
number of particles injected. In addition to that we also consider the particle deposition in
four UAM compartments, - (a) oral, (b) pharynx, (c) larynx, and (d) trachea (Figure 4.4).
This is defined as relative deposition efficiency, defined as the ratio particles deposited in
a particular compartment to the total number of particle deposition in UAM.

Figure 4.4, Simplified UAM model and respective compartments used in UAM

66

The present UAM model is discretized into 0.7 10 6 , 1.2 10 6 and 2.5 10 6 hexahedral
elements with a near wall clustering of elements and a stretching ratio of 1.1. The y+
value of the first layer of cells next to the wall is about 1. Figure 4.5 shows the mesh
corresponding to mesh of 1.2x106 details at different section planes.

Figure 4.5, Mesh at different section planes in UAM

The mesh requirement for the present test case has already been validated in the works of
Brouns [98] and Jayaraju [91]. Here we briefly presents normalized mean velocity
magnitude comparision from the three meshes plotted at four sections of the sagittal
plane namely, five mm above epiglottis and one, two and three thracheal diameters
downstream of glottis, are shown in Figure 4.6. Based on the comparison it can be infered
that the flow solution remain unchanged with the use of 2.5 10 6 or 1.2 10 6 cells.
For the fluid phase, we impose a uniform/top-hat velcoity at inlet and outflow/zero
gradient boundary condition at the outlet. For tubulent kinetic energy and specific
dissipation we use aforemtioned relations given in (4.10). For the particle phase, about
10,000 randomly distributed particles are injected at the inlet of the UAM model
correspeonding to each particle diameter (in the same way as done for bend test case).
The initial velocity of the particles is assumed to be the same as the fluid velocity at the
67

inlet, i.e., particles are carried by the flow. Since the airway walls are in general wet, it is
assumed that a particle is considered deposited as soon as it touches the wall. Using either
1.2 10 6 or 2.5 10 6 hexahedral cells results in less than 2% change for all particle

diameter. Based on the above convergence study we have chosen a mesh of size 1.2 10 6
cells for the rest of our numerical experiments.

(a)

(b)

c)

(d)

Figure 4.6, Comparison of normalized velocity component for different mesh counts. Section (a),
(b), (c) and (d) corresponds to five mm above epiglottis and one, two and three tracheal diameters
downstream of glottis, respectively

68

Further, the turbulence intensity level at the inlet is set at 5% for all simulations;
alternatively using turbulence intensity levels of either 5% or 10% resulted in less than
2% change in computed deposition efficiencies.
Figure 4.7, shows the total deposition efficiency simulation for 30 l/min (left panel) and
60 l/min (right panel) as a function of Stk Re 0.37 as proposed by Grgic et al. [101], where
Stokes (Stk) and Reynolds (Re) numbers are defined as,
p d p2U m
18 d m
U m d m
Re

Stk

(4.11)

In this equation, and p are the fluid phase and particle density, d m is the mean
hydraulic diameter U m is the mean velocity and is the dynamic viscosity of the fluid
phase. The terms d m and U m are defined as,
V
L
Q
Um
LV
dm 2

(4.12)

where Q is the flow rate, L is the mean cast length (326mm) and V is the cast volume
(90.6ml).
The empirical curve of Grgic et al. [101] (dashed line in Figure 4.7) was derived from a
representative set of experimental data for inspiration in different upper airway
geometries, and is shown alongside the experimental data points obtained from our
particular UAM cast (solid squares, corresponding to 3 and 6 m).

69

Figure 4.7, Inspiratory total deposition efficiencies, (left) 30 l/min and (right) 60 l/min; Particle
sizes range from 2 m-10 m; (connected symbols)-simulations; (disconnected symbols)experiments

As expected (Jayaraju et al. [44], Verbanck et al. [45]), the isotropic EIM simulation
largely over predicts the total deposition over the entire Stk Re0.37 range considered The
extent of over prediction for 60 l/min is even greater than for 30 l/min for corresponding

Stk Re0.37 values. In contrast, the helicity EIM and Wang and James EIM method
led to markedly lower total deposition values, bringing the entire deposition curve much
closer to the LES of Jayaraju et al. [44], empirical curve of Grgic et al. [101] and to the
experimental data points obtained in this UAM (solid squares), certainly in the low

Stk Re0.37 range. When actually comparing the helicity EIM simulations with the
experimental data obtained in the UAM (solid squares) the remaining discrepancy is
mostly marked for the largest particle size (6m). While absolute differences in case of
6m are rather large (18% vs 36% for 30l/min and 50% vs 80% for 60 l/min), it should
be borne in mind that these data points are on a very steep part of the deposition curve.

70

Figure 4.8, Inspiratory relative deposition efficiencies, (upper) 30 l/min and (lower) 60 l/min;
(left) 3m and (right) 6m; (connected symbols) simulations; (disconnected solid squares,experiments)

In Figure 4.8, we consider the relative deposition patterns within the UAM
compartments, (a) oral, (b) pharynx, (c) larynx, and (d) trachea, i.e., deposition in each
compartment relative to total deposition for each condition.

Relative deposition

efficiencies are obtained by normalizing the total number of particles deposited in a given
UAM compartment with the total number of particle deposited in the entire UAM. At
lower flow rate (upper panels) the relative deposition trend in experiments shows that one
has a subsequent decrease in relative deposition efficiency moving from oral
compartment to larynx and then a slight increase in trachea. This trend is reproduced by
71

Wang and James EIM and helicity EIM but not by isotropic EIM. In the case of
isotropic EIM for 3m (upper left panel) the relative deposition first decreases from
oral to pharynx and then subsequently increases. This is mainly because we have lower
deposition in oral and pharynx compartment as such more is available for latter
compartments. For the case of 6m (upper right panel), the relative deposition first
decreases from oral to pharynx remains almost same in larynx and subsequently
increases. At higher flow rate (lower panels) the isotropic EIM is able to predict
relative deposition trends observed during experiments better than at lower flow rate
(upper panels). In contrast, both Wang and James EIM and helicity EIM are able to
predict the experimental deposition patterns better than isotropic EIM at both the flow
rates, with both the models giving comparable results.
Altogether, the total and the relative deposition efficiencies in the UAM are simulated
more accurately by the helicity EIM and the Wang and James EIM methods with
isotropic EIM greatly overpredicting total deposition efficiency with relative deposition
trends comparable to experiments only at higher flow rate. The advantage of the helicity
EIM in comparison with the Wang and James EIM is that it does not require any
estimation of y+, which in complicated geometries like UAM is troublesome.

4.4 Conclusions
In the framework of RANS coupled with EIM for studying the particle deposition, an
anisotropic eddy interaction model for the particle deposition has been proposed. The
capacity of the model has been investigated through two test cases, a) 90 degree bend
pipe and b) simplified UAM. The proposed model was compared against the existing
classical EIM and that of Wang and James EIM.
Consistent with the observations made in literature the isotropic assumption involved
with the classical EIM resulted in an overprediction of total deposition efficiency in the
both test cases for the entire stokes range considered. The only case where the results
72

from the classical EIM seem to be comparable with experiments is the relative deposition
efficiency plot for second test case at 60 l/min. But the total deposition efficiency was
greatly overpredicted. In contrast, using anisotropic model of Wang and James EIM
and the proposed EIM model resulted in the reduction of total deposition efficiency and
the results compared well with existing literature.
In the both test cases, Wang and James EIM was seen giving slightly better particle
results than the proposed EIM, however, as discussed before, the damping functions used
in Wang and James EIM is questionable. Especially in cases that involve recirculation
zones, flow separation and where both laminar and turbulent regions exist. Also, the
method requires a priori estimation of y which is not known beforehand. As such the
method utilizes an adhoc approach for accounting flow anisotropies in particle
calculations.
Unlike Wang and James EIM, the proposed EIM do not uses any adhoc approach and
accounts flow anisotropy directly from the calculated flow field. The results obtained
from the proposed method compared well not only in terms of total deposition efficiency
but also in terms of relative deposition patterns. In fact, in the UAM for particle sizes less
4m, the results were very close to experiments for both flow rates. As discussed, the
accuracy of the proposed model is dependent on the computed flow-field, which will be
studied in more details in the following chapter.

73

74

Chapter 5
Performance of helicity eddy interaction
model
Contents
5.1 Introduction.75
5.2 Simulation condition...76
5.3 Results and discussion....77
5.3.1 Inspiration..........................................................................................................................................77
5.3.2 Expiration...........................................................................................................................................81
5.3.3 Influence of inflow condition and flow field data...........................................................................85
5.4 Conclusion...............................................................................................................................................95

5.1 Introduction
The previous chapter aimed at introducing the helicity EIM model in general. It was
discussed that the helicity EIM, by the way it is formulated, is heavily dependent on the
accurate representation of the flow field. This in turn depends upon two upstream flow
conditions, a) inlet velocity profile and initial particle distribution, b) turbulence model
used. For example, Longest and Vinchurkar [102] showed that the validation of
fluid/particle simulation against experimental data is crucially dependent on inlet
conditions, i.e., inlet velocity profile and initial particle distribution profile.
Therefore, in this chapter the quality and capability of the helicity EIM is further studied
model through a sensitivity analysis of the model to the turbulence model and to the
upstream flow conditions. We will include the behavior of the model upon the particle
sizes when the flow configuration is reversed i.e. during the expiration breathing phase. It
is shown experimentally (Verbanck et al. [45]) that for particle sizes 3-6 m, total
75

deposition efficiency in an upper airway model is similar during the inspiratory and
expiratory breathing phase, but the local aerosol deposition patterns for both flow
directions are completely different. During inspiration, the aerosol preferentially deposits
in the oral compartment, while during expiration, local aerosol concentration is highest in
the pharyngeal compartment.
In the present work, we will apply the helicity EIM methodology to the prediction of total
and local deposition patterns during both the inspiratory and expiratory breathing phase.
In particular, we will investigate whether the preferential deposition in the pharynx
region during expiration, which could not be reproduced at all by isotropic EIM, can be
more reliably predicted using helicity EIM. The influence of the inlet boundary
conditions corresponding to inspiratory and expiratory deposition measurements will also
be scrutinized. In particular, the influence of top-hat/uniform inlet velocity and parabolic
inlet velocity profiles on aerosol deposition will be assessed. The choice of considering
top-hat and parabolic inlet velocity profile for numerical experiments is related to the
experimental set-up (schematic shown in Figure 5.1) employed for conducting particle
deposition experiments. It can be seen that inlet to the UAM has additional tube
connectors and we assume flow to be laminar before entering the UAM. The actual
velocity profile is not available because in experiments an additional facility to measure
velocity profile was not in the scope. Finally, we will briefly investigate the dependence
of helicity EIM model upon the turbulence model used. This dependence will be shown
by comparing results from SST k- model and Std. k- model.

5.2 Simulation condition

The mathematical equations used for modeling flow-particle transport are as described in
the previous Chapter 4. The air and particle material properties also remain the same. All
the numerical experiments are conducted on the mesh of 1.2 million hexahedral cells
which has shown to be adequate in the grid refinement study of the previous chapter. In
the subsequent subsections first the comparison between particle simulations and
76

experimental data will be demonstrated for 3 and 6 m, and then results for rest particle
sizes (2, 4, 8 and 10 m) will be demonstrated.

Figure 5.1, Schematic of Aerosol deposition experiments, Verbanck et al. [45]

5.3 Results and discussion

5.3.1 Inspiration
Figure 5.2, shows the total deposition efficiency simulation for 30 l/min (left panel) and
60 l/min (right panel) as a function of Stk Re 0.37 as proposed by Grgic et al. [101]. In
this figure, the plots of isotropic EIM and helicity EIM using top-hat inlet velocity are the
same as in Figure 4.7, Chapter 4. Looking at the comparison, it can be inferred that using
either top-hat or parabolic inlet velocity profile result in almost similar deposition
efficiencies. With respect to the overall agreement, the effect of top hat versus parabolic
inlet profile is of minor importance for the helicity EIM simulations, except maybe for
the highest Stk Re 0.37 values considered here (corresponding to particle size 8-10 m
and 60 l/min).

77

Figure 5.2, Inspiratory total deposition efficiencies, (left) 30 l/min and (right) 60 l/min; Particle
sizes range from 2m-10m; (connected symbols) simulations; (disconnected symbols)experiments

In Figure 5.3, we consider the relative deposition patterns within the UAM
compartments, (a) oral, (b) pharynx, (c) larynx, and (d) trachea, i.e., deposition in each
compartment relative to total deposition for each condition. Considering top hat or
parabolic inlet profiles hardly affects these results. In fact, the influence of the inlet
profile was only noteworthy for helicity EIM simulations, and mainly so for the two
intermediate values, 3 m at 60 l/min (Figure 5.3; lower left panel) and 6 m at 30 l/min
(Figure 5.3; upper right panel). In both these cases, the top-hat inlet velocity profile
resulted in an overestimation of deposition in the pharyngeal compartment, at the expense
of that attributable in the oral compartment. Compared to the top-hat inlet profile, the
parabolic profile showed an increased relative deposition in the oral cavity, such that
relatively less was available for deposition in the pharynx, consistent with experiments.
In order to get a general overview on the effect of the EIM models and inlet profiles,
Figure 5.4 illustrates the relative deposition efficiencies for the rest particle size (2, 4, 8
and 10m) as predicted from particle simulation. As can be seen from these plots the
relative deposition trends and the effect of inlet profiles as observed for 3 and 6 m
(Figure 5.3) also holds true for other particle sizes. In summary of the inspiratory case,
the helicity EIM with parabolic inlet velocity profile gives a satisfactory agreement of
78

simulations with experiments in terms of total deposition (Figure 5.2) and of local
deposition patterns (Figure 5.3).

Figure 5.3, Inspiratory relative deposition efficiencies, (upper) 30 l/min and (lower) 60 l/min;
(left) 3m and (right) 6m; (connected symbols) simulations; (disconnected solid squares,experiments)

5.3.2 Expiration
Figure 5.5 shows the simulations corresponding to Figure 5.2, but for expiration instead
of inspiration. Experimental data points obtained for 3 and 6m particles at 30 l/min (left
panel) and 60 l/min (right panel) are also shown for comparison (solid squares), but
unlike for inspiration, no empirical curve exists for expiration. As was the case for
inspiration, isotropic EIM overestimates total deposition efficiency for 3 m particles at
79

both flow rates, while helicity EIM produces total DE values that are closer to the
experimental data. For the large particle size (6 m), experimental data are better
matched by the isotropic EIM, while helicity EIM consistently underestimates total
deposition. For both isotropic EIM and helicity EIM simulations, the parabolic inlet
velocity profile led to greater total deposition than the top-hat velocity profile, as was the
case for inspiration (Figure 5.2). However, the effect of inlet velocity profile was greater
in magnitude for expiration than for inspiration.

80

Figure 5.4, Inspiratory relative deposition efficiencies, (left) 30 l/min and (right) 60 l/min; All
panels top to bottom correspond to 2, 4, 8and 10 m

81

Figure 5.5, Expiratory total deposition efficiencies, (left) 30 l/min and (right) 60 l/min; Particle
sizes range from 2m-10m; (connected symbols)simulations; (disconnected square symbols)experiments

The impact of using a particular inlet velocity profile in association with a particular EIM
becomes clearer when examining local deposition patterns for expiration (Figure 5.6).
The helicity EIM associated with the parabolic inlet profile was the only modality which
mimicked the experimental observation of greater relative pharyngeal versus oral
deposition efficiency.
In summary of the expiratory case, it can be concluded that for the smaller particles (3
m) helicity EIM adequately simulates both total deposition and local deposition
patterns, if it is used in association with a parabolic inlet profile. This also holds for
relative distribution of the larger particles (6 m) but in this case the total deposition is
largely underestimated.

82

Figure 5.6, Expiratory relative deposition efficiencies, (upper) 30 l/min and (lower) 60 l/min;
(left) 3m and (right) 6m; (connected symbols)simulations; (disconnected square symbols)experiments

As was done for inspiratory flows, an overview of the general behavior of local
deposition for the remaining particle sizes is shown in Figure 5.7. Unlike the inspiratory
flows, the local deposition trend during expiratory flows varies depending on the particle
size and on the flow rate. In the case for 30 l/min flow rate, the preferential deposition as
observed for helicity EIM with a parabolic profile for 3 and 6m (Figure 5.6) exits only
till 8m. Similarly in the case of 60 l/min flow rate, this preferential deposition is
observed only up to 6 m. In the extra-thoracic region, the inertial impaction is the only
dominant mechanism by which particle deposition takes place (Finlay [49]).

83

Figure 5.7, Expiratory relative deposition efficiencies, (left) 30 l/min and (right) 60 l/min; All
panels top to bottom correspond to 2, 4, 8and 10 m; (connected symbols) simulations

84

Further Grgic et al. [101], demonstrated that in the extra-thoracic region the deposition is
influenced by both Stokes and Reynolds number. This implies that for a given Reynolds
number and particle density, the particle deposition is influenced by the Stokes number or
in other words the particle size. Therefore, the variation observed in the simulated local
deposition trend is attributed to the fact that, as the particle size and hence the particle
inertia increases, it becomes less sensitive to flow fluctuations. This will be demonstrated
in the following subsection where a comparison will be made between the helicity EIM
model prediction and the prediction using the mean flow i.e. not accounting flow field
fluctuations in the particle calculations.

5.3.3 Influence of inlet conditions and flow field data

In this section, first the effect of connector tubing (shown in Figure 5.8) in the particle
deposition will be accessed. This will be followed by a comparison between mean flow
and helicity EIM model predictions to access the dependence of higher particle sizes
towards flow fluctuations. Finally, the effect of flow field data on the helicity EIM model
will be shown.
Because of the remaining discrepancies between simulated and experimental results, the
effect of the connector tubing used in the aerosol deposition experiments (Verbanck et al.
[45]) was included in the simulations. With the UAM in the inspiratory mode, the piece
of tubing connecting the aerosol generator to the UAM leads to a constriction as drawn to
scale in Figure 5.8 (left panel). In this particular series of simulations, we impose a
steady top-hat/uniform velocity inlet boundary condition and a 5% turbulent intensity
level at the inlet of the connector. The particles are injected at the inlet in the same way
as was done at the UAM inlet without any connector tubing.

85

Figure 5.8, (left) UAM with connector tubing (shown in shaded area) at the inlet for inspiratory
breathing phase setup; (right) UAM with 900 elbow bend connector (shown in shaded area) at the
inlet for expiratory breathing phase setup. The additional mesh for the connector tubings are
(left) 0.4 x 106 and (right) 0.8 *106 hexahedral cells.

In the left panels of Figure 5.9, we compare for an inspiratory flow of 60 l/min, helicity
EIM simulation in the UAM with connector tube (crosses) to those in the UAM without
connector tube (triangles). In the latter case, we alternatively considered a top-hat (open
triangle) or a parabolic inlet profile (closed triangle); these two latter curves are the same
curves as in Figure 5.2 (right panel). Exactly the same is done for the expiratory mode
setup (Figure 5.8, right panel), where a 90 degree bend needed to be introduced
experimentally, in order to preserve the UAM orientation with respect to gravity. In the
right panels of Figure 5.9, we compare helicity EIM simulations for expiration in the
UAM with connector tubing (crosses), to that in the UAM without connector tube. Again,
in the latter case, a top-hat (open triangle) or parabolic (closed triangle) inlet profile was
considered (same curves as in Figure 5.5 right panel).

86

Figure 5.9, (top) Total deposition efficiency for particle sizes ranging from 2-10m at 60 l/min,
(bot-tom) relative deposition efficiency for 6m at 60 l/min. (left) inspiration, (right) expiration.

While Figure 5.2 and Figure 5.3 had shown that the inspiratory simulations in the UAM
without any connector tube were satisfactory when using helicity EIM with parabolic
inlet velocity profile, this conclusion still holds when validating this against the
simulations with connector tube used in experiments (Figure 5.9, left panels). In fact, the
inspiratory simulations without the connector tube were closest to those with the
connector tube (crosses) if a parabolic profile (closed triangle), as opposed to the top-hat
profile (open triangle), inlet condition was considered.
Figure 5.5 and Figure 5.6 had shown that the expiratory simulations in the UAM without
connector tubing were only satisfactory in terms of relative deposition when using
87

helicity EIM with parabolic inlet velocity profile. Total deposition was still
underestimated for the larger particle size (6m) with the parabolic inlet, but less so than
with the top-hat inlet profile. It is shown in Figure 5.9 (right panel) that simulations
without connector tubing and parabolic inlet condition are the ones most closely
resembling the simulations incorporating experimental connector tubing.
Overall, the addition of connector tubing on both in- and expiratory sides tends to
increase the deposition, as could be expected, especially at the higher flow rates. On the
inspiratory side, this can be attributed to a reduction (of 47%) in cross sectional lumen
area in the tubing connecting the aerosol generator to UAM inlet, increasing local
velocities. On the expiratory side, the bend and some additional constrictions at tube
transitions could be responsible. All things considered, the influence of connector tubing
was relatively small in terms of total deposition, for both in- and expiration (at least when
selecting the parabolic inlet profile when no connector tubing is considered). However, in
terms of relative deposition, the addition of the connector tubing clearly indicates that the
choice of the inlet profile is crucial for adequate simulations in the UAM without
connector tubing, particularly for expiration (Figure 5.9, right lower panel). It is shown
that the parabolic velocity profile is the inlet condition that leads to the experimentally
observed greater relative deposition in the pharynx, which cannot be obtained when
imposing a top-hat inlet condition.

88

Figure 5.10, Impact of top hat and parabolic inspiratory velocity profiles imposed either directly
at UAM inlet (upper and middle panels) or via connector tubing (bottom panel). Velocity
magnitude contours, flow streamlines in the central sagittal plane, and particle distribution
profiles at different UAM stations are shown for 60 l/min

89

Figure 5.10 illustrates for inspiration, the fluid dynamic differences between the UAM
and the UAM with connector tubing, when a top-hat inlet profile is considered in both
cases. It can be noted that, (i) the velocity profile at the UAM inlet is no longer top-hat,
(ii) the velocity at the core of the UAM inlet is greater in the presence of the connector
tubing (6.5 m/s versus 4.9 m/sec without connector), (iii) the streamline convergence due
to the constriction between the connector and the UAM inlet results in particles being
concentrated more in the core region where flow velocities are higher. Hence, the main
impact of the connector tubing in the case of inspiration is that it causes the flow
streamlines to converge rapidly, thereby increasing the velocity magnitude and forcing
the particles into a region with higher velocities. This explains the greater total deposition
(Figure 5.9; upper left panel) and enhanced relative deposition in the oral compartment
(Figure 5.9; lower left panel) for inspiration when top hat inlet velocities are used with
the connector (crosses) versus that directly at the UAM inlet (open triangles). Despite
some apparent differences in flow patterns depending on inlet velocity profiles, aerosol
patterns in crucial locations within the UAM are very similar across all conditions
considered (Figure 5.10). This explains the relatively small differences observed in
Figure 5.9 (left panels) across all conditions considered. This is not the case for
expiration. Figure 5.11 illustrates for expiration, the fluid dynamic differences between
the UAM (leftmost panel) and the UAM with connector (rightmost panel), when a top-hat
inlet profile is considered in both cases. It can be noted that,(i) even though the maximum
velocity magnitude of the jet originating at the glottis are similar, the jet in the UAM
without connector is able to more readily negotiate the sudden change in flow path and is
directed towards the oral compartment (in the UAM with connector, this jet is directed
towards the pharynx top, thereby causing an impingement of the flow on the wall) (ii)
there is marked difference in terms of the flow structures developing into the pharynx
compartment. (iii) due to the differences in flow fields, the particle distribution in section
a-a to d-d is quite different, with more particles being trapped in the pharynx top when
the UAM is considered with connector.

90

Figure 5.11, Impact of expiratory velocity profiles: (left and middle panel) are namely Top-Hat
and Parabolic without connector tubing; (right panel) Top-Hat with connector tubing. Velocity
magnitude contours, streamlines in the central sagittal plane, and particle distribution profiles at
different UAM stations are shown for 60 l/min.

Hence, the main impact of additional tubing on the UAM deposition simulations for
expiration is that the flow structure and particle distribution both contribute to an
91

increased relative deposition in the pharynx compartment. This explains the greater
relative deposition in the pharynx compartment (Figure 5.9; lower right panel) for UAM
with connector rather than without connector, when top hat inlet velocities are used in
both cases. When simulating particle deposition in the UAM without connector, the inlet
condition which best mimics the UAM with connector are the parabolic inlet profile
(closed triangle). Figure 5.11 (middle panel) indeed shows that in the corresponding
particle distribution (sections a-a to d-d) particles are concentrated closer to the
pharyngeal walls, and hence an enhanced pharyngeal deposition during expiration can be
obtained in an UAM without connector, if a parabolic profile at the UAM inlet is used.
In order to examine the sensitivity of the bigger size particle sizes (8 and 10 m) during
inspiratory and expiratory flows, a small numerical experiment was conducted. For this
particular numerical simulation, we include the connector tubing present in during
inspiratory and expiratory breathing phase. We compare the helicity EIM model
prediction with that of mean flow predictions i.e. particle tracking without any
fluctuations.

Figure 5.12, Total deposition efficiency; (left panel) Inspiration; (right panel) Expiration for 60
l/min

Figure 5.12 illustrates the total deposition efficiency for the case of 60 l/min during
inspiratory and expiratory breathing phase. On comparing the predicted total deposition
values between helicity EIM (indicated by crosses) and mean flow (indicated by white
92

squares), we can observe two interesting facts, a) For lower Stk Re 0.37 particles sizes, 26 m, the effect of fluctuations is clearly visible during both inspiratory and expiratory
flows. If one does not account the flow fluctuations in the particle simulation the total
deposition values are severely underpredicted. This underprediction greatly improves
when one account the fluctuation as can be seen from the helicity EIM (crosses), b) for
higher Stk Re 0.37 range (particle size, 8-10m), an interesting observation is that the
underprediction seen in the Mean flow predictions for lower particle sizes improves and
the results becomes closer to those of the helicity EIM. This observation implies that as
the particle sizes increases (and hence the inertia) the particle relaxation time also
increases and as such it becomes less sensitive to the flow fluctuations as expected form
physical arguments. Also, the similarity between the prediction from helicity EIM and
mean flow for bigger particle sizes shows the capability of the proposed model in
predicting the correct behavior i.e. particles with higher inertia should be less sensitive to
flow fluctuations. In order to see whether the proposed model helicity EIM also gives
correct behavior for local deposition for bigger particle sizes when compared against the
mean flow results, Figure 5.13 shows the local deposition for (8-10m) for inspiratory
and the expiratory case.
As seen from Figure 5.13, for the bigger particle sizes, the relatively deposition
efficiencies are similar between the helicity EIM and the mean flow particle simulations.
Thereby, giving more confidence to the proposed model as it correctly captures particle
physics i.e. bigger size particles are less sensitive to flow fluctuations.
The accuracy of helicity EIM model is directly related to the computation of flow field
data via three correction functions (Gx , G y , Gz ) (see equation 4.7-4.8). In the case of
RANS this inturn is dependent on the turbulence model used. As mentioned before, the
Reynolds number range that exits in the present UAM is in the transitional regime i.e.
both laminar and turbulence regions exists. Moreover, the flow field as seen in Figure
5.10 and Figure 5.11 contains recirculation regions; therefore a turbulence model should
be such which can capture such regimes accurately.
93

Figure 5.13, Relative deposition efficiencies,(top) 8m, (lower) 10m; (left) inspiration, (right)
Expiration

Figure 5.14, Dependence of helicity EIM on the turbulence model; (left) Inspiration, (right)
Expiration for 60l/min

94

As described in Chapter 2, in the context of RANS flows with recirculation and

transitional regimes are better captured by LRN SST k- turbulence model and Standard
k- performs the worst ( e.g. [39], [38]). For example, from the literature, (e.g. Stapleton
et al. [20] and Kleinstreuer et al. [15]), showed that the transitional flow behavior is better
predicted using the SST k- turbulence model rather than using standard k-. Therefore,
in the present section we show how the helicity EIM is dependent on the turbulence
model by comparing results from SST k- and Std k- model. This particular numerical
experiment is conducted with constrictions included in the geometry so as negate the
effect of upstream conditions. As illustrated in Figure 5.14 using Std. k- model greatly
underpredicts total deposition efficenciy than compared to SST k- turbulence model.
This is consistent with the observation made by Kleinstreuer et al. [15]. This also
suggests that the underprediction in total deposition data as observed during expiratory
flows (Figure 5.5 and Figure 5.9) may also be attributed to the fact that flow-field data is
not adequately predicted. Since we dont have the PIV measurements for expiratory flow
simulation in the present UAM, this cannot be confirmed with certainty.

5.4 Conclusion
In this chapter we have shown the dependence of previously proposed helicity EIM
model towards the computed flow-field by undertaking various numerical experiments.
The dependence was shown with respect to the upstream flow condition by using TopHat and Parabolic inlet velocity profile and later by adding physical constrictions that
were present during the experiments. We have also discussed the dependence of the
model with respect to the choice of turbulence model used. Additionally we have tested
the performance of helicity EIM on the UAM for a case where flow direction was
reversed i.e. during expiration.
In the case of inspiration, overall the use of top-hat or parabolic inlet velocity profile
resulted in almost similar total deposition efficiency using either isotropic EIM or helicity
EIM for 30 l/min and 60 l/min. However, in terms of relative deposition efficiency,
95

compared to the top-hat inlet velocity profile, the parabolic profile showed an increased
relative deposition in the oral cavity, such that relatively less was available for deposition
in the pharynx, consistent with experiments. The sensitivity is more for helicity EIM
than compared isotropic EIM and is observed either with lower flow rate (30 l/min) and
with bigger particle size or at higher flow rate and with smaller particle size.
The dependence of total deposition efficiency with respect to the choice of inlet velocity
profile was most marked in case for expiration, where using parabolic inlet velocity
profile improves the deposition results for both isotropic and helicity EIM model more
for particle sizes less than 6m. The same is reflected in relative deposition efficiency,
where the preferential deposition in pharynx as observed during experiments is mimicked
only when using parabolic inlet velocity profile and with helicity EIM. The isotropic EIM
performs the worst again giving confidence in the proposed helicity EIM model.
The addition of connector tubing on both inspiratory and expiratory sides was observed
to increase the deposition, more at the higher flow rates. On the inspiratory side, this can
be attributed to a reduction (of 47%) in cross sectional lumen area in the tubing
connecting the aerosol generator to UAM inlet, increasing local velocities. On the
expiratory side, the bend and some additional constrictions at tube transitions could be
responsible. All things considered, the influence of connector tubing was relatively small
in terms of total deposition, for both inspiratory and expiration (at least when selecting
the parabolic inlet profile when no connector tubing is considered). However, in terms of
relative deposition, the addition of the connector tubing clearly indicates that the choice
of the inlet profile is crucial for adequate simulations in the UAM without connector
tubing, particularly for expiration. Finally, the dependence with respect to the choice of
turbulence model was investigated and was shown as the accuracy of flow increases
prediction from helicity EIM improves.
In conclusion in this chapter, we have shown how comparative simulations with different
inlet conditions can lead to a reasonable choice of the inlet profile which most likely
represents that in the experimental aerosol deposition study. In particular, the preferential
deposition in the oral compartment during inspiration and preferential deposition in the
96

pharynx during expiration observed experimentally could be mimicked by the CFD

simulations in the UAM, provided a parabolic inlet profile was used. It can be concluded
that the helicity EIM combined with a proper choice of inlet velocity profile markedly
improves the quality of simulations for respirable particles in complex geometries such as
human upper airway

97

98

Chapter 6
Rotational based Smagorinsky model
Contents
6.1 Introduction.............................................................................................................................................99
6.2 Description of RoSM model.................................................................................................................101
6.3 Results and discussion..........................................................................................................................107
6.3.1 Fully developed turbulent channel flow......................................................................................108
6.3.2Fully developed turbulent flow in a square duct.........................................................................113
6.3.3Flow over a circular cylinder........................................................................................................117
6.3.4Application to a complex geometry..............................................................................................122
6.4 Conclusions ..........................................................................................................................................126

6.1 Introduction
It was seen in Chapter 5, prediction of particle deposition inside human airways is largely
dependent upon the prediction of flow. In case of RANS this is mainly dependent upon
the turbulence and EIM model used. In past few years the use of LES instead of RANS is
gaining popularity. This is mainly because of the availability of larger computing
capabilities. The use of large eddy simulation (LES) is now not limited to academic
geometries only. However, still some issue regarding the use of LES in particular to the
SGS model applied remains. This chapter therefore concerns with flow prediction using
LES and in particular implementation and validation of a new SGS model recently
proposed at VUB.
LES is a methodology wherein, the small scales and their interactions with the large
scales are modeled via the subgrid scale turbulent stress tensor. In regions where
turbulence is inhomogeneous and shear plays a dominant role, the influence of the
subgrid scale (SGS) model is critical. The flow can also be very sensitive to the presence
99

of SGS dissipation in regions where the flow is transitional. For this reason, much effort
has been put into the development of good SGS models. Consequently, a large number of
SGS models are available in the literature. Most of the SGS models are eddy viscosity
based models [74], [76], but also scale similarity models [103] and mixed models [104]
which combine the eddy viscosity and the scale similarity models, have been developed.
The most common approach to date is to use the eddy viscosity based models. The first
proposed eddy viscosity based model is the Smagorinsky model [74]. In the derivation of
this model, the small scales are assumed to be in equilibrium while they dissipate entirely
and instantaneously all the energy received from the large scales. As mentioned in
Chapter 3, following are the major deficiencies of Smagorisnky Model,

Firstly, Smagorinsky model cannot be represented correctly with a single

universal constant. Generally, the value of the model constant is optimized for
different turbulent flow fields.

Secondly, the model is excessively dissipative in laminar or high shear regions,

and its model constant must be decreased. In practice, for near wall flows, this is
successfully dealt with by using the van Driest damping function. However, using
the damping function reduces the subgrid eddy viscosity as a function of wallnormal distance, which is not straightforward to implement in complex
geometries and on unstructured grids.

In order to overcome the inherent deficiency involved with the value for the model
coefficient, the dynamic procedure proposed by Germano et al. [76] is used. This
procedure is too complex and expensive, becoming more difficult to implement when
unstructured grids are used. Another drawback of the dynamic procedure is its numerical
instability when the coefficient values become negative and/or its variation is high in
space and time, which needs some stability control algorithm.
In this work, we implement and evaluate an efficient formulation for determining the
Smagorinsky model coefficient developed at VUB which reduces the computational cost
significantly and is easy to implement on unstructured grids. The coefficient of the
100

subgrid scale model is a function of the flow solution, including the translational and
rotational velocity field contributions. Results of this approach will demonstrate that the
behavior of the model coefficient in practical simulations is actually far from constant.
The method allows vanishing eddy viscosity through the model coefficient vanishing at
walls, and in regions where the flow is laminar. The advantage of this method is that the
coefficient of the SGS model can be optimized without using the dynamic procedure
thereby saving almost 25% on computational cost. In addition, the method guarantees the
model coefficient to be always positive with low fluctuation in space and time. This
property avoids using the numerical stabilization procedure, which becomes very
complicated for complex flows.
The quality of the methodology will be investigated through the following test cases,
I) Fully developed channel flow with Reynolds numbers of 180 based on friction velocity
and channel half-width.
II) Fully developed flow through a rectangular duct of square cross section at Re 300 .
III) A smooth subcritical flow past a stationary circular cylinder, at a Reynolds number of
3900, where the wake is fully turbulent but the cylinder boundary layers remain laminar.
IV) Finally, the ability of the method will be tested in a CT-based simplified human
upper airway model (UAM), where the flow is transitional.

6.2 Description of RoSM model

The derivation for the RoSM SGS model starts with equation 3.43, re-written as below,

(6.1)

In order to determine the model coefficient Cs locally, based on the notion of

Ghorbaniasl et al. [97], a new hypothesis is laid out to determine Smagorinsky length
101

scale. In this hypothesis as pointed out by Ghorbaniasl et al. [97], the Smagorisnky length
scale is determined by evaluating three characteristic length scale lx , l y and lz associated
with the unresolved motions in x-, y- and z- directions in three dimensional flows. These
characteristic lengths are determined by specifying a characteristic velocity U scaled by
a variable with time dimension. The method is explained below and starts by defining
the length scale in x- direction, l x

lx U

(6.2)

The specification of the characteristic velocity U and is given by noting that, the
subgrid scale models in LES approximates the off diagonal component of Reynolds stress
term. This implies that the characteristic length of equation (6.1) can be assumed to be
proportional to the product of cross velocity terms. Here, for the cross velocity terms, we
shall involve the resolved translational velocity and the resolved rotation rate.
Figure 6.1 depicts the large (resolved) and the subgrid (unresolved) scales for a given
grid size Turbulence is generally visualized as consisting of eddies of different sizes that
overlap in space. The resolved unsteady eddies carry unresolved eddies. At a particular
position in the computational domain, the resolved eddy has a resolved translational
velocity u x , u y , u z and a resolved rotation rate x , y , z as shown in Figure 6.1.
Based on the background information of the resolved motion, we specify the
characteristic velocity U * of equation (6.2) as the cross velocity term associated with the

x -component of the translational velocity, u x . One obtains this cross velocity term from
the resolved rotation rate around the x axis as shown in Figure 6.1(b). It is seen that the
cross velocity term can be approximated as the product of the resolved rotation rate x
and the rotation radius / 2 . That is

U * x
102

(6.3)

where the symbol | | stands for the absolute value of the quantity and the resolved
rotation rate x is given by
1 u u
x z y
2 y
z

(6.4)

Now, the variable of equation (6.2) associated with the x direction based on the x component of the translational velocity, ux is specified as

ux
D

(6.5)

where D (with a dimension of m/s2) is a normalizing term (positive) used to be consistent

with the dimension of . Substituting relations (6.3) and (6.5) into equation (6.2) gives,
u

lx x x
2 D

(6.6)

Similarly, the characteristic lengths in the y - and z -directions associated with the
hypothetical eddy are expressed as,
uy

ly y
2 D

lz z z
2 D

(6.7)

1 u x uz

2 z
x
1 u
u
z y x
2 x
y

(6.8)

where

103

(a)

(b)
Figure 6.1, (a) Resolved and unresolved eddies in turbulent flow field. (b) Translational velocity
and rotation rate components in x, y and z directions.

Rearranging equation (6.6) and equation (6.7) yields

2D
lx xu x

2D
l y yu y

2D
lz z uz

(6.9)

g xux i y u y j z uz k

(6.10)

By defining the vector g as

104

Equation (6.9) can be written as follows,

lx
g

i
2D

ly
g

j
2D

lz
g

k
2D

(6.11)

In order to have a dimensionally consistent equation, the quantities on the left- and righthand-side of equation (6.11) should have the same units. Therefore, the normalizing term

D is defined as the norm of the vector g , i.e.

D g

u u u
x x

z z

(6.12)

Finally, the Smagorinsky length scale of the unresolved motions is specified based on the
length scale of the smallest resolved motion. Therefore, the Smagorinsky characteristic
length scale associated with the unresolved motion is taken as the minimum of l x , l y and
l z . That Is

L min l x , l y , l z

(6.13)

Applying the length scale defined in equation (6.13) to the eddy viscosity modeled by
Smagorinsky in equation (3.45), one obtains the eddy viscosity as follows,

t min l x , l y , lz

(6.14)

Further, using equation (6.11) one rewrites equation (6.14) in the following form,

t min Gx , G y , Gz

105

(6.15)

with
lx x ux

2D
yu y
l
Gy y

2D
z uz
l
Gz z

2D
Gx

(6.16)

Equation (6.15) together with equation (6.16) forms the developed subgrid scale model.
Clearly, this relation provides an expression which is easy to implement in an
unstructured NavierStokes solver. The rotation rate being an essential ingredient of this
model, it is referred as rotation rate based Smagorinsky model (RoSM).
When comparing the definition of t in equation (6.15) with the Smagorinsky model
given in equation 3.45, the term min G x , G y , Gz in equation (6.15) is nothing but the
Smagorinsky constant, Cs , i.e.

Cs min Gx , Gy , Gz

(6.17)

In principle, this equation provides an accurate formula which can be used to determine
the Smagorinsky constant. In addition, this relation is easy to use in practice, since
quantities such as the resolved translational velocity and the resolved rotation rate are
readily available in large eddy simulations. As can be seen, the RoSM model evaluates Cs

dynamically but without the need for any dynamic procedure. Also, there is no more need
for damping functions or other special near-wall treatments.
Some obvious limits can now be verified for the model coefficient Cs . First, from
equation (6.16) and (6.17) , we readily see that Cs 0 . Further, from equation (6.12) ,
one can see that the value of

Gx 2 G y 2 Gz 2 const. , therefore, equation (6.17) implies

106

that the maximum value of the coefficient Cs occurs when one has the following
condition,

Gx G y Gz

(6.18)

x u x y u y z uz a

(6.19)

In this case, from equation (6.16) one has

And

Cs ,max

a
2 3a 2

0.2886

(6.20)

Equation (6.20) is independent of the value of a, where a is a nonzero number. The only
case when Cs is not well conditioned numerically is at the walls, since u x , u y and uz
become zero simultaneously (i.e. a 0 ) resulting in a null denominator. This situation is
avoided by scaling up the denominator D by a very small number (epsilon). In this case,
Cs will be zero at the walls. As will be seen from the posteriori analysis, Cs value goes
to zero at the walls and the behavior of the model coefficient near the walls is comparable
with that of the dynamic Smagorinsky.
Overall, in the context of the present procedure, the model coefficient Cs varies between
0 and 0.2886.

6.3 Results and discussion

This section is concerned with the results for the three test cases chosen, turbulent
channel flow with two Reynolds numbers, turbulent flow in a square duct, and flow past
a cylinder. The solution of the first test case is obtained with a structured LES solver for
107

incompressible flows. In order to show the performance of the model with unstructured
methods, the model is also implemented in an unstructured solver. The solutions of the
last two test cases are obtained with the unstructured solver. It is worth adding that the
tendency to use LES in more advanced applications requires LES solvers on unstructured
grids, where the RoSM model is especially suited because of its simplicity, requiring no
test filters, nor distance to the wall.
In order to compare the behavior of the RoSM, additional LES simulations with the
conventional dynamic Smagorinsky model (DSM) were run for each test case. The
models are compared to DNS and/or experimental data for mean velocity profiles,
velocity fluctuations, and turbulence intensities.
In addition, we also run LES simulation without a SGS model (no model). It should be
mentioned that comparison of the results from the SGS models with that of no model
shows the effect of the SGS models on the flow simulations. Hence, the contribution of a
subgrid-scale model to the observed LES simulation quality can be rather incremental,
and conclusions on subgrid-scale-model quality may be misleading.

It has been

demonstrated that the convergence behavior of unresolved coarse-grid DNS is not

monotonic [105]. However, the study of this behavior is outside the scope of the current
thesis. In this chapter, comparison of the RoSM model with that of dynamic Smagorinsky
model is shown. Further the results are compared with coarse-grid DNS (no model) in
order to indicate the SGS model contributions to the resolved turbulent quantities.

6.3.1 Fully developed turbulent channel flow

In order to evaluate the subgrid scale model developed in inhomogeneous flows, a
rectangular channel 0, Lx , 0, Lz 3 shown in Figure 2 was chosen as a
test case. LES calculation was performed for a turbulent channel flow at a Re 180 .
Here, Re u / where u , and denote the friction velocity, the channel halfwidth and the kinematic viscosity, respectively. The computational domain is periodic
in the streamwise ( x ) and spanwise ( z ) directions with no-slip conditions at the channel
108

walls. A fixed mean pressure gradient was used to drive the flow. For spatial
discretization, the second-order central finite volume scheme and for the temporal
discretization a low storage four-stage second order RK scheme was used.
The models were assessed and compared to DNS data of Moser et al. [106] on the basis
of mean profiles of velocity and turbulence intensities.

Figure 6.2, Dimensions of channel flow test case

For the channel at Re 180 , the computational domain was Lx 4 , Ly 2 and

Lz 2 / 3 , where

is the channel half width considered equal to 1. The LES

calculation was performed on a computational grid of 33 49 33 points. The points

were equally spaced in the x - and z - directions, giving cell sizes in wall units of
x 70.68 , and z 11.78 .

The points were stretched in y direction using a

hyperbolic tangent function, leading to 0.5 y 16.32 . We used a computational time

step of t 0.072 , expressed in wall units as t t u2 / .
In order to obtain a statistically stationary state, the computations were carried out for a
period of 30 through flow cycles. Turbulence statistics were obtained by performing the
computations for another period of 30 through flow cycles.

109

In Figure 6.3, the mean streamwise velocity profile (a) and the related absolute error
field (b) are shown. Apart from the present results, results with the dynamic Smagorinsky
model and a coarse DNS (no model) are also plotted for comparison. As can be seen,
the obtained mean streamwise velocity profile is in good agreement with DNS data and is
more accurate than the conventional dynamic Smagorinsky model.
The computed model coefficient with the present approach and with the dynamic
procedure is plotted in Figure 6.3(c). It can be seen that the RoSM model coefficient goes
to zero close to the wall without any damping function or dynamic procedure, enforcing
zero contribution from the subgrid scales in the viscous sublayer. The time averaged
model coefficient for the dynamic Smagorinsky model goes up to 0.0854, compared to
0.027 with the present approach.
Giving emphasis to the non-diagonal Reynolds stresses for assessment of LES in
incompressible flows (carried out here), we present the xy- component of the Reynolds
stress tensor obtained with the models in Figure 6.4. It is noted that the new approach
predicts the results better than does the dynamic Smagorinsky model throughout most of
the channel.
The conclusion drawn from the mean-velocity profile is also valid for the diagonal
Reynolds stresses. Since the models are traceless, we shall study only the deviatoric part
as suggested by Winckelmans et al. [107]. The deviatoric diagonal Reynolds stress (e.g.
streamwise) is computed as

uu

uu

1
uu vv ww
3

(6.21)

where the angle brackets refer to the average over homogeneous directions and time. The
corresponding Reynolds stress components in the streamwise, wall-normal and spanwise
directions and the absolute errors are presented in Figure 6.5. As can be seen, throughout
most of the channel the results of the RoSM model are better than those of the dynamic
Smagorinsky model.
110

(a)

(b)

(c)
Figure 6.3, (a) The calculated mean streamwise velocity profile compared with the dynamic
Smagorinsky, and DNS. (b) The related absolute error field. (c) The model coefficient profiles.

(a)

(b)

Figure 6.4, (a) The calculated xy-component of Reynolds stress tensor compared with the
dynamic Smagorinsky, no model LES and DNS. (b) The related absolute error field.

111

(a)

(b)

(c)

(d)

(e)

(f)

Figure 6.5, (Left) The calculated deviatoric diagonal Reynolds stresses compared with the
dynamic Smagorinsky model and DNS. (Right) The related absolute error field

112

6.3.2 Fully developed turbulent flow in a square duct

The fully developed turbulent flow in a straight duct of square cross section was
simulated using LES. This complex turbulent flow has two inhomogeneous directions
with a remarkable change in flow structure. In this test case, secondary fluid motion
appears near the corners of the geometry, which is a reflection of the anisotropy of the
normal stresses. Prediction of turbulence properties in regions where these secondary
fluid motions occur is a severe test case for subgrid scale models evaluation in large eddy
simulation.
LES results at a Reynolds number of 300, based on the mean friction velocity and the
duct width are presented. A schematic representation of the test case geometry is shown
in Figure 6.6. Here, an aspect ratio of L / D 6.5 ( D being the height and L the length
of the domain) was used. In order to have negligible compressibility effects on the
simulations, the Mach number was set to be 0.054. The mean friction velocity was
obtained from the pressure gradient imposed in the streamwise direction as

dp / dx 4 0u2 / D .

Figure 6.6, Schematic of the duct geometry and the coordinate system is shown

The cells were equally distributed in x - direction and stretched in y - and z -directions.
The grid resolution close to the walls towards the centers was 0.23 y 11.00 , and the
streamwise distance x 30.46 , with 65 65 65 cells in respectively x-, y- and z 113

directions. The grid distributions in the y- and z- directions were identical. Periodic
boundary conditions were applied in the streamwise direction, and no-slip conditions at
the walls. For the initial condition, a parabolic streamwise velocity profile with random,
divergence-free velocity fluctuations was used.
A statistically stationary solution was obtained after 40 dimensionless time units and
thereafter turbulence statistics were obtained by performing the computations for another
40 dimensionless time units. The time was normalized with the friction velocity and the
duct height, i.e. 40 D / u . To assess the RoSM model, results are compared to the DNS
data of Gavrilakis [108] and LES results of Madabhushi and Vanka [109]
The computed mean streamwise velocity profiles of the RoSM model, the dynamic
Smagorinsky and no model along the wall bisector ( y 0.5D ) are plotted in wall
coordinates in Figure 6.7 and compared with DNS data. The mean velocity is
overpredicted, as is often the case with lower resolution LES. However, considerably
better (closer to DNS results) behavior of the RoSM model in comparison with dynamic
Smagorinsky is observed. Away from the wall this improvement is more pronounced,
the two models performing similarly near the wall.

(a)

(b)

Figure 6.7, The calculated mean streamwise velocity compared with the dynamic Smagorinsky
model, no model, and DNS.

114

The rms values of streamwise velocity fluctuations are plotted in Figure 6.8 (a).
Comparison shows a better behavior of the RoSM model throughout the duct. The RoSM
model overshoots the peak DNS value less than the dynamic Smagorinsky model and
shows significantly better agreement with the DNS data than the dynamic Smagorinsky
model elsewhere. In addition, the RoSM model results are in better agreement with DNS
data than the LES results of Madabhushi and Vanka [109].
The rms values of the y-component velocity fluctuations plotted in Figure 6.8 (b) indicate
that, in comparison with dynamic Smagorinsky, the RoSM model performs significantly
better from near the wall region up to the center region of the duct, and is in very good
agreement with the DNS results.
The rms values of the spanwise velocity fluctuations in Figure 6.8 (c) are consistently
better for the RoSM LES model. The agreement with the DNS data of Gavrilakis [108] is
good and much better than DSM and the LES results Madabhushi and Vanka [109].
In Figure 6.9 (a), the yz-component of the Reynolds stress tensor is depicted for the
RoSM model, the dynamic Smagorinsky model, and no model.

As readily seen,

throughout the duct, the results for the RoSM model and for the dynamic Smagorinsky model
are almost on top of each other. Near the wall and in the center region, LES results are close
to DNS for both models, whereas the models overestimate the results elsewhere.

Figure 6.9 (b) shows the profiles of the model parameter Cs for the RoSM model and the
dynamic Smagorinsky model. Near the wall the Cs profiles for both models are
comparable. In the region away from the wall, the Cs profiles behave quite differently
which is an important difference between the two models.

115

(a)

(b)

(c)
Figure 6.8, The calculated rms values of the diagonal Reynolds stresses compared with the
dynamic Smagorinsky, no model, LES of Madabhushi and Vanka and DNS

(a)

(b)

Figure 6.9, (a) the calculated xy-component of Reynolds stress tensor compared with the dynamic
Smagorinsky model, no model and DNS. (b) the model coefficient profiles.

116

6.3.3 Flow past a circular cylinder

The third test case considered is the turbulent flow past a circular cylinder at Re = 3900
(based on cylinder diameter and free stream velocity). This subcritical flow is a severe
test case for LES because it requires the calculation of the transition process to
turbulence. In this study, the cylinder had a diameter of D 1 m and a spanwise length of

D . This length is crucial for capturing the larger spanwise structures which have
wavelengths of the order of / D 1 according to Chyu and Rockwell [110]
An O-type grid was used which extended to 15 diameters in radial direction, as shown in
Figure 6.10. The grid consisted of 128 cells in the circumferential and radial direction and
48 cells in the spanwise direction. This is the same resolution as in the simulations of
Kravchenko et al. [111], Beaudan et al. [112] and Mittal et al. [113]. Also here the mesh
was stretched towards the wall to resolve the viscous sub-layer and towards the wake
zone. Uniform spacing of the mesh was used in the spanwise direction. The physical time
step of the simulation was t 3 105 s. The Mach number of the uniform flow in x direction was M 0.15 . For the simulation a streamwise velocity of U 0 50 m/s was
applied on the boundary.

Figure 6.10, Grid in the x, y-plane

117

The calculation was run until statistical convergence was obtained. In order to investigate
mean flow and turbulent intensities, statistics were compiled over a period of 35 vortex
shedding cycles by averaging in time and in the spanwise direction. The calculated results
from the RoSM model, the dynamic Smagorinsky model and a coarse DNS (no model)
are compared to DNS data of Ma et al. [114], experimental data of Lourenco and Shih
[115] and LES data of Kravchenko and Moin [111].
Figure 6.11 shows the calculated vertical profiles of the mean streamwise velocity at
streamwise locations x / D 1.06 , and x / D 1.54 . A good match of the present
streamwise velocity with DNS and experimental data is found. The present results are
also compared with those obtained from the dynamic Smagorinsky. The results are
almost on top of each other at the two locations. It can also be seen that using LES
without a SGS model underpredicts the results at the second location.

(a)

(b)

Figure 6.11, The calculated mean streamwise velocity compared the dynamic Smagorinsky, no
model, LES data [111], DNS [114] and Experiments [115]. (a) x / D 1.06 , (b) x / D 1.54 .

Figure 6.12 shows the calculated vertical profile of the mean vertical velocity using the
RoSM model at the same downstream locations. The corresponding results from the
dynamic Smagorinsky are also plotted in this figure. At x / D 1.06 , the agreement with
the DNS results is also good, as can also be seen in Figure 6.12 (a), showing better
behavior of the RoSM approach in comparison with the dynamic Smagorinsky model.
118

Again, the results using the RoSM model are in good agreement with the DNS of Ma et
al. [114] at x / D 1.54 and match the results of the dynamic procedure quite well, as
shown in Figure 6.12(b). However, it should be mentioned that the anomalous behavior
seen in comparison to the experimental data of Lourenco and Shih [115] is mainly
attributed to the experimental disturbance as also reported by Kravchenko and Moin
[111], [116] and Meyer et al. [117].

(a)

(b)

Figure 6.12, The calculated mean vertical velocity compared with the dynamic Smagorinsky, no
model, LES data [111] DNS [114] and experiments [115]. (a) x / D 1.06 , (b) x / D 1.54 .

Figure 6.13 to Figure 6.15 show respectively the streamwise Reynolds stress, the
crosswise Reynolds stress and the Reynolds shear stress components from the simulation
and DNS at locations x / D 1.06 , and x / D 1.54 . All the Reynolds stress components
are normalized by the square of the inlet velocity.
For the streamwise Reynolds stress component u u /U 02 , prediction from both the
present model and the dynamic Smagorinsky model compared to DNS is illustrated in
Figure 6.13. As readily seen, agreement between the RoSM results and DNS is fairly
good at location x / D 1.06 . At this location, the uu profile presents two strong
peaks mainly due to the transitional state of the shear layers. The position of these two
peaks is in agreement with the experiment of Lourenco and Shih [115] but the
magnitudes of the peaks are underestimated for the present results. As shown in Figure
119

6.13 (a), an almost similar behavior is observed in LES with the dynamic Smagorinsky
model. Moreover, a good overall agreement, both in amplitude and shape, compared to
the DNS data of Ma et al. [114] is found for the two SGS models. The calculated results
at location x / D 1.54 are plotted in Figure 6.13 (b).

(a)

(b)

Figure 6.13, The calculated streamwise turbulent intensity compared with the dynamic
Smagorinsky model, no model, LES data [111], DNS [114] and experiments [115]. (a)
x / D 1.06 , (b) x / D 1.54 .

(a)

(b)

Figure 6.14, The calculated crosswise turbulent intensity compared with the dynamic
Smagorinsky model, no model, LES data [111], DNS [114] and experiments [115]. (a)
x / D 1.06 , (b) x / D 1.54

120

The position of the two peaks of the uu -profile agrees with the experiment of
Lourenco and Shih [115]. However, there is an underprediction in the magnitude of the
peaks at y 0.5 . Further, the peaks predicted by the present approach are closer to DNS
data, giving better agreement of the streamwise Reynolds stress component with the DNS
data in comparison with the dynamic approach.
In Figure 6.14, comparison of the cross-flow normal Reynolds stresses vv /U 02 is
shown for the above mentioned locations. It is found that near the cylinder the present
approach is closer to DNS results in the peak region (see Figure 6.14 (a)). Also in the
wake center region, the present results are in better agreement with DNS and
experimental data than DSM. Figure 6.14 (b) shows the same comparison for another
downstream location. The results for the RoSM model and the dynamic Smagorinsky are
almost on top of each other. As can be seen, a very good agreement, both in amplitude
and shape, compared to the DNS data of Ma et al. [114]is obtained for the two models.
In comparison with the experimental data of Lourenco and Shih [115] both models
overshoot the peak by the same amount.

(a)

(b)

Figure 6.15, The calculated xy-component of the Reynolds stress tensor compared with the
dynamic Smagorinsky model, no model, LES data [111], DNS [114] and experiments [115]. (a)
x / D 1.06 , (b) x / D 1.54 .

121

In Figure 6.15, the calculated profile of the shear stress is presented at the same locations
as used in the previous plots. A strikingly good agreement with the DNS data of Ma et al.
[114] both in amplitude and shape, is obtained at these locations both for the RoSM
model and DSM.
Overall, the present method predicts the turbulent intensities better than the dynamic
Smagorinsky for the cylinder test case. For the vertical velocity profiles, the RoSM
approach and the dynamic Smagorinsky model show a similar behavior and both LES
results are in good agreement with the reference DNS data.

6.3.4 Application to UAM

We now proceed to check the performance of the RoSM model when used in a more
complicated geometry i.e. the present simplified human upper airway model (UAM).
Based on the available literature, the study on the human airways can be broadly
classified into a fundamental/basic study and an applied study. The available literature on
fundamental/basic studies uses either mathematically derived airway models or
simplified airway models derived from CT-scan of real subjects [46], [44], [25], [62].
Moreover, such studies use idealized breathing conditions such as the steady top-hat
velocity inlet condition. The main focus is to validate the numerical methodologies and to
get some useful insight into the fluid dynamics/particle deposition characteristics. On the
other hand, the literature on application based studies focuses on subject specific airway
geometries and realistic breathing conditions [30], [118] The idea of such studies is to get
meaningful insights which can then be used to either make subject-specific protocols or
to improve drug delivery systems.
The following results show the comparison of the RoSM model with that of DSM purely
from fluid-dynamics perspective. Further, both SGS models are then compared with the
available PIV-experiments done on the same UAM model. Conducting such a study is
crucial because of following reasons,

122

 Understanding of aerosol deposition behavior is of prime importance in pharmaceutical

industries so as to develop efficient drug delivery systems.
It is also used extensively in the medical field to evaluate the toxicological effect of
inhaled particulates.
One of the key parameters in evaluating aerosol deposition characteristics is to describe
the flow-field as accurate as possible, as this has a profound effect on aerosol deposition
in human lungs.
For the flow computations, the UAM model was discretized into 2.2 million cells with a
near wall clustering of elements and a stretching ratio of 1.1. The y of the first cell next
to the wall was about 1. This is the same mesh resolution as was used by Jayaraju et al.
[44] for a similar kind of study. The simulated flow field was validated based on the PIV
experiments conducted on the same UAM for a steady tidal breathing rate of 30 l/min.
For simulations, we considered a top-hat velocity profile corresponding to a flow rate of
30 l/min at inlet and a zero-gradient outflow boundary condition at outlet. The Reynolds
number based on the inlet diameter is ~2500, which is in the transitional regime i.e. the
flow regime inside the UAM varies from laminar to turbulent regime. The computations
for obtaining a time independent averaged solution took approximately 15 through flow
cycles. A through flow cycle defined here was taken as the ratio of length of airway
divided by the averaged cross-sectional velocity at the central plane.
In order to access the performance of the RoSM model, we consider the two-component (
ux and uz ) velocity magnitude obtained through PIV experiments (Brouns et al. [23]).
The experimental data were then compared against the results obtained with the present
approach and the dynamic Smagorinsky subgrid scale model. Figure 6.17 (a), (b), (c) and
(d) show the time averaged two-component velocity magnitude normalized by inlet
velocity magnitude ( U in ) versus normalized distance at four locations respectively one,
two and three tracheal diameters downstream of the larynx. Figure 6.16 shows these
locations as d1, d2, d3 and epi respectively.
123

Figure 6.16, Location of section lines corresponding to Sagittal plane

(a)

(b)

(c)

(d)

Figure 6.17, Normalized two component ( ux and uz ) velocity magnitude corresponding to

central sagittal plane, (a), (b), and (c) are one, two and three tracheal diameter downstream of
larynx, respectively; (d) Five mm above epiglottis.

124

Also shown in Figure 6.17 is the comparison against the RANS simulation. As expected,
LES simulations are in much better agreement with the PIV results than compared to
RANS simulations. In case of RANS, the predicted velocity peaks illustrated in Figure
6.17 (b) and (c) overshoot the PIV experiment whereas, the peaks predicted from RoSM
and DSM are in good agreement. The same is reflected when comparing the two
component velocity magnitude contours at the sagittal plane (shown in Figure 6.18). It
can be seen that shape of the laryngeal jet originating at the glottis is numerically
reproduced by RoSM (c) and DSM (d), while in case of RANS (a) the origin of the jet is
well captured but the tail of the jet is much longer.

(a)

(b)

(c)

(d)

Figure 6.18, Normalized time averaged 2 component velocity magnitude; (a) PIV, (b) RANS; (c)
RoSM; (d) DSM

Figure 6.19, shows the comparison of normalized two component ( u 'x2 and uz2 ' ) turbulent
kinetic energy. Similar to the observations made in Figure 6.17, the results from RoSM
and DSM are similar in nature and in good agreement with the PIV results.

125

(a)

(b)

(c)

(d)
'2

'2

Figure 6.19, Normalized two component ( ux and uz ) kinetic energy corresponding to central
sagittal plane, (a), (b), and (c) are one, two and three tracheal diameter downstream of larynx,
respectively; (d) five mm above epiglottis.

6.4 Conclusions
The RoSM model for estimation of the classical Smagorinsky model coefficient
developed at VUB was implemented and evaluated. We have shown that for the model
126

coefficient estimation, the proposed approach does not need any calibration to obtain the
best fit with the reference or the expensive dynamic approach, and the method finds itself
the right amount of dissipation.
The methodology was tested a posteriori using DNS data available from the literature for
a fully developed turbulent channel flow, a fully developed flow through a rectangular
duct of square cross section and a smooth subcritical flow past a stationary circular
cylinder. A good agreement between the RoSM model and the reference data was found.
Simulations have also been performed with the dynamic Smagorinsky model for each test
case. A good match between the RoSM model and the dynamic Smagorinsky results was
found. In comparison with the dynamic procedure, this improves the quality of the
results. It also has the following advantages,
The model coefficient is a self-adjusting flow dependent parameter.
It reduces the computational time with about 20%-25%.
It is very easy to implement for unstructured grids and complex geometries.
It does not need any stability control algorithm, avoiding all the numerical instability
issues that the dynamic procedure often faces.
These properties are very interesting for industrial applications, where the dynamic
procedure is very difficult to use and does not always provide accurate results.

127

128

Chapter 7
Particle deposition in extrathoracic
airway using LES
Contents
7.1 Introduction.........................................................................................................................................129
7.3 Multiple LES frozen field approach..................................................................................................132
7.3 Mathematical background.................................................................................................................132
7.3.1 Discrete POD method...................................................................................................................133
7.3.2 Procedure to derive optimal set of frozen-fields........................................................................136
7.4 Description of model geometry and sample data sets.....................................................................139
7.5 Results and discussion.......................................................................................................................140
7.5.1 Step 1, Evaluation of sampling period......................................................................................140
7.5.2 Step 2, Evaluation of time interval............................................................................................144
7.6 Particle deposition results.................................................................................................................147
7.6.1 Effect of SGS models...................................................................................................................148
7.6.2 Accounting SGS motions............................................................................................................150
7.6.3 Particle deposition comparision between RANS and LES......................................................154
7.7 Conclusions........................................................................................................................................156

7.1 Introduction
As was noted in Chapter 5, the particle deposition is highly sensitive to the accurate
prediction of transitional flow occurring in human airways. It was demonstrated that
using a different turbulence model results in a large variation in the calculated deposition
efficiencies. Furthermore, it was shown that RANS coupled with EIM overpredicts the
particle deposition data, particularly for smaller sized particles. This prompted to propose
correction functions to the EIM method (such as Wang and James EIM, helicity EIM
described in Chapter 4).

129

Due to the aforementioned problems, LES with an appropriate subgrid scale (SGS) model
is used to potentially improve the accuracy of particle deposition data (Jin et al. [17],
Jayaraju et al. [44], Lambert et al. [47]). In this approach, the LES particle simulations
involve a dynamic procedure wherein the flow and particle governing equations are
solved simultaneously. Although this approach gives a good agreement between the
particle deposition data and experiments it makes the computation very time consuming,
especially when several particle size diameters are involved.
To reduce the computational time involved with the above a second approach exists. In
this approach, first flow field data obtained from LES is stored at different instances of
time (called as frozen fields or snapshots). Then the particle calculation is carried out as a
post-operation procedure over stored frozen fields. Finally, the calculated particle
deposition data is represented as an ensemble average. This approach of calculating
particle deposition is known as multiple LES frozen field and reduces the particle
computation time by almost 70%. This chapter is concerned with the second approach
and gives a more methodical treatment discussed in the following sections.
Particle calculations can also be affected by accounting SGS model contribution,
depending on LES mesh and particle stokes number (Armenio et al. [119] and Berrouk
and Laurence [120]). Since the particle stokes number associated with particle deposition
in the human mouth throat is less than 0.1, in the present study of multiple LES frozen
field procedure, the effect of SGS models is included.
For the validation purposes, the results are compared against the available experimental
particle deposition data on the human respiratory tract (Verbanck et al. [45]).

7.2 Multiple LES frozen field approach

Figure 7.1, illustrates the flow chart for the two approaches mentioned above. In the case
for dynamic approach (Figure 7.1 (a)), the flow and particle calculations are carried out
simultaneously, till all the particles are either deposited on the walls or escaped from the
130

domain. This is computationally expensive, especially when approach is to be repeated

for several particle size diameters. On the other hand the second approach (Figure 7.1
(b)), the particle simulations are carried over stored LES frozen fields as a post-operation
method. The final particle data is represented as an ensemble average. By doing this the
LES flow calculations are carried out only once and this gives an edge over the first
approach. The stored field can then be repeatedly used for other particle size diameters.
In the context of UAM, Matida et al. [46] was the first to use such an approach. He used a
single LES frozen field taken at some random time during LES computation and then
performed particle calculations in that frozen field as a post-operation and without using
any EIM. In an application similar to the idealized human upper respiratory tract, Breuer
et al. [121] found similar particle deposition results when using either a single LES
frozen field or using the dynamic approach. By contrast, Jayaraju et al. [44] did not find a
good agreement when comparing both approaches in an upper respiratory tract model
similar to the one used here. This was attributed to the dissipative nature of using a single
LES frozen field for particle calculations. In any case, a single LES frozen field is
unlikely to capture most of the flow dynamics that would impact on particle simulation
results. In a very different application concerning the study of cyclones, Derksen [122]
used 16 LES frozen fields. In a subsequent study, Derksen [123] reported that the success
of such an approach critically depends upon the resolution of the flow field contained in
the frozen field.
To the best of our knowledge, no systematic procedure has been described that enables
the determination of the number of LES frozen fields required for accurate simulations.
The objective of the present work is to obtain a systematic procedure to determine the
number of multiple (instead of one) LES frozen fields that could greatly reduce the
computational cost while preserving accuracy. The time of duration of each frozen field
and the time interval between subsequent frozen fields will be the crucial parameters for
the success of a multiple LES frozen field approach to the simulation of particle
deposition.

131

Figure 7.1, Schematic of particle calculation procedure, a) dynamic approach, b) multiple LES
frozen field approach

The optimal set of LES frozen fields will be determined by first applying the discrete
proper orthogonal decomposition (POD) method of Sirovich [124] to the frozen field data
base. Then, adequate sample data set will be obtained by an approach similar to that used
by Hekmati et al. [125], where statistical means are used to conduct an indepth POD
mode convergence analysis. However, rather than conducting POD mode convergence,
first discrete POD is used to find characteristic eigenvalues and eigenmodes, to then
obtain an optimal set which captures most of the flow physics. In this investigation, the
final simulation of particle deposition data will be represented as an ensemble average
resulting from the optimal set of frozen fields.

7.3 Mathematical background

The fluid phase is simulated by employing the LES for the filtered Navier Stokes
equations. In this study, two SGS models, namely, a) Rotational Rate based Smagorinsky
model (RoSM), b) the Dynamic Smagorisnky model (DSM) by Lilly [75] have been
132

used. The governing equations relating to RoSM and DSM are as described in Chapter 6
and Chapter 3, respectively.
The particle phase is solved by employing Lagrangian particle equation of motion as
described in chapter 3 rewritten as below

dup
dt

u up

(7.1)

For the similar reasoning as mentioned in Chapter 4, the gravity term is ignored in
equation (7.1)

7.3.1 Discrete POD method

The Proper Orthogonal Decomposition (POD) method is a tool for analyzing a given set
of data by decomposing the data into a best orthogonal basis, giving the best
approximation using as few basis vectors as possible. These basis vectors, called as POD
modes, are ordered by decreasing importance such that the best approximation using
modes is obtained by using the first i modes. This way POD characterizes the dominant
dynamical structures in a given collection of data. This is why POD is extensively used in
fluid mechanics to extract such dominant structures corresponding to large scale turbulent
flow structures.
We start by constructing a sample data base D , consisting of instantaneous velocity field
Ui stored at i 1, 2, , S instances of time drawn at N spatial points. i.e. D is represented
as
DN S [U1 U 2 U S ]

(7.2)

It is common in POD approach to subtract mean from the data i.e.

Ui Ui U

133

(7.3)

where

1 S
U i
S i 1

(7.4)

Whether or not the mean is subtracted this does not changes the basic POD calculation
[126]. However, this changes the geometrical interpretation of the results (see [126]).
On can then define the covariance matrix C as follows,

CN N DD T

(7.5)

The optimal basis is formed by N 1 eigenvectors Vk of the covariance matrix C

obtained by,
CVk kVk ,

k 1,2,3, , S

(7.6)

where, 1 2 3 s denote the eigenvalues of the matrix C . Since the matrix

C is symmetric, the eigenvalues k are real and positive, and eigenvectors Vk are

orthogonal. The remaining N S modes correspond to eigenvalues that are zero.

In this approach, constructing the covariance matrix needs O SN 2 operations. Therefore,

this method is not efficient from computational time point of view when the covariance
matrix has a high dimension, which is usually the case for CFD generated data.
An alternative approach is the discrete POD method introduced by Sirovich [124]. This
method also known as the method of snapshots is based on the property that the
eigenvectors Vk are unique linear combinations of snapshots given by,
Vk DAk ,

134

k 1, 2,3, , S

(7.7)

where the S 1 vector Ak are the temporal coefficients. Using equation (7.7), one can
rewrite the eigenvalue equation (7.6) as follows
CDAk k DAk ,

k 1,2, 3, , S

(7.8)

Substituting equation (7.5) into equation (7.8) yields

DD DA
T

k DAk ,

k 1,2,3, , S

(7.9)

or

D DT DAk D k Ak ,

k 1,2,3, , S

(7.10)

One can remove the matrix D from both sides of equation (7.10) and obtain

D T DAk k Ak ,

k 1, 2, 3, , S

(7.11)

The new S S matrix C is defined as

C S S D T D

(7.12)

Equation (7.12) now becomes,

A ,
CA
k
k k

k 1, 2,3, , S

(7.13)

where the vector Ak represent the coefficients of the eigenvector Vk in the basis of
snapshots (see equation (7.7)). Therefore, the original eigenvalue equation (7.6) is now
shifted to equation (7.13). First, the vector Ak is calculated from equation (7.13) and then
POD modes are obtained from equation (7.7).

135

In the method of snapshots, as shown by equations (7.12) and (7.13), the constructing the
new covariance matrix, C requires O S 2 N operations, which is much cheaper than the

classical POD method ( O SN 2 , N S ). The discrete POD method is basically a

discretization of the classical POD approach in the temporal domain and widely used for
analyzing time dependent CFD data. In the present work, the quantity Ui represents the
fluctuating velocity components in x-, y- and z- directions at the i th snapshot, i.e.
u
U i v

w

(7.14)

where
T

u u1 u2 uN ,
T

v v1 v2 v N ,

(7.15)

w w1 w2 wN

An alternate way of interpreting POD is via Singular Value decomposition (SVD), infact
there is close relationship between POD and SVD. As detailed in [126], the Eigen values
and Eigen vectors could also be calculated using SVD and in such a case one doesnt
requires specification of the covariance matrix (for details please refer to [126]).

7.3.2 Procedure to derive the optimal set of frozen fields

The multiple LES frozen field approach uses a set of several frozen fields and
numerically simulates the particles as a post processing operation. This means that during
the particle calculation for a given snapshot, the LES field remains frozen/static, i.e., the
instantaneous velocity varies only in space and remains constant during the particle
calculation for that frozen time. Moreover, the procedure is the same as performing EIM
in case of RANS i.e. the particle tracking is carried out as a post-operation calculation.
136

Because of the uncoupling of flow and particle simulation only one flow simulation is
needed irrespective of the number of particle sizes. In a coupled approach, coupled
simulations would be needed for all particle sizes. One way to reduce computational time
in a coupled approach is to run simultaneous LES-particle simulations in parallel, but this
requires large computational resource which in may not be available. This restriction can
be overcome by running only multiple particle simulations in parallel and each particle
simulation module is coupled to the single flow solver module. This infact requires the
change in existing solver softwares. Still conducting multiple LES approach is more
efficient, because in this case the one has to run LES flow simulation and then deduce the
optimal set using the steps described below only once. The optimal set can very well be
preserved for future times and can be used for particle simulation of any other particle set
as when requirement arises. This will not be the case with the coupled approach even
using aforementioned suggestions, since for any future requirement one again has to
compute LES-flow simulation.
The accuracy of the multiple LES frozen field methodology strongly depends on the
chosen sample set. This sample set should be an optimal set, capturing most of the flow
dynamics. The optimality of the set is based on two parameters, a) the time duration over
which the frozen fields should be generated i.e. the sampling period Ts and b) the time
interval between each frozen field ts .
In order to determine the optimal set, a reference data bank consisting of S frozen fields
is generated and stored in the matrix D as shown in equation (7.2). Each frozen field
contains instantaneous fluctuating velocities drawn at N mesh points. In the present study
the reference data bank D has S 221 frozen fields. The associated time interval at
which successive frozen fields are stored is taken to be tref 1e 3 , which is 100 fold
of the physical time step used in the LES simulation. After the reference data base is
generated, following two steps are performed,
Step 1, Various reduced sample sets are selected from the reference data bank and for
each set the Relative Information Content (RIC) quantity is defined as,
137

RIC

i 1
m

(7.16)

k 1

where symbol denotes the eigenvalue, m represents the total number of frozen fields
and q 1,2,3, , m . Mathematically, eigenvalues are unique to a particular set and hence
give a signature of that particular set. Physically, eigenvalues represent the relative
energy content of a particular sample set. The sample time period, Ts can then be
determined by the convergence behavior of respective RIC plots. Note that in this step of
the procedure, the evaluated Ts contains sample sets with a fixed time interval of tref .
The results for this step are presented in subsection 3.1.
Step 2, The evaluated Ts from step 1 is used to study the effect of increasing the time
interval ts . Now Ts is fixed and various reduced sample sets are assessed by
progressively increasing the time interval between frozen fields ts , which is greater than
tref . The effect of increasing the time interval can then be quantified by means of an
auto-correlation index. The auto-correlation index for the eigenmode Vk in a given
reduced sample set is defined as,

where the symbol

Vk ,ref Vk
Vk ,ref Vk ,ref

k 1, 2, 3, , S

(7.17)

Vk Vk

is the spatial average and the subscript denotes the reference

sample set. The optimal time interval is then chosen as the one leading to the maximum
correlation index. The results for this step are presented in subsection 3.2.

138

7.4 Description of model geometry and sample data sets

The model geometry is the same as described in Chapter 4 (see Figure 4.4). Figure 7.2
shows the location of the planes with respect to the model geometry, selected for storing
the instantaneous fluctuating velocities. Ideally, we can use the entire flow domain for the
analysis of step 1 and step 2, but this creates a bottleneck on the available disk storage
capacity. This is overcome by choosing a central plane (referred to as sagittal plane) and
five perpendicular planes (as shown in Figure 7.2). The arrangement of all planes was
chosen such that any asymmetry of flow can be adequately captured. To illustrate that the
sagittal plane suffices, and that the extra five planes do not actually affect the overall
findings of subsections 3.1 and 3.2, the results of step 1 and step 2 will be shown by
considering only the sagittal plane (referred to as Case 1) and by considering the sagittal
plane combined with five perpendicular planes (referred to as Case 2). The number of
mesh points for Case 1 and Case 2 is 39567 and 64566, respectively.
All the flow and particle calculations are done for an inspiratory breathing rate of 60
l/min corresponding to a Reynolds number of 5000, and for particle sizes ranging from
2m-10m.The fluid-phase density and dynamic viscosity are set at 1.2 kg/m3 and

1.884e 5 Ns/m2, respectively. The particle density is set a p 912 kg/m3. In total,
about 10,000 randomly distribution particles are injected at the inlet of the UAM model.
A parabolic velocity profile is imposed at the inlet of UAM, whereas a zero gradient
velocity profile is imposed at the outlet.
For the analysis in step 1 and step 2, the RoSM is chosen as a SGS model. The physical
time step for LES computation is set to be 1e-5 which corresponds to a CFL of about 1.
Finally, the same parameters will be used to study the effect of SGS models and their
contributions on the particle deposition behavior.

139

Figure 7.2, UAM model geometry showing location of planes with one sagittal plane and five
perpendicular planes numbered 1-5

7.5 Results and Condition

7.5.1 Step 1, Evaluation of sampling period
Before starting the evaluation of the sampling period, one defines an integral time scale,

Tintegral , as the ratio of mean UAM length L to the average cross-sectional velocity at the
inlet U in,ave . Here, L 0.326 m and U in,ave 4.97 m/sec, and then Tintegral 0.06 sec.
Considering increasing sampling period Ts , Table 1 details the sample sets analyzed,
where the first column denotes various sample sets or test cases, the second column
denotes the number of frozen-fields contained in respective sample sets, and the last
column represents the length over which samples are collected given in terms of the ratio
between Ts and Tintegral

140

Sample Sets

Ts Tintegral

set 1
set 2
set 3
set 4
set 5
set 6
set 7
set 8
set 9
set 10
set 11

21
41
61
81
101
121
141
161
181
201
221
Table 7.1, Sample sets parameters with ts tref

0.5
1.5
2
2.5
3
3.5
4
4.5
5
5.5
6

(a)

(b)

Figure 7.3, Relative Information Content (RIC) for various sample sets detailed in Table 7.1;
considering (a) Case 1 Sagittal plane; (b) Case 2 Sagittal and five perpendicular planes (see
Figure 7.2)

Figure 7.3 shows the relative information content (RIC) curve for each sample set
detailed in Table 1. As mentioned before in section 7.5.1 (step 1), an RIC curve provides
the information regarding the accumulated energy contained in respective sample sets.
Since this curve shows the energy with respect to a given sample set, it can be used to
study the effect of increasing or decreasing sampling period. It can be observed for a
given ts the increase in the number of frozen-fields/snapshots (i.e. increase of Ts ),
141

approaching the maximum available set 11, the difference in RIC curve (energy) for
various sample sets progressively decreases. This observation is hardly affected by the
addition of five additional planes to the sagittal plane, as can be seen from the similarity
between Figure 7.3 (a) and Figure 7.3 (b).
One can also use the RIC curves in Figure 7.3 to determine the number of POD modes
required to represents a certain percentage of the total information content for any given
sample set, which was considered to be 95% in our study. When doing this for each
95%
95%
/ S(11)
sample set, and normalizing this to the reference set 11, the resulting S 95% S(11)

can be evaluated as done in Table 7.2.

Sample Sets
set 1
set 2
set 3
set 4
set 5
set 6
set 7
set 8
set 9
set 10

95%
95%
S 95% S(11)
/ S(11)

95%
95%
S 95% S(11)
/ S(11)

Case 1
0.9
0.81
0.71
0.61
0.52
0.43
0.34
0.26
0.15
0.09
Table 7.2, Sample sets deviation from the reference set 11

Case 2
0.9
0.81
0.71
0.63
0.54
0.46
0.36
0.27
0.14
0.09

As can be seen from Table 7.2, the relative deviation (i.e., loss of energy content with
respect to set 11) in set 10 or set 9 remains below 15%. However, if the sampling period
is decreased to less than 4 Tintegral i.e. set 1 to 7 the deviation becomes larger than 25%.
This observation is hardly affected by the addition of the five perpendicular planes to the
sagittal plane, as can be seen from the similarity between both columns of Table 7.2.

142

(a)

(b)

(c)

(d)

Figure 7.4, Contours and profiles of average velocity magnitude at sagittal plane for selected sets
of Table 1, (a) velocity magnitude contour; (b), (c) and (d) are the velocity profiles at section 1-1

In order to ascertain that set 9 does represent an adequate set, the averaged velocity
magnitude contours and their profiles at the sagittal plane are compared with the solution
from full LES (Figure 7.4). The averaging for full LES was carried out for more than 15
flow through cycles.
As can be observed in Figure 7.4 (a), the shape of the glottal jet is equally well captured
by set 11 and set 9, whereas in set 2 and set 1, the glottal jet is more dissipative as
compared with the full LES. Also, the non-dimensional averaged velocity magnitude
143

profiles around the glottal jet (Figure 7.4 (b), (c), and (d)) show that set 11 and set 9
correspond better to the full LES results than do set 2 and set1. Overall, the above
analysis and flow field assessment suggest that set 9 having a period of Ts 4.5Tintegral
does represent an adequate choice, at least in terms of sample period.

7.5.2 Step 2, Evaluation of time interval

In this step, the effect of increasing the time interval ts between frozen fields is studied
for a fixed evaluated sample period of Ts 4.5Tintegral . The study in this step will yield ts
required for the optimal set. For this purpose, the correlation index defined in equation
(7.17) is determined for sets I-VI detailed in Table 7.3, where set I is now the reference
set.
Sample sets
set I
set II
set III
set IV
set V
set VI

Ts / Tintegral

181
4.5
91
4.5
46
4.5
22
4.5
12
4.5
6
4.5
Table 7.3, Sample sets parameters with Ts 4.5Tintegral

ts / tref
1
2
4
8
16
32

Figure 7.5 illustrates, for Case 1 and Case 2, three chequered plots describing the
correlation index of the first 6 scalar POD modes related to x-, y-, and z- velocity
components of the reduced sample sets (set II to VI). The correlation index indicates how
the sets II-VI are correlated with respect to the reference set I, from which following
observations can be made,
The behavior of the correlation index is comparable for all the three velocity components.
As expected, with smaller ts the correlation of the first and second mode, which correspond to the first two large eigenvalues, is bigger for set II as compared to any other
set.
144

 Sample sets IV, V, and VI give a poor correlation index for all the 6 POD modes
shown, implying that the deviation in terms of retaining flow field with respect to the
reference set I is higher.
Irrespective to the choice of location of the planes, Case 1(top) and Case 2(bottom),
reveals the same information.
Overall, the flow field information retained with respect to the reference set I is better
retained in reduced sample sets II and III.

(a)

(b)

(c)

(d)

(e)

(f)

Figure 7.5, Auto correlation index; (a, d) x-velocity, (b, e) y-velocity, (c, f) z-velocity
respectively; (top) Case 1, (bottom) Case 2

The contours and the profile of the average velocity magnitude at the sagittal plane are
illustrated in Figure 7.6. Sample sets IV to VI retain much less flow field information as
145

can be seen in the dissipation of the glottal jet shown in Figure 7.6(a). In fact, set VI has
the highest dissipation and thus the lowest correlation index (see Figure 7.5). The
dissipation of the glottal jet is the least for set II followed by set III as was reflected in
their respective correlation index.

(a)

(b)

(c)

(d)

Figure 7.6, Contours and profiles of average Velocity Magnitude at sagittal plane for sets detailed
in table 3, (a) Velocity Magnitude Contour; (b), (c) and (d) are the Velocity profiles at section 11, 2-2 and 3-3 respectively.

Also, the velocity profiles in Figure 7.6(b)-(d) essentially reflect the same message.
Based on the analysis of this subsection, it can be concluded that rather than using all 181
frozen fields, 91 or even 46 frozen fields can be used. Overall, from the analysis
presented across subsections 7.5.1 and 7.5.2, it can be inferred that doing POD analysis is
not only intuitive but also beneficial. It can characterize the random data stored in the
146

sample sets and determine an optimal set which captures most of the flow-dynamics. In
the following section, the optimal set for the evaluation of particle deposition will be
used.

7.6 Particle deposition results

In this section, the particle deposition is evaluated for the reference set I, and sets II and
III. The goal of this section is to assess the accuracy of the proposed procedure used in
the present work for the multiple LES frozen field approach. This is done by studying
particle calculations in terms of two quantities namely, a) total deposition efficiency and
b) compartmental relative deposition efficiency for 3m and 6m particle sizes. The total
deposition efficiency (%DE) represents the ratio of number of particles deposited to the
number of particles injected in UAM. The second quantity is evaluated at four
compartments of UAM namely; a) mouth, b) pharynx, c) larynx and d) throat and
represents the ratio of particles deposited in a given compartment to the total number of
particles deposited in UAM. The second quantity serves to illustrate the local variation in
particle deposition.
Figure 7.7 shows the calculated particle deposition from set I, set II and set III. As can be
seen from Figure 7.7(a), the total deposition efficiency curve predicted from sets I, II, and
III follows closely with the empirical curve of Grgic et al. [101]. Also, the calculated
deposition efficiency for 3m and 6m particle sizes is in good agreement with the
available experimental data in UAM (Verbanck et al. [45]). In terms of local deposition
for 3m and 6m particle sizes, the relative deposition efficiency from the three sets is in
very good agreement with the experiments as shown in Figure 7.7(b). In fact, the
predictions from the three sets and experiments overlap each other for 6 m particle size.
A good agreement between the calculated results and experimental data indicates that,
following a systematic approach leads to an optimal set, which captures most of the flow
physics and thus can be used with good accuracy in evaluating particle deposition.

147

(a)

(b)

(c)

Figure 7.7, Particle deposition efficiencies for set I, II and III; (a)Total Deposition
efficiency, (b)-(c) Relative deposition efficiency for 3m and 6m

7.6.1 Effect of SGS models

In this section, the effect of using different SGS models on particle calculations is
compared. For this purpose, the RoSM and DSM subgrid scale models were chosen. In
our recent work (see Ghorbaniasl et al. [97]), we applied both the SGS models to the
UAM and showed that they predict a similar flow field behavior, being in good
148

agreement to the experiments. We will now compare the performance of two SGS models
in terms of the particle deposition. The sampling period Ts and the time interval ts
between frozen fields were kept the same for both the SGS models. A sampling period of
4.5Tintegral and a time interval of ts 2tref (i.e. S 91 ; Table 3) were considered. Figure

7.8 shows the calculated total deposition efficiency (Figure 7.8 (a)) and the relative
deposition efficiencies for 3m and 6m (Figure 7.8 (b) and (c)). It can be seen that both
SGS models predict almost similar values for the particle deposition efficiency.

(a)

(b)

(c)
Figure 7.8, Particle deposition efficiencies comparison ; (a) Total Deposition efficiency, (b)-(c)
Relative deposition efficiency for 3m and 6m

149

This similarity is also reflected when comparing the averaged flow field of the above
used sample set between two SGS models (see Figure 7.9).

(a)

(b)

(c)

(d)

Figure 7.9, Contours and profiles of average Velocity Magnitude at sagittal plane, (a) Velocity
Magnitude Contour; (b), (c) and (d) are the Velocity profiles at section 1-1, 2-2 and 3-3
respectively

7.6.2 Accounting SGS motion

Before going into the details of modeling SGS contribution its worthwhile to discuss the
representation of the instantaneous flow velocities u in equation (7.1). In this equation,
u represents exact or complete instantaneous fluid phase velocity vector. In the case of

LES, the exact or complete instantaneous velocity is given as,

150

u u u'

(7.18)

where u represents resolved velocity vector and u' represents subgrid part. In LES, the
SGS velocity information is not readily available and thus requires modeling. Therefore,
mostly in past studies in equation (7.1) it is assumed that u u . However, in some of the
literature, it has been shown that for particles with Stokes number smaller than 1; it is
worthwhile to account for the SGS contribution as this may lead to different particle
deposition characteristics. There are several ways of accounting the SGS contribution in
equation (7.18) (Fukagata et al. [127], Shotorban & Mashayek [128], Berrouk and
Laurence [120]). On dimensional grounds, subgrid kinetic energy can be modeled as,

sgs

t
L

(7.19)

In equation (7.19), the subgrid length scale is determined as,

L c

(7.20)

where c is the SGS model constant, evaluated based on the SGS model considered and

is the grid size. Finally, the SGS velocity contribution is modeled assuming an
isotropic condition for the SGS part. The SGS velocity components in x, y and z
directions are defined as follows,

u ' N1

2 sgs
k ,
3

v' N2

2 sgs
k ,
3

w' N3

2 sgs
k
3

(7.21)

where N i represents Gaussian random number with zero mean and unity standard
deviation. This approach of modeling SGS velocities is the same as used by Wang and
151

Squires [129]. It is however to be mentioned here that we have chosen a rather

diagnostic/heuristic approach in determining SGS fluctuations. The isotropic assumption
in the above equations is not valid close to the wall. Nevertheless, by determining the
SGS fluctuation, one can get some useful insight into the effect of SGS models involved
and assess how a SGS model part can affect the particle deposition.
In this section, the effect of accounting SGS velocity components in the particle
calculations is discussed and the results are analyzed to determine whether there is a need
to account SGS velocity components, which in the context of simplified airways has not
been scrutinized. The SGS velocity components are accounted by using a heuristic
approach given in subsection 2.2. In addition this effect will be compared between two
SGS models namely, RoSM and DSM, respectively.
Figure 7.10 compares the particle deposition calculation between the two SGS models
and the respective effect of accounting SGS contribution. When comparing this effect on
the total deposition efficiency and the relative deposition efficiency, the overall effect of
accounting SGS contribution in both the SGS models was found to be marginal. We
suspect that this marginal effect can be related to using fine LES mesh, in particular,
close to the walls. This can be qualitatively shown from the contour plots in Figure 7.11
corresponding to the reference databank consisting of 221 LES snapshots. The contours
corresponds respectively, to kinetic energy of resolved field k res (left panel), kinetic
energy contained in the SGS motion ksgs (middle panel), and root mean square kinetic
energy k rms (right panel). The symbol

denotes ensemble average over reference

databank. Also the root mean square is based on this data set. It is clear from the figures
that the subgrid scale kinetic energy are very small compared to the k res . This fact is
also reflected in the particle deposition results. However, further in depth investigation is
required to quantitatively access the effect of SGS fluctuations, which is beyond the
scope of current work, but will be the purpose of future work.

152

(a)

(b)

(c)

(d)

(e)

(f)

153

Figure 7.10, Particle deposition efficiencies comparison ; (a)-(b)Total Deposition efficiency, (c)(d) Relative deposition efficiency for 3m, (e)-(f) Relative deposition efficiency for 6m

(a)
Figure 7.11, Comparison of

(b)

(c)

k res (averaged kinetic energy of the resolved field),

ksgs

(averraged SGS kinetic energy) and k rms (kinetic energy contained in the fluctuation),
corresponding to reference data set;

denotes ensemble average and rms denotes room mean

square fluctuation

7.6.3 Particle deposition comparison between RANS and LES

As was done for the flow field comparison between RANS and LES in the previous
Chapter 6, in this section we compare the particle deposition results previously obtained
for RANS and LES for the case of 60 l/min flow rate. It was seen in the flow field
comparison (see Figure 6.17 and Figure 6.18, Chapter 6) that related peaks and velocity
trends were better matched for LES than compared to RANS. Infact, it was seen for
sections d1, d2 and d3 (see Figure 6.16) the velocity peaks where over predicted in
RANS and also the glottal jet appears more narrower and longer than compared to PIV
experiments (Figure 6.18). The effect of this is also reflected in particle results (Figure
7.12), specifically for the total deposition values where there is an underprediction for 6
m. However, for particle sizes less than 4 m, both LES and RANS (coupleD with
HEIM) appears giving similar results both in terms of total deposition values and for
154

relative deposition values. Overall for it can be inferred that when there is a
computational resource restriction, RANS coupled HEIM can be assumed to give
comparable results with experiments. If there is no such restriction it is preferable to use
more accurate LES approach utilizing the systematic approach described in this chapter
which will further help in reduce computational time.

(a)

(b)

(c)
Figure 7.12, Comparison of particle deposition between RANS and LES at inspiratory flow rate
of 60 l/min; (a) total deposition efficiency, (b)-(c) relative deposition efficiency corresponding to
3m and 6 m

155

7.7 Conclusion
This chapter described a systematic multiple LES frozen field approach that resulted in a
time-efficient yet accurate simulation of particle transport. We tested the approach using
UAM as our test case and considering inspiratory flow rate at 60 l/min. Two SGS models
namely, RoSM and DSM validated in the previous chapter were utilized.
The main idea of the approach was to use several LES frozen fields, stored at different
instances of time and then numerically simulate particle tracking as a post operation step.
This avoids the coupling of LES-particle calculations thereby saving considerable
computational time.
First, a methodical procedure was introduced for determining an optimal set which could
then be used in the multiple LES frozen field approach. The procedure is a two-step
process and involves first generating a data bank containing instantaneous velocities
drawn on the spatial domain of interest. The optimal set was then derived from the flow
field data bank by systematically analyzing eigenvalues and eigenmodes evaluated
through discrete POD methodology. The success of the procedure was demonstrated by
comparing the flow field data contained in the optimal set with the full LES results and
also by comparing the particle calculations with the available experimental data. A good
overall agreement between the results was observed.
In the second part of the work, the effect of subgrid scale models was investigated. It was
noticed that the two SGS models predict similar particle calculations. It was observed that
the effect of accounting SGS motion in particle calculation is negligible. Finally, a
comparison of particle results obtained from LES and RANS was shown. Clearly, as
observed in previous chapter, the accuracy of using LES was also reflected in particle
deposition results. It is concluded that when there is a computational resource restriction,
RANS coupled HEIM can be assumed to give comparable results with experiments. If
there is no such restriction it is preferable to use more accurate LES approach utilizing

156

the systematic approach described in this chapter which will further help in reduce
computational time.

157

158

Chapter 8
Particle deposition in a 5 Generation
intrathoracic airway using LES
Contents
8.1 Introduction..........................................................................................................................................159
8.2 Upper airway geometry.......................................................................................................................160
8.3 Simulation condition............................................................................................................................161
8.4 Results...................................................................................................................................................163
8.4.1 Mesh convergence study...............................................................................................................163
8.4.2 Flow dynamics...............................................................................................................................165
8.4.3 Optimal set determination...........................................................................................................168
8.4.4 Particle deposition results.............................................................................................................171

8.1 Introduction
In order to completely evaluate the drug dose or the toxicity of particulate matter, it is
vital that CFD studies should include fluid/particle simulation in a complete lung model.
Having evaluated the fluid/particle methodologies in previous chapters, this chapter
presents the next logical step which is application of those outcomes to a more complete
airway model. Furthermore, this chapter presents briefly the fluid/particle simulation
results in an upper airway geometry consisting of airways till 5th generation (shown in
Figure 8.1). The success and the accuracy of the multiple LES snapshot methodology
introduced in the previous chapter are, therefore, chosen as the preferred methodology for
this chapter.

159

8.2 Upper airway geometry

As mentioned in Chapter 2, there are several models for lung geometry right from
Weibel, Horsfield to CT-scan based models and each of these have their own complexity
associated to them. While Weibel and Horsfield model are relatively easier to generate,
they include several simplifying assumptions which render them physiologically
incorrect. On the other hand CT-scan captures the airway morphometric details better but
is limited to the resolution of the scanning machine.
The upper airway geometry used in the present chapter is reconstructed by fusing the
scaled UAM model and the CT-scan of a never smoking female adult. Most often not
many image-based data contain both extra- and intrathoracic airways. In such cases,
combining an existing simplified extrathoracic airway model with the available
intrathoracic CT-scan data is a reasonable step. Of course, one can argue that in the event
of unavailability of extra-thoracic data, it is better to restrict fluid/particle simulation only
to intrathoracic airways. However, as recently shown by Lin et al. [30], inclusion of the
oral region is essential in generating the turbulent laryngeal jet that significantly affects
the pressure, velocity and shear stress distribution in the intrathoracic airways. And since
particle transport is directly linked to the flow calculation, therefore including the oral
region becomes important.
Figure 8.1 illustrates the details of the upper airway geometry. The geometry comprise of
extrathoracic region representing oral, pharynx, larynx and trachea. The intrathoracic is
composed of all the airways from generation G0 (representing trachea) till generation G5
(5th generation). A generation includes a parent airway and its two daughter airways. For
example, G1 includes throat together with left and main bronchus. The intrathoracic
region starts at generation G0 and is composed of left and right main bronchus and five
lung lobes. As can be seen, the length of the left main bronchus is greater than the right
main bronchus. Also one can see the curvature at the trachea to the first bifurcation which
is generally induced because of the presence of heart. The left side of the lungs is
composed of two lobes namely, left upper lobe (LUL) and left lower lobe (LLL). On the
160

right side, this is made of right upper lobe (RUL), right middle lobe (RML) and right
lower lobe (RLL). In general, for an upper airway consisting of n generations has 2n
outlets. In the present case, we have n 5 accordingly, 32 airways. Moreover, we have
one extra outlet at LLL. This extra airway is not any superficially added airway but was
observed in the CT-scan and therefore was kept as such. All the outlets of the airways are
extended by 10 times their diameters. This is done to avoid reverse flow and therefore to
have smooth flow exit. The total fluid volume of the air way model is 102.1 liters.

Figure 8.1, Upper airway geometry reconstructed by fusing scaled UAM model with CT-scan
data of a female adult.

The entire geometry is discretized into tetrahedral cells consisting of 2.1, 4.2 and 8.4
million cells. In order to ensure the quality of the resolution in the boundary layer , quad
cells with a stretching ratio of 1.1 are used.

8.3 Simulation condition

A steady parabolic inlet velocity profile corresponding to 30 l/min is imposed at the inlet
of the upper airway. The reason for choosing a parabolic inlet profile is to have the same
161

condition that exits during experiments. All the outlets of the current airway geometry are
imposed with uniform zero gauge pressure. This is because in the experiments, all the
outlets are connected to a single pump. Also in order to have same conditions during the
particle experiments and to have a nearly parabolic velocity profile at the inlet, a long
tube at the inlet of airway cast is used. In order to ascertain with the current imposed
boundary condition, mass fractions going in to each outlet is the same as that of
experiments. Figure 8.6 illustrates measured mass flow rates at each outlet during flow
experiments compared to that obtained with simulation. Clearly, a good correlation can
be observed between experiments and simulations which in turn give confidence with
regards to the choice of the boundary conditions. For the fluid phase, air at density

1.2 kg/m3 and dynamic viscosity 1.884e 5 kg/m-s is considered.

For the particle phase, a total of 8000 randomly distributed particles having particle
density p 912 kg/m3 are injected at the airway inlet. All the particles are initialized
with flow velocity that exists at the inlet. At the airway wall boundary a trap boundary
condition was imposed i.e. particles are considered deposited as soon as they reach the
wall.
The fluid phase is simulated using LES with RoSM as the SGS model. The physical time
step used for flow computation is taken as 5e-5 which returns an average CFL of 1.2 in
the whole domain. Second order implicit formulation for temporal discretization and
central differencing for spatial discretization of momentum. The governing equations are
as explained in Chapter 6. Similarly, the particle phase is calculated by numerically
integrating the equation of motion of each particle as described in Chapter 3. As
mentioned in Chapter 4, unlike in UAM, where regions of high flow velocities exist, the
gravity term is considered in the following particle calculations. This is because of the
presence of low flow velocities in the distal airways [49]. The particle deposition
calculation is carried out using the multiple LES snapshot procedure described in the
previous chapter. For this, a total of 1201 snapshots are considered for the analysis. Each
snapshot is taken at a time interval of 5e-5 which is equal to the physical time step used
in LES computation. In order to derive the optimal set from the 1201 snapshots, a POD
162

analysis is carried for the velocity field stored at the central sagittal plane (shown in
Figure 8.6). The results of the POD analysis are given in the sections below.
Due to the non-availability of particle experiments in the upper airway, at the moment for
the comparison purposes, we consider the work of Lambert et al. [47], where a similar
particle deposition study by imposing a steady parabolic velocity corresponding to 20
l/min is carried out. Further the particle deposition results are also considered with the
experiments of Chan and Lipmann [130] and Zhou and Cheng [131].

8.4 Results and discussion

8.4.1 Mesh convergence study

Figure 8.2, Sagittal plane with section planes used for mesh convergence study

Figure 8.3 illustrates the comparison of u uin at different section shown in Figure 8.2.

163

sec-a

sec-b

sec-c

sec-d

sec-e

sec-f

Figure 8.3, Non dimensional mean velocity magnitude at sections a-f

164

The plotted results represent the solution averaged for 10 through cycles. Apart from
section a and d, the profile for u uin in 4.2 million cells or 8.4 million cells looks
similar. The differences seen in the plots for 4.2 and 8.4 million cells could be attributed
to insufficient averaging. During the LES calculation it was observed that the velocity
profiles in 4.2 million cells and 8.4 million cells get closer from 5 through cycles to 10
through cycles. Overall looking at the velocity profiles in section a till f, it can be
deduced that 4.2 million can be further used for the rest of the calculations.

8.4.2 Flow dynamics

(a)

(b)

Figure 8.4, Contours of mean velocity magnitude (a) and mean turbulent kinetic energy (b)

Before going into the details of the particle deposition results it is worthwhile to briefly
discuss about the computed flow field. The following LES flow results are based on the
165

converged mesh of 4.2 million tetrahedral cells. The flow results presented in this section
are obtained after averaging over 15 integral time scales. The integral time scale is
defined as the ratio of Din / U in where the inlet diameter Din 14mm and U in 2.5 m/s.
Figure 8.4 (a) shows the contours of the mean velocity magnitude and the turbulent
kinetic energy. As observed, the mean flow velocity increases as it passes through the
narrow passage of the mouth. The Laryngeal jet surrounding the glottis (Figure 8.5 (a)) is
formed as a result of acceleration of air caused due to the sudden constriction at the
glottis. Interestingly, the jet is directed more towards the left side (right side in the
picture) as can be seen from the stream lines.

(a)

(b)

Figure 8.5, Contours of mean velocity magnitude and turbulent kinetic energy at central plane

The turbulent kinetic energy contours (Figure8.5 (b)) show that the flow, after initially
being turbulent, laminarizes further down in the trachea.

166

Figure 8.6, Comparison of flow distribution at outlets against experimentally measured data at 30
l/min

The CFD predicted flow rate (shown as circles) distribution at the various lobes is
illustrated in Figure 8.6 and compared against the experimentally measured (shown as
triangles) flow rates. As observed, an overall good correlation can be found between the
experiments and the CFD predicted flow. Comparison of the CFD versus experiments
obtained from Lambert et al. [47] and Darquenne et al. [132] is also shown. The present
data infact lies in between the data obtained with former authors, thereby giving
confidence in the simulated results. Moreover, an interesting observation can also be
made, as mentioned before in the present case we use uniform pressure at all outlets,
where as in the case of Lambert et al. [47] and Darquenne et al. [132] at outlet mass flow
was imposed based on the subtended volmes. In the case of Darquenne et al. [132] it was
based on the data provided by Horsefield et al. [6], while Lambert et al. [47] derived flow
distribution from CT-scan. This do suggests that using uniform pressure at outlets or
imposing flow does not make any huge differences.
Table 8.1 details the flow rate distribution among all the five lobes (CFD and
experiments). It can be noticed that LLL receives greater portion of the inspired air than
167

any other lobe, whereas the RML receives the least. In total the right lung receives
slightly more air than compared to left lung.
Lobes
Q (l/min)- CFD
Q (l/min)- Exp.
Q(l/min)-CFD
Q(l/min)-Exp.
LUL
4.856
4.831
14.057 (left)
14.287 (left)
LLL
9.201
9.447
RUL
5.011
5.180
15.943 (right)
15.722 (right)
RML
2.487
2.724
RLL
8.445
7.817
Table 8.1, Flow rate (l/min) distribution in lobes, CFD versus Experiments

These observations are very much similar to what Lambert et al. [47] observed in their
recent studies.

8.4.3 Optimal set determination

Following the procedure previously explained in Chapter 7, we first determine the
sampling period Ts by analyzing the relative information content (RIC) plots (see
Chapter 7, equation 7.15 for definition).

Figure 8.7, Relative information content RIC i

i 1

, where m=1201 and q denotes

k 1

various reduced sets. The red line corresponds to q=1041 and the blue line corresponds to the
reference data base of q=1201

168

Figure 8.7 illustrates the RIC curve for various reduced sets derived from the reference
data set consisting of 1201 snapshots. In this figure, there are 29 sets, stating from a set
consisting of 41 snapshots leading to reference data set (shown in blue, rightmost in the
figure). As the number of chosen frozen-fields/snapshots increases (i.e. increase of Ts ) the
difference in the RIC curve (energy) for various sample sets progressively decreases.
Similar to the analysis done in Chapter 7, the ratio N q95% / N q95%
1201 is plotted in Figure 8.8,
where N q95% denotes the number of POD modes required to represent 95% of the total
information in a given sample set. The abscissa represents various sample sets plotted in
decreasing order i.e. starting from set q 1201 to set q 49 . As can be observed, by
decreasing sample sets (and hence decreasing sampling period Ts ) the slope of the curve
for the first few sets ( q 1201 to 1001 ) changes slowly, while for the later sample sets it
starts changing dramatically. Based on the on the above analysis of Figure 8.7 and Figure
8.8, we have considered a sample period Ts 0.05s represented by the sample set
q=1041 (shown in red curve in Figure 8.7) as the adequate set for further analysis.

Figure 8.8, N q95% / N q95%

1201 for various sets

The next step in the analysis is to determine the required time interval between each
snapshot ( ts ). This is carried out by the following the POD procedure: wherein Ts is
now fixed and various reduced sample sets are accessed by progressively increasing ts .
169

This is done by calculating the autocorrelation index k (see Chapter 7, equation 7.16 for
the definition). Figure 8.9 illustrates the calculated autocorrelation index for the first 9
POD modes for samples sets II to VII. The sample sets II to VII were derived by
doubling ts . As can be observed, there is a very good correlation among the first 9 POD
modes till set VI, after which the correlation is only seen for the first POD mode. Based
on Figure 8.9, we chose set IV which consists of 66 snapshots having Ts 0.05s and ts
=0.0008s as the set to be used for particle calculation.

(a)

(b)

(c)
Figure 8.9, Auto correlation index; (a) x-velocity, (b) y-velocity, (c) z-velocity

170

8.4.4 Particle deposition results

The particle deposition data is presented in terms of, a) oral deposition efficiency, b) total
deposition efficiency and, c) generational deposition efficiency, and d) Lobar particle
deposition. For the range of particle sizes considered, inertial impaction and
sedimentation are the dominant mechanisms for particle deposition. Inertial impaction
can be characterized by the Stokes number (Finlay [49]). Therefore, in the following
particle deposition plots, the x-axis is represented as function of Stokes number.
I Oral deposition efficiency
Figure 8.10 compares the oral deposition efficiency from the present simulations against
the LES particle simulations of Lambert et al. [47] and the empirical Grgic curve. The
oral deposition comprises the particle deposition occurring in the mouth, the pharynx and
the larynx. The Stokes and Reynolds number used in the x axis of the plot are defined as

p d p2U m
Stk
18d m
U m d m
Re

(8.1)

In the above equation , and p are the fluid phase and particle density, respectively, d m
is the mean hydraulic diameter, U m is the mean velocity and is the dynamic viscosity
of the fluid phase. The terms d m and U m are defined as,
V
L
Q
Um
LV
dm 2

(8.2)

In the present case, L=0.185, V=4.73 ml corresponding to the oral cavity. As can be
observed the present oral deposition correlates well with the Grgic curve. On the other
hand, there is a slight overprediction at the lower end of Stk * Re0.37 range for the case of
171

Lambert et al. [47]. When comparing the present particle calculation with that of Lambert
et al. [47], the x-axis is not the same for a given particle size. This is mainly because of
the difference in flow rates and airway dimensions used. Overall, the oral deposition
correlates well for the present case and that of Lambert et al. [47].

Figure 8.10, Oral Deposition efficiency

II Total Deposition Efficiency

Figure 8.11 examines the total deposition efficiency as a function of the Stokes number.
In this figure, the Stokes number is calculated using the mean flow velocity at the inlet
and the hydraulic diameter of the inlet. For this plot, the Stokes number is chosen since
for particles in which the deposition due to inertial impaction and sedimentation is the
dominant mechanism, inertial impaction can be characterized solely as a function of
Stokes number. The plots show that in respective airway geometries (Lambert et al. [47]
and the present airway model), increasing particle Stokes number increases the particle
deposition. Also, particles with Stokes number less than 0.01 transport much deeper
inside the lungs. Qualitatively, the particle deposition trends in both cases are similar.

172

Figure 8.11, Total deposition efficiency

III Generation deposition efficiency

Figure 8.12, Deposition efficiency at Generation G1

To explore the local deposition behavior, Figure 8.12 presents a qualitative comparison
between the present results with those obtained in experiments reported in literature. The
experimental data of Zhou and Cheng [131] are obtained for a four-generation airway
replica made from an adult cadaver. Their airway replica included an oral cavity,
pharynx, larynx, trachea, and four generations of the bronchi. In addition, the results are
also compared with those of Chan and Lippmann [130], who studied particle deposition
173

in a hollow cast of the human larynx-tracheobronchial tree from the first to the sixth
generations. The generational deposition efficiency plotted is defined as the ratio of the
number of deposited particles divided by the number of particles that entered the airway
branch. In the present case, generation G1 is composed of trachea as the parent branch
and the left and the right main bronchus as the daughter branches (see Figure 8.1).
Regardless of the variability seen among the plotted deposition data, the dependence of
deposition on Stokes number shows similar trends. The Stokes number for generation
deposition efficiency is defined as in equation (8.1), but in this case U m is the mean flow
in the parent branch and d m is the average diameter of the parent branch.
IV Lobar based particle distribution

(a)

(b)

Figure 8.13, Lobar deposition efficiency: (a) 2 m, (b) 10 m

In Figure 8.13, the y-axis represents the particle entering in: i) generation 5 (represented
by square symbol), ii) all fives lobes (represented as circles), and iii) the lungs i.e. the left
and the right lung (represented by triangles). This is compared with the respective flow
distribution, plotted on the x-axis. In both cases, either for 2 m or 10 m, the particles
received are in accordance with the flow distribution, e.g. in the right middle lobe (RML)
174

receives almost 8.2% of the total flow (calculated from Table 8.1) and the percentage of
particles received in the RML is around 8.1% of the particles. Similarly, according to
Table 8.1, the left lung (47%) receives less flow than does the right lung (53%), the same
is also reflected in particle distribution. This observation of particle distribution in
proportion to the flow distribution is observed both for a small and a larger particle size
(as can be seen from Figure 8.13 (a) and (b)). In fact, the results for the smaller particle
size lie almost on the 1:1 ratio line (dashed line). For the larger particle size there is more
scatter, which can be attributed to inertia (since 10 m have higher inertia than compared
to 2 m) possibly combined with a gravitational effect since at the lower airways
deposition due to sedimentation becomes important. One way to check this is to compare
with a simulation where gravity is ignored. This was not further investigated however
because of lack of time.
The observation made in Figure 8.13 (also reported by Darquenne et al. [132]) is in
contrast to the observation of Lambert et al. [47]. According to Lambert et al. [47], even
though the left receives less air flow than the right lung, more particles enter the left lung
as compared to the right lung.

8.5 Conclusions
The present chapter presented the application and the success of multiple LES frozen
field approach (described in Chapter 7) in evaluating particle deposition in a CT-scan
based 5 generation intrathoracic airway for a steady inspiratory flow rate of 30 l/min. The
airway model is generated by merging the earlier simplified UAM representing
extrathoracic region with a CT-scan of an adult female cadaver. The particle deposition
results were compared with the LES results of Lambert et al. [47] and experimental
results of [130] and [131].
It was observed that imposing uniform pressure at the outlets results in a good agreement
of flow distribution with experiments. It was also noted that the left lung receives less
175

flow than does the right lung in consistent with the observation made during experiments
and in the works of Lambert et al. [47] and Darquenne et al. [132].
In terms of particle deposition, an overall good agreement can be seen in the oral region
with the present result being closer to the Grgic curve and comparable with Lambert et al.
[47]. Also, a comparable agreement at least in qualitative terms can be observed for total
and generational deposition. However, it is to be mentioned that the compared geometries
were not geometrically identical and therefore dedicated experiments at 30 l/min is
required which at the moment is not available. This will allow for a more quantitative
validation of the current simulations.
An important observation made, in contrast to Lambert et al. [47] but in agreement with
Darquenne et al. [132], that the particle distribution is in proportion with flow
distribution. This was especially true for smaller particle sizes, where the data falls on the
identity line representing 1:1 ratio. However, there is scatter in deposition data for bigger
sized particles although they still distribute in proportion to flow distribution. This could
either be attributed to the inertia (since bigger sized particle have higher inertia than
compared to smaller sized particle). Or a combination of inertial and gravitational effect
since at lower airways deposition due to sedimentation becomes important. One way to
check this is to compare the present data with a simulation where gravity is ignored. But
this is the scope for future work.

176

Chapter 9
Conclusion and Future work
Contents
9.1RANS......................................................................................................................................................177
9.2 LES........................................................................................................................................................178
9.3 Future work..........................................................................................................................................179
9.3.1 Boundary condition.......................................................................................................................179
9.3.3 Moving larynx..............................................................................................................................181
9.3.4 Uncertainity quantification.........................................................................................................181

The importance for evaluating particle deposition characteristics in realistic human

airway geometries has now been recognized and therefore is the subject of many
researches. This work under the project ASILUNG presents a step forward and mainly
focused in accessing existing numerical methodologies involved in fluid/particle
simulation. The current work involves fluid particle simulation based on RANS and LES
methodologies. Following are the main achievements.

9.1 RANS
Fluid flow and particle simulation in a CT-scan based simplified upper airway model
representing extra-thoracic region were investigated. Inhalation flow rates of 30 l/min and
60 l/min were considered with particle diameters ranging from 2m-10m. Within the
frame work of RANS when using classical EIM, it was observed that numerical particle
data were largely overpredicted in consistent with the observations made in previous
studies on a similar geometry. For respirable range (1m-5m) at 30 l/min, we observe
an overprediction of almost 30% in total deposition efficiency and also inaccurate local
deposition behavior when compared against available experimental data. This deficiency
177

was mainly attributed to the isotropic assumption involved in classical EIM. This
prompted us to propose a new EIM model referred to as helicity EIM (Chapter 4). The
new anisotropic EIM has the advantage over existing Wang and James EIM that it does
not require any priori estimation of y+ and accounts the flow anisotropy directly from the
computed flow field. The method was tested on a 900 bend pipe and against the available
experimental data in UAM. The particle results were in relatively good agreement over
the entire particle range considered. However, some deviations still existed especially at
higher flow rates of 60 l/min (Chapter 5). Several numerical experiments were carried out
in Chapter 5 and the effect of inlet velocity profiles and the effect of additional tubing
during the experiments were assessed. It was shown that helicity EIM based particle
deposition is highly sensitive to inlet velocity profile and in particular to the computed
flow field. It was seen that in the event of non-availability of inlet velocity profile,
parabolic velocity profile mimicked closely the observations made in experiments

9.2 LES
The need for accurate flow simulation is recognized in past studies and was also shown in
Chapter 5. The fact that inaccuracies exists in predicting transitional flow in human
airways when using RANS prompted studies involving LES and DNS methodologies.
However, so far fewer studies involving LES have been conducted, mainly because of
computational cost which is several times higher than that of RANS. In the framework of
eddy viscosity based SGS models for LES, the complete model is the DSM. Models such
as constant Smagorinsky and WALE are not complete in the sense that one needs to
supply the value for constant appearing in the formulation. However, the major
deficiency in DSM is the evaluation of model constant which is not straightforward in
complex unstructured geometries like human airways. Therefore, the recently developed
RoSM SGS model was implemented and verified. The model was posteriori tested on
several test cases before applying on UAM geometry. It was shown that the performance
of this model was similar to DSM and in some cases better but the main advantage is that
it is 25% faster than compared to DSM. This is important saving since one still has to
178

conduct particle calculations. However, when several particle ranges are being
considered, conducting LES/particles simulation is still computationally time consuming.
Therefore, we carried forward the idea proposed by Matida et al. [46] and Jayaraju et al.
[44] of using frozen-field approach for particle simulation. But the above authors still
experienced some deterioratory particle data, infact Jayaraju et al. [44] observed that
RANS coupled with EIM resulted in better results than frozen-field approach. This was
mainly attributed to using only one LES frozen-field. Therefore, there was a need for
using several such frozenfields for which there was no systematic procedure in
estimating number of such field required. In Chapter 7, based on a POD analysis of
frozen fields, a systematic procedure was introduced and tested in UAM, the results were
in good agreement with the available experimental data. This way the computational time
was reduced by 50%. This offers a great advantage especially when several particle sizes
need to be considered. Finally, owing to the success gained in Chapter 7, the multiple
LES frozen-field approach was applied to upper airway geometry consisting of extrathoracic region and tracheobronchial region upto 5th generation (Chapter 8). The
simulated particle results were shown to be at par with the available literature considered.

9.3 Future work

9.3.1 Boundary condition
So far, we have evaluated RANS and LES methodologies using steady velocity profile at
the inlet. Therefore, the results cannot be translated to any physiological conclusions. In
order to do so, time-dependent velocity inlet representing realistic breathing conditions is
required to be applied and tested. In literature, time dependent boundary conditions
assuming a sinusoidal profile exist such as used in [133]:
Qt Q0 sin(t )
with
179

(9.1)

VT w
2
f
w
(Ti / Ttot )

Q0

(9.2)

where VT is the tidal volume, f is the breathing frequency, Ti is the time duration of
single, and Ttot is the time duration of single breadth. One can also use realistic inlet
waveforms such as mentioned in [134].

9.3.2 Detailed comparison of particle data in present tracheobronchial

geometry
Chapter 8 presented the preliminary results for the case of 30 l/min in a tracheobronchial
geometry using the multiple LES frozen field approach. At the time of writing this thesis
since the experimental particle data at 30 l/min were not available as such qualitative
comparison were made based on the recent available literature. Since some interesting
observations were made such as particle distribution in proportion to flow distribution
and variability of this with respect to particle size, it is recommended to have a detailed
comparison of generational (1-5) and lobar based deposition with experiments. Infact, the
preliminary results for the case of 60 l/min do indicate that particle distribution is in
accordance with the flow distribution (see Figure 9.1). Moreover, an additional LES
simulation with 60 l/min, which is currently now underway, should also be compared
with experiments. This way the effect of flow rate for a given particle size and vice versa
can be studied.

180

Figure 9.1, Lobar particle deposition versus lobar flow distribution at 60 l/min, preliminary
experimental results.

9.3.3 Moving Larynx

So far, all the calculation was done based on single glottal opening. However, the glottal
opening varies considerably during quiet breathing. For example, Renotte et al. [135]
observed that glottis dimension changes between 5.70.5 and 10.10.5 mm during
respiratory cycle. This can have considerable effect on particle deposition characteristics
as such numerical experiments involving moving mesh needs to be performed to check
the extent of deviation of the particle deposition compared to single glottis opening.

9.3.4 Uncertainty quantification

Till now all the fluid/particle simulations conducted in this work and those reported in
literature assumes flow and particle conditions (such as particle size, airway geometry,
inflow conditions etc.) to be known with certainty. But in reality variability exists in
these parameters which inturn can have effect on the final result. For example, it was
partially shown in Chapter 5 that particle deposition characteristics vary with different
inlet velocity profiles. Similarly, all calculated particle data were based on assuming
particle sizes to be known with certainty whereas the particle sizes generated during the
experiments have a standard deviation of 1.2. Several methods exist to quantify
uncertainties. The straightforward approach is to use Monte Carlo simulation but this
involves lot of samples which makes the method unsuitable for airway application. From
181

practicality point of view a non-intrusive method should be preferred (for example

Stochastic Collocation method by Mathelin and Hussain [136]).

182

List of Publications
Journal Articles
1) V. Agnihotri, G. Ghorbaniasl, S. Verbanck and C. Lacor. An eddy interaction model
for particle deposition. Journal of Aerosol Science. pp. 39-47, 2012
2) V. Agnihotri, G. Ghorbaniasl, S. Verbanck and C. Lacor. On the multiple LES
frozen field approach for the prediction of particle deposition in the human upper
respiratory tract. Journal of Aerosol Science.pp. 58-72, 2014
3) G. Ghorbaniasl, V. Agnihotri and C. Lacor, "A self-adjusting flow dependent
formulation for the classical Smagorinsky model coefficient," Physics of Fluids, vol. 25,
2013
4) S. Verbanck, H. Kalsi, M. Biddiscombe, V. Agnihotri, B. Belkassem, C. Lacor and O.
Usmani, "Inspiratory and expiratory aerosol deposition in the upper airway," Inhalation
Toxicology, vol. 23, p. 104111, 2011.
Conference and proceedings
1) V. Agnihotri, G. Ghorbaniasl, S. Verbanck, C. Lacor. Performance of helicity eddy
interaction model for simulation of Aerosol deposition Patterns in a Simplified Human
Upper Airway Model, VI European Conference on Computational Fluid Dynamics
ECOOMAS CFD 2012, Vienna,Austria.
2) V.Agnihotri, G.Ghorbaniasl, S.Verbanck, C.Lacor . Numerical Study of Aerosol
deposition in a Simplified Human Mouth-Throat Model. National Congress on
Theoretical and Applied Mechanics (2012), Royal Military Academy, Belgium.

183

3) G. Ghorbaniasl, V. Agnihotri and C. Lacor . On the estimation of subgrid scale

model coefficent in large eddy simulation. VI European Conference on Computational
Fluid Dynamics ECOOMAS CFD 2012, Vienna,Austria.
4)V. Agnihotri, K. Elsayed, C. Lacor and S. Verbanck . Numerical study of Particle
deposition in the human upper airway with emphasis on the hot-spot study and
comparison of LES and RANS results. V European Conference on Computational Fluid
Dynamics ECOOMAS CFD 2010, Lisbon,Portugal.
5) V Agnihotri, S Verbanck, and C Lacor Numerical study of Particle deposition in the
human upper airway with emphasis on the hot-spot study. ERCOFTAC, Cenaero,
Belgium, 10Dec 2011.

184

Bibliography
[1]

I. P. 66, "Annals Of the ICRP, 24, Nos 1-3," Pergamon/Elsevier, Tarrytown, NY,
1994.

[2]

"World health organisation," 2011. [Online]. Available: www.who.int.

[3]

R. V. Lourenco and E. Cotromanes, "Clinical aerosols. 1. characterization of

aerosols and their diagnostic uses," 1982, pp. 142:2163-2172.

[4]

N. R. Labiris and M. B. Dolovich, "Pulmonary drug delivery. part 1: Physiological

factors affecting therapeutic effectiveness of aerosolized medications," Clinical
Pharmacology, pp. 56-6:588599, 2003.

[5]

E. R. Weibel, Morphometry of the human lung, Berlin: Springer-Verlag, 1963.

[6]

K. Horsfield, G. Dart, D. Olson, G. Filley and G. Cumming, "Models of the human

bronchial tree," J. Appl. Physiol, p. 31:207217, 1971.

[7]

"Aerosol Research Laboratory of Alberta," [Online].

[8]

H. Kitaoka, R. Takaki and B. Suki, "A three-dimensional model of the human

airway," Journal of Applied Physiology, vol. 87, p. 22072217, 1999.

[9]

M. Tawhai, A. Pullan and P. Hunter, "Generation of an anatomically based threedimensional model of the conducting airways," Annals of Biomedical Engineering,
vol. 28, p. 793802, 2000.

[10] E. Matida, W. Finlay, C. Lange and B. Grgic, "Improved numerical simulation of

aerosol deposition in an idealized mouth-throat," J. Aerosol Science, vol. 35, pp. 119, 2004.
185

[11] C. van Ertbruggen, C. Hirsch and M. Paiva, "Anatomically based threedimensional model of airways to simulate flow and particle transport using
computational," Journal of Applied Physiology, vol. 98, p. 970980, 2005.
[12] P. Nithiarasu, O. Hassan, K. Morgan, N. P. Weatherill, C. Fielder, H. Whittet, P.
Ebden and K. R. Lewis, "Steady flow through a realistic human upper airway
geometry," Int. J. Numer. Meth. Fluids, vol. 57, p. 631651, 2008.
[13] I. Katz and T. Martonen, "Flow patterns in three-dimensional laryngeal model,"
Journal of Aerosol Medicine, Vols. 9-4, p. 501511, 1996.
[14] T. Corcoran and N. Chigier, "Characterization of the laryngeal jet using phase
doppler interferometery," Journal of Aerosol Medicine, vol. 13, p. 125137, 2000.
[15] C. Kleinstreuer and Z. Zhang, "Laminar-to-turbulent fluid-particle flows in a
human airway model," International Journal of Multiphase Flow, vol. 29, p. 271
289, 2003a.
[16] A. Farkas, I. Balashazy and K. Szocs, "Characterization of regional and local
deposition of inhaled aerosol drugs in the respiratory systemby computational fluid
and particle dynamics methods," Journal of Aerosol Medicine, vol. 19, p. 329343,
2006.
[17] H. Jin, J. Fan, M. Zeng and K. Cen, "Large eddy simulation of inhaled particle
deposition within the human upper respiratory tract," Journal of Aerosol Science,
vol. 38, p. 257268, 2007.
[18] J. Xi and P. Longest, "Transport and deposition of micro-aerosols in realistic and
simplified models of the oral airway," Annals of Biomedical Engineering, vol. 35,
p. 560581, 2007.

186

[19] Y. Cheng, Y. Zhou and B. Chen, "Particle deposition in a cast of human oral
airways," Aerosol Science and Technology, vol. 31, p. 286300, 1999.
[20] K. Stapleton, E. Guentsch, M. Hoskinson and W. Finlay, "On the suitability of k-
turbulence modeling for aerosol deposition in the mouth and throat: A comparison
with experiment," Journal of Aerosol Science, vol. 31, p. 739749, 2000.
[21] W. DeHaan and W. Finlay, "In vitro monodisperse aerosol deposition in a mouth
and throat with six different inhalation devices," Journal of Aerosol Medicine, vol.
14, p. 361367, 2001.
[22] B. Grgic, W. Finlay and A. Heenan, "Regional aerosol deposition and flow
measurements in an idealized mouth and throat," Journal of Aerosol Science, vol.
35, pp. 21-32, 2004.
[23] M. Brouns, S. Jayaraju, C. Lacor, J. Mey, M. Noppen, W. Vincken and S.
Verbanck, "Tracheal stenosis: A flow dynamics study," Journal of Applied
Physiology, vol. 102, p. 11781184, 2007.
[24] N. Nowak, P. P. Kakade and A. V. Annapragada, "Computational fluid dynamics
simulation of airflow and aerosol deposition in human lungs," Annals of
Biomedical Engineering, vol. 31, pp. 373-390, 2003.
[25] H. Shi, C. Kleinstreuer and Z. Zhang, "Nanoparticle transport and depostion in
bifurcating tubes with different inlet conditions," Physics of Fluids, vol. 16, pp.
2199-2213, 2004.
[26] Z. Zhang and C. Kleinstreuer, "Airflow structures and nano-particle deposition in a
human upper airway model," Journal of Computational Physics, vol. 198, pp. 178210, 2004.

187

[27] P. Longest and S. Vinchurkar, "Effects of mesh style and grid convergence on
particle deposition in bifurcating airway models with comparisons to experimental
data," Medical Engineering and Physics, vol. 29, pp. 350-366, 2007.
[28] T. Vasconcelos, B. Sapoval, J. Andrade, J. Grotberg, Y. Hu and M. Filoche,
"Particle capture into the lung made simple," J Appl Physiol, vol. 110, p. 1664
1673, 2011.
[29] L. Holbrook and P. Longest, "Validating CFD predictions of highly localized
aerosol deposition in airway models: In vitro data and effects of surface
properties," Journal of Aerosol Science, vol. 59, pp. 6-21, 2013.
[30] C. Lin, M. H. Tawhai, G. McLennanc and E. Hoffmanc, "Characteristics of the
turbulent laryngeal jet and its effect on airflow in the human intra-thoracic
airways," Respiratory Physiology and Neurobiology, vol. 157, p. 295309, 2007.
[31] B. Ma and K. Lutchen, "An Anatomically Based Hybrid Computational Model of
the Human Lung and its Application to Low Frequency Oscillatory Mechanics,"
Annals of Biomedical Engineering, vol. 34, p. 16911704, 2006.
[32] P. Ghalati, E. Keshavarzian, O. Abouali, A. Faramarzi, J. Tu and A. Shakibafard,
"Numerical analysis of micro-and nano-particle deposition in a realistic human
upper airway," Computers in Biology and Medicine, vol. 42, pp. 39-49, 2012.
[33] W. Hofmann, R. Golser and I. Balashazy, "Inspiratory deposition efficiency of
ultrafine particles in a human airway bifurcation model," Aerosol Science and
Technology, vol. 37, p. 988994, 2003.
[34] J. Comer, C. Kleinstreuer and Z. Zhang, "Flow structures and particle deposition
patterns in double-bifurcation airway models. Part 1. Air flow fields," Journal of
Fluid Mechanics, vol. 435, p. 2554, 2001.

188

[35] Z. Li, C. Kleinstreuer and Z. Zhang, "Particle Deposition in the Human

Tracheobronchial Airways Due to Transient Inspiratory Flow Patterns," J. Aerosol
Science, vol. 38, p. 624644, 2007.
[36] C. Kleinstreuer, Z. Zhang and C. Kim, "Combined inertial and gravitational
deposition of microparticles in small model airways of a human respiratory
system," Journal of Aerosol Science, vol. 38, p. 10471061, 2007.
[37] T. Martonen, Y. Yang and Z. Xue, "Influences of cartilaginous rings on
tracheobronchial fluid dynamics," Inhalation Toxicology, vol. 6, pp. 185-203, 1993.
[38] D. Wilcox, Turbulence modeling for CFD, California: DCW Industries, Inc, 1998.
[39] S. Pope, Turbulent flows, New York: Cambridge university press, 2000.
[40] Z. Zhang and C. Kleinstreuer, "Low-reynolds-number turubulent flows in locally
constricted conduits: A comparison study," AIAA Journal, vol. 41, pp. 831-840,
2002.
[41] Y. Zhang, W. Finlay and E. Matida, "Particle deposition measurements and
numerical simulation in a highly idealized mouth-throat," Journal of Aerosol
Science, vol. 35, p. 789803, 2004..
[42] S. 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, vol. 38, p. 494:508, 2007.
[43] B. Grgic, A. Martin and W. Finlay, "The effect of unsteady flow rate increase on in
vitro mouth-throat deposition of inhaled boluses," Journal of Aerosol Science, vol.
37, p. 12221233, 2006.
[44] S. Jayaraju, M. Brouns, C. Lacor, B. Belkassem and S. Verbanck, "Large eddy and
189

detached eddy simulations of fluid flow and particle deposition in a human

mouththroat," Journal of Aerosol Science, vol. 39, p. 862875, 2008.
[45] S. Verbanck, H. Kalsi, M. Biddiscombe, V. Agnihotri, B. Belkassem, C. Lacor and
O. Usmani, "Inspiratory and expiratory aerosol deposition in the upper airway,"
Inhalation Toxicology, vol. 23, p. 104111, 2011.
[46] E. Matida, W. Finlay, M. Breuer and C. Lange, "Improving prediction of aerosol
deposition in an idealized mouth using large eddy simulation," Journal of Aerosol
Medicine, vol. 19, p. 290300, 2006.
[47] A. Lambert, P. O'shaughnessy, M. Tawhai, E. Hoffman and C. Lin, "Regional
Deposition of Particles in an Image-Based Airway Model: Large-Eddy Simulation
and Left-Right Lung Ventilation Asymmetry," Aerosol Science and Technology,
vol. 45:1, pp. 11-25, 2011.
[48] H. Jin, J. Fan, M. Zeng and K. Cen, "Large eddy simulation of inhaled particle
deposition within the human upper respiratory tract," Aerosol Science, vol. 38, p.
257 268, 2008.
[49] W. Finlay, The mechanics of inhaled pharmaceutical aerosols, London: Academic
press, 2001.
[50] Q. Lu, J. Fontaine and G. Aubertin, "A lagrangian model for solid particles in
turbulent flows," International Journal of Multiphase Flow, Vols. 19-2, p. 347
367, 1993.
[51] M. Maxey, "The gravitational settling of aerosol particles in homogeneous
turbulence and random flow fields," Journal of Fluid Mechanics, vol. 174, pp. 441465, 1987.

190

[52] J. Pozorski and J. Minier, "The pdf method for lagrangian two-phase flow
simulations," Gas Particle Flows, 1995.
[53] P. Hutchinson, G. Hewitt and A. Dukler, "Deposition of liquid or solid dispersions
from turbulent gas streams: a stochastic model," Chemical Engineering Science,
vol. 26, p. 419439, 1971.
[54] A. Gosman and E. Ioannides, "Aspects of computer simulation of liquid-fuelled
combustor," AIAA Journal, pp. 81-0323, 1981.
[55] M. Sommerfeld, A. Ando and D. Wennerberg, "Swirling, particle-laden flows
through a pipe expansion," ASME: Journal of Fluids Engineering, vol. 114, p. 648
656, 1992.
[56] D. Graham and P. James, "Turbulent dispersion of particles using eddy interaction
models," International Journal of Multiphase Flow, Vols. 22-1, pp. 157-175, 1996.
[57] D. Graham, "On the inertia effect in eddy interaction models," International
Journal of Multiphase Flow, Vols. 22-1, p. 177184, 1996.
[58] D. Graham, "Improved eddy interaction models with random length and time
scales," International Journal of Multiphase Flow, Vols. 24-2, p. 335345, 1998.
[59] D. Graham, "Spectral characteristics of eddy interaction models," International
Journal of Multiphase Flow, vol. 27, p. 10651077, 2001.
[60] Z. Zhang, C. Kleinstreuer and C. Kim, "Micro-particle transport and deposition in a
human oral airwaymodel," Journal of Aerosol Science, vol. 33, p. 16351652,
2002.
[61] Y. Wang and P. James, "On the effect of anisotropy on the turbulent dispersion and
deposition of small particles," International Journal of Multiphase Flow, vol. 25, p.
191

551558, 1999.
[62] J. Sandeau, I. Katz, R. Fodil, B. Louis, G. Apiou-Sbirlea, G. Caillibotte and D.
Isabey, "CFD simulation of particle deposition in a reconstructed human oral
extrathoracic airway for air and heliumoxygen mixtures," Journal of Aerosol
Science, vol. 41, p. 281294, 2010.
[63] K. Inthavong, T. Jiyuan, Y. Yong, S. Ding, S. Aleks and F. Thien, "Effects of
airway obstruction induced by asthma attack on particle deposition," Journal of
Aerosol Science, 2010.
[64] Y. Zhang, W. Finlay and E. Matida, "Particle deposition measurements and
numerical simulation in a highly idealized mouththroat," Aerosol Science, vol. 35,
pp. 789-803, 2004.
[65] V. Agnihotri, G. Ghader, S. Verbanck and C. Lacor, "On the multiple LES frozen
field approach for the prediction of particle deposition in the human upper
respiratory tract," Journal of Aerosol Science, vol. 68, pp. 58-72, 2014.
[66] H. Tennekes and J. L. Lumley, A first course in turbulence, MIT press, 1972.
[67] O. Reynolds, "An experimental investigation of the circumstances whcih
determnine whetehr the motion of water shall be direct or sinous and of the lea of
resistance in parallel channels," Philosophical Transactions of the Royal Society of
London, vol. 172, pp. 935-982, 1883.
[68] B. Baldwin and H. Lomax, "Thin-Layer Approximation and Algebraic Model for
Separated Turbulent Flows," aiaa, pp. 78-257, 1978.
[69] F. Menter, "Two equation eddy-viscosity turbulence models for engineering
applications," AIAA Journal, Vols. 32-8, p. 15981605, 1994.

192

[70] F. Menter, J. Ferreira, T. Esch and B. Konno, "The SST Turbulence Model with
ImprovedWall Treatment for Heat Transfer Predictions in Gas Turbines," in
Proceedings of the International Gas Turbine Congress, Tokyo, 2003.
[71] A. Leonard, "Energy cascade in les of turbulent fluid flows," Advances in
Geophysics, vol. 18, p. 237248, 1974.
[72] J. Deardorff, "A numerical study of three-dimensional turbulent channel flow at
large Reynolds number," J.Fluid Mech, vol. 41, pp. 453-480, 1970.
[73] P. Saugut, Large eddy simulation for incompressible flows, Springer, 1998.
[74] J. Smagorinsky, "General circulation experiments with the primitive equation,"
Monthly Weather Report, Vols. 91-3, p. 99106, 1963.
[75] D. Lilly, "The representation of small-scale trubulence in numerical simulation
experiments," in IBM Scientific Computing Symp. on Environmental Sciences,
Yorktown Heights, 1967.
[76] M. Germano, U. Piomelli, P. Moin and W. Cabot, "A dynamic subgridscale eddy
viscosity model," Physics of Fluids, vol. 3, p. 1760, 1991.
[77] L. Fan and C. Zhu, Principles of Gas-Solid Flows, Cambridge Unievrsity Press,
1998.
[78] S. Elghobashi, "On Predicting Particle-Laden Turbulent Flows," Applied Scientific
Research, Netherlands, 1994.
[79] C. Kleiunstreuer, Z. Zhang and Z. Li, "Modeling airflow and particle
transport/deposition

in

pulmonary

airways,"

Neurobiology, vol. 163, pp. 128-138, 2008.

193

Respiratory

Physiology

ans

[80] F. Krause, A. Wenk, C. Lacor, W. Kreyling, W. Moller and S. Verbanck,

"Numerical and experimental study on the deposition of nanoparticles in an
extrathoracic oral airway model," Journal of Aerosol Science, vol. 57, pp. 131-143,
2013.
[81] C. Kleinstreuer and Z. Zhang, "Airflow and Particle transport in the Human
Respiratory System," Annual Review of Fluid Mechanics, vol. 42, pp. 301-3334,
2010.
[82] L. Schiller and A. Neumann, "Uber die grundlegenden berechnungen bei der
schwer kraftaufbereitung," Verein Deutscher Ingenieure, pp. 77-318, 1933.
[83] S. Morsi and A. Alexander, "An investigation of particle trajectories in two-phase
flow systems," Journal of Fluid Mechanics, vol. 55, pp. 193-208, 1972.
[84] A. Mehel, A. Taniere, B. Oesterle and J. Fontaine, "The influence of an anisotropic
Langevin dispersion model on the prediction of micro-and nano particle deposition
in wall-bounded turbulent flows," Journal of Aerosol Science, vol. 41, p. 729744,
2010.
[85] A. Dehbi, "Prediction of Extrathoracic Aerosol Deposition using RANS-Random
Walk and LES Approaches," Aerosol Science and Technology, vol. 45, no. 5, pp.
555-569, 2011.
[86] G. Kallio and M. Reeks, "A numerical simulation of particle deposition in turbulent
boundary layers," International Journal of Multiphase flow, vol. 15, pp. 433-446,
1989.
[87] E. Matida, K. Nishino and K. Torii, "Statistical simulation of particle deposition on
the walls from turbulent dispersed pipe flow," International Journal of Heat and
Fluid, vol. 21, p. 389402, 2000.

194

[88] J. Kim, P. Moin and R. Moser, "Turbulence statistics in fully developed channel
Flow at low Reynolds number," Journal of Fluid Mech, vol. 177, pp. 133-166,
1987.
[89] N. Mansour, J. Kim and P. Moin, "Reynolds-stress and dissipation rate budgets in a
turbulent channel flow," Journal of Fluid Mech, vol. 194, pp. 15-44, 1988.
[90] D. Pui, F. Romay-Novas and B. Liu, "Experimental study of particle deposition in
bends with circular cross section," Aerosol Science and Technology, vol. 7, no. 3,
pp. 301-315, 1987.
[91] S. Jayaraju, Study of the Air Flow and Aerosol Transport in the Human Upper
airway using LES and DES Methodologies, Brussels: VUB press, 2009.
[92] Z. Zhang, C. Kleinstreuer, J. Donohue and C. Kim, "Comparison of micro- and
nano-size particle depositions in a human upper airway model," Journal of Aerosol
Science, vol. 36, pp. 211-233, 2005.
[93] J. Xi and P. Longest, "Transport and deposition of micro-aerosols in realistic and
simplified models of the oral airway," Annals of biomedical engineering, vol. 35,
no. 4, pp. 560-581, 2007.
[94] P. Longest and J. Xi, "Computational investigation of particle inertia effects on
submicron aerosol deposition in the respiratory tract," Journal of Aerosol Science,
vol. 38, no. 1, pp. 111-130, 2007.
[95] R. Robinson, P. Snyder and M. Oldham, "Comparison of particle tracking
algorithms in commercial CFD packages: sedimentation and diffusion," Inhalation
Toxicology, vol. 19, p. 517531, 2007.
[96] P. Longest and S. Vinchurkar, "Inertial deposition of aerosols in bifurcating models
during steady expiratory flow," Jounal of Aerosol Science, vol. 40, pp. 370-378,
195

2009.
[97] G. Ghorbaniasl, V. Agnihotri and C. Lacor, "A self-adjusting flow dependent
formulation for the classical Smagorinsky model coefficient," Physics of Fluids,
vol. 25, 2013.
[98] M. Brouns, Numerical and Experimnetal study of Flow and deposition of Aerosols
in the Upper Human Airwways, Brussels: VUB press, 2007.
[99] C. V. Ertbruggen, Study of aerosol transport and deposition in the lungs using
Compuational fluid dynamics, Brussels: ULB press, 2005.
[100] M. Breuer, H. Baytekin and E. Matida, "Prediction of aerosol deposition in 90
degree bends using LES and an efficient Lagrangian tracking method," Journal of
Aerosol Sci., vol. 37, p. 14071428, 2006.
[101] B. Grgic, W. Finlay, P. Burnell and A. Heenan, "In vitro intersubject and
intrasubject deposition measurements in realistic mouththroat geometries,"
Journal of Aerosol Science, vol. 35, p. 10251040, (2004).
[102] P. Longest and S. Vinchurkar, "Effects of mesh style and grid convergence on
particle deposition in bifurcating airway models with comparisons to experimental
data," Medical engineering & physics, vol. 29, pp. 350-366, 2007.
[103] J. Bardina, J. Ferziger and W. Reynolds, "Improved subgrid scale models for LES,"
AIAA, p. 801357, 1980.
[104] Y. Y. Zang, R. Street and J. Koseff, "A dynamic mixed subgrid-scale model,"
Physics of Fluids A, vol. 5, p. 31863196, 1993.
[105] J. Meyers and P. Sagaut, "Is plane-channel flow a friendly case for the testing of
large-eddy simulation subgrid-scale," Phys. Fluids, vol. 19, 2007.
196

[106] R. Moser, J. Kim and N. Mansour, "Direct numerical simulation of turbulent

channel flow up to Re = 590," Phys. Fluids, vol. 11, no. 4, p. 943945, 1999.
[107] G. Winckelmans, H. Jeanmart and D. Carati, "On the comparison of turbulence
intensities from large-eddy simulation with those from experiment or direct
numerical simulation," Phys. Fluids, vol. 14, no. 5, 2002.
[108] S. Gavrilakis, "Numerical simulation of low-Reynolds-number turbulent flow
through a straight square duct," Journal of Fluid Mech., vol. 244, pp. 101-192,
1992.
[109] R. Madabhushi and S. Vanka, "Large eddy simulation of turbulence-driven
secondary flow in a square duct," Phys. Fluids A, vol. 3, p. 27342745 , 1991.
[110] C. Chyu and D. Rockwell, "Near-wake structure of an oscillating cylinder: Effect
of controlled shear-layer vortices," J. Fluid. Mech, vol. 322, pp. 21-49, 1996.
[111] A. Kravchenko and P. Moin, "Numerical studies of flow over a circular cylinder at
Re = 3900," Phys. Fluids, vol. 12, no. 2, p. 403416, 2000.
[112] P. Beaudan and P. Moin, "Numerical experiments on the flow past a circular
cylinder at sub-critical Reynolds number," Department of Mechanical Engineering,
Stanford University, 1994.
[113] R. Mittal and P. Moin, "Suitability of upwind-biased finite difference schemes for
large-eddy simulation of turbulent flows," AIAA, p. 14151417, 1997.
[114] X. Ma, G. Karamanos and G. Karniadakis, "Dynamics and low-dimensionality in
the turbulent near-wake," J. Fluid mECH, vol. 410, p. 2965, 2000.
[115] L. Lourenco and C. Shih, "Characteristics of the plane turbulent near wake of a
circular cylinder: A particle image velocimetry study," private communication,
197

1993.
[116] P. Parnaudeau, J. Carlie, D. Heitz and E. Lamball, "Experimental and numerical
studies of the flow over a circular at Reynolds number 3900," Phys. Fluids, vol. 20,
pp. 1-14, 2008.
[117] M. Meyer, S. Hickel and N. Adams, "Assessment of implicit large-eddy simulation
with a conservative immersed interface method for turbulent cylinder flow," Int. J.
Heat Fluid Flow, vol. 31, pp. 368-377, 2010.
[118] J. Choi, G. Xia, M. Tawhai, E. Hoffman and C. Ling, "Numerical study of highfrequency oscillatory air flow and convective mixing in a CT based human airway
model," Ann. Biomed. Eng., vol. 38, no. 12, p. 35503571, 2010.
[119] V. Armenio and V. Fiorotto, "The importance of forces acting on particles in
turbulent flows," Phys. Fluids, vol. 13, pp. 2437-2440, 2001.
[120] A. Berrouk and D. Laurence, "Stochastic modelling of aerosol deposition for LES
of 90 degree bend turbulent flow," Intl J. Heat Fluid Flow, vol. 29, pp. 1010-1028,
2008.
[121] M. Breuer, G. Durmus, E. Matida and W. Finlay, "LES and an efficient Lagrangian
tracking method for predicting aerosol deposition in turbulent flows," in Fifth
international symposium on turbulence, heat and mass transfer, Dubrovnik,
Croatia, 2006.
[122] J. Derksen, "Separation Performance Predictions of a Stairmand High-Efficiency
Cyclone," Fluid Mechanics and transport phenomena, vol. 49, 2003.
[123] J. Derksen, "Long-time solids suspension simulations by means of large-eddy
approach," Chemical Engineering Research and Design, vol. 84, p. 3846, 2006.

198

[124] L. Sirovich, "Turbulence and the dynamics of coherent structures, parts I-III,"
Quarterly of Applied Mathematics, vol. 65, pp. 561-590, 1987.
[125] A. Hekmati, D. Ricot and P. Druault, "About the convergence of POD and EPOD
modes computed from CFD simulation," Computer and Fluids, vol. 50, pp. 60-71,
2011.
[126] L. Cordier and M. Bergmann, "Proper orthogonal decomposition: An Overview,"
von Karman Institute for Fluid Dynamics, Brussels, 2003.
[127] K. Fukagata, S. Zahrai and F. Bark, "Dynamics of Brownian particles in a turbulent
channel flow," Heat Mass Transfer, vol. 40, pp. 715-726, 2004.
[128] B. Shotorban and F. Mashayek, "Modeling subgrid-scale effects on particles by
approximate deconvolution," Phys. Fluids, vol. 17, 2005.
[129] Q. Wang and K. Squires, "Large eddy simulation of particle-laden turbulent
channel flow," Phys. Fluids, vol. 8, pp. 1207-1223, 1996.
[130] T. Chan and M. Lippmann, "Experimental Measurements and Empirical Modeling
of the Regional Deposition of Inhaled Particles in Humans," Am. Ind. Hyg. Assoc,
vol. 41, p. 399409, 1980.
[131] Y. Zhou and Y. Cheng, "Particle Deposition in a Cast of Human Tracheobronchial
Airways," Aerosol Sci. Technol, vol. 39, p. 492500, 2005.
[132] C. Darquenne, v. Ertbruggen and G. K. Prisk, "Convective flow dominates aerosol
delivery to the lung segments," J. Appl. Phsyiol., vol. 111, no. 1, pp. 48-54, 2011.
[133] L. Vecellioa, P. Kippax, S. Rouquette and P. Diot, "Influence of realistic airflow
rate on aerosol generation by nebulizers," International Journal of Pharmaceutics,
vol. 371, pp. 99-105, 2009.
199

[134] Z. Zhang, C. Kleinstreuer and C. S. Kim, "Aerosol Deposition Efficiencies and

Upstream Release Positions for Different Inhalation Modes in an Upper Bronchial
Airway Model," Aerosol Science and Technology, pp. 828-844, 2010.
[135] C. Renotte, V. Bouffioux and F. Wilquem, "Numerical 3d oscillatory flow in the
time-varying laryngeal channel," Journal of Biomechanics, vol. 33, p. 16371644,
2000.
[136] L. Mathelin and M. Hussaini, "A Stochastic Collocation algorithm for uncertainity
quantification in CFD Simulations," Numerical Algorithms, vol. 38, pp. 209-236,
2005.
[137] W. Bennett, G. Scheuch, K. Zeman, J. Brown, C. Kim, J. Heyder and W.
Stahlhofen, "Regional Deposition and Retention of Particles in Shallow, Inhaled
Boluses: Effect of Lung Volume," J. Appl. Physiol, vol. 86, pp. 168-173, 1999.
[138] W. Moller, G. Meyer, G. Scheuch, W. Kreyling and W. Bennett, "Left-to-Right
Asymmetry of Aerosol Deposition after Shallow Bolus Inhalation Depends on
Lung Ventilation," J. Aerosol Med. and Pulmonary Drug Deliv., vol. 22, pp. 1-7,
2009.

200