The CFD Simulation of An Axial Ow Fan: Frederick Nicolaas Le Roux
Uploaded by
nassimThe CFD Simulation of An Axial Ow Fan: Frederick Nicolaas Le Roux
Uploaded by
nassimThe CFD simulation of an axial ow fan
by
Frederick Nicolaas le Roux
Master of Science in Mechanical Engineering
Thesis presented in partial ful lment of the requirements for
the degree of Master of Science in Mechanical Engineering at
Stellenbosch University
Department of Mechanical and Mechatronic Engineering,
University of Stellenbosch,
Private Bag X1, Matieland 7602, South Africa.
Supervisor: Mr. S.J. van der Spuy
Co-supervisor: Prof. T.W. von Backström
March 2010
Declaration
By submitting this thesis electronically, I declare that the entirety of the work
contained therein is my own, original work, that I am the owner of the copy-
right thereof (unless to the extent explicitly otherwise stated) and that I have
not previously in its entirety or in part submitted it for obtaining any quali -
cation.
Signature: . . . . . . . . . . . . . . . . . . . . . . . . . . .
F.N. le Roux
Date: ...............................
Copyright © 2010 Stellenbosch University
All rights reserved.
i
Abstract
The CFD simulation of an axial ow fan
F.N. le Roux
Department of Mechanical and Mechatronic Engineering,
University of Stellenbosch,
Private Bag X1, Matieland 7602, South Africa.
Thesis: MScEng (Mech)
March 2010
The purpose of this project is to investigate the method and accuracy of simu-
lating axial ow fans with three-dimensional axisymmetric CFD models. Two
models are evaluated and compared with experimental fan data. Veri cation
data is obtained from a prototype fan tested in a facility conforming to the BS
848 standards. The ow eld over the blade surfaces is investigated further
with a visualization experiment comprising of a stroboscope and wool tufts.
Good correlation is found at medium to high ow rates and recommendations
are made for simulation at lower ow rates as well as test guidelines at the fan
test facility. The results and knowledge gained will be used to amend currently
used actuator disc theory for axial ow fan simulation.
Key words: CFD, NUMECA, axial ow fan, ow visualization
ii
Uittreksel
Die BVD simulasie van `n aksiaalwaaier
(The CFD simulation of an axial ow fan)
F.N. le Roux
Departement Meganiese en Megatroniese Ingenieurswese,
Universiteit van Stellenbosch,
Privaatsak X1, Matieland 7602, Suid Afrika.
Tesis: MScIng (Meg)
Maart 2010
Die doel van hierdie projek is om die metode en akkuraatheid om aksiaalvloei-
waaiers met drie-dimensionele BVM modelle te simuleer, te ondersoek. Twee
modelle word geëvalueer en met eksperimentele waaiertoetse vergelyk. Veri-
kasie data is verkry vanaf `n prototipe waaier wat in `n fasiliteit getoets is en
wat aan die BS 848 standaarde voldoen. Die vloeiveld oor die lemoppervlak-
tes word ondersoek met `n visualisering eksperiment wat uit `n stroboskoop en
wolletjies bestaan. Goeie korrelasie word gevind vir medium tot hoë massa-
vloeie en aanbevelings word gemaak vir die simulasie by laer massavloeie met
riglyne vir toetswerk in die toets-fasiliteit. Die resultate en kennis opgedoen
sal gebruik word in die verbetering van huidige aksieskyfteorie vir numeriese
aksiaalvloeiwaaier simulasies.
Sleutelwoorde: BVM, NUMECA, aksiaalvloeiwaaier, vloei visualisering.
iii
Acknowledgements
I would like to express my sincere gratitude to the following people and organ-
isations:
My two supervisors for their continued support and guidance throughout
the duration of the project. Thank you Mr. SJ van der Spuy for your
motivation and encouragement in tiring times.
Fluxion for their nancial support.
My family for continually pushing me and giving me the determination
to nish.
My friends for their motivation, in particular Andrew de Wet. Thank
you for joining me on this journey.
And nally my Lord the Saviour, without whom this and every achieve-
ment in my life would not be possible.
iv
Dedications
This thesis is dedicated to my parents, for their perpetual love.
v
Contents
Declaration i
Abstract ii
Uittreksel iii
Acknowledgements iv
Dedications v
Contents vi
List of Figures ix
List of Tables xii
Nomenclature xiii
1 Introduction 1
2 Literature survey 6
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.1 Simulation of an axial ow fan . . . . . . . . . . . . . . . 6
2.1.2 Experimental . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Summary of main ndings . . . . . . . . . . . . . . . . . . . . . 12
3 Numerical modelling 14
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.2 Fan blade data . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3 Computational domain . . . . . . . . . . . . . . . . . . . . . . . 16
3.3.1 AutoGrid . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3.2 IGG . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.3.3 Domain speci cations . . . . . . . . . . . . . . . . . . . . 19
3.4 Computational parameters . . . . . . . . . . . . . . . . . . . . . 29
3.4.1 Fluid Model . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4.2 Flow Model . . . . . . . . . . . . . . . . . . . . . . . . . 30
vi
CONTENTS vii
3.4.3 Rotating Machinery . . . . . . . . . . . . . . . . . . . . . 30
3.4.4 Boundary Conditions . . . . . . . . . . . . . . . . . . . . 33
3.4.5 Numerical Model . . . . . . . . . . . . . . . . . . . . . . 34
3.4.6 Initial Solution . . . . . . . . . . . . . . . . . . . . . . . 38
3.4.7 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4.8 Computation Steering and Monitoring . . . . . . . . . . 38
3.5 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5.1 General Navier-Stokes Equations . . . . . . . . . . . . . 39
3.5.2 Time averaging of quantities . . . . . . . . . . . . . . . . 40
3.5.3 Treatment of turbulence in the equations . . . . . . . . . 41
3.5.4 Formulation in rotating frame for the relative velocity . . 41
3.5.5 Formulation in rotating frame for the absolute velocity . 42
4 Experimental evaluation 44
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.2 Fan Test Facility . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.3 Experimental setup parameters . . . . . . . . . . . . . . . . . . 45
4.3.1 Fan axis position . . . . . . . . . . . . . . . . . . . . . . 46
4.3.2 Blade angle . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3.3 Tip clearance . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3.4 Root seals . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.4 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.5 Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.6.1 Repeatability . . . . . . . . . . . . . . . . . . . . . . . . 48
4.6.2 Comparison with previous work . . . . . . . . . . . . . . 49
4.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5 CFD validation with experimental data 54
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.2 Fan performance curve . . . . . . . . . . . . . . . . . . . . . . . 54
5.3 Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.4 Investigation of streamline distribution . . . . . . . . . . . . . . 58
5.5 Flow pattern visualization . . . . . . . . . . . . . . . . . . . . . 60
6 Conclusion and recommendations 69
6.1 Motivation for study . . . . . . . . . . . . . . . . . . . . . . . . 69
6.2 Research ndings . . . . . . . . . . . . . . . . . . . . . . . . . . 70
6.2.1 Experimental . . . . . . . . . . . . . . . . . . . . . . . . 70
6.2.2 Numerical Simulation . . . . . . . . . . . . . . . . . . . . 71
6.3 Recommendations for future research . . . . . . . . . . . . . . . 72
Appendices 75
CONTENTS viii
A Fan blade data for numerical simulation 76
B Sample calculations for experimental data 79
B.1 Sample Calculations . . . . . . . . . . . . . . . . . . . . . . . . 80
C Experimental Data 85
D Experimental Instrumentation 92
D.1 Ambient Conditions . . . . . . . . . . . . . . . . . . . . . . . . . 94
D.2 Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
D.3 Torque . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
D.4 Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
D.5 Bridge ampli er . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
D.6 Blade angle setting jig . . . . . . . . . . . . . . . . . . . . . . . 95
E Calibration of instrumentation 98
E.1 Pressure Calibration . . . . . . . . . . . . . . . . . . . . . . . . 98
E.2 Torque Calibration . . . . . . . . . . . . . . . . . . . . . . . . . 100
E.3 Speed Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . 103
F Fan performance characteristics 105
G Numerical Simulation Data 113
H Streamline distribution diagrams 115
List of References 119
List of Figures
1.1 The Matimba direct dry-cooled power plant (www.eskom.co.za). . . 2
1.2 Direct air-cooled condensing system with A-frame heat exchanger
con guration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1 ADM compared to experimental data (Van der Spuy et al. (2009)). 9
2.2 Main fan rotor dimensions and set-up, adapted from Meyer (1996). 12
2.3 Blade stagger angle and blade setting angle. . . . . . . . . . . . . . 13
3.1 Leading edge grid point distribution. . . . . . . . . . . . . . . . . . 16
3.2 O4H topology blocks and grid points, (AutoGrid Manual, 2007a). 18
3.3 Azimuthal view of computational domain dimensions, Approach-1. . 20
3.4 Isometric grid view of computational domain, Approach-1. . . . . . 20
3.5 Azimuthal view of computational domain dimensions, Approach-2. . 21
3.6 Isometric grid view of computational domain, Approach-2. . . . . . 21
3.7 Isometric grid view of computational domain inlet section. . . . . . 22
3.8 Isometric grid view of computational domain inlet section solid
boundaries. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.9 A single blade generated in AutoGrid. . . . . . . . . . . . . . . . . 24
3.10 y1+ values for the blade section. . . . . . . . . . . . . . . . . . . . . 25
(a) y1+ range 1 to 10. . . . . . . . . . . . . . . . . . . . . . . . 25
(b) y1+ range 1 to 5. . . . . . . . . . . . . . . . . . . . . . . . . 25
(c) y1+ near tip gap. . . . . . . . . . . . . . . . . . . . . . . . . 25
3.11 Axisymetric nature of CFD model. . . . . . . . . . . . . . . . . . . 26
1 th
(a) axisymmetric model, without repetition. . . . . . . . . 26
8
1 th
(b) axisymmetric model, with repetition. . . . . . . . . . . 26
8
3.12 Averaged surfaces for obtaining outlet pressure. . . . . . . . . . . . 27
3.13 Isometric grid view of computational domain outlet section, Approach-
1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.14 Isometric grid view of computational domain outlet section, Approach-
2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.15 Static pressure distribution over the mixing plane approach rotor-
stator interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.16 Static pressure distribution over the frozen rotor rotor-stator inter-
face. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
ix
LIST OF FIGURES x
3.17 Mutigrid Strategy. . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.18 Multigrid resolution for the blade hub. . . . . . . . . . . . . . . . . 36
(a) Finest mesh resolution, 000. . . . . . . . . . . . . . . . . . 36
(b) Coarser mesh resolution, 111. . . . . . . . . . . . . . . . . 36
(c) Coarsest mesh resolution, 222. . . . . . . . . . . . . . . . . 36
4.1 BS 848 Fan Test Facility, Stinnes (1998). . . . . . . . . . . . . . . . 45
4.2 Fan static pressure rise, repeatability. . . . . . . . . . . . . . . . . . 49
4.3 Fan power, repeatability. . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4 Fan static e ciency, repeatability. . . . . . . . . . . . . . . . . . . . 50
4.5 Fan static pressure rise, comparison with Stinnes (1998). . . . . . . 50
4.6 Fan power, comparison with Stinnes (1998). . . . . . . . . . . . . . 51
4.7 Fan static e ciency, comparison with Stinnes (1998). . . . . . . . . 52
5.1 Fan static pressure rise, CFD veri cation. . . . . . . . . . . . . . . 55
5.2 Fan power, CFD veri cation. . . . . . . . . . . . . . . . . . . . . . 56
5.3 Fan static e ciency, CFD veri cation. . . . . . . . . . . . . . . . . 57
3
5.4 Streamline distribution, 16 m /s, Approach-1, relative velocity so-
lution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3
5.5 Streamline distribution, 16 m /s, Approach-1, absolute velocity so-
lution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3
5.6 Streamline distribution, 16 m /s, Approach-2, absolute velocity so-
lution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.7 Streamline distribution close-up views for Approach-1 and Approach-
2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3
(a) Streamline distribution, 16 m /s, Approach-1, close-up. . . 61
3
(b) Streamline distribution, 16 m /s, Approach-2, close-up. . . 61
3
5.8 Streamline distribution, 8 m /s, Approach-2, close-up view. . . . . . 61
3
5.9 Flow pattern visualization at 2 m /s. . . . . . . . . . . . . . . . . . 65
(a) Experimental fan blade ow pattern with throttle device
closed. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3
(b) CFD fan blade ow pattern at 2 m /s (Approach-1). . . . 65
3
5.10 Flow pattern visualization at 10 m /s. . . . . . . . . . . . . . . . . 66
3
(a) Experimental fan blade ow pattern at 10 m /s. . . . . . . 66
3
(b) CFD fan blade ow pattern at 10 m /s (Approach-1). . . . 66
3
(c) CFD fan blade ow pattern at 10 m /s (Approach-2). . . . 66
3
5.11 Flow pattern visualization at 16 m /s. . . . . . . . . . . . . . . . . 67
3
(a) Experimental fan blade ow pattern at 16 m /s. . . . . . . 67
3
(b) CFD fan blade ow pattern at 16 m /s (Approach-1). . . . 67
3
(c) CFD fan blade ow pattern at 16 m /s (Approach-2.) . . . 67
3
5.12 Flow pattern visualization at 20 m /s. . . . . . . . . . . . . . . . . 68
(a) Experimental fan blade ow pattern with throttle device
fully open. . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3
(b) CFD fan blade ow pattern at 20 m /s (Approach-1). . . . 68
LIST OF FIGURES xi
3
(c) CFD fan blade ow pattern at 20 m /s (Approach-2). . . . 68
A.1 Chord length and stagger angle distribution. . . . . . . . . . . . . . 76
A.2 A plot of the GA(W)-2 and LS(1)-0413 pro les. . . . . . . . . . . . 77
D.1 Schematic layout of experimental instrumentation. . . . . . . . . . 93
D.2 Blade stagger angle setting jig. . . . . . . . . . . . . . . . . . . . . 96
D.3 Stagger angle jig chord leading edge. . . . . . . . . . . . . . . . . . 96
D.4 Stagger angle jig chord trailing edge. . . . . . . . . . . . . . . . . . 97
D.5 Dierence between stagger angle and blade setting angle. . . . . . . 97
D.6 Device for clamping rotor shaft. . . . . . . . . . . . . . . . . . . . . 97
E.1 Pressure calibration setup. . . . . . . . . . . . . . . . . . . . . . . . 99
E.2 Pressure calibration bellmouth. . . . . . . . . . . . . . . . . . . . . 100
E.3 Pressure calibration settling chamber. . . . . . . . . . . . . . . . . . 100
E.4 Torque calibration horizontal level measurement. . . . . . . . . . . 101
E.5 Torque calibration curve. . . . . . . . . . . . . . . . . . . . . . . . . 102
E.6 Speed tachometer calibration curve. . . . . . . . . . . . . . . . . . . 104
F.1 Fan static pressure rise at three blade angles. . . . . . . . . . . . . 105
F.2 Fan power at three blade angles. . . . . . . . . . . . . . . . . . . . 106
F.3 Fan static e ciency at three blade angles. . . . . . . . . . . . . . . 106
F.4 Fan static pressure rise, comparison at 59°. . . . . . . . . . . . . . . 107
F.5 Fan power, comparison at 59°. . . . . . . . . . . . . . . . . . . . . . 107
F.6 Fan static e ciency, comparison at 59°. . . . . . . . . . . . . . . . . 108
F.7 Fan static pressure rise, comparison at 58°. . . . . . . . . . . . . . . 108
F.8 Fan power, comparison at 58°. . . . . . . . . . . . . . . . . . . . . . 109
F.9 Fan static e ciency, comparison at 58°. . . . . . . . . . . . . . . . . 109
F.10 Fan static pressure rise, comparison at 60°. . . . . . . . . . . . . . . 110
F.11 Fan power, comparison at 60°. . . . . . . . . . . . . . . . . . . . . . 110
F.12 Fan static e ciency, comparison at 60°. . . . . . . . . . . . . . . . . 111
F.13 Fan static pressure rise, comparison at 1.5 mm tip clearance. . . . . 111
F.14 Fan power, comparison at 1.5 mm tip clearance. . . . . . . . . . . . 112
F.15 Fan static e ciency, comparison at 1.5 mm tip clearance. . . . . . . 112
H.1 Streamline distribution, 8 m3 /s, Approach-2. . . . . . . . . . . . . . 115
3
H.2 Streamline distribution, 20 m /s, Approach-2. . . . . . . . . . . . . 116
3
H.3 Streamline distribution, 2 m /s, Approach-1, relative velocity solu-
tion.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
3
H.4 Streamline distribution, 6 m /s, Approach-1, relative velocity solu-
tion.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3
H.5 Streamline distribution, 10 m /s, Approach-1, relative velocity so-
lution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3
H.6 Streamline distribution, 20 m /s, Approach-1, relative velocity so-
lution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
List of Tables
2.1 B2-fan speci cations . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.1 Grid level distribution. . . . . . . . . . . . . . . . . . . . . . . . . . 37
A.1 Fan blade pro le data for numerical simulation . . . . . . . . . . . 78
B.1 Raw experimental data . . . . . . . . . . . . . . . . . . . . . . . . . 80
C.1 Experimental data constants for 59° stagger angle . . . . . . . . . . 85
C.2 Conversion of voltage data for 59° stagger angle . . . . . . . . . . . 86
C.3 Voltage data corrected for drift for 59° stagger angle . . . . . . . . . 87
C.4 Calculated performance parameters for 59° stagger angle . . . . . . 88
C.5 Scaled performance parameters for 59° stagger angle . . . . . . . . . 89
C.6 Experimental data constants for 58° stagger angle . . . . . . . . . . 89
C.7 Scaled performance parameters for 58° stagger angle . . . . . . . . . 90
C.8 Experimental data constants for 60° stagger angle . . . . . . . . . . 90
C.9 Scaled performance parameters for 60° stagger angle . . . . . . . . . 91
E.1 Calibration of pressure transducers . . . . . . . . . . . . . . . . . . 99
E.2 Torque calibration data . . . . . . . . . . . . . . . . . . . . . . . . 101
E.3 No-load torque test without and with at plate . . . . . . . . . . 102
E.4 Speed calibration data . . . . . . . . . . . . . . . . . . . . . . . . . 103
G.1 Numerical data for Approach-1 . . . . . . . . . . . . . . . . . . . . 113
G.2 Numerical data for Approach-2 . . . . . . . . . . . . . . . . . . . . 114
xii
Nomenclature
Symbols
A Cross-sectional area
b Width
Cp Speci c heat at constant pressure
cb1 , etc Empirical constants for turbulence model
D Drag force
d Diameter
E Energy
e Eectiveness
F Force
h Height
K Thermal conductivity
k Turbulent kinetic energy
L Length, or Lift force
M Torque
ṁ Mass ow rate
N Rotational speed
n Number, normal
P Power
p Pressure
Q Source term, Volumetric ow rate
q Heat ux vector
R Universal gas constant
r Radius
S Magnitude of vorticity
S̃, g, rsp Intermediate variables for turbulence model
Sij Strain-rate tensor
T Temperature
t Axial thickness of the actuator disk model, time
xiii
NOMENCLATURE xiv
U Freestream velocity
u, v, w Velocity components
ui Fluctuating velocity components
V Volume
v Velocity vector
x, y, z Cartesian coordinates
α Angle of attack
β Relative ow angle
χ Intermediate variable for turbulence model
∆ Dierential
δ Incremental
δij Kronecker delta
η E ciency
γ Speci c-heat ratio (Cp /Cv ), Blade angle
κ Kármán constant
µ Viscosity
ν Kinematic molecular viscosity
ν̃ Working variable for turbulence model
ω Angular rotation speed
φ Blade setting angle
π = 3.141 592 654
ψ Momentum source term
ρ Density
σ Blade solidity, constant for turbulence model
τ Boundary-layer shear stress
τij Stress tensor
ξ Stagger angle
ε Turbulent dissipation
Superscripts
r Distance from axis of rotation
+ Law-of-the-wall variable
TM Trademark
Subscripts
amb Ambient
b Blade
NOMENCLATURE xv
bell Bellmouth
c Chord
D Drag
d Dynamic
e External
F Fan
h Hydraulic
I Inviscid
i Numeral index, 1,2,3. . .
L Lift
m Meridional direction
p Pressure
w Wall
R Relative
r Radial direction
ref Reference
S Surface
s Static
setl Settling chamber
t Turbulent
tot Total
V Viscous
z Axial direction
θ Tangential direction
τ Friction
1 First inner cell
Auxiliary Symbols
0
Fan-law scaled variable, turbulent uctuation
¯ Time-mean
~ Vector
Dimensionless Groups
αε Compound calibration constant
Re Reynolds number
Cf Skin-friction coe cient
NOMENCLATURE xvi
Acronyms
ADM Actuator Disk Model
AGS Abu-Ghannam and Shaw
ACHE Air-Cooled Heat Exchanger
ACSC Air-Cooled Steam Condenser
CFD Computational Fluid Dynamics
CFL Courant-Friedrichs-Levy
GUI Graphical User Interface
MAX Maximum
MPA Mixing plane approach
PJM Pressure jump method
RES Residual
RMS Root mean square
USB Universal serial bus
Chapter 1
Introduction
Axial ow fans have a wide range of application in industry. Direct cooling of
Air Cooled Heat Exchangers (ACHE) in the form of fan arrays is one of the
largest industrial axial ow fan applications with appropriate large saving pos-
sibilities in terms of improving fan e ciencies. To illustrate this point, consider
one of the world's largest forced draft dry-cooled power plants, the Matimba
power plant, located in South Africa (Figure 1.1). The power plant consists
of 6 × 665 MW(e) Air Cooled Steam Condenser (ACSC) units, each unit em-
ploying 48 heat exchangers (six rows of eight fans). That amounts to 288 axial
ow fans, 9.145 m in diameter, each driven through a reduction gearbox by a
270 kW electrical motor (Kröger (1998) and Bredell (2005)). With fossil fuels
diminishing, increasing popularity of alternative (or renewable) power gener-
ation and a constant growth in national power demand, large scale electricity
savings at such power plants becomes increasingly more important. An ap-
propriate and well-designed cooling system can have a very signi cant positive
impact on plant pro tability (Kröger, 1998). A rough estimation (neglecting
distorted in ow conditions resulting in dierent fan performance from fans at
the edge of the array) indicates that improving a single axial ow fan's e -
ciency by 5% relates to a total power saving for the 288 fans of 5.6 MW.
Although water is a more appropriate heat sink than air, in arid regions (such
as the Limpopo province of South Africa where the Matimba plant is situ-
ated) the use of air-cooled or dry-cooled systems in power plants is justi ed
where cooling water may be insu cient and very expensive or not available
at all (Kröger, 1998). The Matimba power generation plant uses single pass,
forced draught ACHE's in the con guration displayed in Figure 1.2. In com-
parison to an induced draught ACHE (with the fan located above the heat
exchanger bundle), the forced draught ACHE has the advantage of consuming
less power for a given air mass ow rate since it pumps cooler air with a higher
density. Since the forced draught ACHE's fan drives are located beneath the
heat exchangers, they are more easily maintained and are not subjected to
high air ow temperatures, allowing simpler materials for construction and a
1
CHAPTER 1. INTRODUCTION 2
Figure 1.1: The Matimba direct dry-cooled power plant (www.eskom.co.za).
longer fan drive lifespan. As shown in Figure 1.2, the turbine exhaust steam
is piped directly to the ACSC unit where heat is rejected to the environment
by means of air ow through the nned tube heat exchanger. The nned
tubes are arranged in an A-frame or delta con guration in order to drain the
condensate eectively, reduce the plot area of the ACSC units and reduce
the duct lengths distributing the process uid to the ACSC units. The dy-
namic interaction between the steam turbines and the ACSC, the dierence
between ambient drybulb temperature and the nned tube temperature and
the ow rate delivered by the axial ow fans directly in uence the e ciency
of a typical air-cooled power plant (Bredell, 2005). Therefore any axial ow
fan performance reducing eects should be taken under careful consideration
when optimizing the e ciency of a direct air-cooled power generation plant.
One of the performance reducing eects on axial ow fan units within an
ACSC is distorted in ow to the fan units. This topic has been investigated in
numerous studies (Venter (1990), Thiart and von Backström (1993), Stinnes
(1998) and Bredell (2005)). Causes for inlet ow distortion are the proximity
to structures (turbine house, ground surface), induced cross-drafts from adja-
cent fans and prevailing winds surrounding the fan array. All of these factors
have to be taken into account when designing an ACSC. However, since these
factors vary from one fan unit to the next depending on the location within
the ACSC and relative to other fan units within the fan array, it is necessary
CHAPTER 1. INTRODUCTION 3
Figure 1.2: Direct air-cooled condensing system with A-frame heat exchanger
con guration.
to broaden the individual fan investigation to incorporate the whole fan array.
A very e cient way of investigating the design factors for such an ACSC is
by means of a Computational Fluid Dynamic (CFD) analysis of the complete
system. Unfortunately using a full three-dimensional computational domain
becomes computationally expensive, especially if discrete fan models are used
in the system.
To simplify the numerical modelling of an axial ow fan, dierent models have
been developed, such as the constant velocity fan model, constant static pres-
sure fan model, varying static pressure fan model and the actuator disc model
(ADM) (Bredell, 2005). The latter model was extensively used in CFD anal-
yses by Thiart and von Backström (1993), Meyer (2000) and Bredell (2005).
Unfortunately the model under-predicts the fan's performance at ow rates
less than the fan stalling ow rate. Another draw-back of the model is the
large database that it requires, containing the airfoil characteristics over a
wide range of angles of attack, Reynolds numbers and Mach numbers (Ohad
and Aviv, 2005). In order to improve the model's ability to cover the complete
operating range of the axial ow fan, an investigation is launched to model the
fan with a full discrete model incorporating rotating solid surfaces. Bredell
(2005) recommends a full three-dimensional modelling approach for eective
CHAPTER 1. INTRODUCTION 4
and realistic modelling of wind-eects, and mentions the use of improved fan
blade lift and drag characteristics and including the modelling of radial forces
in the ADM, as possible areas of improvement.
This thesis is motivated by the hypotheses that a detailed investigation of the
ow eld around the blades of an axial ow fan, in a geometrically correct
setup, incorporating the radial ow eects due to centrifugal forces present
on the blade, will provide more accurate lift and drag coe cients for imple-
mentation in the ADM. The lift and drag coe cients in the ADM, used by
Meyer (2000) and Bredell (2005), are veri ed only to empirical data from two-
dimensional ow experiments, where this dissertation takes a three-dimensional
approach.
To ensure that realistic ow conditions are enforced on the rotating blades, the
CFD model should resemble a test facility used for characterising fan perfor-
mance. Such a facility currently exists at Stellenbosch University and conforms
to the British Standards test codes (BS 848 standards, 1980). This facility has
been extensively used for testing scale model fans utilized in air-cooled power
plants (Venter (1990), Bruneau (1994) and Stinnes (1998)).
The B2-fan rotor analysed in this thesis, was designed by Bruneau (1994), and
used in investigations by Stinnes (1998), Meyer (2000) and Bredell (2005). It
1 th
is a scale prototype model of a typical fan unit in an air-cooled power plant
6
with a hub-to-tip ratio of 0.4. The B2-fan was selected for the CFD analysis
because of the large amount of ADM analyses previously done on this fan by
Meyer (2000) and Bredell (2005). To verify the integrity of the CFD results,
the CFD results are compared with experimental data from the test facility.
Unfortunately the original blades developed by Bruneau (1994) no longer exist
and another experimental dataset had to be produced for the new set of blades
manufactured form the original moulds.
The overall objective of this project would be a recommendation regarding the
application and accuracy of CFD to model axial ow fans and to produce a
fan performance dataset for the new B2-fan blades.
An outline of the thesis chapters and a brief discussion on each follows:
Chapter 2 gives a literature review of previous work done in developing
the ADM, and the design and utilisation of the fan test facility.
Chapter 3 provides a detailed outline of the numerical techniques em-
ployed in developing a full three-dimensional CFD model. The simula-
tion steps of grid generation, setting up various computational parame-
CHAPTER 1. INTRODUCTION 5
ters and post processing are discussed.
Chapter 4 discusses the fan test facility, fan rotor setup, experimental
procedure and results.
Chapter 5 illustrates the validation of the CFD results with the experi-
mental results. The ow visualisation experiment performed on the fan
blades in the fan test facility is described and discussed.
Chapter 6 renders a conclusion on developing a three-dimensional CFD
model for amending the ADM and lists recommendations for future work.
Chapter 2
Literature survey
2.1 Introduction
The scope of this thesis is two-fold: Simulating an axial ow fan numerically
using CFD code and verifying the accuracy of the numerical predictions with
experimental results. The literature survey discusses the previous work done
within these two elds and nishes o with a summary on the main ndings.
2.1.1 Simulation of an axial ow fan
The ideal method with which to simulate an axial ow fan in CFD would be
to use a fully discrete model. The amount of detail included in a CFD model
is however limited by the capability of the computation resources. When sim-
ulating special cases like large ACSC's, a simpler CFD model is required that
consumes less computational resources (Van Staden, 1996).
A number of previous projects have been done, using dierent simple CFD
models to simulate an axial ow fan or propeller.
Pericleous and Patel (1986) developed a mathematical model for the simu-
lation of tangential and axial agitators in chemical reactors. The dierent
impellers were represented as a quadratic source of momentum in the tangen-
tial and/or axial directions (depending on the type of impeller); while the other
geometries, such as the ba es, were represented as sinks of momentum. The
lift and drag coe cients from experimental data of the blade cross-sectional
airfoil pro les were used to calculate the blade forces. The momentum con-
tributions from these forces were then introduced as appropriate momentum
sources and sinks in the Navier-Stokes equations and solved using a control
volume formulation. Good agreement was found, but the model showed some
inadequacy in modelling turbulence.
6
CHAPTER 2. LITERATURE SURVEY 7
Pelletier and Schetz (1986) and Schetz et al. (1988) investigated the three-
dimensional ow in close vicinity of a propeller. This three-bladed propeller
typically represents push-prop arrangements on aircraft. In this case the pro-
peller was modelled as an actuator disk of thickness roughly equal to the
physical thickness of the propeller. The thrust and torque were constant in
the tangential direction, but allowed to vary radially. The three-dimensional,
Reynolds-averaged, steady-state, Navier-Stokes equations were solved by a
penalty function nite element method. Turbulence modelling was achieved
through a generalization of an integrated turbulence-kinetic-energy model.
The general-purpose nite element uid dynamics program FIDAP was used
to program the turbulence model. Good agreement was found in comparison
with wind tunnel measurements for the prediction of velocity and pressure
pro les along the radial axis.
Pelletier et al. (1991) developed a new model to compute the axisymmetric vis-
cous through ow past propellers. This was an eort to remove the limitations
(of the previous numerical work on propellers) of relying on experimental val-
ues of thrust and torque to be introduced as momentum sources/sinks in the
actuator disk. An analysis of the kinematics of the rotor provides the thrust,
torque and power input to (or extracted from) the ow. Both a mixing and a
penalty function nite element method were used to solve the time averaged
Navier-Stokes equations. Turbulence was modelled by a simple mixing length
equation and programmed by CADYF (a nite element based computer pro-
gram for solving steady state, incompressible uid ow problems). Again good
agreement was found in comparison with wind tunnel measurements for the
prediction of velocity and pressure.
Lötstedt (1992) developed a model for the computation of the time-averaged
ow eld produced by an aircraft propeller over a nacelle and wing. He used
the panel method and a slip-stream model based on classical propeller the-
ory. The panel method was amended to handle the vertical ow behind the
rotating propeller and had the advantage of being easy to use and relatively
computationally inexpensive. The time-averaged ow behind the propeller
was generated by a system of vortices of which the strength was determined
by combined momentum-blade element theory.
Later Lötstedt (1995) expanded this model for a full aircraft and replaced the
propeller with an actuator disk. The blade forces were still approximated by
combined momentum-blade element theory. Three simpli cations were imple-
mented to reduce the computational complexity: inviscid ow assumption, a
thin propeller disk and using time-averaged variables. The unknown values
at the disk, placed between two layers of cells in the grid, were extrapolated
from the upstream and downstream cells adjacent to the disk. The calcula-
tions performed using the panel method were compared to a one-dimensional
CHAPTER 2. LITERATURE SURVEY 8
Euler code and wind-tunnel experiments. The Euler solution provided better
agreement with the wind-tunnel tests at higher Mach numbers, but required
a much ner grid. The panel method was deemed faster and easier to use
for subsonic ows, but Lötstedt (1995) preferred the Euler method due to its
ability to capture the shape of the slipstream automatically.
Thiart and von Backström (1993) developed an actuator disk model for a low
solidity/low hub-to-tip ratio axial ow fan. The Navier-Stokes equations were
solved with the aid of a k−ε turbulence model and the SIMPLEN method.
The thrust forces and torque applied on the air ow by the fan blades were de-
termined by blade element theory and the eect of distorted in ow conditions
were investigated. Empirical correlations were used to determine the lift and
drag coe cients of the blade elements. The method developed by Pericleous
and Patel (1986) was followed, hence the lift and drag coe cients of the blade
elements were used to calculate the lift and drag forces exerted by that spe-
ci c element on the ow. These forces were then included in the Navier-Stokes
equations as body forces. Good agreement with experiments were found but
the increased power associated with distorted in ow conditions was under pre-
dicted.
Van Staden (1996) integrated a fan performance model into a CFD model for
the performance prediction of a complete air-cooled condenser. Experimental
fan performance curves were used to obtain the momentum source term of the
fan in the axial direction, which was added to the Navier-Stokes equation as a
source term in the CFD model. Good agreement was found with experimental
data at ideal conditions, but for non-ideal conditions (such as distorted in ow)
the model required the fan performance curves at these adverse conditions.
Meyer (2000) and Bredell (2005) utilized the actuator disk model of Thiart and
von Backström (1993) for their numerical investigations of air-cooled heat ex-
changers. As mentioned in the introduction chapter, this model under-predicts
the fan performance at low ow rates (see Figure 2.1). This can be attributed
to the model only taking the axial and tangential velocity components into
consideration for calculating the relative in ow vector, neglecting the radial
component. The radial component of velocity and the Corriolis eect, asso-
ciated with rotation, become more prominent at lower ow rates and induce
a delay in boundary-layer separation that results in retarded stall (Ohad and
Aviv (2005)).
Most of the literature involving axial ow fans/propellers/impellers revolve
around stirred tanks, propeller propulsion (such as ships and helicopters) and
wind turbines. The latter two cases dier from the work presented here since
they involve air ow over unducted fan blades. No previous research in lit-
erature could be found for discrete simulation of ducted axial ow fans used
CHAPTER 2. LITERATURE SURVEY 9
Figure 2.1: ADM compared to experimental data (Van der Spuy et al. (2009)).
in forced-draught heat exchangers. Based on the available literature it can
be concluded that the ADM remains the preferred method for simulating the
eect of an axial ow fan on air ow through ACSC's. The most recent work
on the simulation of a stirred tank and one other case of ducted axial fan sim-
ulation is presented next.
A complete three-dimensional CFD model was used by Ramasubramanian
et al. (2008) to analyse the ber diusion process in the manufacturing of
wet-laid nonwovens. The three-bladed impeller and ba es were modelled in
FLUENT by using the MIXSIM user interface and incorporating the multiple
reference frame model. This meant that for the impeller a rotating reference
frame was used and for the ba es and tank, a stationary reference frame was
used. The impeller has a diameter of 0.2 m and rotates at 350 rpm. A standard
k−ε turbulence model was used for the ow solution. The simulation was
compared to experimental work on a mixing tank with ba es and an impeller
located in the center of the tank. The eect of a rectangular and triangular
cross section on dispersion quality was predicted by the CFD model. The
ba es in this application reduce the recirculating ow by eectively breaking
up the large central vortices produced by the impeller. The authors found
the model useful to predict the location, source and mechanism behind the
formation rope and log defects.
Kelecy (2000) predicted the fan performance of a four-bladed axial ow pro-
peller over a range of ow rates and compared the results with wind tunnel
data. The blade has a rotational speed of 2000 rpm and a diameter of 0.11 m.
The rotating reference frame method of FLUENT was used to simulate the
CHAPTER 2. LITERATURE SURVEY 10
1 th
rotation of the axisymmetric fan model. The ow equations were solved in
4
the rotating reference frame. A zero velocity was imposed on the blade sur-
faces and shaft, while the outer walls of the tunnel were rotated at the speci ed
speed in the opposite direction when viewed from a stationary reference frame.
He observed the sensitivity of the ow solutions to initial conditions, signi -
cant pressure gradients from the blades to the inlet at lower ow rates as well
as radial out ow from the fan as the ow rate decreases.
2.1.2 Experimental
The accuracy of the CFD model is veri ed with experimental results. To en-
sure con dence in a set of experimental data, the data set needs to conform
to certain air performance test methods (standards), in this case for industrial
fans. These standards aim to provide a speci ed environment where the fans
can be tested. When comparing dierent fan designs for a certain application,
the fan test results all need to adhere to similar conditions, standardized setup
and testing procedures to eliminate any uncertainties in the comparison. The
fan test standards, extensively used at the low speed fan test facility at Stellen-
bosch University for previous work as well as in this dissertation, is the British
Standards, BS 848 : Part 1 : 1980. The Type A standard installation type is
used, with free inlet and outlet. The fan exhausts into open atmosphere. The
characteristics required for fan performance set forth in the BS 848 standards
1980, are the fan static pressure rise, the fan power and fan static e ciency (a
relation of the aforementioned two).
Venter (1990) designed the test facility according to the BS 848 standards 1980
for testing axial ow fans. The facility was used to investigate the performance
of axial ow fans in ACSC's. He looked at the in uence of ow distorting fac-
tors (such as support structures and heat exchanger con gurations) and fan
design parameters (such as blade tip clearance and hub plates) on the perfor-
mance of a typical industrial cooling fan that he referred to as the V-fan.
From Venter (1990)'s research it became apparent that the V-fan suered from
a large amount of recirculating ow in the hub region. Bruneau (1994) set out
to resolve this issue and improve on the e ciency, while maintaining the origi-
nal fan pressure rise and volumetric ow rate, by designing two prototype fans
with a higher hub-tip ratio than the V-fan. The B1-fan utilizes the Clark-Y
aerofoil pro le and the B2-fan incorporates the NASA LS series. Bruneau
(1994) mentions that the choice of the NASA LS series blade pro le was mo-
tivated by the particular pro le's exceptional lift/drag characteristics, good
o-design performance and its tolerance to inlet ow distortion. The blades
were designed with the use of simple radial equilibrium equations. The two
fans compared favourably with the fan from Venter (1990) over the entire fan
operating envelope. The fan performance characteristics of the B2-fan with
CHAPTER 2. LITERATURE SURVEY 11
the NASA LS series airfoil pro le serves as reference data for the experimental
work carried out in this thesis.
An incident occurred in which the V-fan was destroyed and it was replaced
by Stinnes (1998) with the geometrically similar S-fan. Stinnes (1998) inves-
tigated the eect of a cross- ow component at the inlet on the performance
of axial ow fans. For this he incorporated the prototype B1-fan and B2-fan
of Bruneau (1994) along with the S-fan and in addition provided a thorough
guideline for performing experimental tests on the test facility at Stellenbosch
University. Furthermore he measured the ow eld immediately up- and down-
stream of the fan with a 5-hole probe and developed a prediction model to
predict local ow patterns for cases that where not examined with the 5-hole
probe experiments. The method of the prediction model is similar to the ac-
tuator disc theory and takes fan blade and local ow patterns at various radial
stations and circumferential positions into account to determine global fan
performance.
Meyer (1996) experimentally investigated a number of component and geomet-
rical changes to ACHE's in order to determine the eect of these changes on
the ACHE performance characteristics. Two of the changes that are of partic-
ular interest were the use of dierent fan types and fan sizes, and the variation
in the fan rotor axial position within the shroud. The necessity to investigate
these two parameters led to another data set of fan performance characteristics
for the B2-fan for possible use as reference regarding the experimental work
within this thesis. Meyer (2000) also performed experimental veri cation of
his axial ow fan model with the B2-fan designed by Bruneau (1994).
The fan test facility stood dormant for ve years, after which it was recom-
missioned by Esterhuyze (2006). Unfortunately Esterhuyze (2006) had an
incident where the original B2-fan was destroyed and a new set of blades from
the same mould designed by Bruneau (1994) was manufactured by another
student (Dippenaar, 2007).
The main fan rotor dimensions and set-up of the facility is shown in Figure 2.2.
Changing the fan axis position (in an upstream direction) had no discernible
eect on the fan performance characteristics (Meyer, 1996); therefore the fan
axis position was kept constant throughout the experiments. The blade stagger
angle (ξ ) is set relative to the plane perpendicular to the axis of rotation (see
Figure 2.3). To set the blades at a prescribed blade stagger angle, a blade
setting angle (φ) was used in conjunction with a digital projector xed onto
a blade angle setting jig. The B2-fan speci cations are given in Table 2.1
(Bruneau, 1994).
CHAPTER 2. LITERATURE SURVEY 12
Figure 2.2: Main fan rotor dimensions and set-up, adapted from Meyer (1996).
Table 2.1: B2-fan speci cations
Number of blades 8
Hub diameter (dH ) [m] 0.617
Fan casing diameter (dFC ) [m] 1.542
hub-tip ratio 0.4
Fan axis position (xF ) [m] 0
Tip clearance [mm] 3
Average chord length [m] 0.167
2.2 Summary of main ndings
The ADM model provides a means to eectively simulate a complete ACHE
consisting of multiple fan units, without using a discrete fan model. This ap-
CHAPTER 2. LITERATURE SURVEY 13
Figure 2.3: Blade stagger angle and blade setting angle.
proach saves computational grid complexity and computer processing power,
resulting in reduced time required to obtain a solution. Unfortunately this
model under-predicts the fan performance at very low ow rates (Ohad and
Aviv (2005) and Van der Spuy et al. (2009)). This necessitates a detailed in-
vestigation of ow over the fan blade surfaces in order to amend the existing
ADM model with the appropriate source terms. Although excellent experi-
mental tests (in particular by Stinnes (1998)) were performed on the B2-fan
in the past, replacement of the B2-fan blades warrants a new fan performance
data set. The data from Stinnes (1998) did not include fan performance tests
at 3 mm tip clearance as referenced by Bruneau (1994). The new fan perform-
nce data set would include both tip clearances for reference with previous work.
Chapter 3
Numerical modelling
3.1 Introduction
The analysis of CFD problems generally involve three steps:
Discretization of the ow domain
Setting up and initiating the ow computation
Visualizing the results
For these three steps NUMECA has developed three software systems. The dis-
cretization is performed in an Automated Grid generator for turbomachinery
systems (AutoGrid) and/or in an Interactive Geometry modeller and Grid
generation system for multiblock structured grids (IGG). The second step is
executed by the ow solver EURANUS (EURopean Aerodynamic NUmerical
Simulator), which is a three-dimensional multiblock ow solver that is able to
simulate Euler or Navier-Stokes (laminar or turbulent) ows. The nal step is
performed with a highly interactive Computational Field Visualization soft-
ware system (CFView). All three software systems have been integrated into
NUMECA's Flow INtegrated Environment (FINE), which is a user friendly
GUI, that operates on a project basis eliminating the need for le manipula-
tion in the consolidation of the three steps. Detailed discussions of the three
steps follow.
3.2 Fan blade data
All analyses were performed on what Bredell (2005) referred to as the B-fan.
The B-fan is actually the B2-fan designed by Bruneau (1994). The reason
for choosing the B2-fan is based on the availability of original design data for
the fan. In order to compare a full three-dimensional CFD simulation of the
axial ow fan to the experimental results, the same data for the blade pro le
14
CHAPTER 3. NUMERICAL MODELLING 15
had to be used. Bruneau (1994) designed the blade with a linear decrease in
blade thickness of 13 % (thickness/chord ratio) at the hub down to 9% at the
tip. To de ne the change in blade properties from hub to tip, the blade was
divided into several two-dimensional sectional blade pro les at dierent radial
stations. To calculate the blade pro le co-ordinates the chord length and stag-
ger angle at each radial station is needed. After scaling the chord length at the
hub and tip radius, Bruneau (1994) used linear interpolation to determine the
chord length at the intermediate radial stations. The stagger angle is deter-
mined from the dierence between the inlet relative ow angle and the angle
of attack. The angle of attack for each radial station was interpolated from a
C-spline tted through experimentally determined angle of attack versus lift
coe cient data, for the particular airfoil pro le used in the fan blade design.
Figure A.1 in Appendix A shows the distribution of chord length and stagger
angle radially across the blade length.
Bruneau (1994) used the General Aviation (Whitcomb)-number two (GA(W)-
2) airfoil designed by McGhee et al. (1977), as a pro le for the blades of the
B2-fan. The GA(W)-2 is a 13 % thick airfoil derived from the 17 % thick
GA(W)-1 airfoil originally designed for propeller driven light aircraft. The
13 % thick GA(W)-2 airfoil is obtained by linearly decreasing the mean thick-
13
ness distribution of the 17 % thick GA(W)-1 airfoil with a ratio of .
17
McGhee et al. (1977) produced an airfoil pro le data set, as used by Bruneau
(1994), of 39 points per surface. Using this data set rendered a coarse mesh,
observed after generating the mesh in IGG. Further investigation revealed
that a whole family of airfoils emerged from the GA(W)-1 airfoil with a new
designation, resulting in the GA(W)-1 airfoil becoming the LS(1)-0417 and the
GA(W)-2 airfoil becoming the LS(1)-0413 airfoil. According to McGhee et al.
(1979) the LS(1) designates low speed ( rst series); the rst two digits specify
the airfoil design lift coe cient in tenths, while the last two digits indicate
the airfoil thickness in percent chord. McGhee et al. (1979) produced a ner
airfoil pro le data set of 45 points per surface. Figure A.2 displays a plot of
the GA(W)-2 pro le and the LS(1)-0413 pro le from the dierent data sets.
Seven airfoil pro le data sets were produced at dierent radial stations from
the 45-point-per-surface data. These radial station blade co-ordinates were
used in the mesh modelling by importing the data set into IGG and tting
a C-spline through the data. From this C-spline 200 data points per surface
were exported to a data le for use in the three-dimensional blade generation
process in AutoGrid. Table A.1 displays the fan blade data set for the hub
pro le. A close-up view of the grid resolution at the blade leading edge is
illustrated in Figure 3.1.
CHAPTER 3. NUMERICAL MODELLING 16
Figure 3.1: Leading edge grid point distribution.
3.3 Computational domain
1 th
A discrete three-dimensional computational grid for an axial ow fan was
8
developed using a combination of the AutoGrid and IGG software from NU-
MECA. The AutoGrid software facilitates fully automated grid generation for
turbomachinery blades, while the IGG software is a powerful structured grid
generator. The model consists of three segments, namely the inlet, blade and
outlet segments. Both the inlet and outlet segments were created in IGG,
while AutoGrid was used for creation of the more complex blade geome-
try. Once all three segments were completed separately, they were combined
within AutoGrid with the use of three-dimensional technological eects to
form a single model. Three-dimensional technological eects in AutoGrid
include all eects de ned by three-dimensional surfaces or three-dimensional
curves whose meshes were created manually in IGG. This single model was
reopened in IGG for the nal speci cation of the rotor-stator interface be-
tween the blade segment and the inlet segment on one side as well as the outlet
segment on the other side.
3.3.1 AutoGrid
The approach stipulated in the user manual (AutoGrid Manual, 2007a) was
followed for semi-automatic three-dimensional blade generation. This included:
1. De nition of the geometry. This includes speci cation of the blade pro-
CHAPTER 3. NUMERICAL MODELLING 17
le surfaces from the imported data of Table A.1 in Appendix A, as
well as specifying the hub and shroud surfaces of revolution. Additional
information such as tip clearance (3 mm in this case) is also speci ed.
2. Generation of meridional ow paths. These paths act as meridional
traces for the surfaces of revolution on which the three-dimensional mesh
is built.
3. Generation and control of the two-dimensional mesh on the surfaces of
revolution. This allows the manipulation of the mesh topology and grid
clustering. The grid quality can be checked at dierent radial stations.
4. Generation of the nal three-dimensional mesh. The three-dimensional
mesh is generated by using conformal mapping between the three-dimensional
Cartesian space and the two-dimensional meshes on the surfaces of rev-
olution that follows the meridional ow paths.
The AutoGrid interface allows checking of the mesh quality at any level, radi-
ally between the hub and the shroud. The mesh quality is measured according
to the following criteria:
Orthogonality, the measure of the skewness of the cells and is determined
by the minimum angle between the edges of the cell. It has a range of
◦ ◦
0 to 90 .
Angular deviation, the measure of the angular variation between two
◦ ◦
adjacent cells in all directions. It has a range of 0 to 180 .
Aspect ratio, the measure of how thin (ratio of width to length of the
cells) a cell is. It has a range of 1 to 50000.
Expansion ratio, the measure of the size variation between two adjacent
cells in a speci c direction. It has a range of 1 to 100.
Cell width, i.e. the height of the cell measured in all directions. It has a
range of 0m to 1 000 000 m.
A check for negative cells is also performed at this stage.
The mesh quality can be improved by controlling the selected topology. Au-
toGrid has three prede ned topologies. The HHOHH (O4H) topology allows
for full automatic meshing for all kinds of turbomachinery geometries; while
the HOH and H&I topologies ensure very high quality grids, but according to
the user manual (AutoGrid Manual, 2007a) are not suitable for all applica-
tions. In the generation of the three-dimensional mesh for the axial ow fan
the default O4H topology is used. The O4H topology allows control of ve
blocks (referring to Figure 3.2):
CHAPTER 3. NUMERICAL MODELLING 18
Figure 3.2: O4H topology blocks and grid points, (AutoGrid Manual, 2007a).
1. an O-block around the blade called the skin block
2. an H-block upstream of the blade leading edge called the inlet block
3. an H-block downstream of the blade trailing edge called the outlet block
4. an H-block above the blade section called the up block
5. an H-block below the blade section called the down block
By altering the number of grid points in the dierent blocks the mesh quality
can be improved accordingly. Once the required mesh quality is achieved, the
AutoGrid le is saved, allowing for automatic regeneration of the blade with
the speci ed grid points at a later stage if needed.
3.3.2 IGG
Both inlet and outlet segments were created by projecting blocks from a sur-
face geometry. These projected blocks were revolved 45° around the Z-axis to
1 th
create a axisymmetric model. The number of grid points in all three direc-
8
tions was speci ed at this stage. Once the required mesh quality is achieved,
the clustering of cells near surfaces to be de ned as walls was attempted. IGG
(IGG Manual, 2007c) uses the same mesh quality criteria as AutoGrid (Au-
toGrid Manual, 2007a). The clustering of cells speci es the cells distribution
CHAPTER 3. NUMERICAL MODELLING 19
+
in any direction normal to a surface. To achieve low y values, dense cell dis-
tribution (clustering) close to a wall surface is necessary. The IGG interface
allows the creation of segment groups with the same clustering to ease and
automate the process of adjusting cell clustering. Some clustering near the
fan exit was implemented where high velocity gradients were expected. This
step requires the mesh quality to be checked again, since the blocks need to
be regenerated to accept the clustering changes.
After completing the geometry block generation and setting the periodicity
for eight blades the various boundaries were speci ed. This includes outlet,
inlet as well as solid wall boundaries. At this stage the connecting surfaces
surrounding the bellmouth blocks on the inlet segment were changed to solid
wall surfaces to accommodate attaching the bellmouth inlet to the fan segment.
3.3.3 Domain speci cations
The complete computational domain consists of three main sections, namely
the inlet to the fan, the fan rotor and the outlet. The inlet and outlet sections
consist of a structured mesh generated in IGG, including all boundary condi-
tions and periodicities. The blade and the endwalls (hub and shroud) forming
the fan rotor part of the model, were generated in AutoGrid, including set-
ting the periodicities. All three complete sections were integrated into a single
model for the complete CFD simulation of the fan test facility by adding the
inlet and outlet to the fan blade model in AutoGrid. Saving this le creates
an *.igg le (which can be manipulated within IGG) that is integrated in the
FINE interface to be coupled with the EURANUS ow solver. A detailed
discussion of the three sections follows.
Essentially two computational domains were evaluated. Both domains shared
the same inlet and blade section, but utilized dierent outlet sections. The
rst CFD model contains a pinched outlet section and will be addressed as
Approach-1. The second model with an open outlet will be addressed as
Approach-2. The two models consists of the same amount of grid cells, namely
4728631 cells. Figure 3.3 displays an azimuthal view of the rst model with
the general grid dimensions. An isometric view of the complete computational
domain for the rst model is illustrated in Figure 3.4. Figure 3.5 presents
the general grid dimensions of the second model and an isometric view of the
second model is presented in Figure 3.6. The fan diameter is denoted by dF .
CHAPTER 3. NUMERICAL MODELLING 20
Figure 3.3: Azimuthal view of computational domain dimensions, Approach-1.
Figure 3.4: Isometric grid view of computational domain, Approach-1.
CHAPTER 3. NUMERICAL MODELLING 21
Figure 3.5: Azimuthal view of computational domain dimensions, Approach-2.
Figure 3.6: Isometric grid view of computational domain, Approach-2.
CHAPTER 3. NUMERICAL MODELLING 22
3.3.3.1 Inlet
The inlet section of the computational domain is geometrically true to the
test facility dimensions beyond the mesh screens inside the settling chamber
and includes the bellmouth inlet to the fan. Instead of modelling the settling
chamber with a square inlet the hydraulic diameter of the settling chamber
walls was calculated and used. This option is implemented to coincide with
the eight-segment axisymmetric nature of the CFD model. Equation (3.3.1)
shows the calculation of the settling chamber hydraulic diameter based on the
principle of equal ow area. An isometric view of the inlet section, containing
1483560 cells, is illustrated in Figure 3.7. Non-rotating solid wall boundaries
are speci ed for all walls, except at the 0.05 m inner radius, where an Euler
(zero-shear) wall is used. According to White (2006) for a xed Euler wall
the no-slip condition is dropped and the tangential velocity allowed to slip.
These solid wall boundaries can be seen in Figure 3.8. The inlet boundary
is mass ow imposed and a rotor-stator interface is assigned to the patch that
is coupled with the fan rotor section. The remaining boundaries are set to
1 th
periodic boundaries for a axisymmetric model.
8
r
4bh
dh =
r π
(4)(3.7)(3.7)
=
π
= 4.175 m (3.3.1)
3.3.3.2 Blade
The grid point distribution on the O4H-topology is shown in Figure 3.2. This
leads to a total of 1106311 grid points for the fan blade, hub and shroud. A
tip-gap of 3 mm is incorporated in the fan blade section. A fully generated
blade is displayed in Figure 3.9. The cell blocks containing the tip gap is not
shown.
In order to capture the high gradients of velocity within the boundary layer,
an adequate amount of cells should be placed within the boundary layer. The
+ +
wall variable commonly used to evaluate this is the y value. The y value
associated with the rst node o the wall is given in equation (3.3.2), accord-
ing to the FINE/Turbo Manual (2007b ).
ρ uτ yw
y1+ = (3.3.2)
µ
CHAPTER 3. NUMERICAL MODELLING 23
Figure 3.7: Isometric grid view of computational domain inlet section.
Figure 3.8: Isometric grid view of computational domain inlet section solid
boundaries.
CHAPTER 3. NUMERICAL MODELLING 24
Figure 3.9: A single blade generated in AutoGrid.
Here yw is the height of the rst grid node o the wall and uτ is the friction
velocity de ned by
r r
τw 1 2
uτ = = U Cf (3.3.3)
ρ 2 ref
with τw the wall shear stress, Uref the reference velocity and Cf the shear
coe cient.
Dierent turbulence models for Low- and High-Re number ows require dier-
+
ent ranges of y1 values to accurately predict the boundary layer. To achieve
+
this, the correct range of y1 values per turbulence model need to be obtained.
The FINE/Turbo Manual (2007b ) speci es that for Low-Re turbulence mod-
+ +
els solving the viscous sublayer (y1 ≤ 5, White (2006)), y1 values between
1 and 10 is acceptable. The FINE/Turbo Manual (2007b ) also recommends
using a Low-Re turbulence model if separation is expected, since the logarith-
+
mic function does not apply for separated ow. Since the y1 value depends
on the value of yw , according to equation (3.3.2), it can be used to control the
+
corresponding y1 values based on the truncated series solution of the Blasius
CHAPTER 3. NUMERICAL MODELLING 25
equation (FINE/Turbo Manual, 2007b):
− 87
1
Uref Lref 8 +
yw = 6 y1
ν 2
− 87 1
1.5 0.167 8
=6 (5)
1.511 × 10−5 2
= 9.335×10−4 (3.3.4)
where Uref can be taken as the average inlet velocity, ν the kinematic viscosity
and Lref the reference length (taken as the blade chord length). The y1+ value
+
for the blade is well within the 1 to 10 range (y1 ≤ 9.551), and only reaches
the limit of 10 near the tip gaps (Figure 3.10). This was deemed su cient
since the tip clearance ow consists of highly accelerating uid and is dierent
from a normal boundary layer.
(a) y1+ range 1 to 10. (b) y1+ range 1 to 5.
(c) y1+ near tip gap.
Figure 3.10: y1+ values for the blade section.
Non-rotating solid wall boundaries are speci ed for the shroud. The blade and
hub are assigned solid surfaces rotating at 750 rpm. The patches connecting
CHAPTER 3. NUMERICAL MODELLING 26
the inlet and outlet sections to the fan rotor sections are speci ed as rotor-
stator interfaces. The remaining boundaries are set to periodic boundaries for
1 th
a axisymmetric model. Figure 3.11 gives an indication of the axisymmetric
8
nature of the CFD model.
1 th 1 th
(a) axisymmetric model, without repe- (b) axisymmetric model, with repeti-
8 8
tition. tion.
Figure 3.11: Axisymetric nature of CFD model.
3.3.3.3 Outlet
The fan test facility on which the computational domain is based, houses a fan
that exhausts into the open atmosphere. De ning such an open atmosphere
outlet within a computational domain is problematic. To keep the integrity
of the model intact would require outlet patches in both the axial and radial
directions. This opens the possibility for a large amount of back ow to enter
the domain at the outlet boundary and severely destabilize the simulation.
The support team at NUMECA strongly advised against using such an open
outlet and deemed any results obtained from such simulations untrustworthy.
To resolve this issue the outlet is pinched for Approach-1, some distance away
from having an in uence on the fan performance. The only objective of the
pinching is to increase the local velocity in this region and to restrict the
back ow over the outlet. The static outlet pressure is not taken at the pinched
outlet, but rather averaged over a series of surfaces, on the constant z-plane,
before the pinching in the model commences. The static pressure value of each
surface is an averaged value for the static pressure over the particular area.
Figure 3.12 shows a plot of these surfaces. To resolve the radial out ow issue,
Euler walls are implemented at the endwalls of the outlet. With the no-slip
condition removed for Euler walls, they only prohibit the ow from exiting the
model, thus aiding the pinching eect. The Euler walls are placed su ciently
CHAPTER 3. NUMERICAL MODELLING 27
far away from the fan rotor.
Figure 3.12: Averaged surfaces for obtaining outlet pressure.
Similar to the inlet, the patch connecting the outlet section with the fan rotor
is speci ed as a rotor-stator patch. The patch coinciding with the outside wall
of the settling chamber is de ned as a solid wall. The remaining patches are
assigned Euler walls, referring to the discussion earlier. A radial equilibrium
static pressure is enforced at the outlet patch with atmospheric pressure im-
posed at the outer circumference. The complete computational domain, of
2138760 cells, is illustrated in Figure 3.13.
The outlet section for Approach-2 avoids the use of a pinched outlet and zero-
shear Euler walls. The circumferential outlet patch employs a static pressure
boundary condition. The outlet patch on the plane normal to the fan axis
imposes a radial equilibrium static pressure boundary condition, similar to the
outlet section of Approach-1. Both outlet sections contain the same amount
of cells and are connected to the fan rotor via a rotor-stator interface. The
complete computational domain is presented in Figure 3.14.
CHAPTER 3. NUMERICAL MODELLING 28
Figure 3.13: Isometric grid view of computational domain outlet section,
Approach-1.
Figure 3.14: Isometric grid view of computational domain outlet section,
Approach-2.
CHAPTER 3. NUMERICAL MODELLING 29
3.4 Computational parameters
All the parameters necessary to set up and facilitate a steady-state ow sim-
ulation are discussed in detail.
3.4.1 Fluid Model
Within FINE it is possible to choose a uid type from a pre-de ned database
or create a new uid type. The pre-de ned database contains the following
types of uids: perfect gas, real gas, incompressible gas or liquid and condens-
able uid.
After researching the user manual it became apparent that FINE's real gas
is actually a thermally perfect gas and the perfect gas a calorically perfect
gas. According to Anderson (1990) a true real gas has an equation of state
that takes compressibility eects, variable heat capacity, Van Der Waals forces,
non-equilibrium thermodynamic eects and issues with molecular dissociation
and elementary reactions into account. This is a very extensive analysis and
various simpler gas laws exist. The ideal gas law simpli es the real gas by
assuming a compressibility factor of one. White (2006) stipulates that all
common gases follow with reasonable accuracy, at least in some nite region,
the so-called ideal or perfect gas law. Furthermore a perfect gas can be de ned
as one where the intermolecular forces of the real gas are neglected to simplify
calculations. Anderson (1990) states that this perfect gas simpli cation can
be further simpli ed and classi ed as a thermally perfect gas and a calorically
perfect gas.
The thermally perfect gas assumes that the gas is in thermodynamic equilib-
rium, not chemically reacting and that internal energy, enthalpy, and speci c
heats are functions of temperature only. The calorically perfect gas enforces
all the assumptions of the thermally perfect gas, but assumes constant speci c
heats. In FINE the perfect gas model assumes constant speci c heat (Cp )
and constant speci c-heat ratio (γ ).
For the incompressible uid type the density can either be constant or a pre-
de ned function of the pressure. Using air as an incompressible gas in the
default FINE setting, the density law is de ned by the Boussinesq law, in
which case the density is kept strictly constant. With no terms in the mass
and momentum equations depending on the temperature, these conservation
equations are decoupled from the energy conservation equation.
Relative stable uid temperature meant that the use of the real gas (thermally
perfect gas) uid type was not considered. The perfect gas (calorically perfect
gas) and incompressible uid types were investigated and a fan static pressure
CHAPTER 3. NUMERICAL MODELLING 30
rise error amongst the two uid types of less than 3% was found. However
the incompressible uid type required almost double the amount of iterations
to give answers that correlate with the perfect gas uid type. Therefore the
perfect gas model was used for the axial ow fan model.
3.4.2 Flow Model
For this model steady ow is assumed and time independent ow calcula-
tions are solved. This investigation considers a one-equation turbulence model,
namely Spalart-Allmaras (Spalart and Allmaras, 1994). The Spalart-Allmaras
turbulence model is NUMECA's default model, providing good convergence for
common turbomachinery applications (FINE/Turbo Manual, 2007b). The
Spalart-Allmaras turbulence model performs considerably better than the al-
gebraic turbulence models, such as the Baldwin-Lomax model, especially in
mapping the performance of separated ow (Ashford and Powell, 1996). The
model is also more robust, computationally less expensive and requires less
memory than the two-equation k−ε turbulence model (Ashford and Powell
(1996). Both models employ constant Prandtl numbers.
3.4.3 Rotating Machinery
FINE allows the user to specify rotating blocks and rotor-stator interfaces
to accommodate rotating and non-rotating sections within a domain. In this
mesh the blocks were all set to be rotating while the relevant solid surfaces,
such as the shroud and settling chamber walls, were kept stationary. This ap-
proach is somewhat similar to that used by Kelecy (2000) who kept the blades
and shaft stationary, while maintaining an angular velocity at the outer walls
in the opposite direction from the supposed blade rotation.
An attempt was made to utilize the classic turbomachinery simulation ap-
proach where only the blocks containing the rotor were allowed to rotate, keep-
ing the remaining blocks stationary. In this instance two rotor-stator interface
types, the mixing plane approach and the frozen rotor, were investigated:
With the mixing plane approach circumferentially averaged ow quanti-
ties (such as mass, momentum and energy uxes) are exchanged at the
rotor-stator interface. According to the FINE/Turbo Manual (2007b )
a more strict global conservation through the interface is achieved when
using a ux based approach instead of exchanging the classical primi-
tive variables. Thus any blade wake or separation is mixed circumfer-
entially before entering the downstream component, resulting in a cir-
cumferentially uniform pressure distribution as well as uniform velocity
components at the interface. The static pressure distribution over the
rotor-stator can be viewed in Figure 3.15.
CHAPTER 3. NUMERICAL MODELLING 31
Figure 3.15: Static pressure distribution over the mixing plane approach rotor-
stator interface.
With the frozen rotor approach no mixing of the ow solution at the in-
terface occurs. Instead the rotor movement is neglected and the upstream
and downstream components can be seen as being literally connected.
Thus the continuity of the velocity components and pressure is imposed
on the interface. The governing equations for the rotor are solved in a
relative reference frame, and that of the stator in an absolute reference
frame. The ow solution is thus dependent on the relative position of
the rotor and stator sections. The static pressure distribution over the
rotor-stator can be viewed in Figure 3.16.
Unfortunately neither of the aforementioned methods provided satisfactory re-
sults. For Approach-1 the mixing plane approach delivered the closest results
compared to the experimental data with a 28.7 % error in fan static pressure
rise at the design operating point. This was only achieved after 18234 itera-
tions with a very unstable simulation. The frozen rotor approach was more
stable, but could only achieve a 53.82 % error in fan static pressure rise com-
pared to experimental data. The aforementioned results were produced with a
zero-order extrapolation of ow quantities from inner cells to the rotor-stator
interface. Using rst-order extrapolation provided somewhat better results at
a 111-grid-level (coarser mesh), but it was too unstable at a 000-grid-level
( nest mesh) and the simulations could not complete more than 1000 itera-
tions before diverging, for both cases.
One of the problems with using the aforementioned approach, is the mixing
CHAPTER 3. NUMERICAL MODELLING 32
Figure 3.16: Static pressure distribution over the frozen rotor rotor-stator
interface.
losses introduced at the rotor-stator interfaces due to the ow being forced
to mix out in a short distance, where normally it would have had an in nite
distance to do so. The mixing losses over the rotor-stator interface can range
anywhere from 10 Pa to 100 Pa; which is small enough to be neglected in tur-
bomachinery applications with a big pressure rise. However, in this application
it is an order of magnitude of the pressure capability of the axial ow fan and
in uences the fan performance considerably.
Another issue unique to the axial ow fan is the amount of back ow observed
at the rotor-stator interface. The back ow increases signi cantly at lower
ow rates and is a major source of unsteadiness in the simulation. With the
mixing plane approach the azimuthal averaged ow quantities on the down-
stream side of the blade outlet rotor-stator interface contains back ow and is
dierent from the ow quantities on the upstream side. To resolve this issue
the rotor-stator interface would need to be placed far enough downstream to
avoid having back ow over the interface. This would also allow the ow to
mix out naturally before reaching the interface, therefore eliminating mixing
losses. In eect this is what has been done in the current model, where the
rotor-stator interfaces have essentially been moved all the way to the inlet and
outlet boundaries. Having no stator-like structures in the geometry of the ax-
ial ow fan, neglecting the need for a rotor-stator interface in the ow domain,
the error introduced by allowing all the blocks to rotate is small.
The aforementioned simulation approach is commonly used for three-dimensional
CHAPTER 3. NUMERICAL MODELLING 33
wind turbine simulations (Sezer-Uzol and Long, 2006).
3.4.4 Boundary Conditions
FINE allows the speci cation of ve dierent types of boundary conditions:
inlet, outlet, solid walls, periodic and external (far eld).
The use of a mass ow and static temperature imposed inlet boundary and
static pressure outlet boundary for the axial ow fan was advised per email
correspondence with the NUMECA support team and mentioned as the ad-
visable setup for compressible or low-speed ows in the FINE/Turbo Manual
(2007b ). According to the manual this setup stabilizes the ow calculations
and provides better initial solutions for multistage calculations.
At the inlet boundary the mass ow is imposed through a speci ed control
surface consisting of related patches grouped together. With this boundary
condition the velocity vector and temperature is imposed on the same control
surface. For the velocity vector either the swirl and the direction of the ve-
locity vector in the meridional plane or the direction of the absolute velocity
vector can be imposed. The latter was enforced since no swirl is expected at
the inlet boundary which coincides with the exit of the mesh screens in the
settling chamber of the fantest facility.
The absolute velocity vector is speci-
Uz
ed in the axial direction: =1 . The static temperature can either be
|U |
speci ed as a constant or as a relation between the inlet static pressure and
the average static pressure along a speci ed outlet. The latter option is not
applicable to the current setup as it requires the mass ow imposed at both
inlet and outlet boundaries. Furthermore, the turbulent viscosity µt for the
Spalart-Allmaras turbulence model need to be speci ed at the inlet boundary.
All speci ed values are absolute quantities.
Imposing static pressure at the outlet boundary can be done with three dier-
ent methods:
1. Imposing a uniform static pressure across the control surface. For this
case the static temperature and velocity components at the boundary
are extrapolated from the nearest control volumes.
2. Imposing an average pressure across the control surface. The average
static pressure is speci ed and instead of a uniform pressure distribution
the pressure pro le is extrapolated from the inner cells and translated to
ensure the average pressure value that was speci ed is met at the control
surface.
CHAPTER 3. NUMERICAL MODELLING 34
3. Using radial equilibrium. Since this model has a cylindrical outlet with
mesh lines at constant radius the static pressure can be imposed at a con-
stant radius (the outer radius in this case). The hub-to-shroud pressure
pro le is then calculated by integrating the radial equilibrium equation
(3.4.1) in the spanwise direction with pitchwise averaged values of vθ and
r.
∂p v2
=ρ· θ (3.4.1)
∂r r
The option of using radial equilibrium provided the best results.
The velocity and thermal conditions for solid wall patches are de ned using
the solid wall boundary type. The boundary can be either de ned as zero-slip
or zero-shear. The settling chamber walls, shroud and bellmouth surfaces are
de ned as zero-slip walls with a zero rotational velocity. All the rotating sur-
faces consisting of the blade and hub surfaces are de ned as zero-slip walls with
a rotational velocity of 750 rpm. All zero-slip walls are de ned as adiabatic.
All rotating surfaces are utilized to compute the force and torque of the sys-
tem. For the cylindrical axial ow fan model the axial thrust is de ned as the
projection of the global force on the rotation axis and the couple exerted by
the global force, around the rotational axis of the fan, as the torque. The axial
thrust is computed from the pressure and velocity elds of the uid acting on
the blade surfaces according to equation (3.4.2):
X
F~ · n~z (3.4.2)
S
The projection of the torque along a given direction ~z is given by equation
(3.4.3).
!
X
~r × F~ · ~z (3.4.3)
S
3.4.5 Numerical Model
Three computational parameters allow the user to improve the convergence
rate of the simulation:
CFL number
Multigrid parameters
Preconditioning parameters
CHAPTER 3. NUMERICAL MODELLING 35
3.4.5.1 CFL number
The Courant-Friedrichs-Levy (CFL) number scales the time-step sizes that are
used for the time-marching scheme of the ow solver. A higher CFL number
leads to faster convergence but can lead to divergence and unstable simulations
(FINE/Turbo Manual, 2007b). The inverse of this is also applicable where
choosing a smaller CFL number in an unstable simulation improves conver-
gence. At the low ow sections of the fan performance curve, a lower CFL
number was chosen to improve the convergence of the simulation.
3.4.5.2 Multigrid parameters
For e ciency and fast convergence the EURANUS ow solver uses a multigrid
strategy in solving the ow equations on dierent grid levels. To complement
this, a Full Multigrid Strategy option is also available wherein the solution of
a coarser grid level is used as an initial solution for the next level of ner grid.
With the ow solver automatically coarsening the initial mesh from IGG an
option is available to run simulations on several sub-meshes. The coarsest grid
level that can be achieved with any mesh depends on the number of times
the grid can be coarsened along each of the (I, J, K) directions. This number
would be represented by n in equation (3.4.4)
cell size = 2n (3.4.4)
For example, if a grid size is de ned by 8 × 16 × 16 cells in the I, J, K directions
respectively, the mesh can be coarsened to 3,4,4 grid levels in the respective
directions. Figure 3.17 displays grid level coarsening in one direction. The
same eect on the hub grid distribution around the fan blade can be viewed in
Figure 3.18. However, using a hybrid grid level structure as described above
may deform the mesh in a preferential direction upon coarsening. This might
increase the aspect ratio for instance and eectively reduce the quality of the
overall mesh. In setting up a mesh in IGG care should be taken to keep
a uniform grid level distribution. The nest grid level is denoted by a 0 in
any (I, J, K) direction and the coarser grid levels follow in a numerical order:
0,1,2. . . (see Table 3.1 for a grid level display of the previous example). In the
axial ow fan model three grid levels 222, 111 and 000 were used to obtain a
ow solution.
With the Multigrid Strategy the ow calculation is performed on all the grid
levels simultaneously in order to speed up convergence. The EURANUS ow
solver implements an explicit multi-stage Runge-Kutta time marching scheme.
The number of times the Runge-Kutta operator is applied (number of sweeps)
can be speci ed for each grid level. The coarse grid level to ne grid level
sweep of the multigrid cycle is initiated once a solution of the coarsest mesh is
CHAPTER 3. NUMERICAL MODELLING 36
Figure 3.17: Mutigrid Strategy.
(a) Finest mesh resolution, 000. (b) Coarser mesh resolution, 111.
(c) Coarsest mesh resolution, 222.
Figure 3.18: Multigrid resolution for the blade hub.
achieved and smoothed. The current solutions on ner grids are updated with
the solution on the next coarser grid level. This approach leads to an e cient
and fast convergence (FINE/Turbo Manual, 2007b).
CHAPTER 3. NUMERICAL MODELLING 37
Table 3.1: Grid level distribution.
Grid level Grid points in Number of
notation I, J, K direction grid points
0 0 0 9×17×17 2601
1 1 1 5×9×9 405
2 2 2 3×5×5 75
3 3 3 2×3×3 18
3 4 4 2×2×2 8
The option of using a Full Multigrid Strategy provides a good initial solution.
A preliminary multi-stage Runge-Kutta explicit scheme is started on the coars-
est grid level, but the solution will not be interpolated to the next ner grid
level, until a speci ed convergence is achieved. This solution is then applied
as an initial solution on the ner grid level for further iterations. The cycle
is repeated until the nest grid level is reached. The grid level at which the
preliminary run should be initiated (level 222 or 111) as well as the number
of iterations to be performed on the corresponding grid level (if convergence
is not achieved) can be speci ed for each grid. It was found that if conver-
gence on coarser grid levels were not obtained it resulted in divergence of the
simulation on ner grid levels. For the axial ow fan model care was taken
to ensure a good level of convergence was achieved before initializing the next
ner mesh with the previous solution. The larger cell size of the coarsest mesh
leads to a fast convergence, which provides a good initial solution and allows
a more robust simulation on the nest grid level.
3.4.5.3 Preconditioning parameters
Within FINE preconditioning is automatically utilized for incompressible
ows and serves as a viable option for other low Mach number uid types
to improve the convergence rate.
According to the FINE user manual when the magnitude of the ow ve-
locity becomes small in comparison with the acoustic speeds, time marching
compressible codes converge very slowly. This problem is addressed with the
use of a low speed preconditioner. A preconditioning matrix is multiplied with
the time derivatives of the unknowns that arise in the ow equations when
time marching algorithms are used to solve steady state applications. These
time derivatives have no physical meaning and can be modi ed without alter-
ing the nal steady state solution. This results in reduced round-o errors at
low Mach numbers (caused by the use of absolute pressure in the momentum
equations) by the introduction of reduced ow variables such as the dynamic
CHAPTER 3. NUMERICAL MODELLING 38
pressure and the dynamic enthalpy. Also, by using a pseudo-wave speed in
order of magnitude of the uid speed, instead of the acoustic wave speed, the
dierence between the convective and the acoustic eigenvalues are lessened.
For the axial ow fan model Merkle preconditioning was used. The Merkle
preconditioner is amongst the most popular used perconditioners and has
been extended, from a very e cient preconditioner for two-dimensional low
Mach number ows, to also solve three-dimensional ows (Turkel et al. (1997),
Zhonghua et al. (2007) and Vigneron et al. (2008)).
3.4.6 Initial Solution
FINE allows, amongst other, the option to either initialize a computation
from constant values or from an initial solution le. In the former case physi-
cal values need to be speci ed that are used uniformly in all the blocks except
on the boundaries. The variables for these values are static pressure, static
temperature and the velocity components. This method of initialization was
used for the coarsest (222) grid level. For the intermediate grid level (111)
coarse grid initialization was used where the 111-grid-level computation was
automatically started from the 222-grid-level computation after a few itera-
tions. The 111-grid-level computation was allowed to converge to provide an
initial solution le to be used in the latter initialization method mentioned
earlier. The nest grid level (000) computation was started from this ini-
tial solution le. Using this approach accelerated the convergence rate and
removed the risk op having a solution diverging at a ner grid level due to
starting from an unconverged initial solution from a coarser grid level. Using
this hybrid approach is especially bene cial at very low ow rates where a
lower CFL number is needed. The 222- and 111-grid-level computations can
still be solved with a higher CFL number, for faster convergence, and with the
switch to the nest grid level (000), a lower CFL number is chosen to stabilize
the computation.
3.4.7 Output
In order to have e cient memory usage and still retain all the ow quantities
of immediate interest, FINE allows specifying the necessary ow quantities to
be saved and lets the post-processing package CFView calculate the derived
quantities. The ow quantities can either be calculated at all the mesh nodes
or along the solid wall boundaries.
3.4.8 Computation Steering and Monitoring
FINE allows multi-process analyses where the convergence histories of mul-
tiple computations can be viewed simultaneously. Utilizing the MonitorTurbo
CHAPTER 3. NUMERICAL MODELLING 39
interface the default quantities for a turbomachinery computation can be
viewed independently. The residuals relative to the transport equations are
also made available.
The residuals are calculated as a ux balance of the uxes summed on all the
faces of each cell:
X
RES = f luxes (3.4.5)
The root mean square and the maximum of the residuals are calculated ac-
cording to the following relations:
RES
RM SRES = log RM S (3.4.6)
cell volume
RES
M AXRES = log M AX (3.4.7)
cell volume
For the axial ow fan model a decrease of three orders with stabilization in
the RMS residual curve was considered to be good convergence, along with a
dierence of less than 5% between mass ow at inlet and outlet.
3.5 Governing Equations
The basic Reynolds-Averaged Navier-Stokes equations solved in the EURANUS
ow solver, as de ned in the FINE/Turbo Manual (2007b ) are presented in
this section.
3.5.1 General Navier-Stokes Equations
The general Navier-Stokes equations can be expressed in a Cartesian frame of
reference as:
∂ ~
X + ∇F~I + ∇F~V = Q
~ (3.5.1)
∂t
~
X is the vector of the conserved variables:
ρ
X = ρ~v (3.5.2)
ρE
CHAPTER 3. NUMERICAL MODELLING 40
F~I and F~V are the respective ux vectors:
ρvi
ρv1 vi + pδ1i
FIi =
ρv2 vi + pδ2i
(3.5.3)
ρv3 vi + pδ3i
(ρE + p)vi
and
0
τi1
−FV i =
τi2
(3.5.4)
τi3
qi + vj τij
The stress and heat ux components are obtained from:
∂ w̃i ∂ w̃j 2
τij = (µ + µt ) + − (∇~
wδij ) (3.5.5)
∂xj ∂xi 3
and
∂
qi = (K + Kt ) T̃ (3.5.6)
∂xi
The source terms are contained within Q:
0
Q = ρf~e (3.5.7)
Wf
f~e expresses the external force eects and Wf the work performed by those
external forces, according to the following relation:
Wf = ρf~e · ~v (3.5.8)
3.5.2 Time averaging of quantities
Time averaging is applied to the Navier-Stokes equations. The instantaneous
value of the density and pressure is time averaged according to the following
relation:
q = q̄ + q 0 (3.5.9)
q̄ is the time averaged (mean) value, q0 the uctuating part and the mean
uctuation
q̄ 0 = 0 (3.5.10)
CHAPTER 3. NUMERICAL MODELLING 41
The energy, temperature and velocity components are density weighted aver-
ages following the relation:
ρq
¯
q̃ = (3.5.11)
ρ̄
3.5.3 Treatment of turbulence in the equations
For treating turbulence a rst-order closure model based on the Boussinesq
assumption for the turbulent shear is used:
∂wi ∂wj 2 −→ 2
−ρwi00 wj00 = µt + − (∇w̃)δij − ρ̄kδij (3.5.12)
∂xj ∂xi 3 3
In this case wi is the xi component of the relative velocity and k the turbulent
kinetic energy de ned as:
1 ρwi00 wi00
k= (3.5.13)
2 ρ̄
The contribution of the turbulent kinetic energy to the static pressure and
total energy is de ned as:
2
p¯∗ = p̄ + ρ̄k (3.5.14)
3
and
1
Ẽ = ẽ + w̃i w̃i + k (3.5.15)
2
There is no term for the angular velocity in equation (3.5.15), however it is
accounted for in the source term. When assuming stationary ow, the term for
the angular velocity should correspond to the last term in equation (3.5.19).
3.5.4 Formulation in rotating frame for the relative
velocity
The time averaged Navier-Stokes equations for the relative velocity compo-
nents in the relative rotating frame of reference are de ned as:
ρ̄
ρ̄w̃1
X=
ρ̄w̃2
(3.5.16)
ρ̄w̃3
ρ̄Ẽ
CHAPTER 3. NUMERICAL MODELLING 42
ρ̄w̃i
¯∗
p δ1i + ρ̄w̃i w̃i
FIi =
p¯∗ δ2i + ρ̄w̃i w̃i
(3.5.17)
p¯∗ δ3i + ρ̄w̃i w̃i
ρ̄Ẽ + p¯∗ w̃i
0
τi1
−FV i =
τi2
(3.5.18)
τi3
qi + w̃j τij
The shorthand notation, i, in the expression for −FV i denotes derivatives with
respect to xi .
The source term vector that contains the contributions of the Coriolis and
centrifugal forces is de ned as:
0
Q = − (ρ̄) [2~ω × w
~ + (~ω × (~ω × ~r))] (3.5.19)
ρw ~ (0.5ω 2 r2 )
~∇
The angular velocity in the relative reference frame is denoted by ω.
3.5.5 Formulation in rotating frame for the absolute
velocity
Normally the governing equations for rotating systems are solved for the rel-
ative velocity components in the relative frame of reference. However, as de-
scribed in Chapter 5, Section 5.4, the excess arti cial dissipation in the far eld
region is lessened by solving the absolute velocity components in the relative
reference frame.
The Reynolds stresses now become:
∂vi ∂vj 2 −→ 2
−ρvi00 vj00 = µt + − (∇ṽ)δij − ρ̄kδij (3.5.20)
∂xj ∂xi 3 3
Here vi is the xi component of the absolute velocity.
The ux vectors mentioned earlier are decomposed into Cartesian components:
FI = fI1 1~x + fI2 1~y + fI3 1~z
Fv = fv1 1~x + fv2 1~y + fv3 1~z (3.5.21)
CHAPTER 3. NUMERICAL MODELLING 43
The resulting time averaged Navier-Stokes equations for the absolute velocities
in the relative rotating frame of reference are de ned as:
ρ̄
ρ̄v˜1
X=
ρ̄v˜2
(3.5.22)
ρ̄v˜3
ρ̄Ẽ
ρ̄w̃i
p¯∗ δ1i + ρ̄w̃i w̃i v˜1
p¯∗ δ2i + ρ̄w̃i w̃i v˜2
FIi = (3.5.23)
p¯∗ δ3i + ρ̄w̃i w̃i v˜3
ρ̄Ẽ w̃i + p¯∗ ṽi
0
τi1
−FV i =
τi2
(3.5.24)
τi3
qi + v˜j τij
The shorthand notation, i, in the expression for −FV i denotes derivatives with
respect to xi . Both the relative and absolute velocity vectors are involved in
this expression.
The source term vector is de ned as:
0
Q = −ρ̄(~ω × ~v̄ ) (3.5.25)
0
When equation (3.5.25), is compared to equation (3.5.19) one can see the rel-
ative velocity component being multiplied to the squared angular velocity has
now been removed, thus eliminating the excess arti cial dissipation in the far
eld region (where the angular velocity that is squared can become very large).
Lastly the stress tensor is now given by
∂ṽi ∂ṽj 2
τij = (µ + µt ) + − (∇~v δij ) (3.5.26)
∂xj ∂xi 3
Chapter 4
Experimental evaluation
4.1 Introduction
The aim of the work detailed in this chapter is to produce a data set of fan
performance characteristics of the new fan blades that replicate the axial ow
fan used by Bruneau (1994) and Stinnes (1998). This new data set would then
provide reference experimental work for future analytical and computational
design work. It also provided the student with fan performance data for his
own computational analysis of the B2-fan. The fan test facility, experimental
setup and processing procedures are discussed in this chapter.
4.2 Fan Test Facility
The measurement of fan performance characteristic was carried out in a fan
test facility that conforms to the speci cations of the British Standards, (BS
848 standards, 1980). This facility was designed by Venter (1990), for type A
tests with a free inlet and outlet to the fan. A schematic layout of the fan test
facility can be seen in Figure 4.1.
The mass ow rate through the facility is measured by means of an inlet bell-
mouth with a diameter of 1.008 m (1). This is accomplished by measuring the
dierence in static pressure in the bellmouth relative to that of the atmosphere.
A throttling device (3) with six opposed blades is used to control the volume
ow rate. Flow straighteners are installed upstream (2) and downstream (4) of
the throttling device to aid in smoothing the air ow. To neutralize the pres-
sure losses due to all the various components in the ow system, a six-bladed
auxiliary fan (5) is installed. The auxiliary fan has an outside diameter of
1.540 m and a set of ow straighteners (6) is installed downstream to eliminate
swirl introduced by the fan rotor. To distribute the air ow more uniformly,
ow guide vanes (7) are installed at the inlet of the settling chamber (8). In-
44
CHAPTER 4. EXPERIMENTAL EVALUATION 45
side the settling chamber (dimensions 3.7 m × 3.7 m × 7 m) three stainless steel
mesh screens (9), which dier in coarseness, are installed to provide a uniform
ow velocity pro le to the inlet of the test fan (10). The fan test facility at the
University of Stellenbosch diers from the (BS 848 standards, 1980) fan test
code in the sense that the hydraulic motor (11) that drives the fan is situated
on the outside of the settling chamber instead of inside. This setup provides
the opportunity to use a stable motor mount that is rigid enough to eliminate
most of the deviations at the design speed and allows for ne tip clearances
to be used on the test fans. Meyer (1996) showed that the change in the po-
sition of the hydraulic motor has no discernible eect on the fan performance
characteristics.
Figure 4.1: BS 848 Fan Test Facility, Stinnes (1998).
4.3 Experimental setup parameters
Bruneau (1994) found no signi cant dierence in fan performance between us-
ing a hemispherical nose cone and a at plate to cover the front of the hub. For
simplicity in the numerical model it was decided to complete the experimental
work with a at plate attached to the hub. This at plate extends upstream
CHAPTER 4. EXPERIMENTAL EVALUATION 46
to the face of the shroud ange, such as the one used by Stinnes (1998) and
consists of a cardboard disk covering the spokes of the hub, such as used by
Bruneau (1994). Special attention was paid to the following parameters:
4.3.1 Fan axis position
This experimental setup utilized the same frame used by Bruneau (1994),
Meyer (1996) and Meyer (2000). The axial position of the fan inside the
shroud was unchanged, with the inlet face of the hub ush with the inlet face
of the shroud annulus. Changing the fan axis within a distance range from
0m to 0.075 m,
Meyer (1996) noted no eect on the fan static pressure rise
3
for ow rates higher than 8 m /s. For lower air ow rates a slight decrease in
the static pressure rise accompanied with a decrease in fan power resulted in
unchanged fan static e ciencies. After every blade angle and/or tip clearance
adjustment the fan was centred circumferentially inside the shroud.
4.3.2 Blade angle
The blade stagger angle was set using a custom-made jig that held a digital
projector in place that measured the blade setting angle. This custom jig
would be placed at the root of the blade and the blade angle would be set
with the fan installed in the test facility. The B2-fan was originally designed
with on-site blade angle adjustment in mind. The blade setting angle was
de ned relative to the blade chord at a 4.2° oset from the stagger angle, as
described in Appendix D Section D.6. With reference to Bruneau (1994) the
blades were set at an angle from the plane perpendicular to the axis of ro-
tation. Stinnes (1998) used the complementary angle to this that coincided
with Venter (1990), although Venter set the blade angle at the tip of the blade.
To set the blade angle the corresponding blade was locked into a horizontal
position with the clamping device. After loosening both the lock nut and
the grub screw the fan was tilted until the correct reading from the digital
projector was obtained. Using a digital projector allowed very ne blade angle
adjustment up to 0.1°. The position was locked by rst tightening the grub
screw and then the lock nut. This order was followed to prevent the blade
shaft from turning when the lock nut is tightened.
4.3.3 Tip clearance
The tip clearance is set with the use of a feeler gauge propped in-between the
blade and the shroud. The process of setting the tip clearance with the B2-fan
can be accomplished with high accuracy due to the design of the B2-fan hub.
A threaded collar protrudes from the hub and presses against a small lip on
the root of the blade. This collar is adjusted until the desired tip clearance is
CHAPTER 4. EXPERIMENTAL EVALUATION 47
achieved. The collar is locked in place with a lock nut.
After setting the tip clearance of all the blades the shaft was released from the
locking device and the tip clearance was checked at various positions around
the shroud. This ensures that the fan shaft is centred in the shroud. If there
were any deviations from the desired tip clearance the process was repeated.
An attempt was made to run experiments at the 1 mm to 1.5 mm tip clearance
used by Stinnes (1998), but the fan blades interfered with the shroud in the
upper right sector of the shroud. This was due to the frame oscillating at
higher speeds, as a result of the frame not being stable enough to limit rotor
circumferential movement. It is believed that the frame that Stinnes (1998)
used was more rigid. The experiments were carried out at a tip clearance of
3 mm, with reference to Bruneau (1994). The experiments were carried out at
700 rpm, which was found to be the maximum safe rotational speed at which
the oscillation of the motor frame did not cause the blades to touch the shroud.
4.3.4 Root seals
Bruneau (1994) provided data illustrating an increase in fan static pressure rise
and an accompanying higher fan static e ciency when root seals were applied
in contrast to not using root seals between the blade and the hub. In order to
provide rm root seals and keep some exibility for when blade angles were set,
expandable foam was used to construct the root seals. The foam adhered to
both the base of the blade surface and the hub surface, thereby giving a 100 %
closed seal from the leading edge to the trailing edge. After allowing the foam
to dry and set, excess foam was carefully removed and the root seal was shaped
to the contour of the blade. The outside surface of the shaped foam was given
a silicone layer to further seal the microscopic gaps and provide a smoothed
closed seal. The 1° angle adjustment in both directions from 59° (the angle at
which the foam was applied) was small enough for the exibility of the foam
to allow the adjustment without shearing loose. It should be mentioned that
setting the blade at any greater angle would result in the expandable foam
shearing at either the hub or bottom of the blade surface.
4.4 Experiment
Experiments were performed to measure static pressure (at the inlet bellmouth
and at the settling chamber), rotational speed and the fan shaft torque. Dif-
ferent blade angles as well as tip clearances were evaluated. To asses very low
volumetric ow rates (particularly with numerical work in mind), the throt-
tling device was closed beyond the lowest setting point. This provided exper-
3
imental data at a volumetric ow rate point of about 1.1 m /s, well beyond
CHAPTER 4. EXPERIMENTAL EVALUATION 48
the ± 3.6 m3 /s value of Bruneau (1994) and Stinnes (1998). Special care has
been taken in all cases of instrumentation calibration. Pressure transducer
calibration was repeated at dierent time periods yielding almost no dier-
ence in calibration value. To account for the speed uctuating when no load
is applied during calibration, a wooden brake was applied to the shaft. Again
no discernible dierences were noted. The torque transducer was calibrated
numerous times and the calibration constant of M = 27.325 V − 0.026 corre-
sponds well with that of Stinnes (1998), being M = 27.527 42 V. In addition
leakage tests were done on all the tubes connecting the pressure transducers
and no leaks were found.
To aid investigation of the velocity pro le over the blade surfaces, small tufts
were attached to the suction side surface of one of the blades. The eect of the
air ow over the blade surface on these tufts was captured by camera. The de-
tails of this investigation and results thereof are discussed further in Chapter 5.
4.5 Data Processing
Fan static pressure rise, fan power consumption and the fan static e ciency
against volumetric ow rate are used to produce curves of the results for the
performance mapping of the axial ow fan.
The BS 848 standards (1980) de nes the fan static pressure rise as the static
pressure at the outlet minus the total pressure at the inlet:
pF s = pamb − (pssett + pdsett )
= ∆pssett − pdsett
(4.5.1)
4.6 Results
4.6.1 Repeatability
Providing good repeatable experimental results creates trust in the ability
of the experimental facility to reproduce accurate experimental results when
reinstating experimental setups. In terms of repeatability, excellent agreement
was found for each of the three fan performance characteristics, as can be seen
in Figures 4.2, 4.3 and 4.4, respectively.
Comparative gures for blade angles 59°, 58° and 60° can be found in Appendix
F.
CHAPTER 4. EXPERIMENTAL EVALUATION 49
500
run 1
450
run 2
400
Fan Static Pressure [Pa]
350
300
250
200
BS 848: 1980: Type A
150
B-fan: 8-bladed
Fan casing diameter: 1.542 m
100
Tip clearance: 3 mm
Blade root stagger angle: 59°
50 Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
Volumetric Flow Rate [m3/s]
Figure 4.2: Fan static pressure rise, repeatability.
7000
run 1
run 2
6000
5000
Fan Power [W]
4000
3000
BS 848: 1980: Type A
B-fan: 8-bladed
2000
Fan casing diameter: 1.542 m
Tip clearance: 3 mm
1000
Blade root stagger angle: 59°
Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
3
Volumetric Flow Rate [m /s]
Figure 4.3: Fan power, repeatability.
CHAPTER 4. EXPERIMENTAL EVALUATION 50
60
run 1
run 2
50
Fan static effeciency [%]
40
30
BS 848: 1980: Type A
20
B-fan: 8-bladed
Fan casing diameter: 1.542 m
Tip clearance: 3 mm
10
Blade root stagger angle: 59°
Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
Volumetric Flow Rate [m3/s]
Figure 4.4: Fan static e ciency, repeatability.
4.6.2 Comparison with previous work
A comparison of the performance mapping for the fan characteristic static
pressure rise, power and static e ciency is illustrated in Figures 4.5, 4.6 and
4.7, respectively.
In comparing the fan performance characteristics to that of Stinnes (1998) fair
agreement is found. Stinnes (1998) did measure higher fan static pressure rise
as well as a lower fan power input, which resulted in higher fan static e ciency.
By examining Figure 4.5 it is clear that there is an almost constant dierence
3
between the corresponding angles. For the data range between 10 m /s and
14 m3 /s the data from this investigation seem to have a 1° o-set from the
data of Stinnes (1998). If this was the case, then beyond this range (to the
lower and higher ow rates) the static pressure rise of Stinnes (1998) would
seem to decrease. Looking at Figures F.4, F.7 and F.10 it can be seen that
the curve shape of Stinnes (1998) is not as uniform and smooth as that of this
investigation.
Figure 4.6 shows that the values for fan power at corresponding angles do
agree, albeit only at very high volumetric ow rates, with the deviation in fan
power increasing at lower volumetric ow rates.
CHAPTER 4. EXPERIMENTAL EVALUATION 51
550
500
450
Fan Static Pressure [Pa]
400
350
300
250
200
Experimental 58 deg
150
Experimental 59 deg
100
Experimental 60 deg
Stinnes 58 deg
50 Stinnes 59deg
Stinnes 60 deg
0
0 4 8 12 16 20 24
Volumetric Flow Rate [m3/s]
Figure 4.5: Fan static pressure rise, comparison with Stinnes (1998).
8000
7000
6000
5000
Power [W]
4000
3000
Experimental 58 deg
2000
Experimental 59 deg
Experimental 60 deg
1000 Stinnes 58deg
Stinnes 59deg
Stinnes 60deg
0
0 4 8 12 16 20 24
Volumetric Flow Rate [m3/s]
Figure 4.6: Fan power, comparison with Stinnes (1998).
CHAPTER 4. EXPERIMENTAL EVALUATION 52
70
60
50
Fan Static Effeciency [%]
40
30
Experimental 58 deg
20
Experimental 59 deg
Experimental 60 deg
10 Stinnes 58deg
Stinnes 59deg
Stinnes 60deg
0
0 4 8 12 16 20 24
3
Volumetric Flow Rate [m /s]
Figure 4.7: Fan static e ciency, comparison with Stinnes (1998).
The aforementioned discrepancies of lower static pressures and higher fan
power lead to lower fan static e ciencies than that of Stinnes (1998). A
test was carried out with a tip clearance of 1.5 mm which provided very good
agreement with that of Stinnes (1998), but the tests where carried out too
late in the day resulting in poor repeatability in torque readings due to higher
temperatures. The results where discarded. Further attempts to test at this
small tip clearance resulted in the fan blades interfering with the shroud. In
order to protect the fan blades it was decided not the run any further tests
at this small tip clearance and further tests were carried out at a 3 mm tip
clearance. From the discussion above and the resulting gures it is fair to
assume that the increased fan performance characteristics of Stinnes (1998) is
due to his experiments being carried out at a 1 mm to 1.5 mm tip clearance.
The 1.5 mm tip clearance results can be viewed in Appendix F.
Figures of the fan performance characteristics for the three dierent blade
angles mentioned above without comparison to Stinnes (1998), giving a better
view of the shape of the curves can be found in Appendix F.
CHAPTER 4. EXPERIMENTAL EVALUATION 53
4.7 Conclusion
The aim of gaining knowledge of the fan test facility and collecting data of the
B2-fan for use in computational analysis was achieved. Unfortunately whether
this provides good reference experimental work for future analysis and com-
putational design work remains debatable. Only fair agreement was obtained
when comparing the fan performance characteristics to that of Stinnes (1998).
The trend of the curves are more uniform and smooth than those of Stinnes
(1998), but the quantitative results are below the reference fan performance
characteristics.
There are some factors that cannot be neglected when comparing the results
to previous work:
A dierent, less sturdy frame than that of Stinnes (1998) was used. The
frame used by Stinnes (1998) had the additional bene t that the ma-
chined shroud was secured to the frame, preventing any independent
movement of the shroud from that of the frame as is the case with
the current frame. This allowed Stinnes (1998) to test at a very small
tip clearance of 1 mm to 1.5 mm. Before the blades interfered with the
shroud, good agreement with the results of Stinnes (1998) were found
when tests were carried out at 1.5 mm tip clearance, con rming the im-
portance of tip clearance eect on fan performance characteristics.
The fan blades used in this experimental setup are not the same fan
blades used by Bruneau (1994), Stinnes (1998) and Meyer (1996). Sadly,
that set of blades was destroyed before this project was initiated, result-
ing in a dierent set of blades that had to be produced to be used in
the test facility. It should however be noted that the new blades were
produced from the same moulds as the previous blades.
The new blades have a rougher surface nish than the previous blades.
This adds to performance losses due to surface drag as well as irregular-
ities on the blade surface that may induce early stall at some volumetric
ow rates.
Chapter 5
CFD validation with experimental
data
5.1 Introduction
The aim of this chapter is to compare the outcome of the CFD analysis of
an axial ow fan with the performance curve obtained from a fan test facility
conforming to the BS 848 standards (1980) as well as discuss relevent ndings
on dierent simulation techniques. Additional work was done to visualize
the ow eld across a single blade surface at four dierent ow settings by
using black wool tufts. This experiment is discussed in detail and results are
provided.
5.2 Fan performance curve
The fan performance curve obtained from the fan test facility provides a means
to verify and validate the accuracy and integrity of the CFD simulation. This
was done rstly to produce a new data set for the newly manufactured fan
blades and secondly to aord the student more insight into the operation of
the axial ow fan in practical use, especially in terms of the manner in which
the air ow exits the fan rotor. This knowledge was deemed important in
analysing the streamlines of the CFD computations.
Two CFD models are compared with experimental data. The rst model in-
corporates a pinch at the outlet boundary to reduce the amount of back ow
over the boundary by eectively increasing the axial velocity. A zero-shear
wall is used for the circumferential boundary surface at the outlet. This model
is referred to as Approach-1. The second model has no pinch at the outlet
boundary and the zero-shear wall boundary is replaced by a second outlet
boundary. This model is referred to as Approach-2.
54
CHAPTER 5. CFD VALIDATION WITH EXPERIMENTAL DATA 55
Figure 5.1 shows the comparison of fan static pressure rise between the CFD
analysis and the experimental data. The two dierent approaches are com-
pared side by side with experimental data obtained at 3 mm tip clearance.
3
Both outlets give exceptional agreement at the design operating point (16 m /s)
with the experimental data obtained by the student. The CFD results also
compare very well at higher ow rates and the deviation between the two
models is small. At lower ow rates the models deviate from each other
as well as from the experimentally measured data. At the lower ow rates
Approach-1 over-predicts contrary to Approach-2. Near 10 m3 /s, close to blade
stall, Approach-2 compares more favourably than Approach-1. In the range
3 3
of 4 m /s to 6 m /s Approach-2 is extremely unstable, to such an extent that
3
the simulation diverged. The data point for 8 m /s, although not diverging,
did not converge fully.
600
Experimental
Approach-1
Approach-2
500
Fan static pressure [Pa]
400
300
Numeca CFD software
200
B-fan: 8-bladed
Fan casing diameter: 1.542 m
Tip clearance: 3 mm
100 Blade root stagger angle: 59°
Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
Volumetric flow rate [m³/s]
Figure 5.1: Fan static pressure rise, CFD veri cation.
Figure 5.2 shows the comparison of fan power between the CFD analysis and
the experimental data. The deviation of the CFD results for both models from
experimental data can be attributed to the fact that the CFD models cannot
accommodate frictional losses from fan shaft bearings. Even though a no load
torque test was performed in the calibration stage, this was a static test and the
frictional forces from the bearings due to axial loading of the fan rotor during
operation is not accounted for. Both CFD models show exceptional trendline
CHAPTER 5. CFD VALIDATION WITH EXPERIMENTAL DATA 56
agreement to the experimental data measured by the student at volumetric
3
ow rates higher than 16 m /s, albeit at a constant dierence.
7000
6000
5000
Fan power [W]
4000
3000
Numeca CFD software
B-fan: 8-bladed
2000
Fan casing diameter: 1.542 m
Tip clearance: 3 mm
1000
Blade root stagger angle: 59°
Experimental
Density: 1.2 kg/m3
Approach-1
Rotational speed: 750 rpm
0
Approach-2
0 4 8 12 16 20 24
Volumetric flow rate [m³/s]
Figure 5.2: Fan power, CFD veri cation.
Figure 5.3 shows the comparison of fan static e ciency between the CFD
analysis and the experimental data. The fan static e ciency is a direct function
of the fan static pressure rise and fan power. Thus the deviations from the
experimental data as discussed above are directly transferred to the fan static
e ciency. The CFD results agree favourably with the experimental results as
3
measured by the student at 20 m /s and show good trendline agreement at a
3 3
volumetric ow rate of 10 m /s to 16 m /s.
Tables providing the data obtained through the simulation of the two CFD
models are presented in Appendix G.
5.3 Transition
A boundary layer starting as laminar ow and converting to fully turbulent ow
goes through a stage in the ow process called transition. According to White
(2006) transition can take place through any of the following three processes:
natural transition, bypass transition and separated- ow transition. Transition
CHAPTER 5. CFD VALIDATION WITH EXPERIMENTAL DATA 57
70
Experimental
Approach-1
60 Approach-2
Fan static efficiency [%]
50
40
30
Numeca CFD software
B-fan: 8-bladed
20
Fan casing diameter: 1.542 m
Tip clearance: 3 mm
10
Blade root stagger angle: 59°
Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
Volumetric flow rate [m³/s]
Figure 5.3: Fan static e ciency, CFD veri cation.
is a very complicated phenomenon typically in uenced by the Reynolds num-
ber, free stream turbulence intensity, surface roughness, shape factor and the
pressure gradient over the surface. White (2006) also mentions that no fun-
damental theory for transition exists and that the prediction of transitional
onset is carried out through experiments and correlations.
FINE allows the user to add a transition model to the Spalar-Allmaras turbu-
lence model. The onset of transition can either be user de ned by an educated
guess or predicted by the Abu-Ghannam and Shaw (AGS) model. Since accu-
rately predicting the onset of transition is beyond the scope of this thesis, the
latter option was employed. The turbulence model applied by FINE uses an
intermittency factor. The intermittency factor de nes the time during which
the ow over any point on the surface is fully turbulent and has a range from
zero, for fully laminar ow, to one for fully turbulent ow. The intermittency is
calculated at every point on the surface, at each iteration (FINE/Turbo Man-
ual, 2007b). The transitional model is incorporated into the Spalart-Allmaras
turbulence model by multiplying the turbulence production term with the in-
termittency factor.
The AGS model uses empirical relations derived from experimental data for
transition on a at plate with pressure gradients. The model predicts the onset
of transition at a momentum thickness Reynolds number that is exponen-
CHAPTER 5. CFD VALIDATION WITH EXPERIMENTAL DATA 58
tially dependent on a function of a dimensionless pressure gradient. The sign of
the dimensionless pressure gradient depends on whether the pressure gradient
is adverse or favourable. For an adverse pressure gradient the transition is pro-
moted and for a favourable pressure gradient the onset of transition is retarded.
Incorporating the transition model for the Spalart-Allmaras model had no
discernable eect on fan performance prediction of either the design operating
point or at a very low ow rate. The investigation on the eect of the transition
model on fan performance for the latter case was spawned from the indication
that the fan blade operates in a Reynolds number range of 200 000 to 700 000.
3
At 16 m /s the dierence in fan static pressure rise for the Spalart-Allmaras
3
model with and without transition is 0.019 % and at 2 m /s the dierence is
0.427 %.
5.4 Investigation of streamline distribution
Investigating the streamline distribution along the domain length indicates
whether the CFD analysis gives a true representation of the actual ow eld
produced by the fan blade. Figure 5.4 shows the streamline distribution at the
design operating point for Approach-1. The extent of radial ow depicted here
is not realistic, since during the experimental work no air ow is experienced
when standing adjacent the fan exit at the design operating point. Large
recirculation cells can clearly be seen in the ow outlet domain of the fan.
The governing equations for rotating systems are usually solved for relative ve-
locity components in the relative reference frame. Solving the absolute velocity
components in the relative reference frame should lead to the same ow solu-
tion. However it was found that by rather solving the latter case for the axial
ow fan greatly improved the streamline distribution. The problem with the
former case is that at far eld locations the relative velocity may become very
high, creating excess arti cial dissipation. According to the FINE/Turbo
Manual (2007b ) this excess arti cial dissipation leads to non-physical rota-
tional ow in the far eld region. Figure 5.5 shows the improvement on the
streamline distribution of the inlet for the simulation where the absolute veloc-
ity is solved in the relative reference frame. No recirculation cells are present
in the inlet upstream of the bellmouth as would be expected from the uniform
in ow provided by the mesh screens and settling chamber design of the fan
test facility. Unfortunately the unrealistic radial out ow distribution is still
evident.
A vast improvement in streamline distribution is apparent from Figure 5.6,
which shows the streamline distribution at the design operating point for
Approach-2. The streamlines are evenly spaced and no radial out ow from
CHAPTER 5. CFD VALIDATION WITH EXPERIMENTAL DATA 59
3
Figure 5.4: Streamline distribution, 16 m /s, Approach-1, relative velocity
solution.
3
Figure 5.5: Streamline distribution, 16 m /s, Approach-1, absolute velocity
solution.
CHAPTER 5. CFD VALIDATION WITH EXPERIMENTAL DATA 60
the fan rotor is evident at the fan outlet, which agrees with the actual setup.
Air in ow from the circumferential outlet boundary can also be seen.
3
Figure 5.6: Streamline distribution, 16 m /s, Approach-2, absolute velocity
solution.
Figure 5.7 gives a close-up view of the streamline distribution across the
blade for both CFD models. The uncharacteristically high radial out ow of
Approach-1 is in clear contrast to the axial out ow of Approach-2. The stream-
line distribution is an important aspect that will be used in the investigation
of the inlet relative velocity for the two-dimensional pro le sections employed
in the ADM. At the design operating point the ADM compares very well with
experimental data, while employing no radial ow component in the momen-
tum source terms. The no radial ow assumption of the ADM correlates
favourably with the streamline distribution of Figure 5.7(b).
It is assumed that the instability of the 8 m3 /s simulation is largely due to the
recirculation cells at the inlet to the blade towards the tip radius, the large
separation at the hub inlet to the fan blade and the large amount of back ow
at the trailing edge of the fan blade near the hub (Figure 5.8). The amount of
turbulence produced by the recirculation cells near the tip radius is the most
probable cause of unsteadiness of this particular simulation.
CHAPTER 5. CFD VALIDATION WITH EXPERIMENTAL DATA 61
(a) Streamline distribution, 16
3
m /s, (b) Streamline distribution, 16
3
m /s,
Approach-1, close-up. Approach-2, close-up.
Figure 5.7: Streamline distribution close-up views for Approach-1 and
Approach-2.
The streamline distribution diagrams for the remaining CFD data points on
Figures 5.1 to 5.3 are presented in Appendix H.
3
Figure 5.8: Streamline distribution, 8 m /s, Approach-2, close-up view.
CHAPTER 5. CFD VALIDATION WITH EXPERIMENTAL DATA 62
5.5 Flow pattern visualization
Additional work has been performed by investigating the ow pattern across
the blade length. This is essential for future inquiry into the shape of the
two-dimensional pro le section to be used in the ADM. It is assumed that at
lower ow rates the increase in radial ow requires the two-dimensional pro le
to slant radially and align itself with the inlet relative velocity. In doing this
the chord length of the particular pro le is eectively lengthened, which would
result in dierent momentum source terms, contrary to the linearly decreasing
chord lengths of the vertically aligned radially stacked pro les used by current
ADM's.
To validate the ow pattern output from the CFD model, small tufts of wool
were used to capture the ow pattern on the experimental blades. The blade
surface was divided (radially) into ten rows containing ve tufts each. A very
light and thin string of wool was used to construct the tufts, which were at-
tached as close as possible to each other. Care was taken to ensure that no
tufts interacted with one another. Since the tufts indicated an outwards de ec-
tion due to the centrifugal force exerted on them, only the change in de ection
was considered when evaluating the occurrence of stall on the fan blades. A
Lucas Dawe Ultrasonics Stobotorch (stroboscope) was used to visually slow
down one blade and keep it stationary for an opportunity to take a photo of
the fan blade ow pattern. To provide a good contrasting background for the
stroboscope the particular blade containing the black tufts was lightly spray
painted white and the experiment was carried out at night time. The ow pat-
tern visualization experiment was carried out at 700 rpm, since at this stage
there had already been an incident with blades connecting the shroud and the
decision was made to run at a lower speed to avoid another accident that might
result in a loss of the fan blades.
Figure 5.9(a) shows the ow pattern of a fully stalled blade. Towards the blade
trailing edge there is no ow over the blade surface, as can be seen by the tufts
only subjected to its own centrifugal force. The trailing edge of the blade
is completely stalled from hub to tip. The ow over the leading edge is very
unstable, typical of turbulent ow. It is not apparent from the photo, but while
the experiment was undertaken at this point, one could clearly see the tuft at
the hub near the leading edge ipping around to the pressure side of the blade
and then back again. This is due to the presence of reverse ow at this point.
The stagnant region on the blade surface indicated by the tufts is mirrored
in Figure 5.9(b) by the ow eld representation from the CFD analyses. The
CFD does not predict stalling of the complete trailing edge. This means that
a larger part of the blade surface is still eectively performing work on the
uid, according to the CFD. Once separation occur the lift force is restricted
and the drag force increased (White, 2006). This would explain the higher fan
CHAPTER 5. CFD VALIDATION WITH EXPERIMENTAL DATA 63
static pressure rise at lower ow rates predicted by the CFD model with the
pinched outlet (Approach-1).
(a) Experimental fan blade ow pattern (b) CFD fan blade ow pattern at 2 m /s
3
with throttle device closed. (Approach-1).
3
Figure 5.9: Flow pattern visualization at 2 m /s.
In the operating range where it is assumed that fan blade stall begins, it is
clear from Figure 5.10(a) that the blade starts stalling at the hub and towards
the trailing edge. Between the hub radius and mean radius roughly 60 % of
the blade is stalled. The CFD models replicate the stall region visible from
the experiments. Comparing Figures 5.10(b) and 5.10(c) it is evident that the
CFD model with the open outlet (Approach-2) predicts a larger stalled region.
This explains the lower predicted pressure rise of Approach-2 as discussed in
the previous paragraph.
CHAPTER 5. CFD VALIDATION WITH EXPERIMENTAL DATA 64
(a) Experimental fan (b) CFD fan blade ow pat- (c) CFD fan blade ow pattern
blade ow pattern at 10
3
tern at 10 m /s (Approach-1).
3
at 10 m /s (Approach-2).
3
m /s.
3
Figure 5.10: Flow pattern visualization at 10 m /s.
Figure 5.11(a) shows that at the design operating point, only a small portion
of the blade near the hub trailing edge stalls. The stalled region is again
replicated by the two CFD models. There is very little discrepancy between
Figures 5.11(b) and 5.11(c). This is in line with the 0.684 % dierence in
predicted fan static pressure rise between the two models.
(a) Experimental fan (b) CFD fan blade ow pat- (c) CFD fan blade ow pattern
blade ow pattern at 16
3
tern at 16 m /s (Approach-1).
3
at 16 m /s (Approach-2.)
3
m /s.
3
Figure 5.11: Flow pattern visualization at 16 m /s.
CHAPTER 5. CFD VALIDATION WITH EXPERIMENTAL DATA 65
There is no stalling of the blade for the throttling device in the fully open
position (Figure 5.12(a)). Figures 5.12(b) and 5.12(c) indicate that only the
small region adjacent to the hub experiences separated air ow.
(a) Experimental fan (b) CFD fan blade ow pat- (c) CFD fan blade ow pat-
blade ow pattern with
3
tern at 20 m /s (Approach-1).
3
tern at 20 m /s (Approach-2).
throttle device fully
open.
3
Figure 5.12: Flow pattern visualization at 20 m /s.
Chapter 6
Conclusion and recommendations
6.1 Motivation for study
Axial ow fans are rotary machines designed to provide a relatively low pres-
sure rise at high ow rates. Their applications range from small scale building
air ventilation and gas extraction to large scale process uid cooling or con-
densation at power plants. The axial ow fan considered in this thesis is a
1 th
scale version of the typical axial ow fans employed in direct air cooled
6
heat exchangers at power generation plants, such as Matimba. These fans
are utilized in arrays beneath the nned tube heat exchangers in an A-frame
construction. The process uid is cooled or condensed by the dierence in
temperature between the ambient temperature of the air ow that is moved
through the heat exchanger bundles (by the axial ow fans) and the temper-
ature of the nned tubes.
E cient design of forced draught dry-cooled steam condenser sites result in
large cost savings, considering the annual total power consumption of the axial
ow fans. The Matimba power plant employs 288 axial ow fans, 9.145 m in
diameter. CFD is a fast and relatively low cost design tool to investigate the
system eects for such ACSC units. Unfortunately simulating the axial ow
fans with full discrete rotating surfaces when numerically modelling the entire
plant is computationally very expensive and simpler methods have been suc-
cessfully used for ideal environments and system eects (Van Staden, 1996).
The actuator disc method is commonly applied for such investigations. How-
ever, the ADM under-predicts fan performance at lower ow rates since it does
not incorporate the eect of radial ow across the blade.
The aim of this dissertation is to model a single axial ow fan with three-
dimensional rotating surfaces to capture the eects of radial ow over the
fan blade at low ow rates. The knowledge gained will be utilized to amend
the ADM. Veri cation with experimental work enforce con dence in the CFD
66
CHAPTER 6. CONCLUSION AND RECOMMENDATIONS 67
results. The dissertation aimed to produce a new database of fan performance
data for the newly manufactured B2-fan blades.
6.2 Research ndings
A new database of fan performance data was produced for the B2-fan. The
experimental research results and main ndings are discussed in Section 6.2.1.
A discrete three-dimensional CFD model was produced and simulated with
NUMECA CFD software. The numerical research results and main ndings
are discussed in Section 6.2.2.
6.2.1 Experimental
Extremely good repeatability was found between subsequent tests. A fan static
3
pressure rise of 207.4 Pa at the design ow rate of 16 m /s was achieved. Stinnes
(1998) achieved a higher fan static pressure rise at the design operating point,
but tested the fan at a tip clearance of 1 mm to 1.5 mm. This was achieved
by using a sturdier frame limiting the radial movement of the fan blades when
the fan shaft starts to oscillate. In an attempt to recreate the fan perfor-
mance data for a tip clearance of 1.5 mm the fan blades interfered with the
shroud annulus. To preserve the fan blades the tip clearance was set back to
3 mm, which was deemed safer since a dierent frame to that used by Stinnes
(1998) was utilized. The one set of fan performance characteristics obtained
at 1.5 mm tip clearance compared well with the results obtained by Stinnes
(1998) (Figures F.13 to F.15).
Most of the work done by the blade on the air is performed near the tip radius.
The higher pressure on the pressure side of the blade causes the ow to leak
through the tip clearance towards the lower pressure suction side along a ow
path of least resistance. This phenomenon is termed a tip vortex, i.e. the air
loss or leakage around the fan blade tip. Larger tip clearances leading to larger
tip vortexes result in loss of performance in terms of pressure rise capability
and air ow (Hudson Products Corporation, 2000).
The proposed method of Venter and Kröger (1992) for determining the eect of
tip clearance on fan performance was used to establish the proposed fan static
pressure rise achievable at a tip clearance of 1.5 mm. This method predicted a
fan static pressure rise of 212 Pa, which correlates well with the measured fan
static pressure rise of 216 Pa at 1.5 mm tip clearance.
All fan performance data curves follow a smooth trend line with no discernible
irregularities observed throughout the volumetric ow rate range. This range
is extended beyond the range of ow rate points tested by Bruneau (1994)
CHAPTER 6. CONCLUSION AND RECOMMENDATIONS 68
and Stinnes (1998). The importance of ambient temperature on the torque
readings became apparent only after extensive calibration and recalibration
of all experimental instrumentation. Testing in the early hours of the morn-
ing (typically two hours before sunrise) when there is no change in ambient
temperature and no heating of the settling chamber by direct sunlight and
radiation from surrounding window panes, vastly improved the repeatability
of the torque readings and as such the fan power characteristic. Arriving at
this conclusion involved at lot of testing and re-testing the B2-fan in the test
facility.
Additionally an experiment was carried out to visualize the ow eld over the
fan blade suction side surface. This was achieved by running the fan at the
dierent volumetric ow settings during the evening and letting one of the
blades become stationary with the aid of a stroboscope. Photographs and
videos were taken of the blade with a digital SLR camera. Good agreement was
found in visualizing the stalled region in the ow eld over the blade surface.
With reference to the CFD streamline representation, the black wool tufts does
not give a good indication of the ow eld distribution in terms of ow angle,
due to the eect of centrifugal forces. The strength and advantage from the
tuft representation experiment lies in the ease of setting up the experiment and
acquiring an indication of the extent of stall occuring on the blade surface. The
examination of the dierence between the tuft representation of the ow eld
and the ow eld predicted by the CFD models provided better understanding
of the dierence in fan static pressure rise predicted at lower ow rates. From
Figure 5.9(b) the large degree of radial ow experienced over the blade is
evident. This con rms the need to amend the ADM to incorporate the eects
of radial ow at lower ow rates.
6.2.2 Numerical Simulation
1 th
A axisymmetric model of the B2-fan was constructed using structured grid
8
cells. The model developed from a learning curve that started with just mod-
elling the fan blades in an annular pipe section and progressed to incorporate
the full dimensions of the settling chamber as well as the bellmouth inlet to
the fan rotor. Two dierent outlets were compared to correctly simulate a fan
rotor exhausting into open atmosphere. Both models compared exceptionally
well at the design operating ow rate as well as higher ow rates.
Approach-1 was developed based on the recommendations of the NUMECA
support team to pinch the ow domain at the outlet boundary. The model
is only allowed an outlet boundary on a plane normal to the fan axis. This
method constricts the amount of back ow over the outlet boundary and im-
proves the stability of the simulation. The fan performance curve obtained
with this model deviated from experimental data at lower ow rates. The
CHAPTER 6. CONCLUSION AND RECOMMENDATIONS 69
streamline distribution of the model along the domain is not realistic. The
recirculation cells in the inlet section, comprising of the settling chamber and
bellmouth, was a particular point of concern.
Switching from the method of solving the relative velocity in the relative refer-
ence frame to one where the absolute velocity is solved in the relative reference
frame resolved the streamline distribution issue in the inlet section. Unfortu-
nately this resulted in a 10.7 % decrease in the predicted fan static pressure
rise. The tuning of the amount of pinch applied at the outlet boundary that
successfully eliminated any back ow over the outlet boundary was also lost.
Pinching the outlet even more removed the new found back ow over the bound-
ary, but resulted in a further adverse eect on the fan static pressure rise. For
both cases the recirculation cells in the outlet sections remained.
Removing the pinch from the previous model resulted in a further 2.9 % de-
crease in fan static pressure rise. The zero-shear walls at the circumferential
boundary in the outlet section still in uenced the out ow from the fan by allow-
ing recirculation cells to form. A separate investigation into this phenomenon
showed that the longer the simulation ran, the stronger these recirculation
cells became (adversely aecting the fan static pressure rise).
In the nal CFD model (Approach-2) the zero-shear wall as well as the pinch
was removed. The recirculation cells were eliminated and the fan static pres-
sure rise was considerably improved. This model slightly over-predicts the fan
3
static pressure rise at the design ow rate of 16 m /s by 1.4 %. A uniform
streamline distribution along the computational domain was achieved with no
radial ow over the blade at the design operating point. Previous attempts
at simulating this particular outlet when solving the relative velocity in the
relative reference frame failed.
Another issue identi ed within this research is the mixing losses introduced at
the rotor-stator interfaces. The mixing losses over the rotor-stator interface
can range anywhere from 10 Pa to 100 Pa, which is an order of magnitude of
the pressure capability of the axial ow fan and in uences the fan performance
considerably.
6.3 Recommendations for future research
All eects were investigated for producing experimental fan performance data
curves. For future test work to be carried out at the test facility at Stellen-
bosch University care should be taken regarding the time of the day that tests
are conducted. A recommendation for future test work would be to use a ther-
CHAPTER 6. CONCLUSION AND RECOMMENDATIONS 70
mocouple to measure the temperature of the air going into the fan.
To facilitate in-depth research in tip clearance beyond the minimum tip clear-
ance investigated by Venter and Kröger (1992) of 3 mm, a more rigid fan motor
frame should be used. The method of centering the fan rotor radially within
the shroud annulus needs to be addressed. With the current setup the can-
tilever beam housing the fan shaft bearings and fan shaft is attached to and
supported by the shroud annulus with three chains. By adjusting the tension
in each chain independently the fan rotor is centered, however the fan rotor
is constrained to move only in these three directions. If the chain setup is
continued it should be replaced by four chains aligned 0°, 90°, 180° and 270°
relative to the horizon, to allow more freedom of movement and centering of
the fan rotor. The method of supporting the cantilever beam by attaching
it to the shroud annulus with chains should be abandoned altogether, since
during testing the whole settling chamber wall normal to the fan axis starts to
move back and forth. The frame shroud be xed to the settling chamber wall
to prevent any independent movement.
Solving a complete performance curve for a low pressure rise, ducted turbo-
machinery application was attempted with CFD software primarily developed
for high pressure rise, high velocity turbomachinery applications. Simulation
of the fan performance curve at volumetric ow rates lower than the design
operating point is constrained by instabilities in the ow solution. Recircula-
tion cells forming at the inlet to the fan rotor produce high turbulence that
the ow solver is unable to disperse.
Recommendations for future research include:
An investigation to ascertain to what extent a transient simulation would
solve the low ow rate instabilities.
The model was unsuccessful in switching from the Spalart-Allmaras tur-
bulence model to any of the low Reynolds k − ε turbulence models
(Launder-Sharma k − ε, Chien k − ε and Extended wall-function k − ε),
with the Yang-Shih k − ε turbulence model being the only exception.
At the design operating point the k − ε turbulence model is less stable
than the Spalart-Allmaras turbulence model and the eect of the k − ε
turbulence model at lower ow rates warrants future investigation.
Implementing a grid that is dependent on the operating point speci -
cation of the fan performance curve. Kelecy (2000) mentions observing
signi cant pressure gradients (from the inlet to the fan) at lower ow
rates. To resolve this, he extended the inlet mesh an additional two fan
diameters upstream from the original location, for lower ow rates. The
same method of a ow-dependent mesh should be investigated.
CHAPTER 6. CONCLUSION AND RECOMMENDATIONS 71
The knowledge gained related to the model setup and the available blade
geometry data obtained should be used for future investigation into the
simulation of an axial ow fan on dierent CFD software packages such
as FLUENT and OpenFoam.
Appendices
72
Appendix A
Fan blade data for numerical
simulation
In this appendix the distribution of chord length and stagger angle radially
across the blade length is illustrated. A plot to exemplify the similarity be-
tween the GA(W)-2 pro le and the LS(1)-0413 pro le from the dierent data
sets is presented. The fan blade pro le data set that de nes the hub pro le is
given.
50 1.15
Stagger angle
Chord length
1.1
40
1.05
Stagger angle [ ° ]
30 Chord length [m]
1
0.95
20
0.9
10
0.85
0 0.8
0.4 0.6 0.8 1 1.2 1.4 1.6
r/rm
Figure A.1: Chord length and stagger angle distribution.
73
APPENDIX A. FAN BLADE DATA FOR NUMERICAL SIMULATION 74
0.2
LS(1)-0413
GA(W)-2
0.1
y/c
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
-0.1
-0.2
x/c
Figure A.2: A plot of the GA(W)-2 and LS(1)-0413 pro les.
APPENDIX A. FAN BLADE DATA FOR NUMERICAL SIMULATION 75
Table A.1: Fan blade pro le data for numerical simulation
x/c y/c, upper y/c, lower
0.0000 0.0000 0.0000
0.0005 0.0034 -0.0015
0.0010 0.0062 -0.0028
0.0015 0.0085 -0.0040
0.0020 0.0104 -0.0050
0.0030 0.0131 -0.0068
0.0040 0.0148 -0.0082
0.0050 0.0159 -0.0094
0.0125 0.0242 -0.0145
0.0250 0.0332 -0.0191
0.0375 0.0397 -0.0223
0.0500 0.0448 -0.0250
0.0750 0.0526 -0.0294
0.1000 0.0586 -0.0328
0.1250 0.0635 -0.0356
0.1500 0.0675 -0.0379
0.1750 0.0710 -0.0398
0.2000 0.0740 -0.0414
0.2250 0.0765 -0.0427
0.2500 0.0786 -0.0437
0.2750 0.0803 -0.0443
0.3000 0.0818 -0.0448
0.3250 0.0830 -0.0451
0.3500 0.0838 -0.0452
0.3750 0.0843 -0.0450
0.4000 0.0846 -0.0447
0.4250 0.0846 -0.0442
0.4500 0.0844 -0.0435
0.4750 0.0838 -0.0426
0.5000 0.0829 -0.0414
0.5250 0.0817 -0.0399
0.5500 0.0802 -0.0381
0.5750 0.0783 -0.0359
0.6000 0.0761 -0.0333
0.6250 0.0733 -0.0305
0.6500 0.0702 -0.0274
0.6750 0.0667 -0.0242
0.7000 0.0629 -0.0210
0.7250 0.0587 -0.0177
0.7500 0.0542 -0.0144
0.7750 0.0495 -0.0113
0.8000 0.0445 -0.0083
0.8250 0.0393 -0.0057
0.8500 0.0340 -0.0035
0.8750 0.0284 -0.0018
0.9000 0.0227 -0.0008
0.9250 0.0169 -0.0006
0.9500 0.0110 -0.0013
0.9750 0.0048 -0.0034
1.0000 -0.0016 -0.0071
Appendix B
Sample calculations for
experimental data
In this appendix the sample calculations are given to illustrate how the fan per-
formance characteristics are derived according to the BS 848 standards (1980).
The B2-fan from Bruneau (1994) was used, set at a 59° stagger angle with a
at plate covering the hub.
To process the raw data points the following steps were followed:
1. The captured data points from the LabVIEW program were saved in a
® spreadsheet where the starting and end zero values were used to
text le. The data points were moved from the text le into an Microsoft
Excel
calculate the drift. The correction for the drift was made linearly between
the starting and end zero values.
2. The corrected values were adjusted by the respective calibration correc-
tion as derived in Chapter E. In the case of the torque values, a further 2
Nm was deducted to account for the no-load torque values as discussed
in Chapter E, Section E.2.
3. The fan laws with the conversion factors given in the BS 848 standards
(1980), were used to scale the results for fan pressure, fan power and
fan e ciency to the standardised conditions. Since the fan size remains
3
unchanged, only the speed (750 rpm) and density (1.2 kg/m ) conversion
factors need to be applied.
4. The results were scaled and ready for representation in graphs with fan
performance curves for fan static pressure rise, fan power consumption
and fan static e ciency plotted against volumetric ow rate respectively.
76
APPENDIX B. SAMPLE CALCULATIONS FOR EXPERIMENTAL DATA 77
B.1 Sample Calculations
A fan performance data set consists of twenty one data points, each of which
contains the static pressure dierential between the settling chamber and the
atmosphere(1), the static pressure dierential across the inlet bellmouth(2) ,
the torque transferred to the fan(3) and the rotational speed of the fan(4).
Before and after each run the ambient pressure and temperature is recorded
and averaged.
Table B.1: Raw experimental data
Data point (1)∆psett (2)∆pbell (3)M (4)N
1 -0.004 -0.006 -0.004 -0.001
...
10 1.699 2.639 2.58 8.252
...
25 0.005 -0.001 -0.001 -0.001
Data points 2 to 9 and 11 to 14 have been excluded from Table B.1, since only
the tenth data point will be used in the sample calculations. See Table C.2 for
the complete data range. Data points 1 and 25 are taken before and after each
experimental run to account for drift during the duration of the experiment.
The drift is assumed to increase linearly from the zero wind still condition
(data point 1) to data point 25 containing the total zero value error drift at
the end of the run. The initial error in obtaining a zero value is also added to
the error due to drift to be deducted from each data point. The raw data is
corrected for drift by using the following relation:
number of data point −1
drift = data(1) + (data(25) − data(1))
number of measurements − 1
(B.1)
APPENDIX B. SAMPLE CALCULATIONS FOR EXPERIMENTAL DATA 78
The corrected values for data point 10:
9
∆psett = (162.248) − (−0.531) + 24 × [(0.329) − (−0.531)]
= 162.457 Pa (B.2)
9
∆pbell = (254.156) − (−0.127) + 24 × [(0.354) − (−0.127)]
= 254.102 Pa (B.3)
9
M = (70.472) − (−0.135) + 24 × [(−0.053) − (−0.135)]
= 70.577 Nm (B.4)
9
N = (700.162) − (−0.460) + 24 × [(−0.460) − (−0.460)]
= 700.622 rpm (B.5)
From equation (B.5) it is apparent that the speed indicator box is not aected
by drift but does render the same error for zero rotation at the beginning and
end of a test run. Therefore only this error value is subtracted from the mea-
sured values in agreement with Meyer (1996).
The ambient pressure was measured as 1007.6 hPa.
pamb = 1007.6 × 100
= 100 760 Pa (B.6)
The ambient temperature was measured as 10.6 °C.
Tamb = 10.6 + 273.15
= 283.75 K (B.7)
The sample calculations below follow the guidelines set out in the BS 848 stan-
dards (1980).
The ambient density is calculated from the perfect gas relation:
pamb
ρamb =
R Tamb
100760
=
287.08 × 283.75
= 1.237 kg/m3 (B.8)
The mass ow rate through the fan test facility is determined by measuring
APPENDIX B. SAMPLE CALCULATIONS FOR EXPERIMENTAL DATA 79
the pressure drop across the inlet bellmouth and using the following relation:
πd2bell p
ṁ = αε 2ρamb ∆pbell
4
π × (1.008)2 √
= 0.9803 2 × 1.237 × 254.102
4
= 19.614 kg/s (B.9)
The compound calibration constant (αε) used here is the one determined by
Venter (1990) to be 0.9803 and diers from the value of 0.985 recommended by
the BS 848 standards (1980). This is due to the transformation piece between
the inlet bellmouth and the remainder of the test facility being shorter (due
to space limitations) than speci ed in the BS 848 standards.
The air density inside the settling chamber is calculated according to the ideal
gas relation:
pamb − ∆psett
ρsett = ρamb
pamb
100760 − 162.457
= 1.237
100760
= 1.235 kg/m3 (B.10)
The mass ow calculated in equation (B.9) is used to determine the settling
chamber's dynamic pressure component:
1 2
pdsett = ρsett Usett
2
2
1 ṁ
= ρsett
2 Asett
2
1 ṁ
=
2ρsett Asett
2
1 19.614
=
2 × 1.235 16
= 0.608 Pa (B.11)
The BS 848 standards (1980) stipulate the fan pressure rise as the dierence
between the fan static pressure at the fan outlet and the total pressure at the
pFs = pamb −(pssett +pdsett ), but since the measured settling
fan inlet. Therefore
chamber pressure (∆psett ) is the static pressure dierence between the settling
chamber pressure and the atmosphere it can be seen as ∆psett = pamb − pssett
APPENDIX B. SAMPLE CALCULATIONS FOR EXPERIMENTAL DATA 80
and the fan pressure rise de ned earlier can be simpli ed accordingly:
pFs = pamb − (pssett + pdsett )
= ∆pssett − pdsett
= 162.457 − 0.608
= 161.849 Pa (B.12)
The fan power input is a product of the rotational speed and the torque applied
to the fan shaft. 2 Nm is deducted from the applied torque to account for
frictional losses as noted in Section E.2.
2πN M
PF =
60
2 × π × 700.622 × (70.577 − 2)
=
60
= 5031.427 W (B.13)
The volumetric ow rate is calculated from the mass ow through the test
facility and the inlet density.
ṁ
Qsett =
ρsett
19.614
=
1.235
= 5.882 m3 /s (B.14)
The fan static e ciency is de ned as the pressure rise per power consumed by
the fan, multiplied by the volumetric ow rate through the fan test facility.
pFs Qsett
ηFs =
PF
161.849 × 15.882
= × 100
5031.427
= 51.09% (B.15)
The above calculated fan test parameters are scaled to a referenced air den-
3
sity of 1.2 kg/m and a rotational speed of 750 rpm by using the fan laws as
stipulated in the BS 848 standards (1980). The prime noted variables refer to
APPENDIX B. SAMPLE CALCULATIONS FOR EXPERIMENTAL DATA 81
the scaled reference parameters:
0 3 0
0 d N
Q = Qsett
d N
3
1.542 750
= 15.882
1.542 700.622
3
= 17.002 m /s (B.16)
0 2 0 2 0
d N ρ
p0Fs = pFs
d N ρsett
2 2
1.542 750 1.2
= 161.849
1.542 700.622 1.235
= 180.217 Pa (B.17)
0 5 0 3 0
d N ρ
PF0 = PF
d N ρsett
5 3
1.542 750 1.2
= 5031.427
1.542 700.622 1.235
= 5997.31 W (B.18)
0
ηFs = ηFs = 51.09% (B.19)
Appendix C
Experimental Data
For 59° the experimental constants, the data from the conversion of raw voltage
readings, the data from the correction for drift, the calculated performance pa-
rameters and the scaled performance parameters are given. These intermediate
data sets are not provided for angles 58° and 60° and only the experimental
constants and scaled performance parameters for these two angles are given.
Table C.1: Experimental data constants for 59° stagger angle
3
Reference density 1.2 kg/m
Reference rotational speed 750 rpm
Reference fan diameter 1.542 m
Shroud diameter 1.542 m
Bellmouth diameter 1.008 m
Tip clearance 0.003 m
◦
Stagger angle 59
Compound calibration constant 0.9803
Universal gas constant 287.08
Actual rotational speed 700 rpm
Ambient pressure 1007.6 hPa
◦
Ambient temperature 10.6 C
82
APPENDIX C. EXPERIMENTAL DATA 83
Table C.2: Conversion of voltage data for 59° stagger angle
Raw voltage data Voltage data converted
∆psett ∆pbell T N ∆psett ∆pbell M N
V V V V Pa Pa Nm rpm
1 -0.004 -0.006 -0.004 -0.001 -0.531 -0.127 -0.135 -0.460
2 -0.127 4.948 1.624 8.264 -12.288 476.136 44.350 701.181
3 -0.089 4.892 1.637 8.251 -8.656 470.752 44.705 700.077
4 -0.06 4.851 1.648 8.241 -5.884 466.811 45.006 699.228
5 0.129 4.589 1.773 8.248 12.181 441.623 48.421 699.823
6 0.241 4.445 1.849 8.256 22.887 427.779 50.498 700.502
7 0.484 3.999 2.003 8.246 46.114 384.902 54.706 699.653
8 0.774 3.612 2.162 8.252 73.833 347.697 59.051 700.162
9 1.18 3.159 2.362 8.25 112.640 304.147 64.516 699.992
10 1.699 2.639 2.58 8.252 162.248 254.156 70.473 700.162
11 2.149 2.12 2.719 8.25 205.261 204.260 74.271 699.992
12 2.478 1.714 2.78 8.253 236.708 165.229 75.938 700.247
13 2.807 1.296 2.802 8.25 268.155 125.044 76.539 699.992
14 3.029 1.012 2.771 8.246 289.375 97.741 75.692 699.653
15 3.215 0.767 2.728 8.248 307.153 74.187 74.517 699.823
16 3.398 0.503 2.644 8.254 324.645 48.807 72.221 700.332
17 3.623 0.334 2.556 8.249 346.152 32.560 69.817 699.907
18 3.869 0.195 2.515 8.255 369.665 19.197 68.696 700.417
19 4.12 0.105 2.459 8.252 393.657 10.544 67.166 700.162
20 4.282 0.058 2.438 8.249 409.142 6.026 66.592 699.907
21 4.374 0.041 2.432 8.255 417.935 4.392 66.429 700.417
22 4.418 0.031 2.419 8.248 422.141 3.430 66.073 699.823
23 4.516 0.021 2.419 8.255 431.508 2.469 66.073 700.417
24 4.631 0.01 2.419 8.253 442.500 1.411 66.073 700.247
25 0.005 -0.001 -0.001 -0.001 0.329 0.354 -0.053 -0.460
APPENDIX C. EXPERIMENTAL DATA 84
Table C.3: Voltage data corrected for drift for 59° stagger angle
Data corrected for drift
∆psett [Pa] ∆pbell [Pa] M [Nm] N [rpm]
-12.324 476.243 44.482 701.641
-8.196 470.839 44.833 700.537
-5.460 466.877 45.131 699.688
12.569 441.669 48.543 700.282
23.239 427.806 50.616 700.962
46.430 384.909 54.821 700.113
74.113 347.683 59.162 700.622
112.885 304.113 64.624 700.452
162.457 254.102 70.577 700.622
205.434 204.187 74.372 700.452
236.845 165.135 76.035 700.707
268.256 124.930 76.633 700.452
289.440 97.607 75.782 700.113
307.183 74.034 74.604 700.282
324.639 48.633 72.305 700.792
346.110 32.366 69.897 700.367
369.587 18.983 68.774 700.877
393.543 10.311 67.240 700.622
408.992 5.772 66.663 700.367
417.750 4.118 66.495 700.877
421.920 3.136 66.137 700.282
431.251 2.155 66.133 700.877
442.207 1.078 66.130 700.707
APPENDIX C. EXPERIMENTAL DATA 85
Table C.4: Calculated performance parameters for 59° stagger angle
ṁ ρsett pdsett pFs PF Qsett ηFs
3 3
kg/s kg/m Pa Pa W m /s %
26.852 1.237 1.138 -13.462 3121.369 21.706 -9.362
26.699 1.237 1.125 -9.322 3142.268 21.583 -6.403
26.586 1.237 1.116 -6.576 3160.233 21.493 -4.472
25.859 1.237 1.056 11.513 3413.146 20.908 7.053
25.450 1.237 1.023 22.216 3568.645 20.579 12.811
24.140 1.236 0.921 45.509 3872.588 19.525 22.945
22.943 1.236 0.832 73.282 4193.919 18.562 32.434
21.457 1.236 0.728 112.157 4593.517 17.367 42.403
19.614 1.235 0.608 161.849 5031.427 15.882 51.090
17.582 1.234 0.489 204.945 5308.558 14.243 54.988
15.812 1.234 0.396 236.450 5432.546 12.813 55.768
13.753 1.234 0.299 267.957 5474.416 11.148 54.567
12.156 1.233 0.234 289.206 5409.407 9.856 52.694
10.587 1.233 0.178 307.006 5324.304 8.585 49.503
8.581 1.233 0.117 324.522 5159.481 6.960 43.774
7.000 1.233 0.078 346.032 4979.747 5.679 39.460
5.361 1.232 0.046 369.542 4900.891 4.350 32.800
3.951 1.232 0.025 393.518 4786.590 3.207 26.363
2.956 1.232 0.014 408.978 4742.514 2.400 20.694
2.497 1.232 0.010 417.740 4733.679 2.027 17.888
2.179 1.232 0.008 421.912 4703.366 1.769 15.870
1.806 1.232 0.005 431.246 4707.106 1.467 13.436
1.277 1.232 0.003 442.205 4705.715 1.037 9.746
APPENDIX C. EXPERIMENTAL DATA 86
Table C.5: Scaled performance parameters for 59° stagger angle
3
Q0sett [m /s] p0Fs [Pa] PF0 [W]
23.202 -14.921 3697.970
23.107 -10.365 3740.505
23.038 -7.330 3775.703
22.392 12.813 4068.227
22.019 24.679 4241.668
20.916 50.690 4620.761
19.870 81.527 4994.639
18.595 124.885 5476.617
17.002 180.217 5997.310
15.251 228.413 6334.950
13.714 263.416 6477.868
11.937 298.827 6536.960
10.558 322.906 6470.101
9.195 342.673 6364.804
7.448 361.762 6155.401
6.081 386.290 5953.054
4.655 412.032 5847.388
3.433 439.189 5718.608
2.570 456.845 5673.006
2.169 465.995 5650.595
1.895 471.468 5628.948
1.569 481.126 5619.630
1.110 493.646 5622.668
Table C.6: Experimental data constants for 58° stagger angle
3
Reference density 1.2 kg/m
Reference rotational speed 750 rpm
Reference fan diameter 1.542 m
Shroud diameter 1.542 m
Bellmouth diameter 1.008 m
Tip clearance 0.003 m
◦
Stagger angle 58
Compound calibration constant 0.9803
Universal gas constant 287.08
Actual rotational speed 700 rpm
Ambient pressure 1007.9 hPa
◦
Ambient temperature 12.5 C
APPENDIX C. EXPERIMENTAL DATA 87
Table C.7: Scaled performance parameters for 58° stagger angle
3
Q0sett [m /s] p0Fs [Pa] PF0 [W] ηFs %
23.566 7.802 4518.528 4.069
23.564 6.008 4479.143 3.161
23.718 4.305 4463.241 2.288
22.869 27.460 4777.109 13.146
22.186 42.089 4999.553 18.678
21.355 63.546 5273.662 25.732
20.366 94.761 5666.742 34.057
18.961 140.071 6135.895 43.285
17.330 192.219 6623.218 50.295
15.650 242.104 6961.444 54.427
13.773 282.148 7100.306 54.731
12.218 310.224 7103.937 53.357
10.741 334.088 6997.754 51.280
9.343 353.842 6839.119 48.337
7.519 369.650 6621.966 41.974
6.193 392.168 6412.283 37.875
4.630 420.932 6317.413 30.852
3.423 447.817 6258.080 24.493
2.645 465.980 6245.329 19.734
2.112 475.657 6228.794 16.125
1.926 481.770 6232.219 14.890
1.514 490.803 6249.711 11.889
1.238 499.808 6282.884 9.846
Table C.8: Experimental data constants for 60° stagger angle
3
Reference density 1.2 kg/m
Reference rotational speed 750 rpm
Reference fan diameter 1.542 m
Shroud diameter 1.542 m
Bellmouth diameter 1.008 m
Tip clearance 0.003 m
◦
Stagger angle 60
Compound calibration constant 0.9803
Universal gas constant 287.08
Actual rotational speed 700 rpm
Ambient pressure 1005.8 hPa
◦
Ambient temperature 16.35 C
APPENDIX C. EXPERIMENTAL DATA 88
Table C.9: Scaled performance parameters for 60° stagger angle
3
Q0sett [m /s] p0Fs [Pa] PF0 [W] ηFs %
22.633 -23.108 3186.913 -16.411
22.613 -24.147 3200.884 -17.059
22.476 -19.072 3233.801 -13.256
22.106 -4.418 3445.759 -2.834
21.456 17.735 3740.938 10.172
20.510 41.769 4077.620 21.009
19.635 66.761 4393.162 29.839
18.263 111.212 4898.502 41.462
16.772 168.176 5449.637 51.758
15.098 211.753 5750.974 55.590
13.570 249.137 5952.692 56.792
12.177 278.952 6019.756 56.430
10.541 309.986 6003.413 54.428
9.020 336.497 5901.827 51.427
7.269 355.033 5715.762 45.154
5.928 381.173 5512.271 40.992
4.665 403.893 5424.308 34.733
3.411 432.610 5283.961 27.925
2.654 451.199 5216.193 22.961
2.135 461.401 5180.472 19.014
1.753 469.589 5131.040 16.047
1.433 479.036 5116.212 13.414
1.167 486.446 5115.879 11.098
Appendix D
Experimental Instrumentation
A schematic representation of the instrumentation that was used during the
experiments can be seen in Figure D.1.
Figure D.1: Schematic layout of experimental instrumentation.
The procedure listed below was followed each time an experimental test was
carried out:
89
APPENDIX D. EXPERIMENTAL INSTRUMENTATION 90
1. The bridge ampli er is switched on and allowed to warm up for at least 30
minutes. While in the control room the frequency counter and personal
computer is switched on. At this time the auxiliary fan and hydraulic
motor's power supply is connected and the tap is turned on to supply
cooling water to the hydraulic motor.
2. The fan test facility is inspected for anything that might seem out of
place. Special care is taken to ensure that the fan can rotate freely and
that both hatch doors of the settling chamber are closed tightly. The
throttling device is placed back to its fully open position.
3. Ambient conditions are measured (see Section D.1). The LabVIEW pro-
gram is started on the personal computer and zero starting readings are
taken, to be used in calculating the drift when the data is processed. The
drift calculations is performed following the method proposed by Stinnes
(1998).
4. The auxiliary fan is switched on. Hereafter the main hydraulic motor is
switched on and left to idle and warm up.
5. The rotational speed of the fan is slowly increased to the test speed
by monitoring the frequency counter and keeping a careful eye on the
rotating fan for any abnormalities. Once a satisfactory steady state has
been reached at the test speed, a data point is captured in the LabView
program.
6. The throttling device is then closed by one notch. Procedure 5 is repeated
until the throttling device is completely closed.
7. A data point is taken with the throttling device at the fully closed po-
sition and the auxiliary fan switched o to investigate the pressure dif-
ference at an even lower volumetric ow rate than that of the closed
position.
8. The oil supply to the hydraulic motor is gradually reduced until an idle
state is achieved and the motor is switched o. The ambient pressure
and temperature are recorded again.
9. In this time both the fans would have come to a standstill and the end
zero values are recorded for drift calculation. All the data captured is
saved in a text le on the computer.
10. All the instrumentation inside the control room is switched o and the
power leads for the auxiliary fan and hydraulic motor is removed and
stored away. The cooling water tap is closed.
APPENDIX D. EXPERIMENTAL INSTRUMENTATION 91
D.1 Ambient Conditions
Ambient pressure was measured with a mercury column barometer situated in
a laboratory room close to the test facility. The barometer is calibrated by the
manufacturer to account for the local gravitational force and provide readings
in hPa.
Ambient temperature was measured with a thermometer located next to the
test facility in the shade. A substantial dierence was noted between the
alcohol column thermometer attached to the barometer inside the nearby lab-
oratory and the thermometer hanging outside next to the facility. The mer-
cury column thermometer measured higher temperatures and uctuated only
slightly during the course of the day. After carrying out numerous experi-
ments during dierent times of the day it was established that testing later
than an hour after sunrise provided big uctuations in torque readings, in u-
encing repeatability. This variation may be the result of the test facility being
heated non-uniformly by direct sunlight and re ection from the windows of
the adjacent buildings as well as part of the facility being under a roof while
the rest is exposed. Stinnes (1998) also noted an indeterminable variation in
a comparison of temperature measurements in dierent locations around the
test facility. For the best test results, experiments were carried out two hours
before sunrise when the atmosphere had cooled down to a constant tempera-
ture with wind still conditions. Temperature readings were taken before and
after each test run and averaged.
D.2 Pressure
To acquire the static pressure from the inlet bellmouth and settling cham-
ber two Höttinger PD1 inductive dierential pressure transducers were used.
These pressure transducers have a range of −1000 Pa to 1000 Pa.
Stinnes (1998) found Venter (1990)'s placement of the transducer adjacent to
the bellmouth unfavourable and moved the transducers into the control room.
The two pressure transducers were left inside the control room and kept in
the same orientation when calibrated. All the plastic tubing connecting the
transducers to the test facility was tested for leaks and was found air-tight.
D.3 Torque
To measure the torque a Höttinger T2 (resistive full bridge strain gauge type)
torque transducer is used. The torque transducer has a nominal range of
−500 Nm to 500 Nm.
APPENDIX D. EXPERIMENTAL INSTRUMENTATION 92
The torque transducer is connected to the hydraulic motor and the fan shaft by
two rubber couplings to account for vibration and misalignment. Two bearing
blocks support the transducer in-between the rubber couplings. The friction
in these couplings and bearings are described in Section E.2 and accounted for
in the calibration of the torque measurements.
D.4 Speed
A magnetic speed pick-up sensor below the fan shaft is used to obtain the rota-
tional speed of the fan by giving an input signal to the dual output frequency
counter. The rst output is to a digital display that is monitored to regulate
the rotational speed by adjusting the supply of hydraulic oil to the fan motor.
The second output is a linear voltage signal fed into the National Instruments
card.
Calibration of all the instrumentation is discussed in Appendix E.
D.5 Bridge ampli er
A Höttinger KWS 3073 bridge ampli er is used to amplify the input signals
from the transducers before the voltages are sent to the personal computer.
The input signals are ampli ed over a range of −10 V to 10 V.
All signals from the bridge ampli er as well as that of the dual output fre-
quency counter are fed to a National Instruments NI USB-6210 bus-powered
multifunction data acquisition module. The signals are then sent to the per-
sonal computer via its USB port where a LabVIEW program is used to capture
the data.
D.6 Blade angle setting jig
The blade stagger angle was set by a custom made jig that provided a at
platform for the SmartTool digital projector. The pressure side blade pro le
was laser cut into two stainless steel plates that were attached to either side
of a plastic block as illustrated in Figure D.2. To determine the oset in angle
from the chord line to the line tangent of the pressure surface, a blade was
laid at on a marble table with the edges supported with Presstick. The chord
line was set horizontal by positioning the leading edge and trailing edge points
at the same height as can be seen in Figures D.3 and D.4. The dierence
between the blade setting angle and the stagger angle was measured to be
4.2°, as can be seen in Figure D.5. To therefore set the blades at a stagger
angle of 59° a blade setting angle of 63.2° was used. To keep the blade at a
APPENDIX D. EXPERIMENTAL INSTRUMENTATION 93
horizontal level while the angle was being set, the shaft was locked in position
by a clamping device designed by a fellow student. The clamping device is
illustrated in Figure D.6.
Figure D.2: Blade stagger angle setting jig.
Figure D.3: Stagger angle jig chord leading edge.
APPENDIX D. EXPERIMENTAL INSTRUMENTATION 94
Figure D.4: Stagger angle jig chord trailing edge.
Figure D.5: Dierence between stagger angle and blade setting angle.
APPENDIX D. EXPERIMENTAL INSTRUMENTATION 95
Figure D.6: Device for clamping rotor shaft.
Appendix E
Calibration of instrumentation
E.1 Pressure Calibration
The two pressure transducers were calibrated by using a Betz water column
manometer. The negative pressure tappings of the pressure transducers were
connected by means of a pipe system and a ball-valve to the negative column of
the Betz manometer. Figure E.1 gives an illustration of the pressure calibration
setup. Before the calibration was started the pressure transducers were stuck
unto the table with Presstick. This prevented the pressure transducers from
moving during calibration and it was left as is to keep the transducers in the
same position during experiments as they were during the calibration process.
The collected data was fed into a Microsoft Excel
® spreadsheet to draw up
a graph of pressure versus voltage distribution. A linear trend line was tted
to the curve and the equation from the graph was used to correct the pres-
®
sure readings obtained from experimental work. The calibration data can be
viewed in Table E.1 with the corresponding graphs from the Microsoft Excel
spreadsheet in Figure E.2 and Figure E.3.
96
APPENDIX E. CALIBRATION OF INSTRUMENTATION 97
Figure E.1: Pressure calibration setup.
Table E.1: Calibration of pressure transducers
(a) Settling chamber pressure transducer (b) Bellmouth pressure transducer
Channel 1 Channel 4
Water level Pressure Voltage Water level Pressure Voltage
mm Pa V mm Pa V
0 0.000 0.000 0 0.000 0.000
10 97.764 1.023 10 97.764 1.011
20 195.528 2.042 20 195.528 2.021
30 293.292 3.074 30 293.292 3.046
40 391.056 4.098 40 391.056 4.064
50 488.820 5.127 50 488.820 5.089
60 586.584 6.137 60 586.584 6.097
70 684.349 7.152 70 684.349 7.111
APPENDIX E. CALIBRATION OF INSTRUMENTATION 98
800
700
600
Pressure [Pa]
500
400
300
200
Calibration pressure
100
Linear (Calibration pressure)
P = 96.137V + 0.45
0
0 1 2 3 4 5 6 7 8
Voltage [V]
Figure E.2: Pressure calibration bellmouth.
800
700
600
Pressure [Pa]
500
400
300
200
Calibration pressure
100
Linear (Calibration pressure)
P = 95.584V - 0.149
0
0 1 2 3 4 5 6 7 8
Voltage [V]
Figure E.3: Pressure calibration settling chamber.
APPENDIX E. CALIBRATION OF INSTRUMENTATION 99
E.2 Torque Calibration
The torque calibration was done by using a cantilever beam. To use the
cantilever the fan and hub rst need to be removed. Hereafter the shaft on
the motor side of the torque transducer is locked in place. This enables the
transducer to only measure the torque due to the de ection of the cantilever
beam. The shaft is locked in place when the cantilever beam is at a horizontal
level. This level is measured with a digital projector as shown in Figure E.4.
Weights weighing 5.020 kg, 5.025 kg, 5.015 kg and 4.975 kg respectively were
placed on the holding disk of the cantilever beam. Each time a weight is
placed on the disk the shaft is rotated back to its horizontal position to account
for the deformation of the shaft, torque transducer and the couplings. The
corresponding voltages were taken from the bridge ampli er and recorded.
Figure E.4: Torque calibration horizontal level measurement.
The collected data was fed into a Microsoft Excel
® spreadsheet to draw up a
graph of torque versus voltage distribution. A linear trend line was tted to
the curve and the equation from the graph was used to calculate the torque
® spread-
readings obtained from experimental work. The calibration data can be viewed
in Table E.2 with the corresponding graphs from the Microsoft Excel
sheet in Figure E.5.
A no-load torque test was also carried out to establish the torque eect of
the shaft bearings and any misalignment. This was done by stripping the fan
blades from the hub and running the test at the design speed with the at
APPENDIX E. CALIBRATION OF INSTRUMENTATION 100
Table E.2: Torque calibration data
Reading Mass [kg] Voltage [V] Torque [Nm]
1 0 0 0
2 5.020 0.158 25.276
3 10.045 0.297 50.578
4 15.060 0.450 75.829
5 20.035 0.602 100.879
plate attached to the hub and without. The results of the no-load test can
be seen in Table E.3. Start and end zero values were taken before the hub
was brought up to design speed. These values were used to eliminate the drift
from the bridge ampli er and produce the corrected voltage as can be seen
in Table E.3. These corrected voltages were then adjusted with the values
obtained from the calibration data. Since the fan will be tested with the at
plate installed, only the data from Table E.3b is taken into account. It can
be seen that the third reading diers from the rst two (which are identical)
and it is therefore disregarded. Bruneau (1994) could also not nd the cause
for the variations in the no-load torques and subtracted a constant torque
value of 2 Nm from all the fan performance torque readings. This value agrees
favourably with the last two values obtained in Table E.3b. Stinnes (1998)
110
100
90
80
70
Torque [Nm]
60
50
40
30
20
Calibration torque
10 Linear (Calibration torque)
M = 27.325V - 0.026
0
0 1 2 3 4
Voltage [V]
Figure E.5: Torque calibration curve.
APPENDIX E. CALIBRATION OF INSTRUMENTATION 101
also subtracted a constant torque value of 2 Nm.
Table E.3: No-load torque test without and with at plate
(a) No-load torque test without at plate (b) No-load torque test with at plate
No-load test without at plate No-load test with at plate
Reading Voltage Corrected Torque Reading Voltage Corrected Torque
[V] Voltage [V] [Nm] [V] Voltage [V] [Nm]
1 -0.003 1 -0.000
2 0.011 0.011 1.817 2 0.009 0.009 1.477
3 -0.002 3 -0.001
1 -0.002 1 0.000
2 0.011 0.011 1.817 2 0.012 0.012 1.987
3 -0.002 3 0.000
1 -0.001 1 0.001
2 0.013 0.013 2.157 2 0.012 0.012 1.987
3 -0.002 3 0.002
E.3 Speed Calibration
To calibrate the speed indicator the motor was run at 100 RPM intervals and
the actual speed was measured inside the plenum chamber with an Ono Sokki
HT-341 digital tachometer. The tachometer was held against the rotating shaft
until a constant reading was obtained. The speed indicator- and tachometer
reading were recorded.
The collected data was fed into a Microsoft Excel
® spreadsheet to draw up
a graph of speed (from the speed indication box and the tachometer) versus
voltage distribution. A linear trend line was tted to the curve and the equa-
tion from the graph was used to correct the speed readings obtained from
® spreadsheet in Figure E.6.
experimental work. The calibration data can be viewed in Table E.4 with the
corresponding graphs from the Microsoft Excel
APPENDIX E. CALIBRATION OF INSTRUMENTATION 102
Table E.4: Speed calibration data
Reading Speed [rpm] Voltage [V]
1 0 0
2 127 1.541
3 169 1.999
4 255 3.024
5 343 4.045
6 425 5.018
7 510 6.018
8 596 7.038
9 680 8.031
10 765 9.030
900
800
700
600
Velocity [RPM]
500
400
300
200
Calibration Speed
100
Linear (Calibration Speed)
RPM = 84.893V - 0.375
0
0 1 2 3 4 5 6 7 8 9 10
Voltage [V]
Figure E.6: Speed tachometer calibration curve.
Appendix F
Fan performance characteristics
In this appendix the fan performance characteristics are given for the three
blade angles, namely 58°, 59° and 60°, without the data from Stinnes to reduce
the clutter. For all three stagger angles the experimental results are compared
to the data from Stinnes (1998) and Bruneau (1994). Finally the fan perfor-
mance characteristics for the tests performed at a tip clearance of 1.5 mm are
shown.
550
58 deg
500 59 deg
60 deg
450
Fan Static Pressure [Pa]
400
350
300
250
200
150
100
50
0
0 4 8 12 16 20 24
Volumetric Flow Rate [m3/s]
Figure F.1: Fan static pressure rise at three blade angles.
103
APPENDIX F. FAN PERFORMANCE CHARACTERISTICS 104
8000
58 deg
59 deg
7000
60 deg
6000
5000
Power [W]
4000
3000
2000
1000
0
0 4 8 12 16 20 24
Volumetric Flow Rate [m3/s]
Figure F.2: Fan power at three blade angles.
60
58 deg
59 deg
50 60 deg
Fan Static Effeciency [%]
40
30
20
10
0
0 4 8 12 16 20 24
Volumetric Flow Rate [m3/s]
Figure F.3: Fan static e ciency at three blade angles.
APPENDIX F. FAN PERFORMANCE CHARACTERISTICS 105
500
Experimental
450 Stinnes
Bruneau
400
Fan Static Pressure [Pa]
350
300
250
200
BS 848: 1980: Type A
150
B-fan: 8-bladed
Fan casing diameter: 1.542 m
100 Tip clearance: 3 mm
Blade root stagger angle: 59°
50 Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
3
Volumetric Flow Rate [m /s]
Figure F.4: Fan static pressure rise, comparison at 59°.
7000
Experimental
Stinnes
6000
Bruneau
5000
Fan Power [W]
4000
3000
BS 848: 1980: Type A
2000
B-fan: 8-bladed
Fan casing diameter: 1.542 m
Tip clearance: 3 mm
1000 Blade root stagger angle: 59°
Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
Volumetric Flow Rate [m3/s]
Figure F.5: Fan power, comparison at 59°.
APPENDIX F. FAN PERFORMANCE CHARACTERISTICS 106
70
Experimental
Stinnes
60
Bruneau
Fan static effeciency [%]
50
40
30
BS 848: 1980: Type A
B-fan: 8-bladed
20
Fan casing diameter: 1.542 m
Tip clearance: 3 mm
10
Blade root stagger angle: 59°
Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
3
Volumetric Flow Rate [m /s]
Figure F.6: Fan static e ciency, comparison at 59°.
500
Experimental
450 Stinnes
Bruneau
400
Fan Static Pressure [Pa]
350
300
250
200
BS 848: 1980: Type A
150
B-fan: 8-bladed
Fan casing diameter: 1.542 m
100 Tip clearance: 3 mm
Blade root stagger angle: 58°
50 Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
Volumetric Flow Rate [m3/s]
Figure F.7: Fan static pressure rise, comparison at 58°.
APPENDIX F. FAN PERFORMANCE CHARACTERISTICS 107
8000
Experimental
Stinnes
7000
Bruneau
6000
Fan Power [W]
5000
4000
3000 BS 848: 1980: Type A
B-fan: 8-bladed
2000 Fan casing diameter: 1.542 m
Tip clearance: 3 mm
Blade root stagger angle: 58°
1000
Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
3
Volumetric Flow Rate [m /s]
Figure F.8: Fan power, comparison at 58°.
70
Experimental
Stinnes
60
Bruneau
Fan static effeciency [%]
50
40
30
BS 848: 1980: Type A
B-fan: 8-bladed
20
Fan casing diameter: 1.542 m
Tip clearance: 3 mm
10
Blade root stagger angle: 58°
Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
Volumetric Flow Rate [m3/s]
Figure F.9: Fan static e ciency, comparison at 58°.
APPENDIX F. FAN PERFORMANCE CHARACTERISTICS 108
500
Experimental
450 Stinnes
Bruneau
400
Fan Static Pressure [Pa]
350
300
250
200
BS 848: 1980: Type A
150
B-fan: 8-bladed
Fan casing diameter: 1.542 m
100
Tip clearance: 3 mm
Blade root stagger angle: 60°
50 Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
3
Volumetric Flow Rate [m /s]
Figure F.10: Fan static pressure rise, comparison at 60°.
7000
Experimental
Stinnes
6000
Bruneau
5000
Fan Power [W]
4000
3000
BS 848: 1980: Type A
B-fan: 8-bladed
2000 Fan casing diameter: 1.542 m
Tip clearance: 3 mm
Blade root stagger angle: 60°
1000
Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
Volumetric Flow Rate [m3/s]
Figure F.11: Fan power, comparison at 60°.
APPENDIX F. FAN PERFORMANCE CHARACTERISTICS 109
70
Experimental
Stinnes
60
Bruneau
Fan static effeciency [%]
50
40
30
BS 848: 1980: Type A
B-fan: 8-bladed
20
Fan casing diameter: 1.542 m
Tip clearance: 3 mm
10
Blade root stagger angle: 60°
Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
3
Volumetric Flow Rate [m /s]
Figure F.12: Fan static e ciency, comparison at 60°.
500
Experimental
450 Stinnes
400
Fan Static Pressure [Pa]
350
300
250
200
BS 848: 1980: Type A
150
B-fan: 8-bladed
Fan casing diameter: 1.542 m
100 Tip clearance: 1.5 mm
Blade root stagger angle: 59°
50 Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
Volumetric Flow Rate [m3/s]
Figure F.13: Fan static pressure rise, comparison at 1.5 mm tip clearance.
APPENDIX F. FAN PERFORMANCE CHARACTERISTICS 110
7000
Experimental
Stinnes
6000
5000
Fan Power [W]
4000
3000
BS 848: 1980: Type A
B-fan: 8-bladed
2000
Fan casing diameter: 1.542 m
Tip clearance: 1.5 mm
1000
Blade root stagger angle: 59°
Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
3
Volumetric Flow Rate [m /s]
Figure F.14: Fan power, comparison at 1.5 mm tip clearance.
70
Experimental
Stinnes
60
Fan static effeciency [%]
50
40
30
BS 848: 1980: Type A
B-fan: 8-bladed
20
Fan casing diameter: 1.542 m
Tip clearance: 1.5 mm
10 Blade root stagger angle: 59°
Density: 1.2 kg/m3
Rotational speed: 750 rpm
0
0 4 8 12 16 20 24
Volumetric Flow Rate [m3/s]
Figure F.15: Fan static e ciency, comparison at 1.5 mm tip clearance.
Appendix G
Numerical Simulation Data
The simulation data of the two CFD models are presented. The rotor-stator
interface mentioned is only used to connect the blade section with the inlet-
and outlet sections. All blocks are rotated as described in Section 3.4.3.
Table G.1: Numerical data for Approach-1
Stagger angle 59°
Tip clearance 3 mm
Fluid model Perfect gas
rotor-stator interface MPA
Outlet Pinched
Velocity components solved in relative reference frame Relative
Mass ow rate Static outlet Total inlet ∆P Fan Power Fan Static
e ciency
3
[m /s] [Pa] [Pa] [Pa] [W] [%]
2 101115 100585.42 529.580 4705.500 22.509
6 101102 100629.42 472.580 5458.400 51.947
10 101156 100767.71 388.290 6223.200 62.394
16 101228 101019.05 208.950 5751.400 58.128
20 101259 101186.11 72.890 4460.500 32.682
111
APPENDIX G. NUMERICAL SIMULATION DATA 112
Table G.2: Numerical data for Approach-2
Stagger angle 59°
Tip clearance 3 mm
Fluid model Perfect gas
rotor-stator interface MPA
Outlet Open
Velocity components solved in relative reference frame Absolute
Mass ow rate Static outlet Total inlet ∆P Fan Power Fan Static
e ciency
3
[m /s] [Pa] [Pa] [Pa] [W] [%]
8 101325 101035.75 289.250 5128.400 45.121
10 101324.98 101018.32 306.660 5197.800 58.998
16 101324.96 101114.57 210.390 5472.600 61.511
20 101324.98 101261.71 63.270 4116.200 30.742
Appendix H
Streamline distribution diagrams
In this appendix the streamline distribution for the remaining ow rates are
provided. This includes all operating points simulated for both Approach-1
and Approach-2.
Figure H.1: Streamline distribution, 8 m3 /s, Approach-2.
113
APPENDIX H. STREAMLINE DISTRIBUTION DIAGRAMS 114
Figure H.2: Streamline distribution, 20 m3 /s, Approach-2.
Figure H.3: Streamline distribution, 2 m3 /s, Approach-1, relative velocity so-
lution.
APPENDIX H. STREAMLINE DISTRIBUTION DIAGRAMS 115
Figure H.4: Streamline distribution, 6 m3 /s, Approach-1, relative velocity so-
lution.
Figure H.5: Streamline distribution, 10 m3 /s, Approach-1, relative velocity
solution.
APPENDIX H. STREAMLINE DISTRIBUTION DIAGRAMS 116
Figure H.6: Streamline distribution, 20 m3 /s, Approach-1, relative velocity
solution.
List of References
Anderson, J. (1990). Modern compressible ow: with historical perspective. 2nd edn.
McRaw-Hill, Inc.
Ashford, G. and Powell, K. (1996). An unstructured grid generation and adaptive so-
lution technique for high-reynolds-number compressible ows. 27th Computational
Fluid Dynamics, vol. VKI LS 1996-06, pp. 184.
Bredell, J. (2005). Numerical investigation of fan performance in a forced draft air-
cooled steam condenser. Master's thesis, Department of Mechanical Engineering,
University of Stellenbosch.
Bruneau, P. (1994). The design of a single rotor axial ow fan for a cooling tower
application. Master's thesis, Department of Mechanical Engineering, University of
Stellenbosch.
Dippenaar, D. (2007). Die sturkturele ontwerp en vervaardiging van `n nuwe ver-
wysingswaaier vir die groot waaier-toetstonnel. Master's thesis, Department of
Mechanical Engineering, University of Stellenbosch.
Esterhuyze, O. (2006). The testing and evaluation of the performance of axial fans
subjected to variable inlet conditions.
Master's thesis, Department of Mechanical
Engineering, University of Stellenbosch.
Kelecy, F. (2000). Study demonstrates that simulation can accurately predict fan
performance. Journal Articles by Fluent Sofware Users, vol. JA108, pp. 14.
Kröger, D. (1998). Air-cooled heat exchangers and cooling towers Thermal- ow per-
formance evaluation and design. Department of Mechanical Engineering Univer-
sity of Stellenbosch.
Lötstedt, P. (1992). Propeller slip-stream model in subsonic linearized potential ow.
Journal of Aircraft, vol. 29 (6), pp. 10981105.
Lötstedt, P. (1995). Accuracy of a propeller model in inviscid ow. Journal of
Aircraft, vol. 32(6), pp. 13121321.
McGhee, R., Beasley, W. and Somers, D. (1977). Low-speed aerodynamic charac-
teristics of a 13-percent-thick airfoil section. NASA Technical Memorandum, vol.
NASA TM X-72679.
117
LIST OF REFERENCES 118
McGhee, R., Beasley, W. and Whitcomb, R. (1979). Nasa low- and medium-speed
airfoil development. NASA Technical Memorandu, vol. NASA-TM-78709.
Meyer, C. (1996). Plenum losses in forced draught air-cooled heat exchangers. Mas-
ter's thesis, Department of Mechanical Engineering, University of Stellenbosch.
Meyer, C. (2000). A numerical investigation of the plenum chamber aerodynamic
behaviour of mechanical draught air-cooled heat exchangers.Ph.D. thesis, Depart-
ment of Mechanical Engineering, University of Stellenbosch.
Ohad, G. and Aviv, R. (2005). Propeller performance at low advance ratio. Journal
of aircraft, vol. 42-2.
Pelletier, D., Garon, A. and Camarero, R. (1991). Finite element method for com-
puting turbulent propeller ow. AIAA Journal, vol. 29(1), pp. 6875.
Pelletier, D. and Schetz, J. (1986). Finite element navier-stokes calculation of three-
dimensional turbulent ow near a propeller. AIAA Journal, vol. 24(9), pp. 1409
1416.
Pericleous, K. and Patel, M. (1986). The modelling of tangential and axial agitators
in chemical reactors. PhysicoChemical Hydrodynamics, vol. 8(2), pp. 105123.
Ramasubramanian, M., Shi er, D. and Jayachandran, A. (2008). A computational
uid dynamics modeling and experimental study of the mixing process for the
dispersion of the synthetic bers in wet-lay forming. Journal of Engineered Fibers
and Fabrics, vol. 3, Issue 1.
Schetz, J., Pelletier, D. and Mallory, D. (1988). Experimental and numerical inves-
tigation of a propeller with three-dimensional in ow. Journal of Propulsion, vol.
4(4), pp. 341349.
Sezer-Uzol, N. and Long, L. (2006). 3d time-accurate cfd simulations of wind turbine
rotor ow elds. AIAA 2006-0394.
Spalart, P. and Allmaras, S. (1994). A one-equation turbulence model for aerody-
namic ows. La Recherche Aérospace, vol. 1, pp. 521.
Stinnes, W. (1998). The performance of axial fans subjected to forced cross- ow
at Inlet. Master's thesis, Department of Mechanical Engineering, University of
Stellenbosch.
Thiart, G. and von Backström, T. (1993). Numerical simulation of the ow eld near
an axial ow fan operating under distorted in ow conditions. Journal of Wind
Engineering and Industrial Aerodynamics, vol. 45, pp. 189214.
Turkel, E., Radespiel, R. and Kroll, N. (1997). Assessment of preconditioning meth-
ods for mutidimensional aerodynamics. Computers and Fluids, vol. 26, No. 6, pp.
613634.
LIST OF REFERENCES 119
Unknown Author (1980). BS 848: Part 1: 1980, Fans for General Puposes, Part 1:
Methods for testing performance. British Standards Institution.
Unknown Author (2000). The basics of axial ow fans. Hudson Products Corpora-
tion.
Unknown Author (2007a ). User Manual AutoGrid v8. NUMECA International, 5,
Avenue Franklin Roosevelt 1050 Brussels Belgium.
Unknown Author (2007b ). User Manual FINE/Turbo v8 (including Euranus). NU-
MECA International, 5, Avenue Franklin Roosevelt 1050 Brussels Belgium.
Unknown Author (2007c ). User Manual IGG v8. NUMECA International, 5,
Avenue Franklin Roosevelt 1050 Brussels Belgium.
Van der Spuy, S., Von Backström, T. and Kröger, D. (2009). An evaluation of simpli-
ed methods for modelling the performance of axial ow fan arrays. Unpublished
Journal.
Van Staden, M. (1996). Integrated approach to cfd modelling of air-cooled con-
densers. In: 1st South African Conference on Applied Mechanics (SACAM).
Venter, S. (1990). The eectiveness of axial ow fans in A-frame plenums. Ph.D.
thesis, Department of Mechanical Engineering, University of Stellenbosch.
Venter, S. and Kröger, D. (1992). The eect of tip clearance on the performance of
an axial ow fan. Energy Convers. Mgmt.
Vigneron, D., Vaassen, J. and Essers, J. (2008). An implicit nite volume method for
the solution of 3d low mach. Journal of Computational and Applied Mathematics,
vol. 215 (2008), pp. 610617.
White, F. (2006). Viscous Fluid Flow. McRaw-Hill, Inc.
Zhonghua, H. and, F.H., Wenping, S. and Zhide, Q. (2007). A preconditioned multi-
grid method for e cient simulation of three-dimensional compressible and incom-
pressible ows. Chinese Journal of Aeronautics, vol. 20(2007), pp. 289296.
You might also like
- COMOROS J1 - EXIM Presentation - With BHEL Additions Rev 00No ratings yetCOMOROS J1 - EXIM Presentation - With BHEL Additions Rev 0021 pages
- Advanced CFD Methods For Wind Turbine AnalysisNo ratings yetAdvanced CFD Methods For Wind Turbine Analysis129 pages
- Johnson Ben Final E-Thesis (Master Copy)No ratings yetJohnson Ben Final E-Thesis (Master Copy)228 pages
- Flow Around Bluff Bodies With Corner Modifications On Cross-SectiNo ratings yetFlow Around Bluff Bodies With Corner Modifications On Cross-Secti104 pages
- Theoretical Study Aeroacoustics Axial Flow FanNo ratings yetTheoretical Study Aeroacoustics Axial Flow Fan91 pages
- AE651 Project Report - 210010008, 22b0060 & 22b0076No ratings yetAE651 Project Report - 210010008, 22b0060 & 22b007618 pages
- 14 - Cooling Fan Optimization For Heavy Electric VehiclesNo ratings yet14 - Cooling Fan Optimization For Heavy Electric Vehicles93 pages
- Predicting Flow Dynamics of An Entire Engine Cooling System Using 3D CFDNo ratings yetPredicting Flow Dynamics of An Entire Engine Cooling System Using 3D CFD66 pages
- Investigation of 2D Airfoils Equipped With A Trailing Edge Flaps MSC in Wind EnergyNo ratings yetInvestigation of 2D Airfoils Equipped With A Trailing Edge Flaps MSC in Wind Energy61 pages
- Design Improvement of Radial Flow Submersible Pump Using CFDNo ratings yetDesign Improvement of Radial Flow Submersible Pump Using CFD61 pages
- Flow and Heat Transfer Dissertation Biegger Online Opt 450No ratings yetFlow and Heat Transfer Dissertation Biegger Online Opt 450221 pages
- The Numerical Investigation of Outlet Guide Vane Heat TransferNo ratings yetThe Numerical Investigation of Outlet Guide Vane Heat Transfer78 pages
- Aeroelastic Analysis and Design of The Turkispain Wind TurbineNo ratings yetAeroelastic Analysis and Design of The Turkispain Wind Turbine62 pages
- (Esch B.P.M. Van, (1997) ) Simulation of Three-Dime PDFNo ratings yet(Esch B.P.M. Van, (1997) ) Simulation of Three-Dime PDF152 pages
- Effects of Surface Roughness and Vortex Generators On The NACA 4415 AirfoilNo ratings yetEffects of Surface Roughness and Vortex Generators On The NACA 4415 Airfoil103 pages
- Engineering Internship Report Houssem REKIKNo ratings yetEngineering Internship Report Houssem REKIK59 pages
- Unsteady Vortex Lattice Method of Moving AircraftsNo ratings yetUnsteady Vortex Lattice Method of Moving Aircrafts70 pages
- CFD Analyses and Performance Comparison of Micro-Hydropowder PlantsNo ratings yetCFD Analyses and Performance Comparison of Micro-Hydropowder Plants109 pages
- Aerodynamic CFD Analysis On Experimental Airplane: Mechanical Engineering 'Fluid Dynamics' Karel Lammers (s1005138)No ratings yetAerodynamic CFD Analysis On Experimental Airplane: Mechanical Engineering 'Fluid Dynamics' Karel Lammers (s1005138)34 pages
- Time-dependent Behaviour and Design of Composite Steel-concrete StructuresFrom EverandTime-dependent Behaviour and Design of Composite Steel-concrete StructuresNo ratings yet
- Sources Renewable Non-Renewable Energy PhilippinesNo ratings yetSources Renewable Non-Renewable Energy Philippines11 pages
- Performance Record: Samsung SEI Tower 467-14 Dogok 2 Dong, Gangnam Gu, Seoul, Rep. of Korea 135-856 Tel. 82.2.3458.3000No ratings yetPerformance Record: Samsung SEI Tower 467-14 Dogok 2 Dong, Gangnam Gu, Seoul, Rep. of Korea 135-856 Tel. 82.2.3458.300022 pages
- General Project Specifications and Requirements 40-0000520 - C - 003 PDFNo ratings yetGeneral Project Specifications and Requirements 40-0000520 - C - 003 PDF29 pages
- International Journal of Electrical Power and Energy SystemsNo ratings yetInternational Journal of Electrical Power and Energy Systems8 pages
- Documents - MX Dissertation Report On PERFORMANCE APPRAISAL BhelNo ratings yetDocuments - MX Dissertation Report On PERFORMANCE APPRAISAL Bhel68 pages
- SC13003-C9001-U10-0002-Commissioning Plan Procedure For MBOP - 24-08-2015 PDFNo ratings yetSC13003-C9001-U10-0002-Commissioning Plan Procedure For MBOP - 24-08-2015 PDF70 pages
- Introduction - : Definition of DevelopmentNo ratings yetIntroduction - : Definition of Development16 pages
- Control & Automation Aspects FOR Reliability of Super-Critical UnitsNo ratings yetControl & Automation Aspects FOR Reliability of Super-Critical Units21 pages