Pre-swirl rotor-stator systems:

Flow and heat transfer

Paul Lewis

A thesis submitted for the degree of Doctor of Philosophy

University of Bath

Department of Mechanical Engineering

July 2008

COPYRIGHT

Attention is drawn to the fact that the copyright of this thesis rests with its author. This copy
of the thesis has been supplied on the condition that anyone who consults it is understood to
recognise that its copyright rests with its author and that no quotation from the thesis and no
information derived from it may be published without the prior written consent of the author.

This thesis may be made available for consultation within the University Library and may be
photocopied or lent to other libraries for the purposes of consultation.

. . . . . . . . . . . .

Abstract

This thesis compares previously published heat transfer measurements from a pre-swirl
rotor-stator experiment with 3D steady state results from a commercial CFD code. The
experimental distribution of Nusselt number on the rotor surface was obtained from a previous
study using a scaled model of a gas turbine rotor-stator cavity, where the flow structure was
representative of that found in the engine. Computations were carried out using a coupled
multigrid RANS solver with a high-Reynolds-number k-ω turbulence model. Previous work
has identified three flow parameters governing heat transfer: rotational Reynolds number (Reφ ),
pre-swirl ratio (βp ) and the turbulent flow parameter (λT ). In addition, geometric parameters
defining the size and shape of the wheelspace affect the flow structure and therefore the heat
transfer characteristic; pre-swirl inlet to receiver hole radius ratio (rp /rb ), gap ratio between the
discs (G) hub to shroud ratio (a/b), number and size of pre-swirl nozzles and receiver holes.
For this study the rotational Reynolds numbers are in the range 0.8 ∗ 106 < Reφ < 1.2 ∗ 106 .
The turbulent flow parameter and pre-swirl ratios varied between 0.12 < λT < 0.38 and
0.5 < βp < 2.0, which are comparable to values that occur in industrial gas turbines.

At high coolant flow rates, the experiments reveal a peak in heat transfer at the radius of the
pre-swirl nozzles, associated with the impingement of the pre-swirl flow. At lower coolant flow
rates, the heat transfer is dominated by boundary-layer effects. The Nusselt number on the
rotating disc increases as either Reφ or λT increases, and is axisymmetric except in the region
of the receiver holes, where significant two-dimensional variations are observed.

The computed velocity field is used to explain the heat transfer distributions observed in the
experiments. The regions of peak heat transfer around the receiver holes are a consequence of
the route taken by the flow. Two routes have been identified: ‘direct’, whereby flow forms a
stream-tube between the inlet and outlet; and ‘indirect’, whereby flow mixes with the rotating
core of fluid. Both of these regimes are present throughout the spectrum of flow conditions.

Two performance parameters have been calculated: the adiabatic effectiveness for the system
(Θb,ad ) and the discharge coefficient for the receiver holes (Cd,b ). Θb,ad increases as βp and λT
increase, however an optimum value of βp is found to maximise Cd,b and hence optimise the
delivery of coolant to the turbine blades.

The effect of the radial location of the inlet nozzles on the performance of the direct-transfer
preswirl system is investigated by performing computations for three inlet-to-outlet radius
ratios, rp /rb = 0.8, 0.9 and 1.0, a range of pre-swirl ratios, 0.5 < βb < 2.0, and varying

ii

flow parameter, 0.12 < λT < 0.36. The rotational Reynolds number for each case is 106 .

The flow structure in the wheel-space and in the region around the receiver holes for each inlet
radius is related to the swirl ratio. The performance of the system is again quantified by the
discharge coefficient for the receiver holes (Cd,b ) and the adiabatic effectiveness for the system
(Θb,ad ).

For each radius ratio the discharge coefficient is found to reach a maximum when the rotating
core of fluid is in synchronous rotation with the receiver holes. As the radius ratio is increased
this condition can be achieved with a smaller value for pre-swirl ratio βb . A simple model
is presented to estimate the discharge coefficient based on the flow rate and swirl ratio in the
system.

The adiabatic effectiveness of the system increases linearly with pre-swirl ratio but is
independent of flow rate. For a given pre-swirl ratio, the effectiveness increases as the radius
ratio increases. Computed values show good agreement with analytical results.

Both performance parameters show improvement with increasing inlet radius ratio, consistent
with earlier research, confirming that for an optimum pre-swirl configuration an engine designer
would place the pre-swirl nozzles at a high radius.

A pilot study is described in the appendix, investigating the fluid dynamics of a rotor-stator
system subject to ingress. Computations are made of a simple wheel-space with an axial
clearance rim seal with non-axisymmetric flow conditions created using a stator vane in an
external mainstream. The sealing effectiveness of the rim seal is calculated from computed
levels of concentration of a tracer scalar variable. Computations are found to be highly grid
sensitive, requiring a large degree of resolution in the region of the seal. Consideration is given
to the design implications of experimental apparatus to measure the ingress phenomenon.

iii

 To My Parents . . .

iv
Acknowledgments

The project was supported by a studentship from the Engineering Innovative Manufacturing
Research Centre at the University of Bath, which is funded by the UK Engineering and Physical
Sciences Research Council; Doctoral Training Account EP/P500036/1. The research into
ingress in gas turbines has been stimulated and supported by collaboration with Mitsubishi
Heavy Industries, Japan.

Thanks to

• My supervisors Dr Mike Wilson, Dr Gary Lock and Prof Chris McMahon for their
perpetual guidance, good humour and motivation

• Prof Mike Owen for his advice and direction

• Drs. Heathcote, Vardaki, Margaris, Marles, Wang, Moussa and ‘Future Drs.’ Tai,
Garcia-Carreras, Jiang, Williams, Gresham, Kakade for their warm company and
friendship over the years

• Giota for her patience, understanding and for always being there

vi

Nomenclature

A element area
a rotor inner radius
b rotor outer radius
cp specific heat capacity at constant temperature
C concentration, mass fraction
Cp pressure coefficient relating wheelspace presure to external mainstream
cw non-dimensional mass flow rate (ṁ/µb)
Cd,b discharge coefficient for receiver holes
d pre-swirl nozzle diameter
G gap ratio (s/b)
h heat transfer coefficient
k thermal conductivity of air, turbulent kinetic energy
ṁ mass flow rate
M disc moment
N number of pre-swirl nozzles
Nu Nusselt number (qw r/k(Tw − Tw,ad ))
n temperature exponent
Pk turbulence production
Pr Prandtl number (µcp /k)
p static pressure
po total pressure
qw rotor wall heat flux
R recovery factor (P r1/3 ), universal gas constant
Reφ rotational Reynolds number (ρΩb2 /µ)
r radius
r p , rb radii of pre-swirl nozzles and receiver holes
S Sutherland constant
s rotor stator separation distance
sc axial seal clearance distance
T absolute static temperature
U velocity vector
uτ friction velocity
v absolute velocity

vii
 x non-dimensional radius (r/b), grid distance
y distance normal to the wall
y+ non-dimensional wall distance (ρyuτ /µ)
β swirl ratio (vφ /Ωr), advection scheme weighting function
Γ ratio of disc angular velocities
γ ratio of specific heats
� turbulence eddy dissipation

θ circumferential angle

Θb,ad adiabatic effectiveness

ηc effectiveness based on concentration

ηt effectiveness based on temperature

λT turbulent flow parameter (cw Re−

φ
0.8
)

µ dynamic viscosity

ρ density

ω turbulent frequency

Ω angular velocity of rotor

Turbulence Model Constants

Cµ , C�1 , C�2 , σk , σ� k-� model constants

α1 , α2 , β � , β1 , β2 , σk1, σk2 , σω1 , σω2 BSL model constants

Subscript

ad adiabatic
b blade-cooling
e external mainstream
ef f effective
f first integration point from wall
i isentropic value
o total value in stationary frame
p pre-swirl
ref reference value
s stator, sealing flow
t total value in rotating frame, turbulent
w rotor, wall

viii
 φ, r, z circumferential, radial and axial direction
∞ value in core at z/s = 0.5
1, 2 upstream and downstream locations in a stream-tube

Acronyms

BSL Baseline turbulence model

LDA Laser doppler anemometry
SST Shear stress transport turbulence model
T ET Turbine entry temperature
T LC Thermochromic liquid crystals

ix

Contents

Abstract ii

Nomenclature vii

1 Introduction 1

1.1 Gas Turbine Engines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Review of Previous Work 6

2.1 Rotating Discs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.2 Direct Transfer Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3 Computational Method 14

3.1 Discretisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3.2 Turbulence Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.2.1 High Reynolds number k-� . . . . . . . . . . . . . . . . . . . . . . . . 15

3.2.2 BSL (Base-Line) Model . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.2.3 Reynolds Stress Model . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.2.4 Wall Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.3 Heat Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

x
 3.4 Fluid Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3.5 Mesh Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3.6 Parallel Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.7 Code Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.7.1 Rotor-Stator with Radial Outflow . . . . . . . . . . . . . . . . . . . . 24

3.7.2 Coverplate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4 Pre-Swirl System 32

4.1 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4.2 Fluid Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

4.2.1 Velocity and Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

4.2.2 Adiabatic Effectiveness . . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.2.3 Discharge Coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4.3 Heat Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

4.3.1 Radial Heat Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

4.3.2 Circumferential Heat Transfer . . . . . . . . . . . . . . . . . . . . . . 45

4.3.3 Heat Transfer - Fluid Dynamics Interaction . . . . . . . . . . . . . . . 46

5 High Radius Pre-Swirl System 52

5.1 Fluid Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5.1.1 Flow Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5.1.2 Discharge Coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

5.1.3 Adiabatic Effectiveness . . . . . . . . . . . . . . . . . . . . . . . . . . 60

5.2 Heat Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

5.2.1 Radial Variation of Heat Transfer . . . . . . . . . . . . . . . . . . . . 75

5.2.2 Circumferential Variation of Heat Transfer . . . . . . . . . . . . . . . 77

xi

6 Conclusions 86

7 Further Work 88

7.1 Investigation of Receiver Hole Geometry . . . . . . . . . . . . . . . . . . . . 88

7.2 Pre-Swirl Nozzle Location . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

7.3 Inlet Velocity Profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

7.4 Scaling to Engine Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

7.5 Derivation of Rotor Temperature Distribution . . . . . . . . . . . . . . . . . . 89

7.6 Derivation of Simple Correlations . . . . . . . . . . . . . . . . . . . . . . . . 90

7.7 Validation of Orifice Model With CFD . . . . . . . . . . . . . . . . . . . . . . 90

Bibliography 91

Appendix 1

Pilot Study
Ingress into rotating disc systems

Appendix 2

Journal of Engineering for Gas Turbines and Power

Physical Interpretation of Flow and Heat Transfer in Pre-Swirl Systems

Appendix 3

Proceedings of ASME Turbo Expo 2008: Power for Land, Sea and Air
ASME GT2008 - 50295: Effect of Radial Location of Nozzles on Performance of Pre-Swirl
Systems

Appendix 4

International Gas Turbine Congress 2007

Three-Dimensional Computations of Ingress in Gas Turbine Cooling Systems

xii

List of Figures

1.1 Gas turbine engine internal air system (Rolls-Royce 1986). . . . . . . . . . . . 3

1.2 Turbine internal air system (Rolls-Royce 1986). . . . . . . . . . . . . . . . . . 4

1.3 Isometric and schematic view of pre-swirl chamber (Rolls-Royce 1986). . . . . 5

2.1 Rotor-stator with radial outflow. . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2 Coverplate model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3.1 Computational domain for pre-swirl rotor-stator configuration. . . . . . . . . . 22

3.2 Top view of a finite volume assuming unit depth (inc., 2001). . . . . . . . . . . 22

3.3 Element of a finite volume (inc., 2001). . . . . . . . . . . . . . . . . . . . . . 23

3.4 Typical radial distribution of y + along the rotor for viscous and inertial regimes 23

3.5 Wall function method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.6 Rotor-stator with radial outflow. . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.7 Coverplate model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.8 Experimental (symbols) and computational (lines) radial and circumferential

velocity distributions for the rotor-stator system with radial outflow. Reφ =

1.25x106 and cw = 6, 100. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

3.9 Rotor-stator system: Rotor temperature distribution. . . . . . . . . . . . . . . . 30

3.10 Rotor-stator system: Radial Nusselt number distribution. . . . . . . . . . . . . 30

3.11 Coverplate system: Rotor temperature distribution. . . . . . . . . . . . . . . . 31

xiii
3.12 Coverplate System: Nusselt number distribution. Reφ = 1.37 ∗ 106 , cw =
28, 354, λT = 0.349 and βp = 2.61 . . . . . . . . . . . . . . . . . . . . . . . . 31

4.1 βp variation with Reφ based on a fixed inlet angle of 20◦ . . . . . . . . . . . . . 33

4.2 Experimental setup: Illustration. . . . . . . . . . . . . . . . . . . . . . . . . . 35

4.3 Radial distribution of swirl ratio: Computation (lines) and experiment

(symbols). (a) Reφ = 0.8 ∗ 106 ; (b) Reφ = 1.0 ∗ 106 ; (c) Reφ = 1.2 ∗ 106 . . . . 40

4.4 Radial distribution of static pressure: Computation (lines) and experiment

(symbols). (a) Reφ = 0.8 ∗ 106 ; (b) Reφ = 1.0 ∗ 106 ; (c) Reφ = 1.2 ∗ 106 . . . . 41

4.5 Radial distribution of total pressure: Computation (lines) and experiment

(symbols). (a) Reφ = 0.8 ∗ 106 ; (b) Reφ = 1.0 ∗ 106 ; (c) Reφ = 1.2 ∗ 106 . . . . 42

4.6 Variation of Θb,ad ,β and Cd,b with βp for 0.8 ∗ 106 < Reφ < 1.2 ∗ 106 and
0.12 < λT < 0.38. (a) Comparison between computed and theoretical Θb,ad .
(b) Comparison between computed and measured β1 and β2 . (c) Comparison

between computed and measured Cd,b . . . . . . . . . . . . . . . . . . . . . . . 43

4.7 Radial Distribution of N uRe−0.8 : Computational (lines) and Experimental

(symbols). (a) βp ≈ 0.5; λT ≈ 0.13, (b) βp ≈ 1.0; λT ≈ 0.24, (c) βp ≈ 1.5;

λT ≈ 0.36. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

4.8 Experimental (a-c) and computational (d-f) Nusselt number contours, Reφ =
0.8 ∗ 106 . (a & d) βp = 0.5, λT = 0.13; (b & e) βp = 1.0, λT = 0.24; (c & f)

βp = 1.5, λT = 0.37. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4.9 Computed streamlines relative to the rotor for a variety of conditions. . . . . . 50

4.10 Computed streamlines superimposed onto experimental heat transfer results.

Reφ = 0.8 ∗ 106 , βp = 1.5, λT = 0.38. . . . . . . . . . . . . . . . . . . . . . . 50

4.11 Nu contours, Reφ = 1.19 ∗ 106 , λT = 0.251 and βp = 0.94. . . . . . . . . . . . 51

5.1 Domain for high radius preswirl study. . . . . . . . . . . . . . . . . . . . . . . 53

5.2 Parameter space for study. Line represents 20◦ nozzle angle. . . . . . . . . . . 54

5.3 Normalised radial velocity versus non-dimensional axial distance. Reφ = 106

and λT = 0.24. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

xiv
5.4 Normalised circumferential velocity versus non-dimensional axial distance.

Reφ = 106 and λT = 0.24. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

5.5 Streamlines in r − z plane, βp = 0.5, Reφ = 106 , λT = 0.24. Inlet A, B & C

respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

5.6 Streamlines in r − z plane, βp = 1.0, Reφ = 106 , λT = 0.24. Inlet A, B & C

respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

5.7 Streamlines in r − z plane, βp = 1.5, Reφ = 106 , λT = 0.24. Inlet A, B & C

respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

5.8 Streamlines in r − z plane, βp = 2.0, Reφ = 106 , λT = 0.24. Inlet A, B & C

respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

5.9 Streamlines in θ − z plane in a rotating frame of reference, Reφ = 106 , λT =

0.24. Disc direction from left to right. . . . . . . . . . . . . . . . . . . . . . . 66

5.10 Swirl ratio in the core with linear best fit for the vortex region (a) Inlet A, (b)

Inlet B and (c) Inlet C. Reφ = 106 and λT = 0.24. . . . . . . . . . . . . . . . . 67

5.11 Swirl ratio in core at rb for λT = 0.24 and Reφ = 106 versus (a) preswirl ratio

based on rp , (b) preswirl ratio based on rb . . . . . . . . . . . . . . . . . . . . . 68

5.12 Swirl ratio in core at rb for Reφ = 106 versus (a) preswirl ratio based on rp , (b)

preswirl ratio based on rb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

5.13 Discharge coefficient for receiver holes versus (a) preswirl ratio based on rp ,
(b) preswirl ratio based on rb , (c) swirl ratio in the core at rb . λT = 0.24 and

Reφ = 106 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5.14 Discharge coefficient for receiver holes versus (a) preswirl ratio based on rp , (b)

preswirl ratio based on rb , (c) swirl ratio in the core at rb . Reφ = 106 . . . . . . 70

5.15 Computed variation of discharge coefficient, Cd,b versus experimental results of

Lewis et al. (2007) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

5.16 Computed variation of discharge coefficient, Cd,b versus equation 5.10 . . . . . 71

5.17 Pressure drop throughout system versus (a) preswirl ratio based on rp , (b)

preswirl ratio based on rb , (c) swirl ratio in the core at rb . λT = 0.24 and

Reφ = 106 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

xv
5.18 Comparison of adiabatic effectiveness from computations and correlation of

Karabay et al. (2001) versus (a) preswirl ratio based on rp , (b) preswirl ratio

based on rb , (c) swirl ratio in the core at rb . λT = 0.24 and Reφ = 106 . . . . . . 73

5.19 Moment coefficients for rotor and stator versus (a) preswirl ratio based on rp ,
(b) preswirl ratio based on rb , (c) swirl ratio in the core at rb . λT = 0.24 and

Reφ = 106 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

5.20 Distribution of Nusselt number on a radial line between receiver holes, λT =

0.24 (a) Inlet A, rp /rb = 0.8 (b) Inlet B, rp /rb = 0.9 (c) Inlet C, rp /rb = 1.0. . 79

5.21 Distribution of Nusselt number on a radial line between receiver holes, inlet
angle = 20◦ (a) Inlet A, rp /rb = 0.8 (b) Inlet B, rp /rb = 0.9 (c) Inlet C, rp /rb =
1.0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

5.22 Distribution of Nusselt number on a radial line between receiver holes, inlet

angle = 20◦ (a) λT = 0.12 (b) λT = 0.24 (c) λT = 0.36. . . . . . . . . . . . . . 81

5.23 Distribution of Nusselt number on a circumferential line between receiver holes

at r = rb , λT = 0.24 (a) Inlet A, rp /rb = 0.8 (b) Inlet B, rp /rb = 0.9 (c) Inlet

C, rp /rb = 1.0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

5.24 Contours of Nusselt number distribution on the rotor. Inlet A, rp /rb = 0.8,

Reφ = 106 and λT = 0.24. (a) βp = 0.5, (b) βp = 1.0, (c) βp = 1.5 and (d)

βp = 2.0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

5.25 Contours of Nusselt number distribution on the rotor. Inlet B, rp /rb = 0.9,

Reφ = 106 and λT = 0.24. (a) βp = 0.5, (b) βp = 1.0, (c) βp = 1.5 and (d)

βp = 2.0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

5.26 Contours of Nusselt number distribution on the rotor. Inlet C, rp /rb = 1.0,

Reφ = 106 and λT = 0.24. (a) βp = 0.5, (b) βp = 1.0, (c) βp = 1.5 and (d)

βp = 2.0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

7.1 Ingress into a rotor stator wheelspace. . . . . . . . . . . . . . . . . . . . . . . 98

7.2 Minimum sealing flow rate (Bayley and Owen 1970). . . . . . . . . . . . . . . 100

xvi

7.3 Cw,min for varying external flow axial Reynolds number and a stationary rotor

disc. Each line shows a different seal clearance gap ratio. (Phadke and Owen

1988b) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

7.4 Cw,min for varying external flow axial Reynolds number. Each line shows a
different rotational Reynolds number. (a) Gc = 0.005, (b) Gc = 0.01,(c) Gc =
0.02. (Phadke and Owen 1988b) . . . . . . . . . . . . . . . . . . . . . . . . . 102

7.5 Effectiveness when Cw < Cw,min (Graber et al. 1987). . . . . . . . . . . . . . 103

7.6 Variation of sealing effectiveness with non-dimensional coolant flow rate for

four external flow conditions. (Green and Turner 1994) . . . . . . . . . . . . . 104

7.7 Comparison of calculated and measured concentration in the cavity. (Hills et al.

2002) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

7.8 Computational domain for ingress study. Red represents stationary surfaces,

blue represents the rotor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

7.9 Schematic of vane geometry in the φ − z plane . . . . . . . . . . . . . . . . . 107

7.10 Pressure coefficient relating static pressure in the wheel-space to that outside

of the seal (a) BSL model with varying mesh size, (b) Reynolds Stress model

with varying mesh size, (c) Reynolds Stress model versus BSL model for given

mesh size. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

7.11 Computed radial velocity averaged in the axial direction in the seal, v¯r (a) BSL

model with varying mesh size, (b) Reynolds Stress model with varying mesh

size, (c) Reynolds Stress model versus BSL model for given mesh size. . . . . 117

7.12 Secondary flow in the seal shown in the r − z plane . . . . . . . . . . . . . . . 118

7.13 Computed effectiveness at the stator using concentration method (a) BSL model

with varying mesh size, (b) Reynolds Stress model with varying mesh size, (c)

Reynolds Stress model versus BSL model for given mesh size. . . . . . . . . . 119

7.14 Computed effectiveness at the rotor using concentration method (a) BSL model

with varying mesh size, (b) Reynolds Stress model with varying mesh size, (c)

Reynolds Stress model versus BSL model for given mesh size. . . . . . . . . . 120

7.15 Computed surface temperatures for various prescribed external flow

temperatures, (a) Stator, (b) Rotor. . . . . . . . . . . . . . . . . . . . . . . . . 121

xvii
7.16 (a) Pressure coefficient relating pressure in wheelspace to that outside seal for
various sealing flow rates. (b) & (c) Effectiveness for various sealing flow rates
at the stator and rotor respectively. . . . . . . . . . . . . . . . . . . . . . . . . 122

7.17 Computed effectiveness using concentration method versus sealing flow rate for
(a) stator and (b) rotor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

7.18 (a) Pressure coefficient relating pressure in wheelspace to that outside seal. (b)
Axially averaged radial velocity in the seal. . . . . . . . . . . . . . . . . . . . 124

7.19 Computed effectiveness using concentration method versus Reynolds number

for (a) stator and (b) rotor. (c) Local value evaluated at x = 0.95. . . . . . . . . 125

xviii

List of Tables

3.1 k-� turbulence model constants. . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.2 BSL turbulence model constants. . . . . . . . . . . . . . . . . . . . . . . . . . 18

4.1 Conditions for experiments and computions . . . . . . . . . . . . . . . . . . . 32

5.1 Inlet dimensions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

5.2 Test cases for parametric study. . . . . . . . . . . . . . . . . . . . . . . . . . . 54

7.1 Mesh Parameters for Ingress Geometry. . . . . . . . . . . . . . . . . . . . . . 108

xix

Chapter 1

Introduction

1.1 Gas Turbine Engines

The first flight of a Whittle engine was in 1941; The W1 which was designed and built by Frank
Whittle’s company, Power Jets Ltd. The aircraft, a Gloster E.28/39, reached speeds of 370 mph
while the engine produced 3.8 kN of thrust. Since then gas turbine development has primarily
been driven by the aerospace industry, with a requirement for engines with greater power output
and lower fuel consumption.

Early engines were limited in power output; due to losses in the system, the work produced by
the turbine was often only just sufficient to overcome the work input required by the compressor.
More recently gas turbine engines have found a wide range of applications. Although not
usually as efficient as alternative power plants such as steam turbines, they can be more compact
and offer greater flexibility in their installation and use.

In order to manufacture an engine with low fuel consumption, it is necessary that the
temperature rise in the combustion chamber be as high as possible. This temperature rise is
limited by the available materials. Stationary components do not experience high loads, so the
material choice is less limited than for rotating components. The most temperature critical part
of the engine is the blades of the first stage turbine. These must sustain high temperatures as
they are immediately downstream of the combustion chamber. The blades also experience both
centrifugal and aerodynamic forces.

Over time, the maximum turbine entry temperature (TET) has increased since the Whittle W1,
which had a TET of 1050 K; up until the 1960’s the maximum temperature was limited to the

metallurgical limit of the alloy used. In order to surpass this ceiling it was necessary to provide
cooling to the blade material to prevent failure.

The earliest blade cooling was achieved by pumping low pressure air radially outwards through
a passage inside the blade. This airflow acted to dissipate thermal energy from the blade,
allowing the TET of the engine to be increased by 150 K.

By the 1970’s the cooling design had been developed to provide film cooling to the blade
surface in addition to internal cooling. For effective film cooling it is necessary to inject air
into the mainstream flow from close to the nozzle leading edge. This requires flow at a higher
pressure than that used for internal cooling. A similar design is used in modern gas turbines,
but with more advanced film cooling and more complex internal cooling passages. Progress
was also being made in the development of the alloys used to manufacture the blades. Modern
turbine blades are made from a single crystal alloy and have excellent material strength and
high metallurgical limit.

The internal air system for a two-stage engine is shown in figure 1.1. Cooling air for the turbine
blades is bled from the compressors upstream of the combustion chamber. Feeds are taken from
different stages to supply high and low pressure air as required. This air, however essential, is
a source of work loss to the system; it is therefore important to minimise the volume of cooling
air to reduce fuel consumption. In addition to cooling, compressed air is used for other purposes
such as pressurised aircraft cabin conditioning.

Figure 1.2 shows in more detail the internal air flows in the turbine. The dark shaded area and
the turbine blades are all rotating, while the remaining parts are stationary. The area of interest
in this study is the chamber directly upstream of the entrance to the blade-cooling passage. An
isometric sketch of this region, called the ‘pre-swirl chamber’, is shown in figure 1.3.

The entrance to the pre-swirl chamber has a row of turning vanes, such that flow is swirled in
the direction of the rotating disc. Flow then leaves the chamber, or ‘wheel-space’, through the
blade-cooling passage entrances, or ‘receiver holes’. There are two reasons why pre-swirling
the air is favourable; firstly the velocity differential between the airflow and the receiver hole
is reduced, increasing the amount of air delivered to the blades for a given pressure drop.
Secondly, work must be performed on the flow to bring it to the same angular velocity as the
receiver holes before the flow can enter the blade-cooling passages. This work input increases
the total temperature of the flow, reducing its cooling effectiveness. By pre-swirling the flow,
the required work input is less, and therefore the cooling air is delivered to the blades at a lower
temperature.

2
 Figure 1.1: Gas turbine engine internal air system (Rolls-Royce 1986).

A cross-section of the pre-swirl chamber is given in figure 1.3. The flow structure in the
wheel-space is complex; flow tends to travel radially outward on the rotating disc and inward on
the stationary disc, while the ‘core’ of the fluid away from the discs rotates in the circumferential
direction.

1.2 Objectives

The objectives of this study are outlined below:

1. Validate a three-dimensional commercial CFD code for application to rotating disc

systems. Use earlier experimental results as a benchmark to investigate the appropriate
use of method, meshing and turbulence models.

2. Use a validated model to perform computations of flow and heat transfer on the pre-swirl
rotor-stator system. Compare with measurements and interpret the computed velocity
field and heat flux patterns to explain the mechanisms driving the delivery of blade cooling
air.

3. Investigate the effect on performance parameters of geometry changes by relocating the

 Figure 1.2: Turbine internal air system (Rolls-Royce 1986).

pre-swirl nozzles to a higher radius. This will provide a basis for comparison with future
experiments to be performed at the University of Bath.

4. Relate results to the appropriate use of pre-swirl and nozzle location to optimise the
temperature of blade cooling air and the pressure drop throughout the system.

5. Investigate computationally the fluid dynamics of a rotating disc system subject to ingress.
In this situation flow from the external main gas path leaks through the outer seal
contaminating the cooling air path, and thus increasing the temperature of cooling air
reaching the blades. It is anticipated that an experimental study will be performed with
the research group and therefore the computational approach developed here will be used
to inform the design of such experiments. The University of Bath has particular research
expertise in the use of Thermochromic Liquid Crystals which will be used to gain new

4
Figure 1.3: Isometric and schematic view of pre-swirl chamber (Rolls-Royce 1986).

insight into disc heat tranfer distributions due to ingested flow.

Chapter 2

Review of Previous Work

2.1 Rotating Discs

The study of rotating disc systems dates back to the end of the 19th century. At this time,
before gas turbines had been conceived, research was driven by the steam turbine industry.
There was much interest in the quantification of drag experienced by turbine plates. An early
summary of disc theory was written by Dorfman (1963), his work included theoretical models
supported by experimental results for a disc rotating in free space (the ‘free disc case’), a disc in
a confined space and a disc in an axial flow. Another work during the same period was produced
by Greenspan (1969) who applied the rotating fluid theory to the circulation of the oceans and
likened it to the contained flow between concentric spheres.

Twenty years later and the wealth of theoretical and experimental literature on the subject of
rotating flows was brought together again in two volumes by Owen and Rogers (1989, 1995).
The research described by the first volume, largely funded by gas turbine companies, starts
with the free disc case, and builds to cover the fluid dynamics and heat transfer for rotor-stator
systems with and without superposed flow. The second volume is concerned with rotating
cavities formed by co-rotating discs and examines the growth of computer modelling techniques
for the solution of these complex flows.

Since this time several research groups have been active in advancing the study of rotating disc
systems. These include the Universität Karlsruhe, University of Surrey, University of Sussex
and the University of Bath.

The group within Bath have investigated a series of configurations that are useful to examine

here briefly, as they are used as validation cases for computational models. The body of research
based on direct transfer pre-swirl systems and ingress is examined in the most detail as this is
the basis of the current study.

The rotor-stator system represents the wheel-space enclosed by the turbine disc and the stator
disc in the turbine stage. This configuration was investigated by Chen et al. (1996) both
computationally and experimentally. A detailed review of the setup is included in section 3.7
as part of the code validation for this study and a schematic of the configuration is shown in
figure 2.1. Laser Doppler Anemometry (LDA) was used to measure circumferential and radial
velocities at different radial locations in the wheel-space. Fluxmeters embedded in the rotor disc
allowed the Nusselt number distribution to be calculated and this was compared with results
from an elliptic solver with a low Reynolds number k-� turbulence model. The findings showed
a decrease in core swirl rate with increasing mass flow rate, and clearly resolved the inflow and
outflow on the stator and rotor respectively.

rotating disc

stationary disc

axis of rotation

Figure 2.1: Rotor-stator with radial outflow.

Contrarotating disc systems are of interest to the engineer due to the possibilities for future
engine technologies (Gan et al., 1994). Ultra-high-bypass-ratio engines could potentially use
contrarotating fans, which in turn would be driven by contrarotating turbines. The most
important parameter defining the flow structure is that of the ratio of the angular velocity of
the slower disc to that of the faster, Γ = Ω1 /Ω2 .

7
Gan et al. (1995) studied the system with Γ = −1, corresponding to discs rotating in opposite
directions at the same angular velocity. Measurements were made, using LDA, of radial and
circumferential velocity distributions. Computations were also performed using both laminar
and turbulent models. It was found that laminar computations cannot predict the flow structure,
even for experiments with very low rotational Reynolds number. Experiments and turbulent
computations show a radial outflow on both discs with a radial inflow throughout the core. The
laminar computations give contra-rotating cores of fluid with a very thin shear layer between
them at the mid-plane. The authors concluded that the flow in the core was always turbulent,
even though the boundary layers could remain laminar at low Reynolds numbers.

Kilic et al. (1996) performed a similar study but for discs rotating at different speeds in the
range −1 ≤ Γ ≤ 0. This was then extended to apply a superposed mass flow rate through the
system, by Gan et al. (1994). Flow entered the system axially at low radius, and exited through
the clearance between the disc shrouds at the outer radius. It was found that the superposed
flow acted to reduce the core rotation of the system, and promoted the transition from Batchelor
to Stewartson-type flow as Γ = −1 is approached.

Heat transfer results for the contrarotating disc configuration were published by Chen et al.
(1997). The downstream disc was heated to approximately 100◦ C and instrumented with
thermocouples and fluxmeters such that wall heat flux distributions and temperature contours
could be measured. They found that the heat flux was only weakly affected by superposed mass
flow through the system, but was heavily dependent on the rotational Reynolds number of the
system.

Kilic and Owen (2003) published results of a computational study with superposed flow
entering radially through the system. The study summarised the three important flow structures
that can occur:

1. Γ = −1. Stewartson-type flow; radial outflow occurs in the entraining boundary layers
on each disc. The core of fluid between the boundary layers is stationary.

2. Γ ≈ 0. Batchelor-type flow; radial outflow in the entraining boundary layer on the

rotating disc, radial inflow on the stationary disc. Between the layers is a rotating core of
fluid.

3. Γ = 1. Ekman-layer flow; non-entraining boundary layers on each disc with a rotating

core of fluid between.

8
They also presented a table of computed disc moment coefficients for a range of Γ values, with
and without superposed flow.

Ekman layers are classically formed when a rotating flow occurs close to a stationary surface.
Within the core of the rotating flow a balance between the radial pressure gradient and
centrifugal forces is achieved, however near the surface the no-slip condition acts to reduce
the flow’s angular momentum. The excess radial pressure gradient therefore creates a radial
inflow of fluid which is ejected from the boundary layer into the core region as a column of
rotating fluid - a mechanism which is a common phenomenon in atmospheric physics.

The flow close to a rotating disc experiences the opposite effect, whereby work is done on the
flow in the boundary layer, increasing its angular momentum. This allows the flow to overcome
the radial pressure gradient and form a radial outflow close to the disc.

An alternative configuration for the chamber between the turbine disc and stator disc is the
coverplate system, a schematic is shown in figure 2.2. A large part of the chamber is contained
by two discs co-rotating to create ‘rotating cavity’.

stationary disc

rotating disc

axis of rotation

Figure 2.2: Coverplate model.

Karabay et al. (1999) showed that the coverplate system could be reduced to a rotor-stator
system and a rotating cavity, as mentioned above. For engine realistic cooling flow rates,
free-vortex flow was shown to occur in the rotating cavity. Pilbrow et al. (1999) extended
the study to include heat transfer from the downstream disc.

Karabay et al. (2000, 2001) describe an ongoing effort to quantify the performance of the

coverplate system. Initially computations were performed using an axisymmetric elliptic

solver with a low Reynolds number k-� and comparing results with those of the experimental
study. A 3D code was also employed when it was found that the three dimensional effects
around the blade-cooling receiver holes were important, this improved the agreement between
computational and experimental results, although the differences were still significant. An
optimum pre-swirl ratio of βp ≈ 1.4 was found, at which the average rotor Nusselt number
was a minimum.

2.2 Direct Transfer Systems

The ‘direct transfer system’ has a rotating turbine disc and an upstream stationary disc, similar
to the rotor-stator system discussed above. The important features are the inlet and outlet
through which is passed the blade-cooling air. The stator disc contains nozzles and the rotor has
discrete receiver holes.

Meierhofer and Franklin (1981), who were the first to measure the effect of pre-swirl on the
temperature drop in a direct-transfer system, showed that swirling the air could significantly
reduce the total temperature in the receiver holes of a turbine disc. El-Oun and Owen (1989)
developed a theoretical model for the so-called adiabatic effectiveness, Θb,ad , based on the
Reynolds analogy. The model, which was in good agreement with the temperatures measured
on their rotating-disc rig, showed that Tt,b , the total temperature in the receiver holes, decreased
monotonically as βp , the pre-swirl ratio, increased even when βp was significantly greater than
unity.

The Reynolds analogy provides an approximate relationship between the skin friction
coefficient and the heat transfer coefficient. The two-dimensional boundary layer equations
for flow over a flat with no pressure gradient and small viscous dissipation are:

∂u ∂u ∂ 2u
u +v =ν 2 (2.1)
∂x ∂y ∂y

∂T ∂T ∂2T
u +v =α 2 (2.2)
∂x ∂y ∂y

where α represents thermal diffusivity.

Reynolds (1874) suggested that the similarities between equations 2.1 and 2.2 should lead to

10

similar distributions of u and T in the boundary layer. The analogy formed by Reynolds is
shown in equation 2.3 below.

� �
|qw | |κ(∂T /∂y)w | κ �� dT ��
= = �� �� (2.3)
τw µ(∂T /∂y)w µ du w

This relationship is commonly used in a simpler form in terms of the non-dimensional

parameters, the Stanton number, St, and the skin friction coefficient, cf :

2St 1
≈ (2.4)
cf Pr

The Reynolds analogy, while applicable to both compressible and incompressible flows,
becomes less accurate for situations where a strong pressure gradient is present and for
non-unity Prandtl numbers.

Geis et al. (2004) made measurements of the adiabatic effectiveness, which showed that the
measured values of Tt,b were significantly higher than the values predicted from their ideal
model. (It should be pointed out that their pre-swirl ratio was based on isentropic values rather
than on measurements.) Chew et al. (2005) made numerical simulations of both the ‘Karlsruhe
rig’, used by Geis et al. (2004), and a ‘Sussex pre-swirl rig’. The computations were in good
agreement with the results of both rigs, and the low adiabatic effectiveness of the Karlsruhe rig
was attributed to the geometry of the pre-swirl chamber; in particular, the Karlsruhe rig had a
much larger stator area, which reduced the effective swirl ratio and consequently reduced the
effectiveness.

Chew et al. (2005) and Farzaneh-Gord et al. (2005) independently derived theoretical models
for the adiabatic effectiveness of a direct-transfer system, taking account of the moment on the
stator. (These models predict lower values of Θb,ad than that of Karabay et al. (2001), who based
their model on a cover-plate system in which the pre-swirl air flows radially outward between
two rotating discs.)

Popp et al. (1998) carried out a CFD analysis of a cover-plate system, computing the
temperature drop and the discharge coefficients for different geometries. They showed that
Cd,b , the discharge coefficient for the receiver holes, became a maximum when the relative
tangential velocity was close to zero. This effect was confirmed experimentally by Dittmann
et al. (2002) who were the first to measure the discharge coefficients in a direct transfer system.

The discharge coefficient is defined as the ratio of the actual mass flow rate to the isentropic

11

mass flow rate, as shown in equation 2.5. It is possible to define several different discharge
coefficients; for the pre-swirl nozzles, wheel-space, receiver holes, and another for the entire
system, however in the studies considered here the parameter of interest is the discharge
coefficient for the receiver holes. In essence the discharge coefficient, which is flow and
geometry dependent, represents the loss in mass flow rate versus ideal for a given pressure
drop through an orifice. A discharge coefficient of 1 (which is the maximum value achievable)
would represent an isentropic system.

ṁb
Cd,b = (2.5)
ṁi

Yan et al. (2003) measured the discharge coefficients for the receiver holes of a direct-transfer
system for a range of rotational speeds and flow rates. For β1 < 1 (where β1 is the measured
swirl ratio upstream of the receiver holes) Cd,b increased monotonically as β1 increased from
β1 ≈ 0.3 to 0.9. They also found, as did Popp et al. (1998), that Cd,b depends on the ratio of
the area of the receiver holes to that of the nozzles; for a given value of the pre-swirl ratio, βp ,
Cd,b increases as the area ratio decreases. (It should be noted that, owing to a printer’s error,
the wrong figures were printed in Yan et al. (2003); the correct figures are given in Lock et al.
(2005a)).

Heat transfer in a direct transfer rig was studied experimentally and computationally by Wilson
et al. (1997) using fluxmeters to determine the local Nusselt numbers. Their axisymmetric
CFD results gave reasonable predictions of the velocity and temperature in the core but
underpredicted the measured Nusselt numbers by up to 25% near the outer shroud..

Accurate measurement of flow fields in the pre-swirl rotor-stator has been achieved by the
group at University of Karlsruhe, the work is published in Geis et al. (2002) and Bricaud
et al. (2004). Particle Image Velocimetry (PIV) was used with two endoscopic cameras looking
through the outer shroud, this stereo PIV configuration allowed all three velocity components
to be measured at tangential planes in the wheel-space. This work highlighted the unsteady
aerodynamics associated with the rotor-stator system, especially the ‘pumping effect’. The
system configuration investigated has pre-swirl nozzles at the same radius as the receiver
holes. As the receiver holes in the rotating frame change position with respect to the nozzles a
fluctuation in the flow field is observed.

Owen and Rogers (1995) showed that, for a rotating cavity of given geometry, the turbulent
flow structure depends on only two nondimensional parameters: the inlet swirl ratio, βp , and

12

the turbulent flow parameter, λT . The inlet pre-swirl ratio is defined as the ratio of the tangential
velocity of the pre-swirl flow at the inlet to the tangential velocity of the disc at the same radius:

vφp
βp = (2.6)
Ωrp

The turbulent flow parameter is defined as the non-dimensional mass flow rate normalised by
the Reynolds number raised to the power of -0.8:

λT = cw Re−0.8 (2.7)

where the non-dimensional mass flow rate is

ṁp
cw = (2.8)
µb

13

Chapter 3

Computational Method

The commercial code used throughout is Ansys-CFX, the relevant version at the beginning
of the study was 5.7. This was upgraded after one year to version 10.0. The versions were
validated against one another and no difference in output from the solver was found. This is
consistent with information given in the release notes from the software provider (inc. 2001).
This commercial code is commonly used for the computation of gas turbine flows, therefore the
novelty of this thesis lies in the application rather than in the computational methodology.

3.1 Discretisation

An algebraic multi-grid method is used to improve the speed of convergence of a given

simulation. The flow is initially solved over a coarse grid, which is gradually refined until a
steady state has been reached for the full grid.

The computational space, shown schematically in figure 3.1, is meshed into finite volumes, as
shown in figure 3.2. The flow properties; velocity, pressure, temperature plus any turbulence
and additional variables are all stored at the nodes. The governing equations are integrated over
each control volume.

The solver uses an extended version of the method of Rhie and Chow (1982), their method,
which was developed and tested in 2D applications, uses a transformation from a staggered
grid to an ordinary grid in order to solve for pressure. The mass continuity equation is shown
one-dimensionally in equation 3.1. A second order accurate scheme is used to calculate the
velocity differential, the second term acts to redistribute the influence of the pressure. It can be

14

seen that as grid refinement is performed and Δx becomes small, the second term rapidly tends
to zero.

� �
∂U Δx3 A ∂ 4p
+ =0 (3.1)
∂x i 4ṁ ∂x4 i

The advection scheme is based on the method of Barth and Jesperson (1989). It is generally
second-order accurate but with a relaxation factor applied to the second-order terms to remain
within the principles of boundedness. The form of the advection scheme is shown in equation
3.2, where 0 < β ≤ 1. Details on the evaluation of β can be found in inc. (2001).

φip = φup + βφ (3.2)

3.2 Turbulence Models

Two turbulence models have been used in this study, a high Reynolds number k-� model and
the BSL model, a blended k-�/k-ω model. Both are two equation eddy-viscosity models and are
used in conjunction with wall functions to reduce computational cost. Details of the models are
can be found in inc. (2001), but are included here for completeness.

3.2.1 High Reynolds number k-�

The k-� model is based on the concept that k, the turbulence kinetic energy, causes an effective
increase in the dynamic viscosity of the flow such that:

µef f = µ + µt (3.3)

The term µt is the turbulence viscosity which is based on k and �, the turbulence eddy
dissipation:

k2
µt = C µ ρ (3.4)
�

The values of k and � are evaluated directly from their respective transport equations:

15
 �� � �
∂(ρk) µt
+ � • (ρU k) = � • µ+ �k + Pk − ρ� (3.5)
∂t σk

�� � �
∂(ρ�) µt �
+ � • (ρU �) = � • µ+ �� + (C�1 Pk − C�2 ρ�) (3.6)
∂t σ� k

Buoyancy is ignored in the computation, since the configuration is assumed to be periodic in

the circumferential direction, there is no defined ‘upward’ direction. Therefore the production
term in the above transport equation reduces to:

� � 2
Pk = µt �U • �U + �U T − � • U (3µt � • U + ρk) (3.7)
3

The constants in the above equations are given in table 3.1 below.

Cµ C�1 C�2 σk σ�
0.09 1.44 1.92 1.00 1.30

Table 3.1: k-� turbulence model constants.

3.2.2 BSL (Base-Line) Model

The BSL, or ‘BaseLine’, model of Menter (1994) is a blended turbulence model. It is generally
recognised (inc., 2001) that k-ω models are sensitive to freestream turbulence levels, therefore
a k-� model is used in the region away from the walls. Close to the wall boundaries the k-ω
formulation of Wilcox (1998) is employed.

In a similar way to the k-� model, the BSL makes an adjustment to the dynamic viscosity of
the flow, as shown in equation 3.3. In this case the turbulent viscosity is defined in terms of the
turbulent kinetic energy, k, and the turbulent frequency, ω:

k
µt = ρ (3.8)
ω

The Wilcox model has transport equations:

�� � �
∂(ρk) µt
+ � • (ρU k) = � • µ+ �k + Pk − β � ρkω (3.9)
∂t σk1

16

 �� � �
∂(ρω) µt ω
+ � • (ρU ω) = � • µ+ �ω + α1 Pk − β1 ρω 2 (3.10)
∂t σω1 k

The k-� model is transformed into terms of ω:

�� � �
∂(ρk) µt
+ � • (ρU k) = � • µ+ �k + Pk − β � ρkω (3.11)
∂t σk2

�� � �
∂(ρω) µt 1 ω
+ � • (ρU ω) = � • µ+ �ω + 2ρ �k�ω + α2 Pk − β2 ρω 2 (3.12)
∂t σω2 σω2 ω k

In order to blend these two sets of similar transport equations together the terms which differ
between them are weighted using the blending factor F1 . The bias on the weighting tends
towards the Wilcox model for higher F1 , i.e. F1 = 0 implies a pure version of the k-� model,
while F1 = 1 implies the Wilcox model is being used.

When this weighting factor is applied, the combined transport equations become:

�� � �
∂(ρk) µt
+ � • (ρU k) = � • µ+ �k + Pk − β � ρkω (3.13)
∂t σk3

�� � �
∂(ρω) µt 1 ω
+ � • (ρU ω) = � • µ+ �ω + (1 − F1 ) 2ρ �k�ω + α3 Pk − β3 ρω 2
∂t σω3 σω2 ω k
(3.14)

Each of the constants in the new transport equations are calculated from their constituents using
the weighting, as shown in 3.15:

Φ3 = F1 Φ1 + (1 − F1 )Φ2 (3.15)

Where F1 is calculated thus:

⎡ ⎛ ⎞4 ⎤
� √ �
⎢ k 500v 4ρk
F1 = tanh ⎣min ⎝max , 2 , � �⎠ ⎥⎦ (3.16)
β � ωy y ω σω2 y ∗ max 2ρ σω12 ω �k�ω, 10−10
2

The constants for the above BSL model transport equations are:

17
 α1 α2 β� β1 β2 σk1 σk2 σω1 σω2
5/9 0.44 0.09 0.075 0.0828 2 1 2 1/0.856

Table 3.2: BSL turbulence model constants.

3.2.3 Reynolds Stress Model

The Reynolds Stress turbulence model (inc. 2001) is used for comparison during the ingress
study included in the appendix. The variation in length scales in each direction in the region of
the seal can give rise to anisotropic turbulence levels which may not be captured sufficiently by
a two equation model. These anisotropic turbulent levels can also become important in flows
under the effect of body forces such as rotating flows or complex geometries where separation
and reattachment occur.

This models introduces six new equations for the turbulent velocity components, ui uj where
i = 1, 2, 3 and j = 1, 2, 3. These replace the isotropic turbulent kinetic energy term, k, used
in the two equation models. The added complexity in this model has a significant effect on the
computation time required and a slight impact on the memory requirements.

3.2.4 Wall Functions

Scaleable wall functions are used to reduce the computational cost associated with integrating
flow properties to the wall. Using wall functions, the spacing between the wall and the first mesh
point can be greatly increased, the shear stress and velocity values at this first point are then
calculated by applying the law-of-the-wall. The method used is that of Launder and Spalding
(1974). In the case of the high Reynolds number k-� and BSL models, the law-of-the-wall
approximation assumes that the first point is located in the log region of the boundary layer
profile. The minimum non-dimensional distance to the wall, defined as y + in figure 3.5, in this
case is y + ≈ 11. Typical y + distributions occurring in the computations are shown in figure 3.4,
values are kept in the range 11 < y + < 60. A sensible upper limit for the parameter is ≈ 300.

The iterative method used to calculate the velocity and shear stress near the wall is shown in
figure 3.5. The transport equations for the relevant turbulence models are solved to produce
an estimate for the turbulence kinetic energy, k. This is then substituted into the equation for
wall shear stress, from which the non-dimensional distance to the wall, y + , can be calculated.
The law-of-wall relates y + to u+ , this value of u+ is then used to calculate the wall tangential
velocity, to finally input back into the transport equations.

18
The performance of the wall functions is particularly sensitive to the value of y + in this rotating
disc system. The relationship of u+ (non-dimensional shear stress) to y + (non-dimensional
distance from the wall) is well understood for boundary layers on a flat plate. However the
boundary layer profile found in the computation has a lower gradient for the outer log-linear
region (y + > 100). Therfore the first integration point near the wall should have y + < 100 in
order not to over-predict u+ and consequently the wall shear stress.

3.3 Heat Transfer

The heat transfer model uses the thermal law-of-the-wall of Kader (1981). The wall heat flux is
calculated based on the temperature difference between the wall and the first integration point,
as shown in equation 3.17.

ρCp u∗
qw = (Tw − Tf ) (3.17)
T+

where T + is calculated thus:

� �2
T + = P ry ∗ e(−Γ) + [2.12ln(y ∗ ) + 3.85P r1/3 − 1.3 + 2.12ln(P r)]e(−1/Γ) (3.18)

0.01(P ry ∗ )4
Γ= (3.19)
1 + 5P r3 y ∗

µCp
Pr = (3.20)
λ

The Nusselt number is the non-dimensionalised measure of heat transfer, the definition used for
the rotating system is shown in equation 3.21. The literature discussed above contains much
work on the definition of a correct reference temperature, the adiabatic wall temperature which
takes account of the frictional heating term by Karabay et al. (2001) is used here and is defined
in equation 3.22.

qw r hr
Nu = = (3.21)
k (Tw − Tw,ad ) k

19

 � �
Vφ,2 ∞ Ω2 r2 Vφ,∞ 2
Tw,ad = To,p − +R 1− (3.22)
2Cp 2Cp Ωr

where R = P r1/3 and represents the thermal recovery factor for the fluid. As the fluid velocity
matches that of the disc at the wall due to the no slip condition the total temperature of the fluid
is recovered. However since the fluid is not brought to rest adiabatically some loss occurs which
is represented by the recovery factor.

3.4 Fluid Properties

The dynamic viscosity of the flow is calculated based on Sutherland’s formula (Sutherland,
1893) with the constant for air as S = 110 and a reference value of viscosity taken at Tref =
288K. n is the temperature exponent for the gas.

� �n
µ Tref + S T
= (3.23)
µref T +S Tref

Density is calculated according to the Ideal Gas Law based on the molecular weight of the gas:

w(p + pref )
ρ= (3.24)
RT

3.5 Mesh Generation

Two mesh generation tools are used in the study. The first is a 3D unstructured automatic
mesher. The surface is meshed using a Delaunay routine followed by an advancing front
algorithm to create the volume mesh. By default the computational domain is entirely
tetrahedral, however this is not ideal for flow near the wall. In general, numerical diffusion
is reduced when element faces are parallel and normal to the flow direction. In the boundary
layer the mean flow direction is parallel to the wall, as such tetrahedral elements would have face
angles acute to the flow and thus suffer from losses in numerical accuracy. Instead prismatic,
or quadrilateral, elements are placed in the wall region, which have faces aligned with the
flow direction. These elements are built up to the prescribed number of layers, with increasing
thickness away from the wall. This gives control over the distance of the first node from the

20

wall boundary (i.e. good control of y + values) and allows a smooth transition in element size
between the wall boundary and the main tetrahedral volume mesh. The pre-swirl model has
these prismatic layers applied to all of the solid walls of the domain. The receiver hole for
the blade-cooling passage represents a region in the configuration where the flow experiences
rapid changes in geometry. To better resolve the flow physics in this area, the mesh is locally
controlled to reduce the mesh length scale and therefore increase node density.

The alternative mesher is used for axisymmetric domains, such as the wheelspace for the ingress
study. It uses the same Delaunay surface mesh routine but only on one axisymmetric plane, this
surface mesh still uses prismatic layers in the near wall region for solid boundaries. The mesh
is then rotated around the central axis and elements created at prescibed angular offsets. In
the specific case of the ingress domain this is attached using GGI (Generic Grid Interface) to a
tetrahedral mesh for the complex geometry of the upstream vane.

A comprehensive grid sensitivity analysis has been performed to ensure that the results of the
computation are insensitive to the meshing parameters and strategy that has been adopted.

3.6 Parallel Approach

A multiprocessor Linux system is used to run CFX in parallel when additional memory or
reductions in computation time is required. The system is a cluster of 20 dual-processor nodes
with distributed memory (1 GB / node) and shared file storage. The limiting factor for this
commercial code is generally the number of parallel licenses purchased.

The domains are generally partitioned in the circumferential direction, such that each partition
is approximately the same size and with similar overlaps.

21

 receiver
hole flow

stationary disc

rotating disc

pre-swirled
inlet flow

O
20

Figure 3.1: Computational domain for pre-swirl rotor-stator configuration.

element face centroid

node

element

finite volume surface

Figure 3.2: Top view of a finite volume assuming unit depth (inc., 2001).

22

 n1 n2

ip1 integration point

element face centroid

ip3 ip2

sectors

n3

Figure 3.3: Element of a finite volume (inc., 2001).

50

40
Non-dimensional Wall Distance, y+

30

20

Viscous Regime, λT = 0.1

Inertial Regime, λT = 0.3

10
0.6 0.7 0.8 0.9 1.0
Radius, x

Figure 3.4: Typical radial distribution of y + along the rotor for viscous and inertial regimes

�
τw
τw = ρkCµ1/2  k from transport equations  Ut = u + ρ
@ �

@ �
R
@ √ �
Δy τw /ρ -
y+ = µ/ρ
u+ = κ1 ln (y + ) + C

Figure 3.5: Wall function method.

23

3.7 Code Validation

Two test cases have been used to validate the commercial code against experimental results for
rotating flows. Both of these experimental studies were performed at the University of Bath and
so a comprehensive set of conditions and results are available.

3.7.1 Rotor-Stator with Radial Outflow

The fluid dynamics of the rotor-stator system with radial outflow has been investigated
experimentally and computationally by Chen et al. (1996). A schematic of the geometry used
is shown in figure 3.6.

The stator was fabricated from transparent polycarbonate allowing optical access such that
LDA measurements of the radial and circumferential velocity components could be made. The
rotor, a steel disc with a glass fibre covering on the inside face, was heated by an array of
heater elements. Thermocouples and fluxmeters were embedded in the glass fibre to measure
temperature and heat transfer. Flow enters the wheel-space through a pipe along the axis of
rotation, and then flows radially outwards through a porous layer. An outer shroud is attached
to each of the discs, which together form the outer radial casing of the wheel-space. The gap
between these shrouds is the outlet for the system.

The computational results are presented using both the high Reynolds number k-� model and
the BSL model (blended k-� / k-ω). Velocity profiles are shown in figure 3.8 for Reφ = 1.25 ∗
106 and cw = 6, 100. Figures 3.8(a)-(d) are the radial velocity distributions (vr /Ωr) between
the rotating discs at four radial locations. The computational predictions are improved at the
locations of greater radius, inaccuracies at low radius are perhaps due to the influence of the
porous layer in the experiment which is not modelled computationally.

At low values of z/s the velocity peak representing the radial jet on the rotor can be seen. The
computations using the k-� model overpredict this feature at each location. This is attributed to
the general overprediction of turbulence kinetic energy by k-� turbulence models, which then
leads to higher than expected near wall velocities. The BSL model is in very good agreement
with the experimental results at this radial velocity peak.

The k-� results show a deficit in the radial velocity component at the edge of the boundary layer
(for example z/s = 0.3 at x = 0.80). This feature is not predicted by the BSL model, which is
again in close agreement with the measurements, with a maximum error of 5%.

24
As z/s increases the measured radial velocity becomes zero. This is consistent with the
predictions from both turbulence models. Near to the stator another peak in the velocity occurs,
this time the flow is inward toward the axis of rotation. Again this feature, at least at high radii,
is predicted within 10% by both turbulence models.

Figures 3.8(e)-(h) show the circumferential velocity distributions (vφ /Ωr) between the discs.
Similar to the radial component, an improvement in prediction is seen at higher radii where the
flow is less dependent on the structure in the source region. Both turbulence models overpredict
the flow velocity very close to the rotor.

Towards the edge of the boundary layer, which appears much thinner in the tangential direction
than the radial direction, the k-� model predicts a significant deficit in velocity of approximately
20% at high radius. Measurements show the existence of this feature and confirm that the deficit
is approximately 15%-20% below the core circumferential velocity. This core rotation velocity
is accurately predicted by the BSL model but under predicted by 20% by the k-� model. In this
case the values of vφ /Ωr are fairly low, this is due to the moderately high superposed flow rate.
With no through flow (cw = 0) the normalised core circumferential velocity is approximately
0.45 at high radius and slightly lower (≈ 0.40) at lower radii.

On the stator side of the cavity the circumferential velocity drops to zero. Both turbulence
models predict slightly too much momentum in the near wall region, but the agreement is within
10% away from the boundary layer.

In order to compute the heat transfer in the system, it is necessary to apply an assumed
temperature distribution on the rotor. Discrete measurements were taken during the experiment,
this is represented in the computation by a radial temperature distribution. A constant
temperature was used for the range 0.0 > x > 0.38 and a fifth order polynomial has been
fitted between 0.38 > x > 1.0. The distribution can be seen in figure 3.9.

Figure 3.10 shows the non-dimensional wall heat flux from the rotor. Four datasets are shown;
the two experimental sets are the uncorrected and corrected results. The correction is calculated
by evaluating the radiative heat flux, this is subtracted from the fluxmeter measurements such
that the Nusselt number includes only the convective component of heat flux.

Generally the Nusselt number increases with radius, a maximum occurs at x ≈ 0.82 and a
sharp drop occurs close to the outer shroud. The primary mechanism for the increase in Nusselt
number is the development of the turbulent boundary layer on the rotor. It should also be
remembered that the Nusselt number also contains an r term, so would increase with radius

25

even if the heat flux remained constant.

The computational results should be compared with the corrected results as the code is only
configured to evaluate the convective component of heat transfer. Throughout the radial range
the k-� model gives approximately 5% higher heat transfer predictions than the BSL model.
This is consistent with the evidence that the k-� overpredicts the turbulence kinetic energy in
the boundary layer. The shape of the two computational heat transfer distributions is the same,
and follows that of the experimental results. The agreement between the computations and
corrected experiments is generally good, with a maximum discrepancy of 10%.

3.7.2 Coverplate

The coverplate system has been investigated by Pilbrow et al. (1999). The system is made of
three component regions: a source region, rotor-stator cavity and rotating cavity. A schematic
is shown in figure 3.7. The experiment used a steel rotor with thermocouples and flux meters
embedded in a fibre-glass ‘mat’ mounted on the front. Stationary heaters were used to produce
a temperature distribution across the disc. Cool air enters the system through the pre-swirl
nozzles, which are stationary and at a low radius. The main flow path exits through receiver
holes at high radius.

To model this system computationally, the receiver holes and pre-swirl nozzles were both
assumed to be axisymmetric slots. This reduces the computation to a steady state and,
essentially, 2D axisymmetric simulation. In practice, the solver being used does not have any
2D capability, therefore a small angular sector is represented. The sealing flow, which exits
through the rotor-stator wheel-space, is assumed to be zero. Therefore only the source region,
at low radius, and the rotating chamber on the right of figure 3.7 is modelled.

In a similar way to the rotor-stator system described above, a temperature distribution must
be assumed on the rotor surface in order to produce heat transfer predictions. Figure 3.11
shows the discrete measured data points and the fitted fifth order polynomial curve used as a
boundary condition in the computation. The conditions for this study were Reφ = 1.37x106 ,
cw = 28, 354, λT = 0.349 and βp = 2.61.

Figure 3.12 shows a comparison of the measured radial variation of Nusselt number with the
computational predictions using the two turbulence models. There are a limited number of
experimental data points, all four of which are in the turbulent boundary layer, at a higher
radius than the impingement from the pre-swirl nozzles (x = 0.52), and at a lower radius than

26

the receiver holes (x = 0.97). These experimental data points show that the Nusselt number
distribution should be approximately constant, with a magnitude of 1400 throughout the range
0.65 < x < 0.90.

Pilbrow et al. (1999) investigated the configuration computationally using two different low
Reynolds number k-� models, that of Launder and Sharma (1974) and that of Morse (1988).
The results they achieved are very similar to those from the high Reynolds number k-� model
presented here, except their impingement peak was slightly greater.

The k-� results produced here show relatively good heat transfer magnitudes in the turbulent
boundary layer region, although the secondary peak predicted at x = 0.67 is not a feature in
the experimental results. The BSL model generally underpredicts the measurements by up to
25%, but arguably shows a more constant distribution throughout the turbulent boundary layer
region.

rotating disc

stationary disc

axis of rotation

Figure 3.6: Rotor-stator with radial outflow.

27

stationary disc

rotating disc

axis of rotation

Figure 3.7: Coverplate model.

28

 (a) x = 0.60 (e) x = 0.60

0.15 0.50

0.45

0.40
0.10
Experiment
Computation (k-ε) 0.35
Computation (BSL)
0.30

vφ / Ωr
vr / Ωr
0.05 0.25

0.20

0.15
0.00
0.10

0.05

-0.05 0.00
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
z/s z/s
(b) x = 0.70 (f) x = 0.70
0.15 0.50

0.45

0.40
0.10
0.35

0.30

vφ / Ωr
vr / Ωr

0.05 0.25

0.20

0.15
0.00
0.10

0.05

-0.05 0.00
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
z/s z/s
(c) x = 0.80 (g) x = 0.80
0.15 0.50

0.45

0.40
0.10
0.35

0.30
vφ / Ωr
vr / Ωr

0.05 0.25

0.20

0.15
0.00
0.10

0.05

-0.05 0.00
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
z/s z/s
(d) x = 0.85 (h) x = 0.85
0.15 0.50

0.45

0.40
0.10
0.35

0.30
vφ / Ωr
vr / Ωr

0.05 0.25

0.20

0.15
0.00
0.10

0.05

-0.05 0.00
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
z/s z/s

Figure 3.8: Experimental (symbols) and computational (lines) radial and circumferential
velocity distributions for the rotor-stator system with radial outflow. Reφ = 1.25x106 and
cw = 6, 100.

29
 400
Experiment: Rotor Measurements
Computation: Fitted Distribution

380

Rotor Temperature, Tw [K]

360

340

320

300
0.0 0.2 0.4 0.6 0.8 1.0
Radius, x

Figure 3.9: Rotor-stator system: Rotor temperature distribution.

1200
Experiment: Corrected Measurements
Experiment: Uncorrected Measurements
Computation: BSL Turbulence Model
Computation: k-ε Turbulence Model
Nusselt number, Nu

800

400

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Radius, x

Figure 3.10: Rotor-stator system: Radial Nusselt number distribution.

30

 330
Experiment: Rotor Measurements
Computation: Fitted Distribution
325

Rotor Temperature, Tw [K]

320

315

310

305

300

295

290
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Radius, x

Figure 3.11: Coverplate system: Rotor temperature distribution.

3000
Experiment
Computation (BSL model)
Computation (k-ε model)
2500
Nusselt Number, Nu

2000

1500

1000

500

0
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Radius, x

Figure 3.12: Coverplate System: Nusselt number distribution. Reφ = 1.37 ∗ 106 , cw = 28, 354,

λT = 0.349 and βp = 2.61

31

Chapter 4

Pre-Swirl System

The pre-swirl geometry, as discussed previously, is a representation of the cavity which may be
found in an engine between the stator and rotor discs. The model includes pre-swirl nozzles
at low radius located in the stator disc and blade receiver holes at higher radius on the rotor
disc. For this configuration rp /rb = 0.8. The flow is swirled in an attempt to reduce the total
temperature of the flow when it reaches the blades by increasing the adiabatic effectiveness of
the system. The amount of cooling flow available to the blades can also be affected by the
geometry and flow conditions for a given pressure difference.

The following cases have been computed. The conditions were chosen to coincide with the
experimental results available.

Case Ω[rpm] Reφ /106 λT cw βp To,p [K] Patm [P a]

1 3010 0.78 0.127 6600 0.52 328.6 982
2 3750 0.97 0.123 7600 0.48 330.5 991
3 4500 1.21 0.130 9500 0.48 324.7 995
4 3002 0.78 0.235 12200 0.96 327.9 985
5 3748 0.96 0.243 14800 0.95 331.0 983
6 4500 1.19 0.251 18200 0.94 327.5 990
7 3003 0.79 0.369 19200 1.49 329.2 996
8 3762 0.97 0.353 21800 1.37 322.6 999
9 4500 1.18 0.376 27200 1.41 330.1 995

Table 4.1: Conditions for experiments and computions

The λT values classify cases 1−3 as being in the viscous regime and cases 4−9 as being within

32

the inertial regime. These regimes are defined by whether the flow from the pre-swirl nozzle
has enough momentum to ‘punch’ through the core and cause an impingement region on the
opposite disc. As discussed above the turbulent flow parameter, λT , and pre-swirl ratio, βp , are
coupled due to the fixed inlet angle of the pre-swirl nozzles. In order to increase the tangential
velocity at the inlet the flow rate must be increased, which in turn impacts λT .

This relationship between the governing parameters for a fixed inlet angle is highlighted in
figure 4.1. Typically in an engine λT ≈ 0.4, implying pre-swirl ratios of just greater than
unity. In the experiment, with lower Reynolds number, the same conditions can not be acheived
therefore it is important to understand how each of these parameters affects the flow.

2.5
λT = 0.4

Experiment Engine
2.0

λT = 0.3
Pre-Swirl Ratio, βp

1.5

λT = 0.2

1.0

λT = 0.1
0.5

0.0 5 6 7
10 10 10
Reynolds Number, Reϕ

Figure 4.1: βp variation with Reφ based on a fixed inlet angle of 20◦ .

4.1 Experimental Results

The experimental results were produced by Yan et al. (2003) and Lock et al. (2005a,b) but the
salient points of the experimental method are presented here for convenience.

The rotor is a transparent polycarbonate disc with a radius of 0.216 m, allowing optical access

33

to the wheel space. The disc has 60 circular receiver holes with centres at a disc radius of 0.200
m. To reduce the heat transfer from the air in the receiver holes, the holes are filled with opaque
lightweight machinable Rohacell (low-conducting foam) bushes producing an effective receiver
hole diameter of 8.0 mm. The disc has a thickness of 10 mm, and the receiver holes, which have
a length-to-diameter ratio of 1.25, vent directly into the laboratory. A shroud of carbon fibre
surrounds the rim of the disc and rotates with it.

The stator is also a polycarbonate disc, which is in turn mounted onto an aluminium disc. The
gap between the rotor and stator is 11 mm (G = 0.051). The pre-swirl nozzles comprise 24
circular holes, of 7.1 mm diameter, drilled at an angle of 20◦ to the tangential direction and at a
radius of 160 mm. The stator has a Rohacell shroud which is aligned with the carbon fibre one
on the rotor. A 1.0 mm gap exists in the centre between the shrouds. The air pressure in the
wheel space is balanced by sealing air on the outside of this gap to restrict leakage or ingress.
A stationary Rohacell hub forms the inner boundary of the wheel space at a radius of 0.145 m.

The main air supply to the system is passed through a mesh heating element which creates
a step change in temperature. A strobe light is used to freeze the rotor and the resulting
transient disc temperature distribution is captured on video at 25 frames per second. The
RGB (red-green-blue) signals of each frame are converted to hue and analysed to calculate
the temperature and heat transfer coefficient.

Numerous experimenters have used thermochromic liquid crystal (TLC) to determine heat
transfer coefficients on purpose-built test sections. A common technique, which is used here, is
to solve Fourier’s transient conduction equation to calculate h for a semi-infinite solid exposed
to a step-change in air temperature. As it is virtually impossible to achieve a step-change in the
air temperature of pre-swirl rigs, Newton et al. (2003) developed the so-called ‘slow transient’
technique. Lock et al. (2005a,b) used this technique to measure the local Nusselt numbers on
the rotating disc.

4.2 Fluid Dynamics

4.2.1 Velocity and Pressure

Figure 4.3 shows a comparison between the computed and measured radial variation of β
(= vφ /Ωr), the non-dimensional swirl ratio, at the mid-plane (z/s = 0.5). The maximum

34

 Figure 4.2: Experimental setup: Illustration.

value of β, which occurs at the inlet radius due to the flow from the nozzles, is well predicted
by the computations. However, at the larger radii, the computations over-predict by 20% the
measured values. This is likely to be due to the over-prediction of turbulence kinetic energy by
the high-Reynolds-number turbulence model, therefore increasing the tranfer of momentum to
the flow.

The experiments contain a feature at a radius of x = 0.81 which is not apparent in the
computations, a local increase in β occurs. This may be an effect caused by the complex mixing
of the jet from the pre-swirl nozzles, since it is more apparent at high βp when the jet is stronger.
The same flow structure may not be captured in the computation due to the modelling of the
inlet as a slot. Alternatively it may highlight an error in the measuring of the total pressure at
this location.

Figure 4.3 demonstrates that the flow structure is independent of Reφ , the magnitude and
distribution of β does not change for the range 0.8 ∗ 106 < Reφ < 1.2 ∗ 106 .

Figure 4.4 shows good agreement between the computed and measured radial distribution of
static pressure. In the rotating core of fluid, away from the rotor and stator, dp/dr = ρvφ2 /r
(Owen and Rogers, 1995), and as a consequence the static pressure increases radially. The
over-prediction of the total pressure in figure 4.5 is caused by the over-prediction of β referred
to above.

The agreement between computations and measurements is best for the lowest value of λT
shown, for which the flow is in the viscous regime. The computed mixing at the higher
values of λT (for which the flow is in the inertial regime) may be affected by the use the
high-Reynolds-number turbulence model and also by the use of the simplified slot geometry

35

at inlet. (Yan et al. (2003) obtained better agreement with measurements than that shown in
figure 4.3 using a discrete inlet and a low-Reynolds-number k-� turbulence model.)

The rotor-stator system studied to validate the code, which had no pre-swirl, gave excellent
predictions of vφ . This is consistent with the data in figure 4.3; the computations give improved
predictions for lower values of βp .

4.2.2 Adiabatic Effectiveness

The adiabatic effectiveness, Θb,ad , is defined in equation 4.1. It is a measure of the work done
on, or by, the flow between the pre-swirl inlet and receiver holes. At low pre-swirl ratios, when
the flow is travelling slower than the rotor disc, the rotor disc will accelerate the flow. This
increases the total temperature of the blade-cooling air, Tt,b , and reduces the effectiveness. The
opposite effect occurs for high pre-swirl flows; the rotor disc slows the flow and is effective in
reducing the total temperature. This effectiveness is discussed in further detail in section 5.1.3.

cp (To,p − Tt,b )
Θb,ad = (4.1)
1/2Ω2 rb2

For given inlet conditions the total temperature of the air in the rotating receiver holes, Tt,b ,
decreases linearly as Θb,ad increases.

Karabay et al. (2001) derived a theoretical value for Θb,ad using the First Law of
Thermodynamics; the work done on, or by, the air was proportional to the moment required
to change the tangential velocity of the air from βp Ωrp at the pre-swirl nozzles, to Ωrb in the
receiver holes. Their equation, which was derived for a cover-plate system (this system was
described in section 3.7) in which there is no stator to reduce the swirl, is given by equation 4.2.
They performed a joint experimental and computational study which showed good agreement
for the range 0 ≤ βp ≤ 3.

� �2
rp
Θb,ad = 2βp −1 (4.2)
rb

This equation was modified independently by Farzaneh-Gord et al. (2005) and Chew et al.
(2005) to account for the moment on the stator which acts to reduce the effectiveness of the
system. Their model gives

36
 � �2
rp Ms
Θb,ad = 2βp −1− (4.3)
rb 1/2ṁΩrb2

where Ms in the moment on the stator disc.

Figure 4.6(a) shows a comparison between the computed and theoretical values of Θb,ad
for the present case where rp /rb = 0.8. The computations based on equation 4.1 use the
bulk-average relative total temperature computed at the outlet from the receiver holes. These
computations are in excellent agreement with the theoretical values of Θb,ad based on equation
4.3. The theoretical values for the cover-plate system are significantly higher than those for the
direct-transfer system.

The evaluation of this parameter is sensitive to the conditions within the receiver hole. In a
long enough blade-cooling passage solid body rotation would be achieved at some downstream
location. This may occur abruptly if the passage turns a corner, in which case the flow is subject
to normal forces rather than merely viscous damping. Figure 4.6(b) shows that the outlet swirl
ratio is indeed very close to unity in the computation. Experimentally this is not measured, but
since the receiver hole is relatively short L/D = 1.25, solid body rotation may possibly not be
achieved.

It should be noted that both the computed and theoretical values rely on the computational
prediction of Ms . It is therefore possible that the values of Θb,ad which could be measured
experimentally would not agree with these. The measurement of Θb,ad is nontrivial; the
conditions in the receiver holes must be measured accurately and the wheel-space must be
thermally insulated to restrict heat transfer to or from the flow.

An extension of the computational study to include higher values of βp is shown in figure 4.6(a).
This demonstrates that the linear relationship between βp and Θb,ad is valid for βp ≤ 3.

4.2.3 Discharge Coefficient

The discharge coefficient is defined as the ratio of the actual mass flow rate to the isentropic
mass flow rate, as shown in equation 4.4. For the current system it would be possible to
define several different discharge coefficients; for the pre-swirl nozzles, wheel-space, receiver
holes, and another for the entire system. In this study the parameter of interest is the discharge
coefficient for the receiver holes as in equation 4.4. In essence the discharge coefficient, which
is flow and geometry dependent, represents the loss in mass flow rate for a given pressure drop

37

over the orifice. A discharge coefficient of 1 would represent an isentropic system.

ṁb
Cd,b = (4.4)
ṁi

The isentropic mass flow, ṁi , for a rotating system can be calculated from the First Law of
Thermodynamics. Yan et al. (2003) derived equation 4.5, where locations 1 and 2 represent
upstream and downstream locations within a streamtube. For the receiver holes these points are
located in the mid-plane (z/s = 0.5) and at the outlet respectively. The radial position of both
locations is r = 0.20 m.

� � 1 ⎡� � ⎧ � � γ−1 ⎫ ⎤ 12
ṁi p2 γ
2γ
p ⎨ p γ ⎬
⎣ 0,1 2 2 ⎦
= ρ0,1 1− + 2Ω(r2 Vφ,2 − r1 Vφ,1 ) − Vφ,2 (4.5)

A2 p0,1 γ − 1 ρ0,1 ⎩ p0,1 ⎭

Equation 4.5 contains three terms inside the square brackets; the first is the general term for
isentropic expansion through a nozzle. The second term is the work term resulting from the
change in angular momentum of the air (note for the receiver holes no change in radius occurs,
r1 = r2 ). The final term is due to the fact that the air in the receiver holes has an absolute
tangential, as well as axial, velocity.

The experimental values of Yan et al. (2003) and the computations presented here use the same
locations and measurements for all properties. vφ,1 and p0,1 are taken at in the mid-plane,
z/s = 0.5, and at the receiver hole radius, r = rb . p2 is atmospheric pressure measured
during the experiment, and then prescribed at the outlet for the computation. vφ,2 = Ωrb in
both cases. Figure 4.6(b) shows that this assumption of solid-body rotation is reasonable; β2 is
approximately unity for all βp in the range of interest.

Figure 4.6(c) shows that the measured values of Cd,b are over-predicted by the computations,
although both data sets show that Cd,b increases with βp over the range plotted. An analysis
of the terms in equation 4.5 shows that this over-prediction is due to the over-prediction of the
swirl ratio in the wheel-space, β1 .

Dittmann et al. (2002) measured Cd,b and showed that a maximum should occur when β1 = 1,
i.e. when the core flow is in synchronous rotation with the receiver hole. The critical value of
βp , for which this occurs has practical significance. Although Θb,ad increases as βp increases,
ṁb will decrease for βp > βp,crit . In most engines, where βp < 1, this is unlikely to be a
problem.

38
Figure 4.6(c) shows that Cd,b does indeed reach a maximum. The pre-swirl ratio at which this
occurs is βp = 1.8, which does coincide with β1 = 1 (as shown in figure 4.6(b)). Remembering
that the computations over-predict β, in reality βp,crit will be greater than 1.8. Yan et al. (2003)
did not measure a maximum value of Cd,b even though, with a 12 nozzle configuration, they
tested up to βp ≈ 2.8. At these conditions β1 = 0.94, just below that required for βp,crit .

In order to achieve high values of βp , with a fixed inlet geometry, it is necessary to use large
mass flows through the system. This fluid will have a high radial velocity at the rotor, which also
implies a strong radial component of velocity across the receiver hole. It is therefore postulated
that the maximum value of Cd,b should actually occur when the resultant velocity relative to the
receiver hole is at a minimum, rather than just when the tangential velocity is synchronous with
the hole.

39

 (a)
1.5 βp = 0.52; λT = 0.127
βp = 0.96; λT = 0.235
βp = 1.49; λT = 0.369

1.0

Swirl Ratio, β
0.5

inlet outlet

0.0
0.6 0.7 0.8 0.9 1.0
(b) Radius, x
1.5 βp = 0.48; λT = 0.123
βp = 0.95; λT = 0.243
βp = 1.37; λT = 0.353

1.0
Swirl Ratio, β

0.5

0.0
0.6 0.7 0.8 0.9 1.0
(c) Radius, x
1.5 βp = 0.48; λT = 0.130
βp = 0.94; λT = 0.251
βp = 1.41; λT = 0.376

1.0
Swirl Ratio, β

0.5

0.0
0.6 0.7 0.8 0.9 1.0
Radius, x

Figure 4.3: Radial distribution of swirl ratio: Computation (lines) and experiment (symbols).
(a) Reφ = 0.8 ∗ 106 ; (b) Reφ = 1.0 ∗ 106 ; (c) Reφ = 1.2 ∗ 106 .

40
 (a)
0.6 βp = 0.52; λT = 0.127
βp = 0.96; λT = 0.235
βp = 1.49; λT = 0.369

Static Pressure, (p-pp)/(½ρΩ2r2p)

0.4

0.2

0.0
0.6 0.7 0.8 0.9 1.0

-0.2
(b) Radius, x
0.6 βp = 0.48; λT = 0.123
βp = 0.95; λT = 0.243
βp = 1.37; λT = 0.353
Static Pressure, (p-pp)/(½ρΩ2r2p)

0.4

0.2

0.0
0.6 0.7 0.8 0.9 1.0

-0.2
(c) Radius, x
0.6 βp = 0.48; λT = 0.130
βp = 0.94; λT = 0.251
βp = 1.41; λT = 0.376
Static Pressure, (p-pp)/(½ρΩ2r2p)

0.4

0.2

0.0
0.6 0.7 0.8 0.9 1.0

-0.2
Radius, x

Figure 4.4: Radial distribution of static pressure: Computation (lines) and experiment
(symbols). (a) Reφ = 0.8 ∗ 106 ; (b) Reφ = 1.0 ∗ 106 ; (c) Reφ = 1.2 ∗ 106 .

41
 (a)
1.0 βp = 0.52; λT = 0.127
βp = 0.96; λT = 0.235
0.8 βp = 1.49; λT = 0.369

0.6

Total Pressure, (p0-p0,p)/p0,p

0.4

0.2

0.0
0.6 0.7 0.8 0.9 1.0
-0.2

-0.4

-0.6

-0.8

-1.0
(b) Radius, x
1.0 βp = 0.48; λT = 0.123
βp = 0.95; λT = 0.243
0.8 βp = 1.37; λT = 0.353

0.6
Total Pressure, (p0-p0,p)/p0,p

0.4

0.2

0.0
0.6 0.7 0.8 0.9 1.0
-0.2

-0.4

-0.6

-0.8

-1.0
(c) Radius, x
1.0 βp = 0.48; λT = 0.130
βp = 0.94; λT = 0.251
0.8 βp = 1.41; λT = 0.376

0.6
Total Pressure, (p0-p0,p)/p0,p

0.4

0.2

0.0
0.6 0.7 0.8 0.9 1.0
-0.2

-0.4

-0.6

-0.8

-1.0
Radius, x

Figure 4.5: Radial distribution of total pressure: Computation (lines) and experiment (symbols).
(a) Reφ = 0.8 ∗ 106 ; (b) Reφ = 1.0 ∗ 106 ; (c) Reφ = 1.2 ∗ 106 .

42
 3.0
(a)

2.5

2.0

1.5

Θb,ad
1.0

0.5

0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
βp
-0.5
Computed (eqn. 4.1)
Coverplate System (eqn. 4.2)
-1.0 Direct Transfer System (eqn. 4.3)

1.8
(b)
1.6
β1 (computed)
β2 (computed)
1.4

1.2

1.0
β

0.8

0.6

0.4

0.2

0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
βp
0.7
(c)

0.6

0.5

0.4
Cd,b

Measured (N=24)
0.3
Measured (N=12)
Computed
0.2

0.1

0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
β1

Figure 4.6: Variation of Θb,ad ,β and Cd,b with βp for 0.8 ∗ 106 < Reφ < 1.2 ∗ 106 and 0.12 <
λT < 0.38. (a) Comparison between computed and theoretical Θb,ad . (b) Comparison between
computed and measured β1 and β2 . (c) Comparison between computed and measured Cd,b .

43
4.3 Heat Transfer

4.3.1 Radial Heat Transfer

The computed heat flux is non-dimensionalised to form the local Nusselt number, equation 4.7,
based on the adiabatic wall temperature formulation by Karabay et al. (2001), shown in equation
4.6.

2 � �
Vφ,∞ Ω2 r2 Vφ,∞ 2
Tw,ad = To,p − +R 1− (4.6)
2Cp 2Cp Ωr

qw r hr
Nu = = (4.7)
k (Tw − Tw,ad ) k

This adiabatic wall equation is based on the Reynolds analogy and is found to be in good
agreement with computations by Karabay et al. (2001) for the cover-plate system and with
additional adiabatic computations performed in this study for the direct-transfer system.
Experimentally this calculation for Tw,ad has been validated by Newton et al. (2003), who used
wide-band TLC to measure the adiabatic wall temperature. The measurements and theory were
generally within 0.5◦ C. Results of Nusselt number presented here use the theoretical result of
Tw,ad rather than the experimental measurement or computed value.

Lock et al. (2004) presented data of Nusselt number versus radius and described the differences
between the inertial and viscous regimes. Figure 4.7(a) shows computed and measured
distribution of Nu for the viscous regime, βp ≈ 0.5 and λT ≈ 0.12. The difference in level
between the computed heat transfer and experiment is clear, but the features of the distributions
are in good agreement. A peak in heat transfer occurs at the receiver hole radius (x = 0.93)
but there is no sign of impingement in the region of the pre-swirl nozzles (x = 0.74). The
difference in absolute level may be due to over-prediction of turbulence intensity near the wall
discussed earlier.

The data follows a power law with respect to Reynolds number and collapses to a single line
when scaled by Re0φ.8 . This shows that the flow is controlled by turbulent boundary-layer flow
and is consistent with free-disc case, N u ∝ Re0.8
φ (Owen and Rogers, 1995).

In figure 4.7(b) the experimental results show a distinct peak near the pre-swirl nozzles,
signifying that the flow is in the inertial regime. The computational results show only a small
peak at this radius. For these conditions the parameter N uRe−
φ
0.8
only collapses the data in the

44

region of the receiver holes, not the impingement region at the pre-swirl nozzle radius. The
magnitude of Nusselt number at this intermediate level of λT is in better agreement than the
inertial regime, but the computations do not predict the features of the distribution as accurately.

The impingement peak is apparent in the computational results in figure 4.7(c) for which λT ≈
0.35 and βp ≈ 1.5. The radial variations of the computations and measurements are qualitatively
similar, but here the parameter N uRe−
φ
0.8
does not collapse the data well at any radius.

4.3.2 Circumferential Heat Transfer

Figures 4.8(a)-(f) show comparisons for Nusselt number across an 18◦ sector of the rotor, as
studied experimentally. Results on the left were produced by experiment and those on the right
are computational results. The conditions for case 4.8(a) & (d) classify them within the viscous
regime; the two other examples, 4.8(b) & (e) and 4.8(c) & (f), relate to the inertial regime.

Comparing figure 4.8(a) with the results in figure 4.7(a), the same difference in Nusselt number
magnitude is visible. The improved agreement at the receiver hole radius is also apparent. In
the region close to the edge of the hole there is an absence of experimental data caused by the
presence of the Rohacell bush described in section 4.1. (Since the bush is opaque it also causes
shadows over the otherwise transparent rotor, and obscures the results in this region.)

In figures 4.8(b) and 4.8(c), representing the inertial regime, the experiments and computations
agree well, except for the impingement region opposite the inlet, which is not captured by the
computation.

A small region of high heat transfer is observable around the receiver holes in each case in
figure 4.8. At low λT and βp this region is located at the ‘9 o’clock’ position with respect to
the holes, figure 4.8(d). As λT and βp are increased the region moves around towards the ‘11
o’clock’ position, figure 4.8(f). Luo et al. (2004) performed temperature measurements around
a rotating disc with receiver holes and found a similar region of high heat transfer around the
holes. (In an engine, high heat transfer in this region could result in thermal stresses within the
rotor.)

45

4.3.3 Heat Transfer - Fluid Dynamics Interaction

Based on results from axisymmetric computations, Wilson and Owen (1994) showed that air
entered the receiver holes by ‘direct’ and ‘indirect’ routes. The former refers to flow travelling
directly along a streamline connecting the inlet and the outlet, and therefore not mixing with
the core flow. Indirect flow mixes with the core flow before entering the receiver holes.

This idea can be extended to the study of non-axisymmetric systems by considering that the
direct flow travels in a stream-tube between the pre-swirl nozzles and receiver holes. Computing
the streamlines for the direct flow allows the path of the stream-tube to be evaluated. Figure 4.9
shows the stream-tube relative to the rotor for a variety of conditions. The inner location is at
the radius of the pre-swirl nozzles, and the outer location is at the radius of the receiver holes.
These results show only a weak effect of Reφ , and have an analogy with the calculation of Cd,b .
It has been shown that Cd,b is a maximum when β1 = 1, this would occur when the streamlines
in figure 4.9 point in the radial direction at the outlet radius (i.e. having only a component of
radial velocity with respect to the rotor).

Figure 4.10 shows measured heat transfer results combined with streamlines calculated using
the full velocity field from the computations. Figure 4.10(a) is a radial section with the pre-swirl
inlet and stator on the left and the receiver hole and rotor on the right. The orange streamline
shows that flow from the nozzle can be either direct or indirect: The direct flow exits through
the receiver hole; the indirect flow continues to a higher radius and will recirculate in the system
and mix with the core flow. The black streamline shows that indirect flow, which has entered
the core, can either exit through the receiver hole or continue circulating in the core.

Figure 4.10(b) shows the same streamlines in a circumferential section. It can be seen that
the flow at the rotor surface that is aligned with the receiver holes becomes direct flow. The
remaining flow follows the indirect route. The black streamlines show that the core flow
replaces the boundary layer flow that has entered the receiver holes.

Figure 4.10(c) shows an isometric view of the same streamlines. It is flow from the core,
replacing the boundary layer flow entering the receiver holes, which gives rise to the region of
high heat transfer.

This variation in flow ‘route’ is also seen to have an impact upstream of the receiver holes.
Figure 4.11 shows the computed Nusselt number distribution for a case with a near-unity
pre-swirl ratio. Streaks are visible on the rotor apparently showing a non-axisymmetric flow
in this region. As the inlet is modeled as an annular slot this effect must be produced by the

46

non-axisymmetric nature of the receiver holes. The angle of the streaks is also consistent with
the direction of the flow across the rotor shown in figure 4.9.

47

 0.040
(a)
Reφ = 0.78 x 106
6
0.035 Reφ = 0.97 x 10
Reφ = 1.21 x 106
0.030

0.025

Nu Re-0.8
0.020

0.015

0.010

0.005
inlet outlet

0.000
0.6 0.7 0.8 0.9 1.0
Radius, x
0.040
(b)
Reφ = 0.78 x 106
6
0.035 Reφ = 0.96 x 10
Reφ = 1.19 x 106
0.030

0.025
Nu Re-0.8

0.020

0.015

0.010

0.005

0.000
0.6 0.7 0.8 0.9 1.0
Radius, x
0.040
(c)
Reφ = 0.79 x 106
6
0.035 Reφ = 0.97 x 10
Reφ = 1.18 x 106
0.030

0.025
Nu Re-0.8

0.020

0.015

0.010

0.005

0.000
0.6 0.7 0.8 0.9 1.0
Radius, x

Figure 4.7: Radial Distribution of N uRe−0.8 : Computational (lines) and Experimental

(symbols). (a) βp ≈ 0.5; λT ≈ 0.13, (b) βp ≈ 1.0; λT ≈ 0.24, (c) βp ≈ 1.5; λT ≈ 0.36.

48
 (a) (d)

pre-swirl inlet radius

(b) (e)

(c) (f) Ω

Figure 4.8: Experimental (a-c) and computational (d-f) Nusselt number contours, Reφ = 0.8 ∗
106 . (a & d) βp = 0.5, λT = 0.13; (b & e) βp = 1.0, λT = 0.24; (c & f) βp = 1.5, λT = 0.37.

49
 Figure 4.9: Computed streamlines relative to the rotor for a variety of conditions.

Figure 4.10: Computed streamlines superimposed onto experimental heat transfer results.
Reφ = 0.8 ∗ 106 , βp = 1.5, λT = 0.38.

50

Figure 4.11: Nu contours, Reφ = 1.19 ∗ 106 , λT = 0.251 and βp = 0.94.

51

Chapter 5

High Radius Pre-Swirl System

In the previous chapter the effect of varying the governing flow parameters was shown for a
generic geometry which had been studied previously in experiments. The ratio of the pre-swirl
nozzle radius to the blade cooling flow receiver hole radius was fixed at rp /rb = 0.8. This
ratio is the key geometric parameter for the system impacting both the flow structure within
the wheelspace and the performance of the system as a whole. According to the theory of
Karabay et al. (2001) increasing the pre-swirl ratio should have a direct effect of the adiabatic
effectiveness of the system.

The domain is shown in figure 5.1. It is identical to that used earlier, except for the addition
of two further annular slots at higher radii. These can be used as inlets, or amalgamated with
the stator surface as required, in order to specify a variety of geometries. Table 5.1 shows the
dimensions of the inlet slots used for the study. Note that the area of each annular slot is constant
in order to maintain the relationship between velocity and mass flow rate at the preswirl inlet.

Inlet Mean Radius [m] Area [m2 ] Inner Radius [m] Outer Radius [m]
A 0.1600 0.0028 0.1586 0.1614
B 0.1800 0.0028 0.1788 0.1812
C 0.2000 0.0028 0.1989 0.2011

Table 5.1: Inlet dimensions.

Table 5.2 shows the conditions to be used for the study. It is divided into two sections; The first
maintains a constant flow rate and focuses on the effect of varying the inlet swirl magnitude,
therefore changing the inlet flow angle as necessary. The second section fixes the inlet flow
angle at 20◦ , as previously, and focuses on the coupled effect of flow rate and swirl ratio. The

52

Reynolds number is constant at Reφ = 106 as the scaling effect of heat transfer with Reynolds
number has already been investigated.

In previous work in which only one location for the pre-swirl nozzles was considered, the
pre-swirl ratio βp was defined as:

vφ,p
βp = (5.1)
Ωrp

In an engine where the total pressure upstream of the pre-swirl nozzles is fixed, the pre-swirl
velocity Vφ,p is approximately invariant with the radial location of the nozzles. Although the
static pressure in the wheel-space varies with radius, this should only have a small effect on the
pressure drop across the nozzles; consequently the pre-swirl velocity should not be significantly
dependent on the radial location of the nozzles.

It is convenient therefore to define a new pre-swirl ratio βb where:

vφ,p
βb = (5.2)
Ωrb

such that βb is invariant with rp . This will make it easier to identify the effect of nozzle location
on pre-swirl performance.

Figure 5.1: Domain for high radius preswirl study.

Figure 5.2 shows the combination of flow rates and swirl ratios used in the study.

53
 Inlet rp /rb βp βb Reφ λT
A 0.8 0.50 0.40 106 0.24
Inlet rp /rb βp βb Reφ λT
A 0.8 1.00 0.80 106 0.24
6
A 0.8 0.50 0.40 106 0.12
A 0.8 1.50 1.20 10 0.24
A 0.8 1.00 0.80 106 0.24
A 0.8 2.00 1.60 106 0.24
A 0.8 1.50 1.20 106 0.36
B 0.9 0.50 0.45 106 0.24
B 0.9 0.44 0.40 106 0.12
B 0.9 1.00 0.90 106 0.24
B 0.9 0.89 0.80 106 0.24
B 0.9 1.50 1.35 106 0.24 6
B 0.9 1.33 1.20 10 0.36
B 0.9 2.00 1.80 106 0.24 6
C 1.0 0.40 0.40 10 0.12
C 1.0 0.50 0.50 106 0.24
6
C 1.0 0.80 0.80 106 0.24
C 1.0 1.00 1.00 10 0.24
6
C 1.0 1.20 1.20 106 0.36
C 1.0 1.50 1.50 10 0.24
C 1.0 2.00 2.00 106 0.24

Table 5.2: Test cases for parametric study.

0.4 0.4

0.3 0.3
λT

λT

0.2 0.2

0.1 0.1
Inlet A
Inlet B
Inlet C

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
βp βb

Figure 5.2: Parameter space for study. Line represents 20◦ nozzle angle.

5.1 Fluid Dynamics

5.1.1 Flow Structure

The variation in radial velocity between the stator and rotor is shown in figure 5.3. The left
hand column are computed at a radius of x = 0.8, midway between the low- and mid-radius

54

inlets. Those in the right hand column are at a radius of x = 0.9, midway between the mid- and
high-radius inlets. The horizontal axis is the axial location, where zero is coincident with the
stator and unity is the rotor.

In general the rotating disc system has an outflow on the rotor and inflow on the stator with
any net flux being due to the mass flow rate through the system. Results for inlets B and C at
x = 0.8 show that the magnitude of the radial velocity at radii lower than that of the inlet are
largely independent of pre-swirl ratio. For inlet A (rp /rb = 0.8) the magnitude of the velocity
increases with increasing pre-swirl ratio.

Figures 5.3(d) and (f) suggest that when rp /rb = 1 inflow on the rotor is possible due to the
inlet flow impinging upon the rotor. Some of this fluid travels radially inwards before meeting
the rotor boundary layer and separating. That this only occurs for particular pre-swirl ratios
suggests some circumferential asymmetry of the flow in the rotor region, likely caused by the
presence of the receiver hole.

The circumferential velocity distribution, computed at the same locations as that for radial
velocity above, is shown in figure 5.4. Part (a) shows that for high radius inlet locations the
core swirl is similar to the theoretical value of 0.43 (Owen and Rogers 1989). The low radius
inlet causes this to reduce due to the radial mass flux past the location.

For high pre-swirl ratios, such as figure 5.4(g), the circumferential velocity of the flow is greater
than that of the rotor. This causes a peak at z = 0.9 and then a tendency towards vφ /Ωr = 1 at
the rotor.

The computed flow structure in the radial (r − z) plane for various pre-swirl ratios are shown in
figures 5.5-5.8. Parts (a)-(c) of each figure represent the inlet located at rp /rb = 0.8, 0.9 and 1.0
respectively. The circumferential location φ of the plane shown coincides with the centre-line of
the receiver holes. There is little effect of circumferential location on the flow structure except
in the immediate region of the receiver hole.

For rp /rb = 0.8, part (a) of each figure, the inlet flow impinges upon the rotating disc and
travels radially outwards, forming the rotor boundary layer. Radial inflow occurs on the stator
and a pair of counter-rotating vortices can be observed inboard of the inlet.

As the inlet is moved radially outwards to rp /rb = 0.9, part (b) of each figure, the circulation
in the outer part of the system becomes more compressed. The pair of counter-rotating vortices
inward of the inlet expands to fill the available space, the larger of the two vortices being that
with outflow on the rotor.

55
The flow is most complex for the case where rp /rb is unity, shown in part (c) of each figure.
Some of the inlet flow enters the receiver holes directly, while the remaining flow impinges upon
the region between the holes. The impinging flow spreads both radially inwards and radially
outwards from the impingement region. The inward flow encounters the rotor boundary layer
flow and separates from the disc, creating the small recirculation on the rotor side inward of the
receiver hole.

Figure 5.9 shows flow streamlines in the tangential (θ − z) plane at the receiver hole radius rb
in a frame of reference rotating at the speed of the rotor (in the left to right direction). In each
image the stator is at the bottom and the receiver hole and outlet is at the top. The three columns
represent the three inlet positions, rp /rb = 0.8, 0.9 and 1.0 respectively, and the rows represent
increasing values of pre-swirl ratio, βp = 0.5, 1.0, 1.5 and 2.0 respectively.

For rp /rb = 0.8, figure 5.9(a) shows a case for which βp = 0.5 and is therefore ‘under-swirled’.
The receiver hole rotates more quickly than the flow in the core, therefore the flow enters at
an acute angle, separating at the leading edge of the hole and causing a recirculation inside the
hole. As the pre-swirl ratio is increased the angle at which the flow enters the receiver hole
tends towards the axial direction. At the point where synchronous rotation between the flow
and the hole occurs the flow would be expected to flow axially into the receiver hole, as can be
seen in Fig. 4(j), for which βp = 2.0.

As the inlet radius is increased, the inlet pre-swirl ratio required to produce this synchronous
rotation is reduced. For rp /rb = 0.9, synchronous rotation occurs when βp = 1.5 as shown in
figure 5.9(h) and for rp /rb = 1.0, when βp = 1.0 as shown in figure 5.9(f). When the swirl
ratio is increased further, or ‘overswirled’, the flow rotates more quickly than the receiver hole,
causing separation and a region of recirculation at the trailing edge of the hole.

The array of streamlines in figure 5.9 can therefore be divided into three groups:

1. Top left (a, b, c, d, e, g): Flow is ‘underswirled’ and enters the receiver hole at a significant
angle from the axial direction.

2. Diagonal (f, h, j): Flow is approximately synchronous with receiver hole and enters
almost axially.

3. Bottom right (i, k, l): Flow is ‘overswirled’ and enters the receiver hole at a significant
angle from the axial direction, but from the opposite direction to the ‘underswirled’ case
above.

56
The variation of swirl ratio β = vφ /Ωr midway between the rotor and stator (z/s = 0.5) and
on a radial line midway between receiver holes is shown in figure 5.10. Figure 5.10(a), (b) and
(c) again correspond to the inlet at rp /rb = 0.8, 0.9 and 1.0 respectively. The horizontal axis is
the non-dimensional radius x = r/b.

In each case a peak is seen in the swirl ratio at the inlet radius due to the high momentum
inlet fluid. At higher radius (i.e. lower values of x−2 ) there is a linear variation of β with x−2 ,
consistent with free vortex behaviour. The dashed line on each plot is a least squares best fit of
the data in the linear region.

There is considerable uncertainty associated with the fit (especially as the region of linear
behaviour becomes smaller), however these results suggest that the flow is related to a Rankine
(combined free and forced) vortex, for which:

β = Ax−2 + B (5.3)

where A and B are constants. This behaviour was found by Mirzaee et al. (1988) to occur in a
rotating cavity with a stationary outer casing.

Figures 5.11(a) and (b) shows the variation of swirl ratio in the core at the radius of the receiver
hole, β∞ , with the pre-swirl ratios at inlet βp and βb . Each line represents a different inlet
location and is approximately linear. Extrapolating back to the βp = 0 condition, the value for
β∞ would be expected to lie between 0.43, the value for turbulent flow in a sealed rotor-stator
system (see Owen and Rogers 1989) and zero, due to the effect of a zeroswirl superposed flow
on the swirl in the rotating core of fluid between the discs. It can be seen that a significant
increase in inlet pre-swirl is required for the inlets at lower radii to achieve the synchronous
rotation condition discussed above, and illustrated by the horizontal dashed line in figures 5.11.
The cases which achieve this synchronous rotation are consistent with those identified in the
streamline plots of figure 5.9.

Figure 5.12 additionally shows the variation for cases where the inlet nozzle angle is fixed at
20◦ and the flow rate, λT is adjusted to generate the appropriate inlet pre-swirl ratio. It can be
seen that the swirl ratio at the receiver holes, β∞ , is largely independent of the flow rate.

57

5.1.2 Discharge Coefficient

The discharge coefficient for the receiver holes has been analysed for the low radius preswirl
nozzles in chapter 4. The definition is recalled as the ratio of the actual mass flow rate through
the receiver holes, ṁb , to the isentropic mass flow rate, ṁi , such that:

ṁb
Cd,b = (5.4)
ṁi

The isentropic mass flow rate through the receiver holes derived by Yan et al. (2003) using the
First Law of Thermodynamics for an adiabatic system taking into account the work done by or
on the fluid as it passes from station 1 in the fluid core to station 2 in the receiver holes. It is
given by:

� � 1 ⎡� � ⎧ � � γ−1 ⎫ ⎤ 12
ṁi p2 γ
2γ
p0,1 ⎨ p2 γ ⎬
⎣ 2 ⎦
= ρ0,1 1 − + 2Ω(r2 Vφ,2 − r1 Vφ,1 ) − Vφ,2 (5.5)

Ab p0,1 γ − 1 ρ0,1 ⎩ p0,1 ⎭

As described earlier, the first term inside the curly brackets is the standard result for
compressible flow in a stationary nozzle; the second term is the work term resulting from the
change of angular momentum of the air; the last term is due to the fact that the air in the receiver
holes has an absolute tangential, as well as an axial, component of velocity. It should be noted
that failure to use the correct equation for ṁi can result in Cd,b values greater than unity, which
is clearly nonsensical.

Figure 5.13 shows this discharge coefficient calculated using locations 1 and 2 as a point in the
core at the radius of the receiver holes and the outlet plane respectively. Figure 5.13(a) and
(b), where βp and βb are on the horizontal axes, clearly shows that to maximise the discharge
coefficient for low radius inlets, the pre-swirl ratio must be greater than unity. The effect of
varying pre-swirl ratio at a fixed non-dimensional flow rate λT (the equivalent of physically
altering the pre-swirl angle) is to produce a sharp change in the relationship between Cd,b and
βp or βb when the maximum value of Cd,b is reached. Figure 5.13(c) shows the swirl ratio in the
core at the receiver hole radius versus Cd,b . The relationship is independent of the inlet radius
and shows a sharp change when β∞ = 1.

The discharge ceofficient for cases with a fixed inlet nozzle angle is shown in figure 5.14. The
peak in Cd,b for these cases is less pronounced than that for the constant flow rate although the
maximum value is similar. Figure 5.14(c) demonstrates the dependence of discharge coefficient

58

on flow rate. At the lowest values of β∞ for variable inlet angle cases λT = 0.24 and for fixed
inlet angle cases λT = 0.12; the lower flow rate corresponding to a lower value of Cd,b .

Lewis et al. (2007) presented measurements, for the rp /rb = 0.8 configuration for a fixed inlet
flow angle of 20◦ to the tangential in the direction of rotation of the disc and variable flow rate.
In this case, the inlet pre-swirl ratio is proportional to the non-dimensional flow rate λT used
(at fixed Reφ ). These measurements are compared with the current computations in figure 5.15.
There is reasonably good agreement between the computations and the measurements for
rp /rb = 0.8 and fixed inlet flow angle.

It was shown above that the effect of ‘under-swirling’ (flow is not swirled sufficiently for the
core flow to be synchronous with the receiver hole) or ‘over-swirling’ (core flow rotating faster
than the receiver hole) the core body of fluid caused the flow to enter the receiver hole at an
angle to the axial direction. This means that the effective area of the receiver hole, as ‘seen’ by
the flow, is reduced. It is logical that, since the flow rate is proportional to the orifice area, Cd,b
will reduce linearly as the effective area is reduced.

The reduction in effective area Ae can be expressed as a function of the flow angle at the receiver
hole, shown in equation 5.6, where α is the flow angle measured from the axial direction.

Ae
= cos α (5.6)
Ab

An equation can be formed for α by considering the ratio of the tangential velocity in the
rotating frame and the axial velocity. The radial component of velocity is ignored as a large
volume of the flow enters the hole from the core rather than from the boundary layer and
therefore has very low radial velocity (see Lewis et al. (2007)) discussion on ‘direct’ and
‘indirect’ routes). Hence:

|Vφ,∞ − Ωrb |
tan α = (5.7)
Vz,b

where:

ṁb
Vz,b = (5.8)
ρAB

and

59

 |Vφ,∞ − Ωb | = |β∞ − 1|Ωrb (5.9)

Using the definition of λT given in the nomenclature it follows that

Cd,b Ae 1
= =� � �2 (5.10)
Cd,b,max Ab Ab ρ|β∞ −1|Ωb
1 + µbλT Re0.8
φ

where Cd,b,max is the value of Cd,b when β∞ = 1.

Figure 5.16 shows Cd,b /Cd,b,max for a range of cases for which λT = 0.24. The model
underpredicts the computational data, which suggests that the predicted flow angle is too large,
so that the predicted effective area is too small. At the entrance to the hole the circumferential
velocity will be somewhere between that of the fluid in the core and that of the receiver hole,
thus using the core cicumferential velocity would be expected to overestimate the angle.

Figures 5.17(a), (b) and (c) show the total pressure loss through the system versus βp , βb and
β∞ respectively. The loss is computed using the total pressure in the stationary frame at inlet
and in the rotating frame at outlet. There is a significant non-linear increase in pressure loss
as the pre-swirl ratio is increased. This relationship, when plotted againts βb , is independent of
inlet radius suggesting it is driven by the absolute level of circumferential velocity at the inlet.
The comparison with the fixed inlet angle results (figure 5.17) shows that this pressure drop is
largely independent of flow rate.

5.1.3 Adiabatic Effectiveness

The adiabatic effectiveness Θb,ad is defined as the nondimensional change in total temperature
between the nozzles in the stationary frame and the receiver holes in the rotating frame:

cp (To,p − Tt,b )
Θb,ad = (5.11)
1/2Ω2 rb2

A theoretical value for Θb,ad was derived by Karabay et al. (2001) for a cover-plate system
using the First Law of Thermodynamics. This equation was modified independently by
Farzaneh-Gord et al. (2005) and Chew et al. (2005) for the direct-transfer system to account
for the moment on the stator which acts to reduce the effectiveness of the system. Their model
gives:

60
 � �2
rp Ms
Θb,ad = 2βp −1− (5.12)
rb 1/2ṁΩrb2

Lewis et al. (2007) expressed this relationship in terms of βb , the pre-swirl ratio based on the
receiver hole radius:

� �
rp Ms
Θb,ad = 2βb −1− (5.13)
rb 1/2ṁΩrb2

Figure 5.18 shows the effectiveness plotted against the relationship in equation 5.13. Note only
discrete values can be evaluated as a computed moment on the rotor disc is required. The
agreement between the two is excellent.

For turbine blade cooling the effectiveness should be as high as possible as this ensures that the
fluid reaching the blades has the lowest possible total temperature. In general, configurations
with low pre-swirl ratios require work to be performed on the flow by the rotor to bring its
tangential velocity to that of the receiver holes, and this work input raises the total temperature
of the flow. Conversely, configurations with high pre-swirl ratios perform work on the rotor,
thus reducing the total temperature.

As shown in equation 5.13, the relationship between the pre-swirl ratio and the adiabatic
effectiveness is approximately linear, with the gradient dependent on rp /rb . As the radius of
the inlet increases there is consequently a resulting increase in effectiveness.

A secondary effect adding to the improvement in effectiveness for high radius inlets is the
reduction in moment on the stator. This is shown in figure 5.19; as the pre-swirl ratio is
increased, the moment on the stator increases and that on the rotor decreases. The change
in moment coefficient is slightly less for rp /rb = 1 than for the other two locations.

Part (c) of the figure shows us that the moments on the discs are equal when the core swirl
ratio is slightly above 0.4. The value for a sealed rotor-stator system is 0.43, and this is reduced
by the presence of a net radial outflow from the system. In this configuration the two lower
inlets may be considered to give a radial mass flux, but the inlet with rp /rb at unity would not.
However the inlet location does not seem to affect this equilibrium value.

61

 1 (a) 1 (b)

0.8 0.8
βp = 0.5 βp = 0.5
x = 0.8 Inlet A x = 0.8
0.6 Inlet B 0.6
Inlet C

0.4 0.4
V r / Ωr

V r / Ωr
0.2 0.2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
z/s z/s
-0.2 -0.2

-0.4 -0.4

1 (c) 1 (d)

0.8 0.8
βp = 1.0 βp = 1.0
x = 0.8 x = 0.9
0.6 0.6

0.4 0.4
V r / Ωr

V r / Ωr
0.2 0.2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
z/s z/s
-0.2 -0.2

-0.4 -0.4

1 (e) 1 (f)

0.8 0.8
βp = 1.5 βp = 1.5
x = 0.8 x = 0.9
0.6 0.6

0.4 0.4
V r / Ωr

V r / Ωr

0.2 0.2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
z/s z/s
-0.2 -0.2

-0.4 -0.4

1 (g) 1 (h)

0.8 0.8
βp = 2.0 βp = 2.0
x = 0.8 x = 0.9
0.6 0.6

0.4 0.4
V r / Ωr

V r / Ωr

0.2 0.2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
z/s z/s
-0.2 -0.2

-0.4 -0.4

Figure 5.3: Normalised radial velocity versus non-dimensional axial distance. Reφ = 106 and
λT = 0.24.

62
 2 2
(a) (b)
1.8 1.8

1.6
βp = 0.5 Inlet A
1.6
βp = 0.5
1.4 x = 0.8 Inlet B 1.4 x = 0.9
Inlet C
1.2 1.2
V φ /Ωr

V φ /Ωr
1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
2 z/s 2 z/s
(c) (d)
1.8 1.8

1.6
βp = 1.0 1.6
βp = 1.0
1.4 x = 0.8 1.4 x = 0.9
1.2 1.2
V φ /Ωr

V φ /Ωr
1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
2 z/s 2 z/s
(e) (f)
1.8 1.8

1.6
βp = 1.5 1.6
βp = 1.5
1.4 x = 0.8 1.4 x = 0.9
1.2 1.2
V φ /Ωr

V φ /Ωr

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
2 z/s 2 z/s
(g) (h)
1.8 1.8
βp = 2.0
1.6
βp = 2.0 1.6
x = 0.9
1.4 x = 0.8 1.4

1.2 1.2
V φ /Ωr

V φ /Ωr

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
z/s z/s

Figure 5.4: Normalised circumferential velocity versus non-dimensional axial distance. Reφ =
106 and λT = 0.24.

63
 (a) (b) (c)

Figure 5.5: Streamlines in r − z plane, βp = 0.5, Reφ = 106 , λT = 0.24. Inlet A, B & C
respectively.

(a) (b) (c)

Figure 5.6: Streamlines in r − z plane, βp = 1.0, Reφ = 106 , λT = 0.24. Inlet A, B & C
respectively.

64
 (a) (b) (c)

Figure 5.7: Streamlines in r − z plane, βp = 1.5, Reφ = 106 , λT = 0.24. Inlet A, B & C
respectively.

(a) (b) (c)

Figure 5.8: Streamlines in r − z plane, βp = 2.0, Reφ = 106 , λT = 0.24. Inlet A, B & C
respectively.

65
 inlet A inlet B inlet C

βp = 0.5
(a) (b) (c)

βp = 1.0
(d) (e) (f)

βp = 1.5
(g) (h) (i)

βp = 2.0
(j) (k) (l)

Figure 5.9: Streamlines in θ − z plane in a rotating frame of reference, Reφ = 106 , λT = 0.24.
Disc direction from left to right.

66

 2
(a)

1.5 βb = 1.60

βb = 1.20
1

β
βb = 0.80

0.5 βb = 0.40

outlet inlet
0
0 0.5 1 1.5 2 2.5
-2
x
2
(b)
βb = 1.80

1.5

βb = 1.35

1
β

βb = 0.90

βb = 0.45
0.5

outlet inlet
0
0 0.5 1 1.5 2 2.5
-2
x
2
βb = 2.00
(c) βb = 1.50

1.5

βb = 1.00

1
β

βb = 0.50

0.5

inlet
outlet
0
0 0.5 1 1.5 2 2.5
-2
x

Figure 5.10: Swirl ratio in the core with linear best fit for the vortex region (a) Inlet A, (b) Inlet
B and (c) Inlet C. Reφ = 106 and λT = 0.24.

67

 (a) (b)
2.5 2.5

Inlet A
2 Inlet B 2
Inlet C

1.5 1.5
β∞

β∞
1 1

0.5 0.5

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
βp βb

Figure 5.11: Swirl ratio in core at rb for λT = 0.24 and Reφ = 106 versus (a) preswirl ratio
based on rp , (b) preswirl ratio based on rb .

(a) (b)
2.5 2.5
Inlet A (λT = 0.24)
Inlet B (λT = 0.24)
Inlet C (λT = 0.24)
o
2 Inlet A (pre-swirl angle = 20 ) 2
Inlet B (pre-swirl angle = 20 o)
o
Inlet C (pre-swirl angle = 20 )

1.5 1.5
β∞

β∞

1 1

0.5 0.5

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
βp βb

Figure 5.12: Swirl ratio in core at rb for Reφ = 106 versus (a) preswirl ratio based on rp , (b)
preswirl ratio based on rb .

68
 1
(a)

Inlet A
Inlet B
0.8 Inlet C

0.6

Cd,b
0.4

0.2

0
0 0.5 1 1.5 2
βp
1
(b)
0.8

0.6
Cd,b

0.4

0.2

0
0 0.5 1 1.5 2
βb
1

(c)
0.8

0.6
Cd,b

0.4

0.2

0
0 0.5 1 1.5 2
β∞

Figure 5.13: Discharge coefficient for receiver holes versus (a) preswirl ratio based on rp , (b)

preswirl ratio based on rb , (c) swirl ratio in the core at rb . λT = 0.24 and Reφ = 106 .

69

 1
(a)
Inlet A (λT = 0.24)
Inlet B (λT = 0.24)
Inlet C (λT = 0.24)
o
Inlet A (Pre-swirl inlet angle = 20 )
0.8 o
Inlet B (Pre-swirl inlet angle = 20 )
Inlet C (Pre-swirl inlet angle = 20 o)

0.6

Cd,b
0.4

0.2

0
0 0.5 1 1.5 2
βp
1
(b)
0.8

0.6
Cd,b

0.4

0.2

0
0 0.5 1 1.5 2
βb
1

(c)
0.8

0.6
Cd,b

0.4

0.2

0
0 0.5 1 1.5 2
β∞

Figure 5.14: Discharge coefficient for receiver holes versus (a) preswirl ratio based on rp , (b)

preswirl ratio based on rb , (c) swirl ratio in the core at rb . Reφ = 106 .

70

 1 rp/rb=0.8 (λT=0.24)
rp/rb=0.9 (λT=0.24)
rp/rb=1.0 (λT=0.24)
o
rp/rb=0.8 (angle = 20 )
o
rp/rb=0.9 (angle = 20 )
0.8 o
rp/rb=1.0 (angle = 20 )
Measurements (Lewis et al.)

0.6
Cd,b

0.4

0.2

0
0 0.5 1 1.5 2
β∞

Figure 5.15: Computed variation of discharge coefficient, Cd,b versus experimental results of
Lewis et al. (2007)

0.8
Cd,b/Cd,max

0.6

0.4

rp/rb=0.8 (λT=0.24)
0.2 rp/rb=0.9 (λT=0.24)
rp/rb=1.0 (λT=0.24)
Analytical relationship

0
0 0.5 1 1.5 2
β∞

Figure 5.16: Computed variation of discharge coefficient, Cd,b versus equation 5.10

71

 0.2
(a)

Inlet A
Inlet B
0.15 Inlet C

(po,p-pt,out)/pt,out
0.1

0.05

0
0 0.5 1 1.5 2
βp
0.2
(b)

0.15
(po,p-pt,out)/pt,out

0.1

0.05

0
0 0.5 1 1.5 2
βb
0.2

(c)

0.15
(po,p-pt,out)/pt,out

0.1

0.05

0
0 0.5 1 1.5 2
β∞

Figure 5.17: Pressure drop throughout system versus (a) preswirl ratio based on rp , (b) preswirl

ratio based on rb , (c) swirl ratio in the core at rb . λT = 0.24 and Reφ = 106 .

72

 3
(a)

Inlet A - Computations
2.5 Inlet B - Computations
Inlet C - Computations
Inlet A - Karabay et al.
2
Inlet B - Karabay et al.
Inlet C - Karabay et al.
1.5

Θb,ad
1

0.5

0
0 0.5 1 1.5 2
βp
-0.5

-1

3
(b)
2.5

1.5
Θb,ad

0.5

0
0 0.5 1 1.5 2
βb
-0.5

-1

(c) 2.5

1.5
Θb,ad

0.5

0
0 0.5 1 1.5 2
β∞
-0.5

-1

Figure 5.18: Comparison of adiabatic effectiveness from computations and correlation of

Karabay et al. (2001) versus (a) preswirl ratio based on rp , (b) preswirl ratio based on rb , (c)

swirl ratio in the core at rb . λT = 0.24 and Reφ = 106 .

73

 0.01
(a)

Inlet A - Rotor
0.008 Inlet B - Rotor
Inlet C - Rotor
Inlet A - Stator
0.006 Inlet B - Stator
Inlet C - Stator

0.004

Cm
0.002

0
0 0.5 1 1.5 2
βp
-0.002

-0.004

0.01
(b)
0.008

0.006

0.004
Cm

0.002

0
0 0.5 1 1.5 2
βb
-0.002

-0.004

0.01

(c)
0.008

0.006

0.004
Cm

0.002

0
0 0.5 1 1.5 2
β∞
-0.002

-0.004

Figure 5.19: Moment coefficients for rotor and stator versus (a) preswirl ratio based on rp , (b)

preswirl ratio based on rb , (c) swirl ratio in the core at rb . λT = 0.24 and Reφ = 106 .

74

5.2 Heat Transfer

This section will examine the heat transfer characteristics created by the use of the three inlet
locations described above. The Nusselt number used is defined consistently with the previous
chapter and is shown in equation 5.15 for clarity. The reference temperature is the formulation
for the adiabatic wall temperature derived by Karabay et al. (2001) and shown in equation 5.14.

� �
Vφ,2 ∞ Ω2 r2 Vφ,∞ 2
Tw,ad = To,p − +R 1− (5.14)
2Cp 2Cp Ωr

qw r hr
Nu = = (5.15)
k (Tw − Tw,ad ) k

Two sample lines are defined on the rotor; the first is a radial line from the inner hub to the rotor
shroud midway between the receiver holes. The second is a circumferential line at a radius of
r = 0.2m, equal to the radial location of the receiver hole centre.

The core velocity term required for the adiabatic wall temperature for the radial line is evaluated
at an equivalent radial location midway between the rotor and stator and between the receiver
holes. For the circumferential calculation a single location at the correct radius, in-line with the
receiver hole centre and midway between the discs is used as Vφ is largely dependent on radial
position and independent of circumferential position.

5.2.1 Radial Variation of Heat Transfer

Figure 5.20 shows the radial distribution of Nusselt number, part (a) is that for the the lowest
radius inlet, rp /rb = 0.8. The flow rate for all cases is constant at λT = 0.24, thus the axial
velocity at inlet is also constant. For pre-swirl ratios up to βp = 1.5 the major characteristics of
the viscous regime are unchanged. An indistinct peak opposite the pre-swirl nozzles, followed
a flat constant region and then a significant peak at the receiver hole radius.

The lowest value of Nu near the inlet occurs for the case of βp = 1.0, by definition this case
has the lowest velocity differential between the inlet flow and the rotor. Consequently it has the
least shear stress on the rotor and therefore reduced heat transfer coefficent.

It has been shown in the previous chapter that the peak in the region of the receiver hole is due to
the three-dimensional nature of the flow as it enters the hole. Recalling figure 5.9, the left hand

75

column shows the reduction in relative tangential velocity as the pre-swirl ratio is increased.
Synchronous rotation is achieved at approximately βp = 2.0, this corresponds to the lowest
Nusselt number in the region of the receiver hole.

Part (b) of figure 5.20 demonstrates a similar pattern for the geometry using the inlet at rp /rb =
0.9. The absolute level of Nusselt number is comparable to that produced by the low radius inlet
and similarly the largest peak at inlet radius is for the case where βp = 2.0. In the region of the
receiver hole a peak occurs for each case except where βp = 1.5. This was identified earlier as
that where synchronous rotation occurs for the mid-level inlet, intuitively this will also have the
lowest level of Nusselt number as shear stress at the wall due to velocity differential will be at
a minimum.

Figure 5.20(c) shows perhaps the most interesting result, that for the inlet and outlet at the same
radius. A peak in the Nusselt number occurs around the receiver hole. The magnitude of the
peak for cases with pre-swirl βp ≤ 1.5 is comparable to that of the lower inlets, but much larger
for the case where βp = 2.0.

A similar set of plots is shown in figure 5.21 for cases where the flow rate parameter, λ, is not
fixed. Instead the pre-swirl flow angle is kept constant at 20◦ to the tangential direction. Part (a)
shows the transition from the viscous regime to the inertial regime as the flow rate is increased
from λT = 0.12 to λT = 0.36. This is observable from the peak in Nusselt number on the
rotor caused by impingement opposite the pre-swirl nozzles. All cases have a peak around the
receiver holes as the highest inlet swirl ratio of βb = 1.2 (βp = 1.5) is not large enough to cause
synchronous rotation at this point.

The mid-height inlet configuration shows this same viscous to inertial transition. In this case
the highest swirl ratio, βb = 1.2 is sufficient for the synchronous swirl condtion to be met and
therefore no heat transfer peak appears in the receiver hole region.

Part (c), in which the pre-swirl radius is equal to that of the receiver holes, highlights the
combination of the two effects driving the heat transfer distribution; the increasing impingement
velocity due to the increasing flow rate and the wall shear on the rotor due to the circumferential
velocity difference between the flow and the rotor. As the pre-swirl ratio is increased from
βb = 0.4 to βb = 0.8 a reduction in the peak is observed due to lower shear, even though the
impingement velocity will have increased. For the cases βb = 0.8 and βb = 1.2, the effect of
shear may be expected to be similar as the velocity differential between pre-swirl flow and the
rotor disc are the same. Instead the effect of the impingement velocity appears to be dominant
and causes a large increase for the higher flow rate case.

76
Figure 5.22 compares Nusselt number distributions for each of the inlets whilst the flow rate,
λT , is held constant. Part (a), for which λT = 0.12, represents the viscous regime and clearly
shows that the inlet position affects the distribution but not the magnitude of Nusselt number on
the rotor.

As the flow rate is increased, shown in part (b), the position of the peak around the receiver hole
has a variable radial position. Inlet B, for which rp /rb = 0.9, has a peak at a higher radius than
inlet A. This is consistent with the higher radial velocity near the rotor, as shown in figure 5.3(d),
(f) and (h).

For the cases with the highest flow rate, λT = 0.36, the inertial impingement peak opposite the
inlet can be seen for the low and mid height inlets. Although there is little in common between
the three cases in the receiver hole region.

5.2.2 Circumferential Variation of Heat Transfer

The circumferential variation of heat transfer was examined in section 4.3 for the case of the
low radius pre-swirl inlet. It was found that at low radius there was little variation in the
circumferential direction, mainly attributed to the axisymmetric inlet slot used. In the receiver
hole region an ‘eye-brow’ (a region around the rim of the hole) of high heat transfer is seen.
This occurs on the leeward edge of the hole as the boundary layer flow is refreshed by fluid from
the core, described by Lewis et al. (2007) as fluid in the ‘indirect’ route. These distributions of
Nusselt number are consistent with those found from experiment (Yan et al. 2003).

In the earlier work the computations and the experimental results had pre-swirl ratios below the
critical ratio for synchronous rotation at the receiver hole radius. Therefore the ‘eye-brow’ of
heat tranfer would always occur on the trailing rim of the hole. These results can be seen in 1D
form in figure 5.23(a) and as a contour plot in figure 5.24(a)-(c). In figure 5.23(a) the left hand
side of the plot represents the leading rim of the receiver hole and has much lower heat transfer
than the trailing rim shown on the right. The peaks begin to equalise for increasing swirl ratio.

This asymetry in the receiver hole region is an important feature of the heat transfer as it may
lead to high thermal gradients and in addition transfers heat to the cooling flow. Fluid in the
disc boundary layer travels radially outward with a component of velocity in the circumferential
direction dependent on the swirl ratio. As this flow reaches the receiver hole the boundary layer
enters the hole and must be replaced by flow from the core. Some of this core flow will also exit
through the receiver hole, however some replaces the lost boundary layer. This core flow has an

77

additional component of axial velocity as it enters the boundary layer and therefore creates an
impingement on the downstream side of the receiver hole. This is the mechanism which gives
rise to the ‘eye-brow’.

Figure 5.24(d) shows the synchronous rotation case for which no high heat transfer region
is visible in the receiver hole region. Thus the condition providing the highest discharge
coefficient also provides the lowest heat transfer variations across the rotor disc. The relatively
low importance of the radial velocity component in driving the heat transfer characteristics is
illustrated by the lack of variation around the upper rim of the receiver hole.

The mid-height inlet is shown in figure 5.23(b) and figure 5.25(a)-(d). Four sets of conditions
are displayed, each with the same Reφ and λT , but with increasing values of βp . Similar to
the low-radius inlet; for low values of βp a peak in heat transfer is seen on the trailing rim of
the receiver hole. The pattern becomes symmetric at the point where the synchronous rotation
condition is satisfied, βp = 1.5. As the pre-swirl ratio is further increased, to βp = 2.0, the peak
in heat transfer reverses to appear on the leading rim of the receiver hole.

The inversion effect on the heat transfer pattern is most obvious with the high radius inlet where
the highest swirl is achieved at the receiver hole radius (figure 5.11). Figure 5.23(c) shows that
at βp = 0.5 the trailing rim of the receiver hole (RHS of graph) experiences a larger peak than
the leading edge (LHS of graph). The case for which βp = 1.0 the peaks are of equal height.
The two cases where βp > 1.0 show the leading rim to have a higher Nusselt number. This
effect of the ‘eye-brow’ moving to the other side of the receiver hole can be seen clearly in
figure 5.26(a)-(d). It is interesting to note that the magnitudes of the Nusselt number peaks for
βp = 1.5 are slightly larger than those for which βp = 0.5.

78

 (a)
3000

βp = 0.5
βp = 1.0
2500
βp = 1.5
βp = 2.0

2000

Nu
1500

1000

500

0
0.7 0.75 0.8 0.85 0.9 0.95 1
X
3000
(b)
2500

2000
Nu

1500

1000

500

0
0.7 0.75 0.8 0.85 0.9 0.95 1
X
3000

(c)
2500

2000
Nu

1500

1000

500

0
0.7 0.75 0.8 0.85 0.9 0.95 1
X

Figure 5.20: Distribution of Nusselt number on a radial line between receiver holes, λT = 0.24
(a) Inlet A, rp /rb = 0.8 (b) Inlet B, rp /rb = 0.9 (c) Inlet C, rp /rb = 1.0.
79
 (a)
3000

2500 βb = 0.4, λT = 0.12

βb = 0.8, λT = 0.24
βb = 1.2, λT = 0.36

2000

Nu
1500

1000

500

0
0.7 0.75 0.8 0.85 0.9 0.95 1
X
3000
(b)
2500

2000
Nu

1500

1000

500

0
0.7 0.75 0.8 0.85 0.9 0.95 1
X
3000

(c)
2500

2000
Nu

1500

1000

500

0
0.7 0.75 0.8 0.85 0.9 0.95 1
X

Figure 5.21: Distribution of Nusselt number on a radial line between receiver holes, inlet angle
= 20◦ (a) Inlet A, rp /rb = 0.8 (b) Inlet B, rp /rb = 0.9 (c) Inlet C, rp /rb = 1.0.
80
 (a)
3000

Inlet A
2500 Inlet B
Inlet C

2000

Nu
1500

1000

500

0
0.7 0.75 0.8 0.85 0.9 0.95 1
X
3000
(b)
2500

2000
Nu

1500

1000

500

0
0.7 0.75 0.8 0.85 0.9 0.95 1
X
3000

(c)
2500

2000
Nu

1500

1000

500

0
0.7 0.75 0.8 0.85 0.9 0.95 1
X

Figure 5.22: Distribution of Nusselt number on a radial line between receiver holes, inlet angle
= 20◦ (a) λT = 0.12 (b) λT = 0.24 (c) λT = 0.36.
81
 (a)
4000

3500 βp = 0.5
βp = 1.0
βp = 1.5
3000 βp = 2.0

2500

Nu
2000

1500

1000

500

0
0 1 2 3 4 5 6
φ [degree]
4000
(b)
3500

3000

2500
Nu

2000

1500

1000

500

0
0 1 2 3 4 5 6
φ [degree]
4000

(c)
3500

3000

2500
Nu

2000

1500

1000

500

0
0 1 2 3 4 5 6
φ [degree]

Figure 5.23: Distribution of Nusselt number on a circumferential line between receiver holes at
r = rb , λT = 0.24 (a) Inlet A, rp /rb = 0.8 (b) Inlet B, rp /rb = 0.9 (c) Inlet C, rp /rb = 1.0.
82
 (a) (b)

(c) (d)

Figure 5.24: Contours of Nusselt number distribution on the rotor. Inlet A, rp /rb = 0.8, Reφ =
106 and λT = 0.24. (a) βp = 0.5, (b) βp = 1.0, (c) βp = 1.5 and (d) βp = 2.0.

83

 (a) (b)

(c) (d)

Figure 5.25: Contours of Nusselt number distribution on the rotor. Inlet B, rp /rb = 0.9, Reφ =
106 and λT = 0.24. (a) βp = 0.5, (b) βp = 1.0, (c) βp = 1.5 and (d) βp = 2.0.

84

 (a) (b)

(c) (d)

Figure 5.26: Contours of Nusselt number distribution on the rotor. Inlet C, rp /rb = 1.0, Reφ =
106 and λT = 0.24. (a) βp = 0.5, (b) βp = 1.0, (c) βp = 1.5 and (d) βp = 2.0.

85

Chapter 6

Conclusions

Flow and heat transfer characteristics of a pre-swirl rotor-stator system have been computed and
compared with measurements from previous studies. The computed velocity fields have been
used to interpret the heat transfer patterns observed experimentally. The following conclusions
have been drawn:

1. The computed static pressure distribution agrees to within 5% with measured values but
the tangential velocity, and hence the total pressure, is over-predicted by 15% - 20%.

2. The computed values of the adiabatic effectiveness, Θb,ad , increase linearly with βp and
consistently agreed with the theoretical model of Karabay et al. (2001) to within 2%.

3. The computed discharge coefficients show a maximum value of Cd,b = 0.65 occurs at
β1 = 1, which corresponds to synchronous rotation of the fluid core adjacent to the
receiver holes. The experimental measurements straddle the computed curve of Cd,b
versus β1 , but there are no measurements for β1 > 1 to confirm the computational result
that Cd,b will decrease as β1 increases.

4. The computed and measured radial distributions of Nusselt number, N u, on the rotating
disc show evidence of the viscous and inertial regimes. Although N u tends to increase as
Reφ increases, the parameter N uRe−0.8 is only weakly dependent on Reφ in the viscous
regime. The computations are qualitatively similar to the measurements but, apart from
the region near the receiver holes, however the the absolute levels are overpredicted by
up to 50%.

5. The computed and measured contours of Nu show that there is a small region of high heat

86

 transfer close to the receiver holes. This is due to the two routes by which flow enters the
holes: a ‘direct’ route from the pre-swirl nozzles and an ‘indirect’ route from the core.
The regions of high heat transfer are of importance for designers as they may result in
thermal stresses around the receiver holes in turbine discs.

Further computations have been performed to investigate the effect of nozzle location on the
pre-swirl system fluid dynamics and heat transfer. Inlet radius ratios of rp /rb = 0.8, 0.9 and 1.0
have been used and the effect of varying the pre-swirl nozzle angle and flow rate, λT , have been
investigated, from which the following can be concluded:

1. Cd,b , the discharge coefficient for the receiver holes, is maximized when the core flow is
in synchronous rotation with the holes (β∞ = 1).

2. A simple model based on the effective receiver-hole area can be used to estimate the
reduction in Cd,b when β∞ �= 1.

3. The maximum Cd,b is achieved at a value of βb that decreases as rp /rb increases, with a
corresponding slight increase in the pressure drop in the system.

4. The adiabatic effectiveness increases as rp /rb increases, and computed values are in
excellent agreement with the theoretical analysis.

5. The magnitude of Nusselt number in the receiver hole region is independent of pre-swirl
radius for the viscous regime, however the heat transfer patterns in the region of receiver
hole are dependent on pre-swirl ratio and the heat transfer pattern is seen to reverse as the
flow is over-swirled.

87

Chapter 7

Further Work

7.1 Investigation of Receiver Hole Geometry

This study has considered the fluid dynamics and heat transfer for the rotor-stator cavity with
a circular receiver hole normal to the rotor surface. The optimal conditions to maximise the
discharge coefficient and minimise thermal gradients for this configuration have been identified.

To extend these results to more complex engine geometries it would be interesting to investigate
the optimisation of different shaped receiver holes, both in cross-section and including various
chamfering on the rim. Adding a chamfer is likely to improve the discharge coefficient whilst
altering the shape of the hole may be expected to change the heat transfer footprint in the
receiver hole region.

Emphasis has been placed on the requirement to achieve synchronous rotation at the receiver
hole such that the flow is aligned with the axis of the receiver hole. Changing the angle of
the blade cooling passage will affect the swirl rate at which the minimum discharge coefficient
occurs and may impact the adiabatic effectiveness through the work term required to achieve
solid body rotation.

7.2 Pre-Swirl Nozzle Location

Three locations for the pre-swirl inlet have been investigated. Each have been at equal or lower
radius than the receiver holes and have shown that increased inlet radial location provides an

88

improvement in adiabatic effectiveness. With the current receiver hole location it is not possible
to acheive rp /rb > 1, but by redesigning the geometry this effectiveness relationship could be
further investigated.

7.3 Inlet Velocity Profile

A uniform velocity and pressure profile has been applied to the pre-swirl inlet which is
appropriate to provide the swirl and flow rate characteristics required in the wheelspace. To
aid application to more complex geometries and inlet designs it would be interesting to model
further upstream to investigate the dynamics of the nozzles or slots required to pre-swirl the air
in this way.

7.4 Scaling to Engine Conditions

A comparison of the governing parameters for an engine versus those from experiments and
computations was shown in figure 4.1. An engine has Reynolds numbers of the order 107 ,
λT = 0.4 and βp ≈ 1. Having validated the model for lower Reynolds number flows it would
be interesting, and useful to the engine designer, to apply the modelling method to engine
conditions. This would require additional validation to ensure features such as the turbulence
model and fluid parameters relationships remain realistic. At this stage it may be applicable to
consider further alternatives to the turbulence models considered in this study.

7.5 Derivation of Rotor Temperature Distribution

The computation of heat transfer characteristics on the rotor disc has been a key output of the
current study. Further consideration could be given to the material properties and boundary
conditions of the disc itself. The modelling of this solid part using the heat transfer coefficients
as an input would allow computation of the temperature distribution within the disc. This could
further be used in the prediction of likely failure modes and inform future rotor disc design.
Should the interaction between the temperature distribution of the disc and the fluid dynamics
be considered important then the two components could be computed simulaneously using a
conjugate heat transfer approach.

89
7.6 Derivation of Simple Correlations

A methodology has been established to compute the fluid and heat transfer characteristics with
the rotor-stator system. However, it would be useful to an engine designer to have simple tools
based on correlations of the these results. This would allow a geometric optimisation process to
be performed without resorting to time consuming CFD for each configuration.

7.7 Validation of Orifice Model With CFD

A pilot study to investigate potential modelling approaches for the ingress study has been
presented here. A large number of difficulties have been highlighted which make simple orifice
type models far more attractive as a solution. These orifice models required validation in their
own right to establish the accuracy of elements such as the driving pressure distributions and
orifice areas.

The application of CFD to the parameterisation of such orifice models would form an important
future step and be performed in tandem with experimental studies such as that proposed by the
group at University of Bath.

90

References

T. Abe, H. Kikuchi, and H. Takeuchi. An investigation of turbine disk cooling. CIMAC

Conference, Vienna, 1979.

T. J. Barth and D. C. Jesperson. The design and application of upwind schemes on unstructured
meshes. Proc. 27th AIAA Aerospace Sciences Meeting, 1989.

F. J. Bayley and J. M. Owen. The fluid dynamics of a shrouded disk system with a radial outflow
of coolant. ASME Journal of Engineering for Power, pages 335–341, 1970.

D. E. Bohn, A. Decker, H. Ma, and M. Wolff. Influence of sealing air mass flow on the velocity
distribution in and inside the rim seal of the upstream cavity of a 1.5-stage turbine. Proc.
ASME Turbo Expo 2003: Power for Land, Sea and Air, pages GT2003–38459, 2003.

C. Bricaud, B. Richter, K. Dullenkopf, and H.J. Bauer. Stereo piv-measurements in an enclosed

rotor-stator system with pre-swirled cooling air. Proc. 12th International Symposium on
Applications of Laser Techniques to Fluid Mechanics, 2004.

J.-X. Chen, X. Gan, and J.M. Owen. Heat transfer in an air-cooled rotor-stator system. ASME
J. Turbomachinery, 118:444–451, 1996.

J.-X. Chen, X. Gan, and J.M. Owen. Heat transfer from air-cooled contrarotating disks. ASME
J. Turbomachinery, 119:61–67, 1997.

J. W. Chew, T. Green, and A. B. Turner. Rim sealing of rotor-stator wheelspaces in the presence
of external flow. Proc. ASME Turbo Expo 1994: Power for Land, Sea and Air, pages
94–GT–126, 1994.

J.W. Chew, F. Ciampoli, N.J. Hills, and T. Scanlon. Pre-swirled cooling air delivery system
performance. Proc. ASME Turbo Expo 2005: Power for Land, Sea and Air, page 68323,
2005.

91
H. Cohen, G. F. C. Rogers, and H. I. Saravanamuttoo. Gas turbine theory. Longman, 4. ed.
edition, 1996. ISBN 0-582-23632-0.

W. A. Daniels, B. V. Johnson, D. J. Graber, and R. J. Martin. Rim seal experiments and analysis
for turbine applications. ASME Journal of Turbomachinery, 114:426–432, 1992.

M. Dittmann, T. Geis, V. Schramm, S. Kim, and S. Wittig. Discharge coefficients of a preswirl

system in secondary air systems. ASME J. Turbomachinery, 124:119–124, 2002.

L. A. Dorfman. Hydrodynamic Resistance and the Heat Loss of Rotating Fluids. Oliver and
Boyd, London, 1963.

Z.B. El-Oun and J.M. Owen. Preswirl blade-cooling effectiveness in an adiabatic rotor-stator
system. ASME J. Turbomachinery, 111:522–529, 1989.

M. Farzaneh-Gord, M. Wilson, and J. M. Owen. Numerical and theoretical study of flow and
heat transfer in a pre-swirl rotor-stator system. Proc. ASME Turbo Expo 2005: Power for
Land, Sea and Air, page 68135, 2005.

X. Gan, M. Kilic, and J.M. Owen. Superposed flow between two discs contrarotating at
differential speeds. Int. J. Heat Fluid Flow, 15(6):438–446, 1994.

X. Gan, M. Kilic, and J. M. Owen. Flow between contrarotating discs. ASME J.

Turbomachinery, 117:298–305, 1995.

T. Geis, G. Rottenkolber, M. Dittmann, B. Richter, K. Dullenkopf, and S. Wittig. Endoscopic

piv-measurements in an enclosed rotor stator system with pre-swirled cooling air. Proc. 11th
International Symposium on Applications of Laser Techniques to Fluid Mechanics, 2002.

T. Geis, M. Dittmann, and K. Dullenkopf. Cooling air temperature reduction in a direct transfer
preswirl system. ASME J. Eng. for Gas Turbines Power, 126:809–815, 2004.

D. J. Graber, W. A. Daniels, and B. V. Johnson. Disk pumping test. Air Force Wright
Aeronautical Laboratories Report AFWAL-TR-87-2050, 1987.

T. Green and A. B. Turner. Ingestion into the upstream wheelspace of an axial turbince stage.
ASME Journal of Turbomachinery, 116:327–332, 1994.

H. P. Greenspan. The Theory of Rotating Fluids. Cambridge University Press, 1969.

92

K. Hamabe and K. Ishida. Rim seal experiments and analysis of a rotor=stator system with
non-axisymmetric main flow. Proc. ASME Turbo Expo 1992: Power for Land, Sea and Air,
pages 92–GT–160, 1992.

N. J. Hills, J. W. Chew, T. Green, and A. B. Turner. Aerodynamics of turbine rim-seal ingestion.

Proc. ASME Turbo Expo 1997: Power for Land, Sea and Air, pages 97–GT–268, 1997.

N. J. Hills, J. W. Chew, and A. B. Turner. Computational and mathematical modeling of turbine

rim seal ingestion. ASME Journal of Turbomachinery, 124:306–315, 2002.

ANSYS inc. Cfx user documentation, 5.7 edition. 2001.

B. V. Johnson, G. J. Mack, R. E. Paolillo, and W. A. Daniels. Turbine rim seal gas path
flow ingestion mechanisms. 30th AIAA/AMSE/SAE/ASEE Joint Propulsion Conference,
Indianapolis, pages AIAA 94–2703, 1994.

B. V. Johnson, R. Jakoby, D. E. Bohn, and D. Cunat. A method for estimating the influence
of time-dependent vane and blade pressure fields on turbine rim seal ingestion. Proc. ASME
Turbo Expo 2006: Power for Land, Sea and Air, pages GT2006–90853, 2006.

B. A. Kader. Temperature and concentration profiles in fully turbulent boundary layers. Int. J.
Heat Mass Transfer, 9(24):1541–1544, 1981.

H. Karabay, J.-X. Chen, R. Pilbrow, M. Wilson, and J.M. Owen. Flow in a cover-plate preswirl
rotor-stator system. ASME J. Turbomachinery, 121:160–166, 1999.

H. Karabay, R. Pilbrow, M. Wilson, and J.M. Owen. Performance of pre-swirl rotating-disc

systems. ASME J. Eng. Gas Turbines Power, 122:442–450, 2000.

H. Karabay, M. Wilson, and J.M. Owen. Predictions of effect of swirl on flow and heat transfer
in a rotating cavity. Int. J. Heat Fluid Flow, 22:143–155, 2001.

M. Kilic and J.M. Owen. Computation of flow between two disks rotating at different speeds.
ASME J. Turbomachinery., 125:394–400, 2003.

M. Kilic, X. Gan, and J.M. Owen. Turbulent flow between two discs contrarotating at different
speeds. ASME J. Turbomachinery, 118:408–413, 1996.

N. Kobayashi, M. Matsumato, and M. Shizuya. An experimental investigation of a gas turbine

disk cooling system. ASME Journal of Engineering for Gas Turbines and Power, 106:
136–141, 1984.

93
B. E. Launder and B. L. Sharma. Application of the energy dissipation model of turbulence to
the calculation of flow near a spinning disc. Lett. Heat Mass Transfer, 1:131–138, 1974.

B. E. Launder and D. B. Spalding. The numerical computation of turbulent flows. Comp. Meth.
Appl. Mech. Eng., 3:269–289, 1974.

P. Lewis, M. Wilsom, G. D. Lock, and J.M. Owen. Physical interpretation of flow and heat
transfer in pre-swirl systems. ASME J. Eng. Gas Turbines Power, 129:769–777, 2007.

G. D. Lock, M. Wilson, and J. M. Owen. Influence of fluid dynamics on heat transfer in a

pre-swirl rotating disc system. Proc. ASME Turbo Expo 2004: Power for Land, Sea and Air,
page 53158, 2004.

G. D. Lock, M. Wilson, and J. M. Owen. Influence of fluid dynamics on heat transfer in a

pre-swirl rotating disc system. ASME J. Eng. Gas Turbines Power, 127:791–797, 2005a.

G. D. Lock, Y. Yan, P. J. Newton, M. Wilson, and J. M. Owen. Heat transfer measurements

using liquid crystals in a preswirl rotating-disk system. ASME J. Eng. Gas Turbines Power,
127:375–382, 2005b.

G. D. Lock, M. Wilson, and J. M. Owen. Epsrc case for support: Measurement and modelling
of ingress through gas turbine rim seals. Unpublished, 2007.

X. Luo, C. Zhang, G. Xu, Z. Tao, and S. Ding. Measurements of surface temperature distribution
on a rotating disk with blade cooling holes using thermochromic liquid crystal. Proc. ASME
Turbo Expo 2004: Power for Land, Sea and Air, page 53519, 2004.

B. Meierhofer and C. J. Franklin. An investigation of a preswirled cooling airflow to a turbine

disc by measuring the air temperature in the rotating channels. Proc. ASME Gas Turbines
Conference and Products Show, pages 81–GT–132, 1981.

F. R. Menter. Two-equation eddy viscosity turbulence models for engineering applications.

AIAA-Journal, 8(32):269–289, 1994.

I. Mirzaee, X. Gan, M. Wilson, and J. M. Owen. Heat transfer in a rotating cavity with a
peripheral inflow and outflow of cooling air. ASME J. Turbomachinery, 120:818–823, 1988.

A. P. Morse. Numerical prediction of turbulent flow in cavities. ASME Journal of

Turbomachinery, 110:202–215, 1988.

94
P. J. Newton, Y. Yan, N. E. Stevens, S. T. Evatt, G. D. Lock, and J. M. Owen. Transient heat
transfer measurements using thermochromic liquid crystal. part 1: An improved technique.
Int. J. Heat Fluid Flow, 24:14–22, 2003.

J. M. Owen and R. H. Rogers. Flow and Heat Transfer in Rotating-Disc Systems, Volume 1:
Rotor-Stator Systems. Research Studies Press, Taunton, UK / Wiley, New York., 1989.

J. M. Owen and R. H. Rogers. Flow and Heat Transfer in Rotating-Disc Systems, Volume 2:
Rotating Cavities. Research Studies Press, Taunton, UK / Wiley, New York., 1995.

U. P. Phadke and J. M. Owen. An investigation of ingress for an air-cooled shrouded rotating

disk system with radial clearance seals. ASME Journal of Engineering for Power, 105:
178–183, 1983.

U. P. Phadke and J. M. Owen. Aerodynamic aspects of the sealing of gas turbine rotor-stator
systems. part 1: The behaviour of simple shrouded rotating disc systems in a quiescent
environment. International Journal of Heat and Fluid Flow, 9(2):98–105, 1988a.

U. P. Phadke and J. M. Owen. Aerodynamic aspects of the sealing of gas turbine rotor-stator
systems, part 2: The performance of simple seals in a quasi-axisymmetric external flow.
International Journal of Heat and Fluid Flow, 9(2):106–112, 1988b.

U. P. Phadke and J. M. Owen. Aerodynamic aspects of the sealing of gas turbine rotor-stator
systems, part 2: The effect of nonaxisymmetric external flow on seal performance.
International Journal of Heat and Fluid Flow, 9(2):113–117, 1988c.

R. Pilbrow, H. Karabay, M. Wilson, and J. M. Owen. Heat transfer in a ”cover-plate” preswirl

rotating-disk system. ASME J. Turbomachinery, 121:249–256, 1999.

O. Popp, H. Zimmermann, and J. Kutz. Cfd analysis of coverplate receiver flow. ASME J.
Turbomachinery, 120:43–49, 1998.

O. Reynolds. On the extent and action of the heating surface for steam boilers. Manchester Lit.
Phil. Soc., 14:7–12, 1874.

C. M. Rhie and W. L. Chow. A numerical study of the turbulent flow past an isolated airfoil
with trailing edge separation. AIAA Paper 82-0998, 1982.

Rolls-Royce. The Jet Engine. Fourth edition, 1986.

95
Z. Sun, K. Lindblad, J. W. Chew, and C. Young. Les and rans investigations into buoyancy
affected convection in a rotating cavity with a central axial throughflow. Proc. ASME Turbo
Expo 2006: Power for Land, Sea and Air, pages GT2006–90251, 2006.

W. Sutherland. The viscosity of gases and molecular force. Phil. Mag., 5:507–531, 1893.

C. Z. Wang, B. V. Johnson, F. D. Jong, and T. K. Vashist. Comparison of flow characteristics

in axial-gap seals for close- and wide-spaced turbine stages. Proc. ASME Turbo Expo 2007:
Power for Land, Sea and Air, pages GT2007–27909, 2007.

D. C. Wilcox. Turbulence Modelling for CFD. DCW Industries, 2nd edition, 1998.

M. Wilson and J. M. Owen. Axisymmetric computations of flow and heat transfer in a pre-swirl
rotor-stator system. Proc. 1st International Conference on Flow Interaction, Hong Kong.,
pages 447–530, 1994.

M. Wilson, R. Pilbrow, and J. M. Owen. Flow and heat transfer in a pre-swirl rotor-stator
system. ASME J. Turbomachinery, 119:364–373, 1997.

Y. Yan, M. Farzaneh-Gord, G. D. Lock, M. Wilson, and J. M. Owen. Fluid dynamics of a

pre-swirl rotor-stator system. ASME J. Turbomachinery., 125:641–647, 2003.
Appendix 1

Pilot Study

Ingress into rotating disc systems

Review of Ingress

The importance of delivering low temperature cooling air to the turbine blades has been
introduced above. The minimising of any temperature rise of the fluid due to frictional heating is
one part of the problem, a second is the contamination of the secondary gas path. The interfaces
between rotating and stationary surfaces, such as that between the stator vane and rotor blade
platforms, must have a finite gap to prevent contact. This gap, usually of the order of 1 mm,
allows interchange of fluid between the primary and secondary gas paths and thus a potential
contamination of the cooling flow by hot fluid from the mainstream, so called ‘ingress’.

With turbine inlet temperatures nearing 1800 K and typical cooling temperatures of around
320K (Cohen et al. 1996), small amounts of ingress can have a significant effect. Using these
figures, for each 1 per cent of external flow ingested compared to the sealing flow rate, the
temperature of the blade-cooling air would be expected to increase by approximately 15 K.
Thus reducing the life of the component.

Typically, ingress into wheelspaces between discs is prevented by providing a sealing flow rate
outwards through the sealing gap. This reduces the overall efficiency of the engine as additional
fluid must be made available in the secondary air system. It is therefore an important area of
research for engine designers. A schematic of the flow structure in the cavity downstream on
the stator vane is shown in figure 7.1.

Figure 7.1: Ingress into a rotor stator wheelspace.

Early work was performed by Bayley and Owen (1970), who considered a simple rotor stator
system with a superposed flow through the central axis of the stator in a quiescent atmosphere.
They measured the pressure distribution inside the wheelspace for various rotational Reynolds
numbers and superposed flow rates. A shroud was also attached to the stator to create an axial
clearance seal for comparison with the open system.

In terms of ingress, it was assumed that whilst the measured pressure just inside the shroud
was sub-atmospheric external flow would be drawn into the system. As the sealing flow rate
was increased the pressure within the wheelspace also rose, until the pressure difference across
the outer seal became zero. This critical sealing flow rate to prevent ingress into the system,
Cw,min , can be seen in figure 7.2 to be a linear function of Reφ . The data is correlated according
to equation 7.1.

Cw,min = 0.61Gc Reφ (7.1)

Phadke and Owen (1983) experimented with five different simple seal geometries, including
both axial and radial overlap types. Attention was paid to the pressure distribution in the
wheelspace, and Cw,min measured for each. Two methods were used; both the pressure criterion
described above and flow visualisation, seeding the quiescent surroundings. Results from both
methods were found to be in good agreement. A linear relationship between Cw,min and Reφ
was found, but the ratio between them was different for each seal configuration.

A thorough investigation of the effect of external flow was published in a three part paper
by Phadke and Owen (1988a,b,c). These examined seal performance for configurations with
zero, quasi-axisymmetric and non-axisymmetric external flow respectively. In addition to the
‘pressure’ and ‘flow visualisation’ criteria used by the group previously, the external flow, or
quisescent atmophere, was seeded with nitrous oxide, N2 O. A gas analyser was then used to
detect when external flow entered the wheelspace.

Figure 7.3 shows the variation of Cw,min for a simple axial clearance seal, the Reynolds number
for the axial mainstream is varied whilst the rotor disc remains stationary. Each line represents
a different sealing gap ratio and shows a linear relationship between the two variables. When
a combination of external flow and rotation of the rotor disc was used, figure 7.4, rather than a
super-position of the two, an improvement in the sealing is seen for a range of axial Reynolds
number. As the axial Reynolds number is increased the effect of rotation decreases and Cw,min
asymptotes to its zero rotation value. This is consistent with the work of Abe et al. (1979) who
 Figure 7.2: Minimum sealing flow rate (Bayley and Owen 1970).

found that an external flow over the seal was dominant in affecting the critical sealing flow for
ingress.

The effect of hot ingress on the temperature of the disc was measured experimentally by
Kobayashi et al. (1984). A hot mainstream was passed over a recessed radial overlap seal, cool
sealing flow entered the wheelspace at low radius and surface temperatures were recorded. Both
the rotational speed of the disc and the sealing flow rate were shown to affect the temperature
distribution. These temperature measurements were also used as a criteria to measure whether
hot gas entered the system. This thermal criteria generally gave much larger estimates for
Cw,min than the pressure based results of Phadke and Owen (1983).

In addition to understanding how to prevent ingress, it was also important to predict how much
hot gas may enter through the seal should lower rates of sealing flow be used. Graber et al.
(1987), and later Daniels et al. (1992), describe an experiment to evaluate a cooling effectiveness
 Cw,min / 103
15

Gc = 0.02

10 Gc = 0.01
Gc = 0.005

Rew / 106
0
0 0.2 0.4 0.6 0.8 1 1.2

Figure 7.3: Cw,min for varying external flow axial Reynolds number and a stationary rotor disc.
Each line shows a different seal clearance gap ratio. (Phadke and Owen 1988b)

for a selection of seals. The rig provided an external mainstream flow, from a set of nozzles,
over the seal, at an angle such that the flow had a significant and controllable swirl ratio. This
external flow was seeded with CO2 and the resultant CO2 level in the wheelspace, assumed to
be a bulk average, was measured.

Figure 7.5 shows the variation in effectiveness found by Graber et al. (1987), where their
effectiveness parameter φ is defined in terms of CO2 concentrations as shown in equation 7.2.
The subscripts (c, e, s) refer to the concentrations in the cavity, external flow and sealing flow
respectively. The nondimensional flow parameter, ηt , is defined in equation 7.3.

Cc − Ce
φ= (7.2)
Cs − Ce

ṁs
ηt = (7.3)
4πµbRe0φ.8

The results show a non-linear relationship between cooling mass flow and effectiveness. This
suggested that it may be inefficient to completely seal the wheelspace, by significantly reducing
the seal flow, the amount of hot gas entering the system could still be relatively small. The effect
of changing the swirl ratio of the flow is shown to be negligable.

Johnson et al. (1994) identifies a list of mechanisms causing ingress into the wheelspace,
shown below, and provides a good description of each. The importance of circumferential
nonuniformities in pressure (from blades and vanes) and geometry (bolt heads, etc.) is
highlighted.
 a)
8 3
Cw,min / 10
7
6
5
4
Re / 106
3
0 0.2 0.4 0.6
2
1 0.8 1.0 1.2 Rew / 106
0
0 0.2 0.4 0.6 0.8 1 1.2

b)
12
Cw,min / 103
10

2 6
Rew / 10
0
0 0.2 0.4 0.6 0.8 1 1.2

c)
16 3
Cw,min / 10
14
12
10
8
6
4
2 Rew / 10
6

0
0 0.2 0.4 0.6 0.8 1 1.2

Figure 7.4: Cw,min for varying external flow axial Reynolds number. Each line shows a different
rotational Reynolds number. (a) Gc = 0.005, (b) Gc = 0.01,(c) Gc = 0.02. (Phadke and Owen
1988b)
 Figure 7.5: Effectiveness when Cw < Cw,min (Graber et al. 1987).

• Disc pumping

• Periodic vane / blade pressure field (3-D and time dependent)

• 3-D geometry in the rim seal region

• Asymmetries in the rim seal region

• Turbulent transport in the platform overlap region

• Flow entrainment

The effect of circumferential pressure distributions caused by NGVs and rotor blades was
investigated further by Green and Turner (1994). Tests were conducted using the concentration
method, the mainstream seeded with nitrous oxide, and the resulting concentration in the
wheelspace measured. Four configurations were used; the complete stage, NGVs only,
axisymmetric flow and no external flow. Results are shown with the Bayley and Owen (1970)
criteria in figure 7.6. The most effective seal for a given sealing flow rate is, unsurprisingly,
that with no external flow. The complete stage, with NGVs and blades, although having
the most complex external pressure field is found to seal at a lower flow rate than the
axisymmetric system. The configuration with NGVs only produces the lowest effectivness, and
therefore highest ingress. For axisymmetric flow the Bayley and Owen (1970) criteria appears
conservative.
Figure 7.6: Variation of sealing effectiveness with non-dimensional coolant flow rate for four
external flow conditions. (Green and Turner 1994)

A simple analytical model to calculate ingress was published by Hamabe and Ishida (1992).
Based on the assumption that the length scale in the circumferential direction is much larger
than that in the axial direction, a simple two-dimensional orifice model was proposed. However
Chew et al. (1994) performed a combined CFD and experimental study and found the model
to underestimate the effectiveness. It was speculated this was due to the exclusion of inertial
terms, being based purely on pressure differences. The 3D steady CFD model showed good
agreement with experiments especially for low values of sealing flow rate.

Hills et al. (2002) describes a comparison of CFD approaches and compares results with an
earlier experiment (Hills et al. 1997). The four CFD configurations used are: NGVs only, blades
only, NGVs and blades with a mixing plane and NGVs and blades in a full unsteady model.
These are compared with the experiment which has NGVs and a rotor ‘peg’ to produce a rotating
pressure field. The results of the comparison are shown in figure 7.7, the species concentration
is equivalent to effectiveness in figure 7.6. Model 1 (producing a stationary pressure field) and
model 2 (producing a rotating pressure field) show broadly similar levels of effectiveness, a
large over-estimation compared with the experiment. Slightly improved results were gained
with the mixing plane model, although this would be expected to ‘circumferentially smear’ the
external pressure variations in the two frames of reference. The best results were produced
 Species concentration
1

0.8

0.6

0.4 Experimental
Model 1 (NGV)
Model 2 (Rotor)
0.2 Model 3 (Mixing Plane)
Model 4 (Unsteady)
radius ratio (= r / b)
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 7.7: Comparison of calculated and measured concentration in the cavity. (Hills et al.
2002)

using the full unsteady model, although these were still an underprediction of the measured
ingress levels. It is interesting to note that the computed concentration levels at the stator are
constant over a large area of the disc.

Johnson et al. (2006) more recently attempted to bring the ingress problem back to basics and
presented a simple orifice model. The input to the model was the time dependent difference
in pressure between an azimuthal location and that inside the cavity from measurements made
using the Aachen University rig described by Bohn et al. (2003). The model also requires
a discharge coefficient to be known for the particular seal geometry, and then predicts the
effectiveness versus sealing flow rate curve to within 5%.
Pilot Study

A summary of previous investigations into the ingress phenomena along with its importance to
the gas turbine designer has been presented in section 7.7. The topic is closely related to that of
rotating cavities and pre-swirl systems as the ingested flow is generally entering the secondary
air system through the gap between rotating and stationary machinery parts. Thus it has the flow
characteristics of the type dealt with earlier in this thesis. Additional complexity is provided by
the interaction between the secondary airstream and the external mainstream gas path. Interest
in the area has been raised at the University of Bath by collaboration with a visiting researcher
from Mitsubishi and proposed collaboration with Siemens in a joint experimental study (Lock
et al. 2007).

This piece of work is a pilot study to investigate the potential for using a similar method for
the study of ingress, to that which has been applied to the pre-swirl problem and validated for
the rotating cavity. An understanding of the flow and pressure field should also help inform
the design of future experimental rigs. The results of the study, although interesting, could
only be partially validated against existing experimental datasets. They are therefore useful in
identifying key mechanisms driving the fluid dynamics and identifying trends in temperature
distribution, but are not intended to be used for direct comparisons with future experiments.

Computational Domain and Model

The computational domain, illustrated in figure 7.8, comprises a rotor-stator wheel-space having
an outer radius b = 0.216m and an axial-clearance rotor-side rim seal leading to an outer
annulus representing the mainstream gas path through the turbine stage. The annulus height
is 10mm and the wheel-space gap ratio and seal gap ratio are G = 0.07 and GC = 0.01
respectively.

Several designs were considered and discarded as a result of preliminary computations. Some
variants had the seal located at different axial locations compared to the wheelspace, however
computations showed that the axial flow near the outer shroud stagnated on the inside lip of the
seal. This essentially sealed the system and prevented any ingress, thus the rotor side seal was
chosen for the study.

The radial offset to the annulus boundary surface on the rotor side was included as the external
mainstream flow stagnated at the outermost rim of the seal on the rotor side, again leading to
very low computed levels of ingress.

These low ingress characteristics may be desirable in practice, however experimental apparatus
would most likely be designed to promote ingress at the chosen flow conditions for the purpose
of making accurate and informative measurements.

Figure 7.8: Computational domain for ingress study. Red represents stationary surfaces, blue
represents the rotor

Figure 7.9: Schematic of vane geometry in the φ − z plane

A stator vane of generic geometry is included in the mainstream annulus upstream of the
seal, see figure 7.9, to provide circumferential variations of pressure and velocity. A 15◦
circumferential sector has been modelled computationally, representing the pitch between each
of twenty four stator vanes on a proposed future experimental rig. Two geometric configurations
have been tested, having different axial spacings of 7.92mm and 15.42mm between the vane
trailing edge and the seal. The larger spacing was found to dampen the pressure field variation at
the seal to such an extent that this design was also discarded as inappropriate for an experimental
study.

The system has two inlets; the sealing air inlet at the the inner radius of the wheel-space (at r =
a) and the mainstream inlet upstream of the vane, at both of which uniform values for velocity
components and temperature are prescribed as described below. An average static pressure
is prescribed at the mainstream outlet boundary. Cyclic symmetry and no-slip conditions are
applied at other boundaries as appropriate and all solid boundary surfaces are assumed to be
adiabatic.

An unstructured mesh has been used, with a blend of quadrilateral elements near wall surfaces
and a Delaunay triangulation in the core away from them, rotated around the central axis. The
mesh around the stator vane was generated using regular layers in the near wall region and an
advancing front scheme in the core. Sensitivity to mesh size was tested over a wide range, see
table 7.1, the results are described below.

Mesh
Code
Total
Elements in
Elements in

(used in fig
Elements
Circumferential
Axial Direction

legends)
Direction
Across Seal

Coarse
FL
583,200 45 15
Regular
FLS90
1,526,200 90 25
Fine
VL135
2,777,355 135 33
Very Fine
EL180
3,963,600 180 39
Very Fine (90)
EL90
1,993,140 90 39

Table 7.1: Mesh Parameters for Ingress Geometry.

As illustrated in figure 7.8, all of the wall boundary surfaces are stationary with the exception of
the rotor. Other computations were carried out for which the annulus boundary surface attached
to the rotor also rotated. In this case, it was found that Taylor vortices were set up in the
mainstream flow downstream of the seal. These vortices affected the computed flow structure
in the seal and initiated unsteady flow in the wheel-space. These effects would need to be
considered in the design of experimental apparatus.

The commercial code Ansys-CFX (v.10) used for the computations was described and validated
in preceding chapters for pre-swirl applications. In addition to the RANS momentum
and energy equations, an additional transport equation was solved for conservation of a
non-interacting scalar quantity. This allowed a ‘tracer’ to be introduced at the inlet to the
mainstream in order to calculate the amount of ingress and hence sealing effectiveness. A
similar computational approach was taken by Wang et al. (2007), and similar methods have
also been used by, for example, Sun et al. (2006) to compute the unstable flows inside rotating
cavities.

The Baseline (BSL) turbulence model of Menter (1994) is used in addition to a Reynolds
Stress model. The comparison is performed to identify whether an anisotropic model may
perform differently in the region of the seal where large differences in the magnitude of velocity
gradients are computed.

Governing Parameters

The rotational Reynolds number for the wheelspace, consistent with the pre-swirl work, is:

ρΩb2
Reφ = (7.4)
µ

and the mainstream annulus Reynolds number is:

ρVabs b
Rez = (7.5)
µ

where Vabs is the velocity magnitude at the vane trailing edge.

The values of the parameters used for the majority of computations are shown below, where
these are varied they will be discussed individually.

Ω = 848rad/s(≈ 8, 000rpm)
Cw = 1, 600
Vabs = 196ms−1
 Reφ = 2.5 ∗ 106

Rez = 2.7 ∗ 106

The stator vane used gives an average flow angle of around 24◦ to the circumferential direction
in the mainstream region outward of the seal. The values of Ω and CW were selected as likely
test conditions in planned future experiments, and the value of Rez used is that which gives rise
to a swirl ratio βe ≈ 1 in the mainstream outward of the seal (matching qualitatively conditions
in the experiments by Graber et al. 1987). The sealing flow rate used is much lower than the
value Cw,min = 15, 250 suggested by equation 7.1 for this configuration. The matching of the
sealing flow rate to the rotational speed through the parameters Reφ and Cw allow the findings
for fluid dynamics to be extrapolated with some confidence to the engine situation, see Owen
and Rogers (1989).

Fluid Dynamics

The stator vane produces a circumferentially varying static pressure field in the external
mainstream as illustrated in figure 7.10. The pressure coefficient, Cp , based on pe , the computed
(external) mainstream static pressure at the half height of the annulus radially outward of
the seal, and pi , the spatially averaged pressure on the stator inside the wheel-space at a
non-dimensional radial location x = 0.95, is defined as follows:

pe − p̄i
Cp = (7.6)
0.5ρΩ2 b2

The circumferential distribution for Cp shown in figure 7.10 shows a minimum at a

non-dimensional circumferential location θ ≈ 0.07 and a distinct peak at θ ≈ 0.61 for each
case. Part (a) of the figure shows good agreement to within 5% for each of the mesh using the
BSL model, equally good agreement is also found for the Reynolds Stress model shown in part
(b). Note the legend relates to the mesh code shown in table 7.1 with a prefix relating to the
turbulence model. Part (c) of the figure compares the turbulence models for an intermediate
sized mesh; the peak-to-peak variation and distribution of Cp is equal in each case, although the
BSL model predicts an 8% larger difference in internal versus external pressure. Computations
using alternative geometries have shown that the circumferential variation of Cp decreases
with increasing distance downstream from the vane trailing edge. With the seal in the further
downstream position the peak magnitude for Cp is approximately half that for the configuration
shown here, largely due to mixing and pressure recovery in the mainstream. In an engine
changing the seal location relative to the vane trailing edge in this way would increase the
relative effect on ingress of the pressure distribution due to the rotating turbine blades. The
mainstream pressure variations for the downstream configuration were found to give rise to very
low levels of ingress, and are therefore not considered useful in the design of an experimental
apparatus.

The influence of the mainstream pressure variations in driving fluid into the wheel-space can be
characterised using v¯r , the mass flow averaged radial velocity across the seal, defined as:

�
vr ṁdz
v¯r = � (7.7)
ṁdz

The variation of v¯r calculated at the seal half-height (x ≈ 1.01) with θ is shown in figure 7.11.
The distribution shows a peak negative value at θ ≈ 0.78, a location shifted circumferentially
by about one-eighth of the stator-vane pitch in the direction of rotation of the disc from that
(θ ≈ 0.61) for the corresponding maximum driving pressure shown in figure 7.10. Similar to
the previous figure, parts (a) and (b) show the comparison of results on each mesh for the BSL
and Reynolds Stress model respectively. Each shows little sensitivity to mesh density in either
the circumferential or axial direction. Part (c) compares computations for an intermediate mesh
using each of the models; the Reynolds Stress models predicts a similar distribution of v¯r to
the BSL model although with a lower peak magnitude. This is consistent with the smaller seal
pressure drop shown in figure 7.10.

The velocity vectors shown in figure 7.12 illustrate the secondary flow (i.e. the flow in the
axial-radial plane) in the seal at the circumferential locations 1 to 4 identified below:

1. θ = 0.29: location of maximum radial outflow

2. θ = 0.57: location of equal inflow and outflow

3. θ = 0.78: location of maximum radial inflow

4. θ = 0.98: location of equal inflow and outflow

In Plane 1, where v¯r is a maximum, this net flow radially outward is due to the boundary
layer on the rotor having sufficient momentum to overcome the adverse radial pressure gradient
through the seal. In Plane 2, where v¯r ≈ 0, the velocity magnitude is small in the seal region,
with some outward flow on the rotor side and some flow drawn inwards from the mainstream
onto the stator side of the seal. Ingress into the system is a maximum at Plane 3, where a
powerful recirculation is formed at the stator side of the seal. This recirculation transports fluid
ingested into the seal from the mainstream inward and towards the stator as it flows into the
wheel-space. In Plane 4, where again v¯r ≈ 0, the strong recirculation apparent in Plane 3
is maintained, but now acts to seal the system. This is due to the smaller pressure difference
between the wheel-space and the mainstream at this location, see figure 7.10. Similar underlying
flow structures to those described here are expected also to be found in engines, modified by
additional unsteady pressure variations due the rotating turbine blades.

Effectiveness Using Non-Interacting Scalar

The sealing effectiveness of the system, ηc , is evaluated using the computed local concentration
(mass-fraction) C of the non-participating scalar ‘tracer’ variable transported throughout the
system:

C − Cs
ηc = 1 − (7.8)
Ce − Cs

where Cs and Ce are the local concentrations of the sealing flow and mainstream flow
respectively. (If no ingestion occurs, C = Cs inside the wheel-space and the sealing
effectiveness is unity.)

Values for sealing effectiveness are shown in figures 7.13 and 7.14 for near-wall solution points
adjacent to the stator and rotor respectively, point results having been averaged circumferentially
to give these radial distributions. The results show significant grid sensitivity; using the BSL
turbulence model some convergence for larger grids is visible, but the Reynolds Stress model
predicts systematically larger effectiveness for larger grids. Agreement to within 2% is found
between the ‘EL180’ and ‘EL90’ mesh suggesting that the sensitivity is due to the resolution
across the seal in the axial direction.

There are no similarly significant differences for the computed velocity field (see figure 7.11),
suggesting that the computed transport of the scalar variable is more sensitive to the grid than
the associated velocities. figure 7.13 shows that (on each of the different meshes tested)
the computed sealing effectiveness is approximately constant with radius near the stator.
This behaviour is consistent with ingested mainstream fluid flowing radially inward in the
wheel-space within the boundary layer on the stator. The ingested fluid then migrates axially
across the wheel-space towards the rotor, however there is no entrainment of fresh fluid to
dilute the concentration. (The spatially averaged computed flowfield within the wheel-space
was found to be very similar to that observed in a rotor-stator system with a radial outflow of
fluid, where for the value of gap ratio G considered here there are boundary layers on the rotor
and stator separated by a rotating core of fluid, see Owen and Rogers 1989).

The radial variation of averaged effectiveness with radial location near the rotor (see figure 7.14)
is more significant than for the stator. At the inner radius, where sealing flow enters the
system, effectiveness approaches unity. As radius increases, fluid is entrained into the rotor
boundary layer from the stator, increasing the concentration of ingested fluid and thus reducing
the effectiveness.

Figures 7.13 and 7.14 part (c) show a comparison of the computed effectiveness on the finest
mesh from each of the turbulence models. Good agreement is found, suggesting that perhaps
the finest grid is approaching mesh independence with respect to the effectiveness parameter.
Limitations in computational resource prevented further mesh investigations.

Temperature Distribution

The inlet total temperature of the wheel-space sealing flow was set at Ts = 293K, while that
of the main gas path was varied between Te = 273K and Te = 393K. In order to limit the
computational requirements the BSL turbulence model (the least computationally intensive of
the two) and the FLS90 mesh are used to compute heat transfer. Figure 7.15 shows computed,
circumferentially averaged, profiles of (static) temperature T on the stator and rotor respectively.

Two mechanisms contribute to the elevated temperatures within the wheel-space shown in
figure 7.15; the ingress of higher temperature fluid from the mainstream and frictional heating
(windage) due to the rotating disc.

Figure 7.15 shows that the temperatures reach values greater than that of the external
maintstream at both the rotor and stator surfaces, indicating that for this situation the fluid
in the wheel-space is heated above that of the external mainstream. This is caused by frictional
heating, exaggerated by the use of adiabatic boundary conditions at solid surfaces and the low
value of sealing flow rate used.

Typically in engines (Te − Ts ) ≈ 1000K and the windage heating may be up to around
50K, however the values used in the computations are characteristic of test conditions that
might be used in simplified experiments. The results shown in figure 7.15 are a concern
in the design of experimental apparatus to study the thermal effects of ingress using modest
differences in temperature, in order for example to make heat transfer measurements using
thermochromic liquid crystal (TLC), as it may be difficult to separate the combined effects of
ingestion and windage. Experiments such as this are likely to require further computations
and/or complementary measurements of concentration in order to interpret measurements of
temperature.

Dependence on Sealing Flow Rate

Figure 7.16(a) shows the variation in pressure coefficient Cp (defined in equation 7.6) for a
variety of sealing flow rates, ṁs . The pressure increase across the seal is the net of two effects;
that of the positive radial pressure gradient versus the orifice pressure drop. As the sealing flow
rate is increased the orifice pressure drop is increasingly important, thus reducing Cp .

The computed effectiveness on the stator and rotor using the concentration technique is shown
in figures 7.16(b) and (c). The system is fully sealed (i.e. ingress levels are negligable) when
the sealing flow rate is increased to Cw = 3200. A third case with reduced sealing flow rate,
Cw = 800, shows similar effectiveness distribution on each disc as that for Cw = 1600, however
the magnitude is reduced.

A direct comparison of the effectiveness parameter versus sealing flow rate is shown in
figure 7.17. Part (a) shows four radial locations on the stator, at each the value of effectiveness
is coincident due to the absence of any dilution of the stator boundary layer described above.
The effectiveness decreases with decreasing sealing flow rate and by definition µc = 0 when
Cw = 0. Part (b) shows the same for locations on the rotor, however the effectiveness is not
independent of radial location. At high radius the rotor boundary layer is contaminated by
entrainment of flow transported axially from the stator. The effect is less significant at low
radius where sealing air is a more dominant component of the boundary layer.
Dependence on Reynolds Number

The effect on the pressure distribution Cp of reducing Reynolds number is shown in

figure 7.18(a). The radial pressure gradient decreases as the disc speed decreases consequently
reducing the pressure difference. However, the peak to trough variation in Cp driven by the
external flow does not change. Part (b) of the same figure shows that the lower Reynolds
number, with lower Cp has an increased variation in axially averaged radial velocity, v¯r .

This increase in v¯r does not translate into increased ingress. Rather, as shown in figure 7.19,
effectiveness decreases with Reynolds number. High disc speeds, associated with high radial
pressure gradient, increase the tendency for inward flow on the stator side of the seal.

A pilot study has also been conducted to investigate the complexities associated with
studying the ingress phenomenon. Three-dimensional steady turbulent flow computations of
a rotor-stator system and an external mainstream have been carried out at conditions typical of
those likely to be used in simplified experiments devised to monitor and measure the effects of
ingress of hot fluid from the mainstream into the rotor-stator wheel-space. From this part of the
study the following conclusions have been drawn:

1. A stator vane in the mainstream, with its trailing edge sufficiently close to the rotor-stator
axial seal, produces a non-axisymmetric flow distribution sufficient to cause significant
levels of ingress into the wheel-space, as deduced from sealing effectiveness values
calculated using computed concentrations of a tracer scalar variable.

2. Investigation of the fluid dynamics within the axial clearance seal shows that ingested
fluid is transported radially inwards, and towards the stator boundary layer inside the
wheel-space.

3. The action of a recirculating flow established in the seal which acts to seal the system
locally from ingress at some circumferential locations, even when the superposed
wheel-space sealing flow rate is low compared with the minimum expected theoretically
to be required to prevent ingress.

4. Greater grid sensitivity was observed for computed results for effectiveness compared
with velocity distributions.

5. Computations of the thermal field suggest that identification of the thermal effects of
ingress in simplified experiments may be complicated by the frictional heating (windage)
of the fluid in the wheel-space due to the rotating disc.
 (a)
0.15

0.1

Cp
0.05 BSL FL
BSL FLS90
BSL VL135
BSL EL180
BSL EL90

0
0 0.2 0.4 0.6 0.8 1
θ

(b) 0.15

0.1
Cp

0.05 RS FL
RS FLS90
RS VL135
RS EL180
RS EL90

0
0 0.2 0.4 0.6 0.8 1
θ

(c) 0.15

0.1
Cp

0.05 BSL FLS90

RS FLS90

0
0 0.2 0.4 0.6 0.8 1
θ

Figure 7.10: Pressure coefficient relating static pressure in the wheel-space to that outside of
the seal (a) BSL model with varying mesh size, (b) Reynolds Stress model with varying mesh
size, (c) Reynolds Stress model versus BSL model for given mesh size.
 15
(a)

10 BSL FL
BSL FLS90
BSL VL135
5 BSL EL180
BSL EL90

Vr
0

-5

-10

-15
0 0.2 0.4 0.6 0.8 1
θ
15
(b)
10 RS FL
RS FLS90
RS VL135
5 RS EL180
RS EL90
Vr

-5

-10

-15
0 0.2 0.4 0.6 0.8 1
θ
15
(c)
10 BSL FLS90
RS FLS90

5
Vr

-5

-10

-15
0 0.2 0.4 0.6 0.8 1
θ

Figure 7.11: Computed radial velocity averaged in the axial direction in the seal, v¯r (a) BSL
model with varying mesh size, (b) Reynolds Stress model with varying mesh size, (c) Reynolds
Stress model versus BSL model for given mesh size.
Figure 7.12: Secondary flow in the seal shown in the r − z plane

 1
(a)

0.8

0.6

ηc
BSL FL St
0.4
BSL FLS90 St
BSL VL135 St
BSL EL180 St
BSL EL90 St
0.2

0
0.7 0.8 0.9 1
X
1
(b)
0.8

0.6
ηc

RS FL St
0.4 RS FLS90 St
RS VL135 St
RS EL180 St
RS EL90 St
0.2

0
0.7 0.8 0.9 1
X
1
(c)
0.8

0.6
ηc

BSL EL180 St
0.4
RS EL180 St

0.2

0
0.7 0.8 0.9 1
X

Figure 7.13: Computed effectiveness at the stator using concentration method (a) BSL model
with varying mesh size, (b) Reynolds Stress model with varying mesh size, (c) Reynolds Stress
model versus BSL model for given mesh size.
 1
(a)

0.8

0.6

ηc
BSL FL Ro
0.4
BSL FLS90 Ro
BSL VL135 Ro
BSL EL180 Ro
BSL EL90 Ro
0.2

0
0.7 0.8 0.9 1
X
1
(b)
0.8

0.6
ηc

RS FL Ro
0.4
RS FLS90 Ro
RS VL135 Ro
RS EL180 Ro
RS EL90 Ro
0.2

0
0.7 0.8 0.9 1
X
1
(c)
0.8

0.6
ηc

BSL EL180 Ro
0.4
RS EL180 Ro

0.2

0
0.7 0.8 0.9 1
X

Figure 7.14: Computed effectiveness at the rotor using concentration method (a) BSL model
with varying mesh size, (b) Reynolds Stress model with varying mesh size, (c) Reynolds Stress
model versus BSL model for given mesh size.
 350
(a)
345

340

335

330
T [K]

325
Te = 273 K, St
320
Te = 293 K, St
Te = 313 K, St
315 Te = 333 K, St
Te = 353 K, St
310 Te = 373 K, St
Te = 393 K, St
305

300
0.7 0.8 0.9 1
X
350

(b) 345

340

335

330
T [K]

325
Te = 273 K, Ro
320
Te = 293 K, Ro
Te = 313 K, Ro
315 Te = 333 K, Ro
Te = 353 K, Ro
310 Te = 373 K, Ro
Te = 393 K, Ro
305

300
0.7 0.8 0.9 1
X

Figure 7.15: Computed surface temperatures for various prescribed external flow temperatures,
(a) Stator, (b) Rotor.
 (a)
0.15

0.1

Cp
0.05 cw = 800
cw = 1600
cw = 3200

0
0 0.2 0.4 0.6 0.8 1
θ
1
(b)
0.9

0.8

0.7

0.6
ηc

0.5

0.4

0.3 cw = 800
cw = 1600
0.2 cw = 3200

0.1

0
0.7 0.8 0.9 1
X
1
(c) 0.9

0.8

0.7

0.6
ηc

0.5

0.4

0.3 cw = 800
cw = 1600
0.2 cw = 3200

0.1

0
0.7 0.8 0.9 1
X

Figure 7.16: (a) Pressure coefficient relating pressure in wheelspace to that outside seal for
various sealing flow rates. (b) & (c) Effectiveness for various sealing flow rates at the stator and
rotor respectively.
 1
(a)

0.8

0.6
ηc

0.4

X = 0.70 St
X = 0.80 St
0.2 X = 0.90 St
X = 0.95 St

0
0 1000 2000 3000 4000
Cw
1

(b)
0.8

0.6
ηc

0.4

X = 0.70 Ro
X = 0.80 Ro
0.2 X = 0.90 Ro
X = 0.95 Ro

0
0 1000 2000 3000 4000
Cw

Figure 7.17: Computed effectiveness using concentration method versus sealing flow rate for
(a) stator and (b) rotor.
 (a) 0.15

0.1
Cp

0.05

6
Re = 2.50 x 10
6
Re = 1.90 x 10
Re = 1.25 x 10 6
0
0 0.2 0.4 0.6 0.8 1
θ
15

(b)
10

5
Vr

-5

-10

-15
0 0.2 0.4 0.6 0.8 1
θ

Figure 7.18: (a) Pressure coefficient relating pressure in wheelspace to that outside seal. (b)
Axially averaged radial velocity in the seal.
 1
(a)

0.8

0.6

ηc
0.4

6
Re = 2.50 x 10
6
Re = 1.90 x 10
0.2 Re = 1.25 x 10 6

0
0.7 0.8 0.9 1
X
1
(b)
0.8

0.6
ηc

0.4

6
Re = 2.50 x 10
6
Re = 1.90 x 10
0.2 Re = 1.25 x 10 6

0
0.7 0.8 0.9 1
X
1
(c)
0.8

0.6
ηc

0.4

Stator X=0.95
Rotor X=0.95
0.2

0
400 500 600 700 800 900
Disc Speed [rads / s]

Figure 7.19: Computed effectiveness using concentration method versus Reynolds number for
(a) stator and (b) rotor. (c) Local value evaluated at x = 0.95.
Appendix 2

Journal of Engineering for Gas Turbines and Power

Physical Interpretation of Flow and Heat Transfer in Pre-Swirl Systems

 Physical Interpretation of Flow
and Heat Transfer in Preswirl
Systems
This paper compares heat transfer measurements from a preswirl rotor–stator experiment
with three-dimensional (3D) steady-state results from a commercial computational fluid
dynamics (CFD) code. The measured distribution of Nusselt number on the rotor surface
Paul Lewis was obtained from a scaled model of a gas turbine rotor–stator system, where the flow
e-mail: p.r.lewis@bath.ac.uk
structure is representative of that found in an engine. Computations were carried out
using a coupled multigrid Reynolds-averaged Navier-Stokes (RANS) solver with a high
Mike Wilson Reynolds number k-� / k-� turbulence model. Previous work has identified three param­
e-mail: m.wilson@bath.ac.uk
eters governing heat transfer: rotational Reynolds number �Re��, preswirl ratio �� p�, and
Gary Lock the turbulent flow parameter ��T�. For this study rotational Reynolds numbers are in the
e-mail: g.d.lock@bath.ac.uk range 0.8� 106 � Re� � 1.2� 106. The turbulent flow parameter and preswirl ratios var­
ied between 0.12� �T � 0.38 and 0.5� � p � 1.5, which are comparable to values that
J. Michael Owen occur in industrial gas turbines. Two performance parameters have been calculated: the
e-mail: j.m.owen@bath.ac.uk adiabatic effectiveness for the system, �b,ad, and the discharge coefficient for the receiver
holes, CD. The computations show that, although �b,ad increases monotonically as � p
University of Bath, increases, there is a critical value of � p at which CD is a maximum. At high coolant flow
Bath BA2 7AY, UK rates, computations have predicted peaks in heat transfer at the radius of the preswirl
nozzles. These were discovered during earlier experiments and are associated with the
impingement of the preswirl flow on the rotor disk. At lower flow rates, the heat transfer
is controlled by boundary-layer effects. The Nusselt number on the rotating disk increases
as either Re� or �T increases, and is axisymmetric except in the region of the receiver
holes, where significant two-dimensional variations are observed. The computed velocity
field is used to explain the heat transfer distributions observed in the experiments. The
regions of peak heat transfer around the receiver holes are a consequence of the route
taken by the flow. Two routes have been identified: “direct,” whereby flow forms a stream
tube between the inlet and outlet; and “indirect,” whereby flow mixes with the rotating
core of fluid. �DOI: 10.1115/1.2436572�

1 Introduction perature in the receiver holes, decreased monotonically as � p, the

The blade-cooling air in gas turbines is usually supplied to the preswirl ratio, increased even when � p was significantly greater
rotating high-pressure blades by stationary preswirl nozzles. The than unity.
cooling air is swirled, which reduces the work done by the rotat- Geis et al. �3� made measurements of the adiabatic effective­
ing turbine disk in accelerating the air to the disk speed. This in ness, which showed that the measured values of Tt,b were signifi­
turn reduces the total temperature of the air entering the receiver cantly higher than the values predicted from their ideal model. �It
holes in the disk. A simplified diagram of the so-called direct should be pointed out that their preswirl ratio was based on isen­
transfer preswirl system is shown in Fig. 1. tropic values rather than on measurements.� Chew et al. �4� made
The designer is interested in calculating the pressure drop and numerical simulations of both the “Karlsruhe rig,” used by Geis et
cooling effectiveness of the preswirl system. There is also a need al., and a “Sussex preswirl rig.” The computations were in good
to understand the heat transfer between the cooling air and the agreement with the results of both rigs, and the low adiabatic
turbine disk, particularly the possible creation of local nonuniform effectiveness of the Karlsruhe rig was attributed to the geometry
temperatures in the metal that could lead to large thermal stresses. of the preswirl chamber; in particular, the Karlsruhe rig had a
Meierhofer and Franklin �1�, who were the first to measure the much larger stator area, which reduced the effective swirl ratio
effect of preswirl on the temperature drop in a direct-transfer sys- and consequently reduced the effectiveness.
tem, showed that swirling the air could significantly reduce the Chew et al. �4� and Farzaneh-Gord et al. �5� independently
total temperature in the receiver holes of a turbine disk. El-Oun derived theoretical models for the adiabatic effectiveness of a
and Owen �2� developed a theoretical model for the so-called direct-transfer system, taking account of the moment on the stator.
adiabatic effectiveness, �b,ad, based on the Reynolds analogy. The �These models predict lower values of �b,ad than that of Karabay
model, which was in good agreement with the temperatures mea- et al. �6�, who based their model on a cover-plate system in which
sured on their rotating-disk rig, showed that Tt,b, the total tem- the preswirl air flows radially outward between two rotating
disks.�
Popp et al. �7� carried out a CFD analysis of a cover-plate
Contributed by the International Gas Turbine Institute of ASME for publication in system, computing the temperature drop and the discharge coeffi­
the JOURNAL OF ENGINEERING FOR GAS TURBINES AND POWER. Manuscript received July cients for different geometries. They showed that CD, the dis­
19, 2006; final manuscript received July 20, 2006. Review conducted by Dilip R.
Ballal. Paper presented at the ASME Turbo Expo 2006: Land, Sea and Air �GT2006�, charge coefficient for the receiver holes, became a maximum
Barcelona, Spain, May 8–11, 2006. Paper No. GT2006-90132. when the relative tangential velocity was close to zero. This effect

Journal of Engineering for Gas Turbines and Power JULY 2007, Vol. 129 / 769
Copyright © 2007 by ASME

Downloaded 29 May 2008 to 138.38.0.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 �T = cw Re�−0.8 �2�
where
ṁ p
cw = �3�
�b
and
��b2
Re� = �4�
�
The value �T = 0.22 corresponds to the flow rate entrained by a
disk rotating in an infinite environment, the so-called free disk.
For turbine-blade cooling systems, �T � 0.4 and � p � 1.
Fig. 1 Schematic diagram of test section For the Bath rig �described in Sec. 3� it follows that
C cw C �T
was confirmed experimentally by Dittmann et al. �8� who were the �p = = �5�
N Re� N Re�0.2
first to measure the discharge coefficients in a direct-transfer sys­
tem. where C is a geometric constant given by
Yan et al. �9� measured the discharge coefficients for the re­ 4b3 cos �
ceiver holes of a direct-transfer system for a range of rotational C= �6�
speeds and flow rates. For �1 � 1 �where �1 is the measured swirl
� d 2r p
ratio upstream of the receiver holes� CD increased monotonically It can be seen that the preswirl ratio is not independently vari­
as �1 increased from �1 � 0.3 to 0.9. They also found, as did Popp able: it depends upon N, the number of preswirl nozzles, and upon
et al., that CD depends on the ratio of the area of the receiver holes �T and Re�. In the experiments of Yan et al. �9�, two values of N
to that of the nozzles; for a given value of the preswirl ratio, � p, were used: N = 12 and 24. In the experiments discussed below,
CD increases as the area ratio decreases. �It should be noted that, N = 24.
owing to a printer’s error, the wrong figures were printed in Ref. The Bath rig uses a simplified engine geometry, and tests were
�9�; the correct figures are given in Ref. �10��. conducted at representative values of � p and �T, thereby produc­
Heat transfer in a direct transfer rig was studied experimentally ing flow structures typical of those found in engines. However, in
and computationally by Wilson et al. �11� using fluxmeters to engines Re� is on the order of 107, which is an order of magnitude
determine the local Nusselt numbers. Their axisymmetric CFD greater than that achievable in the rig. As the heat transfer depends
results gave reasonable predictions of the velocity and tempera­ strongly on Re�, as well as on � p and �T, the rig Nusselt numbers
ture in the core but underpredicted the measured Nusselt numbers. will be much smaller than those found in engines. This is dis­
Numerous experimenters have used thermochromic liquid crys­ cussed further in Sec. 6.
tal �TLC� to determine heat transfer coefficients on purpose-built
test sections. A common technique is to solve Fourier’s transient 3 Experimental Method
conduction equation to calculate h for a semi-infinite solid ex­ Experimental results were produced by Yan et al. �9� and Lock
posed to a step change in air temperature. As it is virtually impos­ et al. �13,10� using the “slow transient” TLC technique described
sible to achieve a step change in the air temperature of preswirl by Newton et al. �12�. The salient points of the experimental
rigs, Newton et al. �12� developed the so-called “slow transient” method are presented here for convenience and a section of the
technique. Lock et al. �13,10� used this technique to measure the experimental rig is illustrated in Fig. 1.
local Nusselt numbers, on the rotating disk of a direct-transfer rig, The rotor is a transparent polycarbonate disk with a radius of
for a range of rotational speeds, flow rates, and preswirl ratios. 0.216 m, allowing optical access to the wheel space. The disk has
The measurements showed that Nu was virtually axisymmetric 60 circular receiver holes with centers at a disk radius of 0.200 m.
except near the receiver holes, where large variations occurred. To reduce heat transfer from the air inside the receiver holes, the
They also found that there were two flow regimes: at the larger holes are filled with Rohacell �low-conductivity foam� bushes pro­
preswirl flow rates, inertial effects dominated and the flow im­ ducing an effective receiver hole diameter of 8.0 mm. The disk
pinged on the rotating disk creating a peak in Nu; at the smaller has a thickness of 10 mm, and the receiver holes, which have a
flow rates, viscous effects dominated and boundary-layer flow length to diameter ratio of 1.25, vent directly into the laboratory.
controlled the heat transfer. A shroud of carbon fiber surrounds the rim of the disk and rotates
In this paper, a commercial three-dimensional �3D� CFD code with it. A section of the rotor surface inside the wheelspace is
is used to compute the flow and heat transfer in the “Bath rig” painted with thermochromic liquid crystal.
used by Lock et al. The computations are compared with mea­ The stator is also a polycarbonate disk, which is mounted onto
sured values of the discharge coefficients, with theoretical values an aluminum disk. The gap between the rotor and stator is 11 mm
of adiabatic effectiveness and with measured local Nusselt num­
�G = 0.051�, and the clearance between the rotating and stationary
bers. In particular, the computations are used to give a physical
shrouds is 1 mm. The air pressure in the wheel space is balanced
insight into the complex flow and heat transfer that occurs in these
by sealing air to restrict leakage or ingress. The preswirl nozzles
direct-transfer preswirl systems.
comprise 24 circular holes, of 7.1 mm diameter, drilled at an
2 Governing Parameters angle of 20 deg to the tangential direction and at a radius of
160 mm. A stationary Rohacell hub forms the inner boundary of
Owen and Rogers �14� showed that, for a rotating cavity, the the wheel space at a radius of 0.145 m.
turbulent flow structure depends on only two nondimensional pa­ The radial variation of pressure and tangential velocity is mea­
rameters: the inlet swirl ratio, � p, and the turbulent flow param­ sured by a combination of pitot tubes, located at nine radial sta­
eter, �T. These are defined as tions in the midplane �z / s = 0.5�, and static pressure tappings at the
v�,p same radii on the stator. A total-temperature probe and pitot tube
�p = �1� are also located in a nozzle outlet to measure the temperature and
�r p velocity of the inlet flow. More details can be found in Yan et al.
and �9�.

770 / Vol. 129, JULY 2007 Transactions of the ASME

Downloaded 29 May 2008 to 138.38.0.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 3 Typical radial distribution of y+ on the rotor for inertial
and viscous regime computations

The turbulence model used is the high-Reynolds number base­

line �BSL� model of Menter �16�. This is a blend of a k-� formu­
lation with wall functions �Ref. �17��, in the near wall region, and
a k-� model away from the wall. This overcomes sensitivities to
freestream turbulence levels normally experienced by k-� models
Fig. 2 Schematic diagram of computational domain �18�. The convective heat transfer model applied to the rotor wall
is based on the method of Kader �19�; and other surfaces were
assumed adiabatic.
The main air supply to the system is passed through a mesh Additional computations were performed to confirm that the
heating element which essentially creates a step change in tem­ results were not sensitive to inlet turbulence level or to the uni­
perature immediately downstream of the heater; owing to heat form rotor temperature boundary condition used to approximate
losses in the ducting, the temperature rise of the air downstream of the time varying distribution that occurred in the experiment.
the preswirl nozzles is exponential. A strobe light is used to illu­
minate the rotor and the resulting transient disk temperature dis­
5 Fluid Dynamics
tribution is captured on digital video at 25 frames/ s. The red–
green–blue �RGB� signals of each frame are converted to hue and 5.1 Velocity and Pressure. Figure 4�a� shows a comparison
analyzed to calculate the temperature history and steady-state heat between the computed and measured radial variation of ��, the
transfer coefficient, as described by Newton et al. �12�. nondimensional swirl ratio at the midplane �z / s = 0.5�. The maxi­
mum value of ��, which occurs at the inlet radius due to the flow
4 Computational Method from the nozzles, is well predicted by the computations. However,
The computational domain is designed to be a realistic repre­ at the larger radii, the computations overpredict the measured val­
sentation of the experimental rig described above, the only excep­ ues. This may be due in part to overprediction of turbulence levels
tion being the preswirl nozzles. To enable steady-state computa­ by the high-Reynolds number turbulence model.
tions, the preswirl nozzles are modeled as an axisymmetric slot Figure 4�b� shows good agreement between the computed and
having the same inlet area, and therefore the same inlet velocity, measured radial distribution of static pressure. In the rotating core
as the nozzles. The model contains a 1 / 60th sector of the experi­ of fluid, away from the rotor and stator, dp / dr = ���2 / r, and as a
mental rig, i.e., a 6 deg section enclosing one of the receiver consequence the static pressure increases radially. The overpredic­
holes. Boundaries have either a no-slip condition or a periodic tion of the total pressure in Fig. 4�c� is caused mainly by the
interface applied, and walls are defined as rotating or stationary as overprediction of �� referred to above.
required. Clearances at no-flow boundaries were taken to be zero. The agreement between computations and measurements
Figure 2 shows the computational domain, the red shaded areas shown in Fig. 4�a� is best for the lowest value of �T shown, for
being stationary while the blue shaded areas rotate with angular which the flow is in the viscous regime. The computed mixing at
velocity �. Axial and circumferential velocities are prescribed at the higher values of �T �for which the flow is in the inertial re­
the inlet to give the flow angle of 20 deg to the tangential direc­ gime� may be affected by the use of a high-Reynolds-number
tion. Velocity and static temperature values were prescribed to turbulence model and the simplified slot geometry at inlet. �Yan et
match measurements made at the nozzle. At the outlet a static al. �9�, who used a discrete nozzle inlet and a low-Reynolds­
pressure boundary condition was used. number k-� turbulence model, obtained better agreement with
The commercial code used for this investigation is CFX5.7, a measurements than that shown in Fig. 4�a�.�
finite volume, coupled algebraic multigrid solver. The advection
scheme is second-order accurate based on the method of Barth 5.2 Adiabatic Effectiveness. The adiabatic effectiveness,
and Jesperson �15�. The energy equation is solved, including the �b,ad, is defined as
viscous work term, and variable density effects taken into account. c p�T0,p − Tt,b�
Buoyancy effects within the wheel space are ignored. �b,ad = �7�
0.5�2r2b
The mesh is a hybrid of unstructured tetrahedral elements, with
prismatic elements near the wall. Delaunay triangulation is used to For given inlet conditions the total temperature of the air in the
create the surface mesh followed by an advancing front volume rotating receiver holes, Tt,b, decreases linearly as �b,ad increases.
mesher. A mesh sensitivity study was carried out to confirm that Karabay et al. �6� derived a theoretical value for �b,ad using the
the fluid dynamics and heat transfer results are not grid dependent. first law of thermodynamics; the work done on, or by, the air was
Typical y + distributions, for computations both in the viscous and proportional to the moment required to change the tangential ve­
inertial regime, are shown in Fig. 3, and are within the required locity of the air from � p�r p, at the preswirl nozzles, to �rb in the
range for the turbulence model. receiver holes. Their equation, which was derived for a cover-

Journal of Engineering for Gas Turbines and Power JULY 2007, Vol. 129 / 771

Downloaded 29 May 2008 to 138.38.0.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 also used CFD to compute �b,ad in a cover-plate system for 0
� � p � 3, and their computations were in good agreement with
this equation.
Farzaneh-Gord et al. �5� modified Eq. �8� for a direct-transfer
system to account for the moment on the stator, M s. Their model
gives

�b,ad = 2� p ��
rp
rb
2
−1−
Ms
1/2ṁ�r2b
�9�

where the effect of M s �a positive quantity� is to reduce �b,ad.

Figure 5�a� shows a comparison between the computed and
theoretical values of �b,ad for the present case, where r p / rb = 0.8.
The computations based on Eq. �7� use the bulk-average relative
total temperature computed at the outlet from the receiver holes.
These computations are in excellent agreement with the theoreti­
cal values of �b,ad based on Eq. �9� for the direct-transfer system;
the values of M S in this equation were obtained by computing the
tangential shear stress over all of the stationary surfaces in the rig
geometry. The theoretical values of �b,ad calculated using Eq. �8�
for a cover-plate system are significantly higher than those for the
direct-transfer system.
The theoretical values of �b,ad are based on the assumption that
the air achieves solid-body rotation in the receiver holes. Figure
5�b� shows that �2, the computed bulk-average value of � at the
outlet from the holes, is indeed close to unity. �The fact that the
computed values of �2 � 1 could be due to the assumed constant
pressure condition at outlet from the receiver holes. In an engine,
where the blade-cooling air flows radially outward from the re­
ceiver holes, solid-body rotation would be expected to occur; for a
rig with short receiver holes, it is possible that solid-body rotation
would not be achieved.�
It should be pointed out that both the computed and theoretical
values of �b,ad for the direct-transfer rig depend, explicitly or
implicitly, on the computed value of M s. It is, therefore, uncertain
that the theory will agree with experimental measurements of
�b,ad, which were not made in the tests reported here. The mea­
surement of �b,ad is nontrivial: the actual �not the isentropic�
value of � p must be known; the bulk-average relative velocity and
total temperature inside the receiver holes must be measured ac­
curately; and the rotating disk should be made from a thermal
insulator to reduce the heat transfer to the cooling air. These three
conditions are seldom, if ever, achieved in rotating-disk rigs.
5.3 Discharge Coefficients. So as to be consistent with other
research workers, the discharge coefficient for the receiver holes
is defined as

ṁb
CD = �10�
ṁi
where ṁi is the isentropic mass flow rate. This can be calculated
from the first law of thermodynamics, for an adiabatic system,
taking into account the rate of work done by or on the air from
Stations 1–2 in a stream tube. The equation derived by Yan et al.
�9� is

Fig. 4 Comparison of computed „lines… and measured „sym­

bols… results for swirl ratio and pressure: Re� = 0.8Ã 106
ṁi
A2
= �0,1� � �� � � � � �
p2
p0,1
1/�
2� p0,1
� − 1 �0,1
1−
p2
p0,1
�−1/�

plate system in which there is no stator to reduce the swirl, is

+ 2��r2V�,2 − r1V�,1� − V�2 ,2 � 1/2
�11�

given by The first term inside the curly brackets is the standard result for
compressible flow in a stationary nozzle; the second term is the

�b,ad = 2� p ��
rp
rb
2
−1 �8�
work term resulting from the change of angular momentum of the
air; the last term is due to the fact that the air in the receiver holes
has an absolute tangential, as well as an axial, component of ve­
It should be noted that �b,ad � � p : �b,ad increases, and Tt,b de­ locity. In their measurements, Yan et al. based p0,1 and V�,1 on
creases, monotonically with � p even when � p � 1. Karabay et al. their pitot-tube measurements at r = rb and z / s = 0.5. They took p2

772 / Vol. 129, JULY 2007 Transactions of the ASME

Downloaded 29 May 2008 to 138.38.0.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 unity, supporting the assumption of solid-body rotation at the out­
let from the receiver holes. The computed values of �1, which are
based on the bulk-average values of V�,1 calculated at r = r1 = rb
and z / s = 0.5, are significantly smaller than � p, and it should be
noted that �1 = 1 when � p � 1.8.
Figure 5�c� shows a comparison between the computed and
measured variation of CD with �1. The measured values were
obtained with 24 and 12 preswirl nozzles �N = 24 and N = 12�; the
N = 12 results were obtained by blocking alternate nozzles. As for
all computations presented in this paper, the computed results
were obtained for the case where the annular slot in the stator had
an area equivalent to the N = 24 tests.
As expected, the computations and experiments show that CD
reaches a maximum value at �1 = 1; the computed maximum of
CD = 0.65 is lower than that in the experiments, CD = 0.70. Dittman
et al. �20� measured discharge coefficients for a rotating short
orifice, with the same length to diameter ratio as the receiver
holes, and found that CD had a maximum value of 0.78 when the
fluid and the orifice had the same circumferential velocity.
The maximum value of CD occurs at a critical value of � p,
which will depend strongly on the system geometry. For the sys­
tem used here, the critical value of � p was approximately 1.8 for
the computations and 2.3 for the experiments. The important prac­
tical consequence is that, although the cooling effectiveness in­
creases monotonically as � p increases, the flow rate of the blade
cooling air will decrease if the critical value of � p is exceeded.

6 Heat Transfer
6.1 Radial Variation of Nusselt Number. The computed
heat flux is nondimensionalised to form the local Nusselt number,
Eq. �13�, based on the adiabatic disc temperature derived by Kara-
bay et al. �6� for a cover-plate system and which is shown in Eq.
�12�

Tw,ad = T0,p −
V�2 ,�
2C p
+R
2C p
�
� 2r 2
1−
V�,�
�r
� 2
�12�

q wr
Nu = �13�
k�Tw − Tw,ad�
Equation �12� is a theoretical value based on the Reynolds anal­
ogy, and the values of Tw,ad computed by Karabay et al. �6� were
in excellent agreement with this equation. Newton et al. �12� used
wide-band TLC to measure the adiabatic-disk temperature on the
Bath rig. Apart from the region near the preswirl nozzles, the
differences between the measured and theoretical values of Tw,ad
were mostly less than 0.5° C. Owing to the uncertainty in the
wide-band TLC measurements, Eq. �12� was used to calculate
Tw,ad for the measured Nusselt numbers.
Owen and Rogers �21� showed that, for turbulent boundary-
layer flow in rotor-stator systems, Nu� Re�0.8. If the heat transfer
in the Bath rig is controlled by turbulent boundary-layer flow then
the parameter Nu Re�−0.8 would be expected to be independent of
Re�.
Figure 6 shows a comparison between computed and experi­
mental values of Nu Re�−0.8 for three values of �T and three values
Fig. 5 Variation of �b,ad, �, and CD with �p for 0.8Ã 106 < Re� of Re� �As Eq. �5� shows, � p � �T�. The measured local Nusselt
< 1.2Ã 106 and 0.12< �T < 0.38: „a… comparison between com­ numbers were obtained by Lock et al. �13�, and those shown in
puted and theoretical �b,ad; „b… comparison between computed Fig. 6 were evaluated along a radial line midway between two
�1 and �2; and „c… comparison between computed and mea­
adjacent receiver holes.
sured CD
Figure 6�a�, for �T = 0.13 and � p = 0.5, shows a distinct peak in
heat transfer near the receiver holes �x = 0.93� but there is no sign
of impingement near the preswirl nozzles �x = 0.74�. Lock et al.
as the atmospheric pressure at outlet from the receiver holes, and defined this as the viscous regime, and it can be seen that Nu
assumed that V�,2 = �rb. The same assumptions and locations Re�−0.8 is a good correlating parameter for both the computational
were used here in determining the computed values of ṁi. and experimental results. Although the radial variation of the com­
As shown in Fig. 5�b� the computed values of �2 are close to putations and measurements are qualitatively similar, the compu-

Journal of Engineering for Gas Turbines and Power JULY 2007, Vol. 129 / 773

Downloaded 29 May 2008 to 138.38.0.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 variation of the computations and measurements are qualitatively
similar, but again the parameter Nu Re�−0.8 fails to collapse the
data.
6.2 Circumferential Variation of Nusselt Number. Figure 7
shows comparisons of Nusselt number contours across an 18 deg
sector of the rotor, as studied experimentally. Results on the right
were produced by computation and those on the left are experi­
mental results. The conditions for case 7�a� classify it within the
viscous regime; the two other examples, 7�b� and 7�c�, relate to
the inertial regime.
Comparing Fig. 7�a� with the results in Fig. 6, the same differ­
ence in Nusselt number magnitude is visible and the improved
agreement at the receiver hole radius is also apparent. In the re­
gion close to the edge of the holes there is an absence of experi­
mental data caused by the presence of the Rohacell bushes de­
scribed in Sec. 3. �The opaque bushes also cause shadows over the
transparent rotor, obscuring the results in this region.�
In Fig. 7�b� and 7�c�, representing the inertial regime, there is
good qualitative agreement between computations and measure­
ments at high radii.
A small region of high heat transfer is observable around the
receiver holes in each case in Fig. 7. At low �T and � p this region
is located at the “9 o’clock position” with respect to the holes, Fig.
7�a�. As �T and � p are increased the region moves around towards
the “11 o’clock position,” Fig. 7�c�. Luo et al. �22� performed
temperature measurements around a rotating disk with receiver
holes and observed similar behavior around the holes. �In an en­
gine, high heat transfer in this region could result in thermal
stresses within the rotor.�
6.3 Physical Interpretation of Heat Transfer Results. Us­
ing an axisymmetric CFD code, Wilson et al. �11� showed that air
entered the receiver holes by “direct” and “indirect” routes. The
former refers to flow traveling directly along a streamline con­
necting the inlet and the outlet, and therefore not mixing with the
core flow. Indirect flow mixes with the core flow before entering
the receiver holes.
This idea can be extended to the study of nonaxisymmetric
systems by considering that the direct flow travels in a stream tube
between the preswirl nozzles and the receiver holes. Computing
streamlines for the direct flow allows the path of the stream tube
to be evaluated. Figure 8 shows the stream tube relative to the
rotor for a variety of conditions. The inner location is at the radius
of the preswirl nozzles, and the outer location is at the radius of
the receiver holes. These results show only a weak effect of Re�.
Figure 9 shows measured heat transfer results combined with
streamlines calculated using the full velocity field from the com­
putations. Figure 9�a� is a radial section with the preswirl inlet and
stator on the left and the receiver hole and rotor on the right. The
orange streamline shows that flow from the nozzle can be either
direct or indirect: the direct flow exits through the receiver hole;
the indirect flow continues to a higher radius and will recirculate
in the system and mix with the core flow. The black streamline
shows that indirect flow, which has entered the core, can either
exit through the receiver hole or continue circulating in the core.
Figure 9�b� shows the same streamlines in a circumferential
Fig. 6 Radial variation of Nu Re� −0.8
: „a… �T = 0.12, �p = 0.5; „b…
section. It can be seen that the flow at the rotor surface that is
�T = 0.24, �p = 1.0; and „c… �T = 0.35, �p = 1.5
aligned with the receiver holes becomes direct flow. The remain­
ing flow follows the indirect route.
Figure 9�c� shows an isometric view of the same streamlines. It
tations overpredict the measured values except in the region near is the flow from the core, replacing the boundary layer flow en­
the receiver holes. tering the receiver holes, which gives rise to the region of high
In Fig. 6�b� for �T � 0.24 and � p � 1.0, the experimental results heat transfer.
show a distinct peak near the nozzles, signifying that the flow is in
the inertial regime. The computations show only a small peak at
this radius. For these conditions, the parameter Nu Re�−0.8 fails to 7 Conclusions
collapse either the experimental or computational data. Flow and heat transfer measurements from an experimental
In Fig. 6�c� for �T � 0.35 and � p � 1.5, both sets of results study of a preswirl rotor–stator system have been compared with
exhibit the inertial peak near the preswirl nozzles. The radial the results of computations. The experiments were conducted on

774 / Vol. 129, JULY 2007 Transactions of the ASME

Downloaded 29 May 2008 to 138.38.0.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 7 Experimental „left… and computational „right… Nusselt number contours, Re� = 0.8Ã 106: „a… �p = 0.5, �T
= 0.13; „b… �p = 1.0, �T = 0.24; and „c… �p = 1.5, �T = 0.37

the “Bath rig:” a purpose-built direct-transfer rig. The measure- The conclusions based on this rig geometry are as follows:
ments were made for 0.8� 106 � Re� � 1.2� 106, 0.1� �T � 0.4
and 0.5� � p � 1.5, and the flow structure was considered to be 1. The computed static pressure distribution agrees well with
representative of that found in gas turbines. The steady-state com- measured values but the tangential velocity, and hence the
putations used a 3D commercial CFD code. total pressure, is overpredicted.

Journal of Engineering for Gas Turbines and Power JULY 2007, Vol. 129 / 775

Downloaded 29 May 2008 to 138.38.0.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 8 Streamline plots for flow in the direct route between
inlet and outlet

2. The computed values of �b,ad were in good agreement with

a theoretical model for direct-transfer systems.
3. The discharge coefficient, CD, reached a maximum value
when �1, the core swirl ratio adjacent to the receiver holes,
was unity. The computed maximum was CD � 0.65, and the
experimental maximum was CD � 0.70.
4. The computed and measured radial distributions of Nusselt
number, Nu, on the rotating disk show evidence of the vis­
cous and inertial regimes. Although Nu tends to increase as
Re� increases, the parameter Nu Re�−0.8 is only weakly de­
pendent on Re� in the viscous regime. The computations are
qualitatively similar to the measurements but, apart from the
region near the receiver holes, they do not show good quan­
titative agreement.
5. The computed and measured contours of Nu show that there
is a small region of high heat transfer close to the receiver
holes. This is due to the two routes by which flow enters the
holes: a “direct” route from the preswirl nozzles and an “in­
direct” route from the core. The regions of high heat transfer
are of importance for designers as they may result in thermal
stresses around the receiver holes in turbine disks.

Acknowledgment
Paul Lewis is a Ph.D. student funded by the UK Engineering
and Physical Sciences Research Council �EPSRC� through a Doc­
toral Training Account �DTA�. The experimental work was per­
formed by Dr Y. Yan, who was funded by Alstom Power Ltd �now
Siemens Industrial Turbines� and EPSRC, and additional dis­
charge coefficient measurements were made by Vinod Kakade
from the University of Bath. The authors would like to thank the
reviewers for their constructive feedback during the review
process.
Fig. 9 Computed streamlines superimposed onto experimen­
tal heat transfer results: Re� = 0.8Ã 106, �p = 1.5, �T = 0.38
Nomenclature
a � rotor inner radius
b � rotor outer radius
cp � specific heat capacity at constant pressure R � recovery factor �=Pr1/3�

cw � nondimensional mass flow rate �=ṁ / �b� Re� � rotational Reynolds number �=��b2 / ��

CD � discharge coefficient for receiver holes r � radius

d � preswirl nozzle diameter r p , rb � radii of preswirl nozzles and receiver holes

G � gap ratio �=s / b� s � rotor–stator separation distance
h � heat transfer coefficient T � static temperature
k � thermal conductivity of air v � velocity
ṁ � mass flow rate u� � friction velocity

M � disk moment x � nondimensional radius �=r / b�

N � number of preswirl nozzles y � distance normal to the wall

Nu � Nusselt number �=qwr / k�Tw − Tw,ad�� y+ � nondimensional wall distance �=�yu� / ��

Pr � Prandtl number �=�c p / k� � � swirl ratio �=v� / �r�
qw � rotor wall heat flux � � ratio of specific heats

776 / Vol. 129, JULY 2007 Transactions of the ASME

Downloaded 29 May 2008 to 138.38.0.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 �b,ad � adiabatic effectiveness 22, pp. 143–155.
�7� Popp, O., Zimmermann, H., and Kutz, J., 1998, “CFD Analysis of Coverplate
�T � turbulent flow parameter �=cw Re�−0.8� Receiver Flow,” ASME J. Turbomach., 120, pp. 43–49.
� � dynamic viscosity �8� Dittmann, M., Geis, T., Schramm, V., Kim, S., and Wittig, S., 2002, “Dis­
� � density charge Coefficients of a Preswirl System in Secondary Air Systems,” ASME J.
Turbomach., 124, pp. 119–124.
� � angular velocity of rotor �9� Yan, Y., Farzaneh-Gord, M., Lock, G., Wilson, M., and Owen, J. M., 2003,
“Fluid Dynamics of a Pre-Swirl Rotor-Stator System,” ASME J. Turbomach.,
Subscripts 125, pp. 641–647.
ad � adiabatic �10� Lock, G. D., Wilson, M., and Owen, J. M., 2005, “Influence of Fluid Dynam­
b � blade-cooling ics on Heat Transfer in a Pre-Swirl Rotating Disc System,” ASME J. Eng. Gas
i � isentropic value Turbines Power, 127, pp. 791–797.
�11� Wilson, M., Pilbrow, R., and Owen, J. M., 1997, “Flow and Heat Transfer in a
o � total value in stationary frame Pre-Swirl Rotor-Stator System,” ASME J. Turbomach., 119, pp. 364–373.
p � preswirl �12� Newton, P. J., Yan, Y., Stevens, N. E., Evatt, S. T., Lock, G. D., and Owen, J.
s � stator M., 2003, “Transient Heat Transfer Measurements Using Thermochromic Liq­
t � total value in rotating frame uid Crystal. Part 1: An Improved Technique,” Int. J. Heat Fluid Flow, 24, pp.
14–22.
w � rotor �13� Lock, G. D., Yan, Y., Newton, P. J., Wilson, M., and Owen, J. M., 2005, “Heat
�,r,z � circumferential, radial, axial direction Transfer Measurements Using Liquid Crystals in a Preswirl Rotating-Disk
� � value in core at z / s = 0.5 System,” ASME J. Eng. Gas Turbines Power, 127, pp. 375–382.
�14� Owen, J. M., and Rogers, R. H., 1995, Flow and Heat Transfer in Rotating-
1,2 � upstream, downstream locations in a stream Disc Systems �Vol. 2, Rotating Cavities�, Research Studies Press, Taunton, UK.
tube �15� Barth, T. J., and Jesperson, D. C., 1989, “The Design and Application of
Upwind Schemes on Unstructured Meshes,” Proc. 27th AIAA Aerospace Sci­
ences Meeting, Reno, NV, January 9–12.
References �16� Menter, F. R., 1994, “Two-Equation Eddy Viscosity Turbulence Models for
�1� Meierhofer, B., and Franklin, C. J., 1981, “An Investigation of a Preswirled Engineering Applications,” AIAA J., 32�8�, pp. 269–289.
Cooling Airflow to a Turbine Disc by Measuring the Air Temperature in the �17� Wilcox, D. C., 1998, Turbulence Modelling for CFD, 2nd ed., DCW Indus­
Rotating Channels,” ASME Paper 81-GT-132. tries, La Canada, CA.
�2� El-Oun, Z. B., and Owen, J. M., 1989, “Preswirl Blade-Cooling Effectiveness �18� Ansys Inc., “CFX User Documentation,” Version 5.7, Ansys, Inc., Canons-
in an Adiabatic Rotor-Stator System,” ASME J. Turbomach., 111, pp. 522– burg, PA.
529. �19� Kader, B. A., 1981, “Temperature and Concentration Profiles in Fully Turbu­
�3� Geis, T., Dittmann, M., and Dullenkopf, K., 2003, “Cooling Air Temperature lent Boundary Layers,” Int. J. Heat Mass Transfer, 24�9�, pp. 1541–1544.
Reduction in a Direct Transfer Preswirl System,” ASME Paper GT2003­ �20� Dittmann, M., Dullenkopf, K., and Wittig, S., 2004, “Discharge Coefficients of
38231. Rotating Short Orifices With Radiused and Chamfered Inlets,” ASME J. Eng.
�4� Chew, J. W., Ciampoli, F., Hills, N. J., and Scanlon, T., 2005, “Pre-Swirled Gas Turbines Power, 126, pp. 803–808.
Cooling Air Delivery System Performance,” ASME Paper GT2005-68323. �21� Owen, J. M., and Rogers, R. H., 1989, “Flow and Heat Transfer in Rotating-
�5� Farzaneh-Gord, M., Wilson, M., and Owen, J. M., 2005, “Numerical and The­ Disc Systems,” Rotor-Stator Systems, Research Studies Press, Taunton, UK,
oretical Study of Flow and Heat Transfer in a Pre-Swirl Rotor-Stator System,” Vol. 10.
ASME Paper GT2005-68135. �22� Luo, X., Zhang, C., Xu, G., Tao, Z., and Ding, S., 2004, “Measurements of
�6� Karabay, H., Wilson, M., and Owen, J. M., 2001, “Predictions of Effect of Surface Temperature Distribution on a Rotating Disk With Blade Cooling
Swirl on Flow and Heat Transfer in a Rotating Cavity,” Int. J. Heat Fluid Flow, Holes Using Thermochromic Liquid Crystal,” ASME Paper GT2004-53519.

Journal of Engineering for Gas Turbines and Power JULY 2007, Vol. 129 / 777

Downloaded 29 May 2008 to 138.38.0.54. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Appendix 3

Proceedings of ASME Turbo Expo 2008: Power for Land, Sea

and Air

Effect of Radial Location of Nozzles on Performance of Pre-Swirl Systems

 Proceedings of ASME Turbo Expo 2008: Power for Land, Sea and Air
GT2008
June 9­13, 2008, Berlin, Germany

GT2008 ­ 50295

EFFECT OF RADIAL LOCATION OF NOZZLES ON PERFORMANCE

OF PRE­SWIRL SYSTEMS

Paul Lewis Mike Wilson Gary Lock J. Michael Owen

p.r.lewis@bath.ac.uk m.wilson@bath.ac.uk g.d.lock@bath.ac.uk j.m.owen@bath.ac.uk

Dept of Mechanical Engineering

University of Bath
Bath BA2 7AY
United Kingdom

ABSTRACT for an optimum pre­swirl configuration an engine designer

This paper investigates the effect of the radial location of would place the pre­swirl nozzles at a high radius.
the inlet nozzles on the performance of a direct­transfer pre­
swirl system in a rotor­stator wheel­space. A commercial code NOMENCLATURE
is used to solve the Reynolds Averaged Navier Stokes (RANS) a,b rotor inner radius, rotor outer radius
equations using a high­Reynolds­number k­ε / k­ω turbulence A ,B combined free and forced vortex coefficients
model with wall functions near the boundary. The 3D steady Ab receiver hole area
state model has previously been validated against experimental cw non­dimensional mass flow rate ( = m � / µb )
results from a scale model of a gas turbine rotor­stator system. cp specific heat capacity at constant pressure
Computations are performed for three inlet­to­outlet radius Cd discharge coefficient
ratios, rp/rb = 0.8, 0.9 and 1.0, a range of pre­swirl ratios,
CM disc moment coefficient ( = M / 12 ρΩ 2 b 5 )
0.5_<_βb_< 2.0, and varying flow parameter, 0.12 < λT < 0.36.
The rotational Reynolds number for each case is 106. G gap ratio ( = s / b )
The flow structure in the wheel­space and in the region m� mass flow rate
around the receiver holes for each inlet radius is related to the M moment on one side of the disc
swirl ratio. The performance of the system is quantified by two Reφ rotational Reynolds number ( = ρΩb2/µ)
parameters: the discharge coefficient for the receiver holes r radius
(Cd,b) and the adiabatic effectiveness for the system (Θb,ad). rp , r b radii of pre­swirl nozzles and receiver holes
As in previous work, the discharge coefficient is found to s rotor­stator separation distance
reach a maximum when the rotating core of fluid is in T temperature
synchronous rotation with the receiver holes. As the radius ratio V velocity
is increased this condition can be achieved with a smaller value x non­dimensional radius ( = r / b)
for pre­swirl ratio β b. A simple model is presented to estimate α flow angle relative to axial direction
the discharge coefficient based on the flow rate and swirl ratio β swirl ratio ( = Vφ / Ωr )
in the system.
The adiabatic effectiveness of the system increases linearly βb pre­swirl ratio based on rb (= Vφ , p / Ωrb )
with pre­swirl ratio but is independent of flow rate. For a given βp pre­swirl ratio based on rp (= Vφ , p / Ωr p )
pre­swirl ratio, the effectiveness increases as the radius ratio
γ ratio of specific heats
increases. Computed values show good agreement with
analytical results. Both performance parameters show θ inlet nozzle angle to the tangential direction
improvement with increasing inlet radius ratio, suggesting that Θ b,ad adiabatic effectiveness ( cp(To,p−Tt,b)/ 12 Ω 2 rb2 )

1 Copyright © 2008 by ASME

 λT turbulent flow parameter ( = c w Reφ−0.8 )
µ dynamic viscosity
ρ density
Ω angular velocity of rotor
Subscripts
ad adiabatic
b blade­cooling
e effective
i isentropic value
max maxium value
o total value in stationary frame

out at receiver hole outlet

p pre­swirl
s stator
t total value in rotating frame
w rotor
φ ,r ,z circumferential, radial, axial direction
∞ value in core at z/s = 0.5, r/rb = 1
1,2 upstream, downstream locations in streamtube Fig 1 Simplified diagram of a direct­transfer pre­swirl system

INTRODUCTION Lewis et al. [8] carried out a combined computational and

A simplified diagram of the so­called direct­transfer pre­ experimental study of a direct­transfer system. The computed
swirl system, where the blade­cooling air is supplied to the adiabatic effectiveness was in good agreement with the
rotating blades by stationary angled pre­swirl nozzles, is shown theoretical expression derived by Farzaneh­Gord et al., and the
in Fig 1. The nozzles swirl the air, and this reduces the work computed values of Cd,b, the discharge coefficient for the
done by the rotating turbine disc in accelerating the air to the receiver holes, reached a maximum value at a critical value of
disc speed. This consequently reduces the total temperature of βp, the pre­swirl ratio of the cooling air.
the air entering the receiver holes in the disc, Meierhofer and Lewis et al studied a system having the pre­swirl inlet at a
Franklin [1]. The designer is interested in calculating the lower radius than that of the receiver holes (as illustrated in
pressure drop and cooling effectiveness of the pre­swirl system, Fig._1), while the 'Karlsruhe' rig used by Dittman et al involved
and there is also a need to understand the heat transfer between pre­swirl nozzles and receiver holes at the same high radius.
the cooling air and the turbine disc. Jarzombek et al [9] studied computationally a configuration
Heat transfer in a direct transfer rig was studied with the pre­swirl nozzles located radially outward of the
experimentally and computationally by Wilson et al. [2]. Total receiver holes, finding the flow to conform to free vortex
temperature probes were used to measure the temperature of the behaviour. It is the object of the present paper to determine the
air entering the receiver holes, which was consistently under­ effect of the radial location of the pre­swirl nozzles on system
predicted by axisymmetric CFD computations. performance.
Geis et al. [3] made measurements of adiabatic In previous papers by the authors, in which only one
effectiveness which showed that the measured values of Tt,b, the location for the pre­swirl nozzles was considered, the pre­swirl
total temperature of the air in the rotating frame of reference ratio β p was defined as:
entering the receiver holes, were significantly higher than the
values predicted from their ideal model. Chew et al. [4] made β p =Vφ , p / Ωr p (1)
numerical simulations that were in good agreement with results
from both the 'Karlsruhe' pre­swirl rig used by Geis et al. and a
In an engine, the total pressure upstream of the pre­swirl
pre­swirl rig at Sussex University. Chew et al. and Farzaneh­
nozzles is fixed. If the static pressure in the core was also fixed
Gord et al. [5] derived independently theoretical models for the
then the pre­swirl velocity Vφ,,p would be invariant with radius.
adiabatic effectiveness.
Under these conditions, it is convenient to define a new pre­
Dittmann et al. [6] measured the discharge coefficients for
swirl ratio, βb say, where:
the pre­swirl nozzles and receiver holes in a direct­transfer
system, and Yan et al. [7] measured the discharge coefficients
for receiver holes. β b =Vφ , p / Ωrb (2)

2 Copyright © 2008 by ASME

 (where b = 0.216 m) by a shroud attached to each disc, and the
r=b centre of the system is sealed by a stationary hub. Clearances
between rotating and stationary surfaces are set to zero.
A cylinder, representing the blade cooling passage, of
r = rb diameter 8 mm and length 10 mm is attached to the rotor at a
radius rb_=_0.200 m and rotates with the disc. The pre­swirl
nozzles are represented by an annular slot which can be placed
at one of three radial locations such that rp/rb = 0.8, 0.9 or 1.0.
The height of the annular slot is adjusted so that the pre­swirl
inlet area remains the same at each of these three radial
locations. The same set of values for inlet mass flow rate, in the
range 0.12_<_λT_<_0.36, is used for each configuration, and
the pre­swirl ratio β b is varied as a parameter. The rotational
Reynolds number considered is Reφ = 106 (corresponding to a
rotor disc speed of 3800 rpm).
The surface mesh for the geometry is created using a
Delaunay triangulation with prismatic elements near the
boundaries for better near wall resolution. An advancing front
mesher is then used to resolve the volume.
CFX­10, a commercial 3D finite volume multi­grid
r computational fluid dynamics (CFD) package is used to solve
z the Reynolds Averaged Navier­Stokes (RANS) equations. The
second order accurate advection scheme is based on the method
of Barth and Jesperson [10]. The energy equation is solved
including the viscous work term, and the effects of variable
density are taken into account. Buoyancy effects within the
wheel­space are ignored.
Fig. 2 Schematic diagram of the flow domain used for the The turbulence model used is the high­Reynolds­number
computational study BSL model of Menter [11]. This is a blend of a k­ω formulation
with wall functions in the near wall region, Wilcox [12], and a
k­ε model away from the wall. This overcomes sensitivities to
free stream turbulence levels experienced by k­ω models.
The equations were solved in the rotating frame such that a
such that βb is invariant with rp, which is the assumption made steady state analysis could be used. Axial and circumferential
for the computations presented below. This will make it easier velocities were specified at the annular slot being used for the
to identify the effect of nozzle location on pre­swirl inlet. The two remaining slots in each analysis were treated as
performance. (As shown in the Appendix, the static pressure in part of the solid stator. A static pressure boundary condition was
the core does vary with radius, and this has an effect on βb used at the outlet.
which reduces, but does not negate, the advantages of locating
the pre­swirl nozzles at as high a radius as practicable.)
The computational method is described below, and
subsequent sections consider the effect of nozzle location on the FLOW STRUCTURE
flow structure, the discharge coefficient for the receiver holes, The computed flow structure in the radial (r­z) plane for
the pressure drop in the system and the adiabatic effectiveness. typical flow rates and pre­swirl ratios is shown in Fig. 3(a­c) for
case where the inlet is located at rp/rb = 0.8, 0.9 and 1.0
respectively. The circumferential location φ of the plane shown
coincides with the centre­line of the receiver hole. (There is
COMPUTATIONAL METHOD little effect of circumferential location on the flow structure
The computational method and configuration is similar to except in the immediate region of the receiver hole.) For
that described in Lewis et al. [8], with adjustments made for rp/rb_=_0.8 the inlet flow impinges upon the rotating disc and
multiple inlets; the salient details are included here for travels radially outwards, forming the rotor boundary layer.
completeness. The computational domain, Fig. 2, is a 6o section Radial inflow occurs on the stator and a pair of counter­rotating
of a wheel­space bounded by a rotor and a stator disc with axial vortices can be observed inboard of the inlet.
gap ratio G = 0.051. Cyclic symmetry is imposed at the
circumferential boundaries. The system is sealed at its periphery

3 Copyright © 2008 by ASME

      Inlet A                        Inlet B                        Inlet C 
          rp/rb = 0.8                   rp/rb = 0.9                    rp/rb = 1.0 
 
 

βb = 0.40 βb = 0.45 βb = 0.50

(a) (b) (c)

  
βb = 0.80 βb = 0.90 βb = 1.00
   a) rp/rb = 0.8               b) rp/rb = 0.9                c) rp/rb = 1.0  (d) (e) (f)
 

Fig. 3 Computed flowfields in the axial­radial plane 
 

 
As  the  inlet  is  moved  radially  outwards  the  circulation  in 
the  outer  part  of  the  system  becomes  more  compressed.  The 
pair of counter­rotating vortices inward  of the inlet expands  to 
fill the available space, the larger of the two vortices being that 
with outflow on the rotor. 
βb = 1.20 βb = 1.35 βb = 1.50
The flow is most complex for the  case when rp/rb is unity, 
(g) (h) (i)
Fig.  3(c).  Some  of  the  inlet  flow  enters  the  receiver  holes 
directly,  while  the  remaining  flow  impinges  upon  the  region 
between  the  holes.  The  impinging  flow  spreads  both  radially 
inwards  and  radially  outwards  from  the  impingement  region. 
The inward flow encounters the rotor boundary layer flow and 
separates  from  the  disc,  creating  the  small  recirculation  on the 
rotor side inward of the receiver hole. 
Fig. 4 shows flow streamlines in the tangential (φ­z) plane 
at the receiver hole radius rb in a frame of reference rotating at  βb = 1.60 βb = 1.80 βb = 2.00
the  speed  of  the  rotor  (in  the  left  to  right  direction).  In  each  (j) (k) (l)
image the stator is at the bottom and the receiver hole and outlet   

is  at  the  top.  The  three  columns  represent  the  three  inlet  Fig.  4  Computed  flow  structure  in  the  tangential  plane  at  the 
positions,  rp/rb  =  0.8,  0.9  and  1.0  respectively,  and  the  rows  receiver  hole  radius,  in  a  frame  of  reference  rotating  with  the 
represent increasing values of pre­swirl ratio.  rotor (direction of motion from left to right). λT = 0.24. 
For rp/rb = 0.8, Fig. 4(a) shows a case for which  βb = 0.40   
and is therefore ‘under­swirled’. The receiver hole rotates more   
quickly than the flow in the core, therefore the flow enters at an  As  the  inlet  radius  is  increased,  the  inlet  pre­swirl  ratio 
acute  angle,  separating  at  the  leading  edge  of  the  hole  and  required  to  produce  this  synchronous  rotation  is  reduced.  For 
causing a recirculation inside the hole. As the pre­swirl ratio is  rp/rb  =  0.9,  synchronous  rotation  occurs  when  βb  =  1.35  as 
increased  the  angle  at  which  the  flow  enters  the  receiver  hole  shown in Fig. 4(h) and for rp/rb = 1.0, when βb = 1.0 as shown in 
tends  towards  the  axial  direction.  At  the  point  where  Fig.  4(f).  When  the  swirl  ratio  is  increased  further,  or  ‘over­
synchronous rotation between the flow and the hole occurs the  swirled’,  the  flow  rotates  more  quickly  than  the  receiver  hole, 
flow would be expected to flow axially into the receiver hole, as  causing  separation  and  a  region  of  recirculation  at  the  trailing 
can be seen in Fig. 4(j), for which βb = 1.6.  edge of the hole. 

  4  Copyright © 2008 by ASME 
 2
2 rp/rb= 0.8 (Inlet A)
rp/rb= 0.9 (Inlet B)
a) rp/rb= 1.0 (Inlet C)

1.5 βb = 1.60
1.5

β∞
βb = 1.20
1
β

1
βb = 0.80

0.5 βb = 0.40 0.5

outlet inlet
0 0
0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2
x­2
βb
2
Fig. 6 Computed variation of β∞ with βb for the three inlet
b) βb = 1.80
locations, λT = 0.24.

1.5

βb = 1.35
The variation of swirl ratio β = Vφ/Ωr midway between the
rotor and stator (z/s=0.5) and on a radial line midway between
receiver holes is shown in Fig. 5. Fig. 5(a), (b) and (c) again
β

1
βb = 0.90
correspond to the inlet at rp/rb = 0.8, 0.9 and 1.0 respectively.
The horizontal axis is the non­dimensional radius x = r/b.

In each case a peak is seen in the swirl ratio at the inlet

βb = 0.45 radius due to the high momentum inlet fluid. At higher radius
0.5 (i.e. lower values of x­2 ) there is a linear variation of β with x­2,
consistent with free vortex behaviour. The dashed line on each
plot is a least squares best fit of the data in the linear region.
outlet inlet There is considerable uncertainty associated with the fit
0 (especially as the region of linear behaviour becomes smaller),
0 0.5 1 1.5 2 2.5
x ­2 however these results suggest that the flow is related to a
2
βb = 2.00 Rankine (combined free and forced) vortex, for which:
c) βb = 1.50
β = Ax −2 + B (3)
1.5

βb = 1.00 where A and B are constants. This behaviour was found by

Mizaee et al [13] to occur in a rotating cavity with a stationary
outer casing.
1
β

Fig. 6 shows the variation of swirl ratio in the core at the

βb = 0.50 radius of the receiver hole, β∞, with the pre­swirl ratio at inlet
βb. Each line represents a different inlet location and is
approximately linear. Extrapolating back to the βb_=_0
0.5
condition, the value for β∞ would be expected to lie between
0.43, the value for turbulent flow in a sealed rotor­stator system
inlet
(see Owen and Rogers, [14]) and zero, due to the effect of a
outlet
zero­swirl superposed flow on the swirl in the rotating core of
0
0 0.5 1 1.5 2 2.5 fluid between the discs. It can be seen that a significant increase
x­2 in inlet pre­swirl is required for the inlets at lower radii to
Fig. 5 Computed tangential velocity distributions on the axial achieve the synchronous rotation condition discussed above,
mid­plane, a) rp/rb = 0.8, b) rp/rb = 0.9, c) rp/rb = 1.0 and illustrated by the horizontal dashed line in Fig. 6.

5 Copyright © 2008 by ASME

 1 rp/rb=0.8 (λT =0.24) The isentropic mass flow rate through the receiver holes
rp/rb=0.9 (λT =0.24) was derived by Yan et al [7] using the First Law of
a) rp/rb=1.0 (λT =0.24) Thermodynamics for an adiabatic system, taking into account
rp/rb=0.8 (Inlet angle=20 o) the work done by or on the fluid as it passes from station 1 in
o
0.8 rp/rb=0.9 (Inlet angle=20 )
o the fluid core to station 2 in the receiver holes. It is given by:
rp/rb=1.0 (Inlet angle=20 )
Measurements, Lewis et al. [8]
1/ 2
⎧ ⎡ γ −1 ⎤ ⎫
0.6 1 ⎪⎛ 2γ ⎞ p
0,1 ⎢ ⎛⎜ p 2 ⎞⎟ γ ⎥ ⎪
m� i ⎛ p 2 ⎞⎟ γ ⎪⎪⎜⎜ γ − 1 ⎟⎟ ρ ⎢1 − ⎜ ⎟ ⎥ ⎪⎪
= ρ 0,1 ⎜ ⎨ ⎝ ⎠ ⎢ ⎝ p 0,1 ⎠ ⎥ ⎬ (5)
CD

Cd,b ⎜ 0,1
Ab ⎝ p 0,1 ⎟⎠ ⎪ ⎣ ⎦⎪
0.4
⎪
(
⎪⎩+ 2Ω r2 Vφ ,2 ) 2 ⎪
− r1 Vφ ,1 − Vφ ,2 ⎪⎭

The first term inside the curly brackets is the standard

0.2 result for compressible flow in a stationary nozzle; the second
term is the work term resulting from the change of angular
momentum of the air; the last term is due to the fact that the air
in the receiver holes has an absolute tangential, as well as an
0 axial, component of velocity. It should be noted that failure to
0 0.5 1 1.5 2
βb use the correct equation for m� i can result in calculated values of
1
Cd,b exceeding unity, which is clearly nonsensical.

Fig. 7 shows this discharge coefficient calculated using

b) locations 1 and 2 as a point in the core at the radius of the

0.8 receiver holes and the outlet plane respectively. Fig 7a, where
βb is on the horizontal axis, clearly shows that to maximize the
discharge coefficient for low radius inlets, the pre­swirl ratio
must be greater than unity. The effect of varying pre­swirl ratio
0.6 at a fixed non­dimensional flow rate λT (the equivalent of
physically altering the inlet pre­swirl angle) is to produce a
CD

Cd,b
sharp change in the relationship between Cd,b and βb when the
0.4
maximum value of Cd,b is reached.
Lewis et al [8] presented computations and measurements,
for the rp/rb = 0.8 configuration also considered here, for a fixed
inlet flow angle of 20º to the tangential in the direction of
0.2 rotation of the disc and variable flow rate. In this case, the inlet
pre­swirl ratio is proportional to the non­dimensional flow rate
λT used (at fixed Reφ). Results for computations at this same
fixed inlet flow angle (for 0.12 < λT < 0.36) are also shown in
0
0 0.5 1 1.5 2 Fig. 7a, and show a smaller effect of rp/rb on the variation of Cd,b
β∞ with βb. Fig. 7b shows that, when the pre­swirl ratio is varied at
Fig. 7 Computed variation of discharge coefficient Cd,b a fixed value of λT, a nearly symmetric variation of Cd,b around
a) with βb, b) with β∞ the point of synchronous rotation (β∞_=_1) is obtained. (The
values of inlet pre­swirl ratio βb required to achieve
synchronous rotation at the receiver hole radius are discussed
above in connection with Fig. 4.) There is reasonably good
DISCHARGE COEFFICIENT agreement between the computations and the measurements
The discharge coefficient Cd,b is defined here as the ratio of made by Lewis et al for rp/rb = 0.8 and fixed inlet flow angle.
the actual mass flow rate through the receiver holes m� b to the It was shown above that the effect of ‘under­swirling ‘ or
‘over­swirling’ the core body of fluid caused the flow to enter
isentropic mass flow rate m� i such that:
the receiver hole at an angle to the axial direction. This means
that the effective area of the receiver hole, as ‘seen’ by the flow,
�b
m is reduced. It is logical that, since the flow rate is proportional
C d,b = (4)
�i
m to the orifice area, Cd,b will reduce linearly as the effective area
is reduced.

6 Copyright © 2008 by ASME

 0.2
1 rp/rb=0.8 (λT=0.24)
rp/rb=0.9 (λT=0.24)
rp/rb=1.0 (λT=0.24)
o
rp/rb=0.8 (Inlet angle=20 )
o
0.8 rp/rb=0.9 (Inlet angle=20 )
0.15 rp/rb=1.0
o
(Inlet angle=20 )

(po,p­pt,out)/pt,out
Cd,b/C d,max

0.6
0.1

0.4

rp/rb=0.8 (λT=0.24) 0.05

rp/rb=0.9 (λT=0.24)
0.2
rp/rb=1.0 (λT=0.24)
eq. (8) (λT=0.24)

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
β∞ βb
Fig. 8 Variation of Cd,b /Cd,max with β∞ Fig. 9 Computed variation of pressure drop from pre­swirl
inlet to receiver­hole outlet with βb

The reduction in effective area Ae can be expressed as a C d,b Ae 1

function of the flow angle at the receiver hole, shown in eq. (6), = = (8)
C d,max Ab 2
where α is the flow angle measured from the axial direction. ⎛ A ρ β − 1 Ωr ⎞
⎜ b ∞ b ⎟
1+ ⎜ ⎟
⎜ µbλT Reφ
0.8
Ae ⎟
= cosα (6) ⎝ ⎠
Ab
where Cd,max is the value of Cd,b when β∞ = 1.
An equation can be formed for α by considering the ratio of Fig. 8 shows Cd,b /Cd,max for a range of conditions and flow
the tangential velocity in the rotating frame and the axial rates. The model underpredicts the computational data, which
velocity. The radial component of velocity is ignored as a large suggests that the predicted flow angle is too large, so that the
volume of the flow enters the hole from the core rather than predicted effective area is too small. At the entrance to the hole
from the boundary layer and therefore has very low radial the circumferential velocity will be somewhere between that of
velocity (see Lewis et al discussion on ‘direct’ and ‘indirect’ the fluid in the core and that of the receiver hole.
routes). Hence: Fig. 9 shows the total pressure loss through the system,
using the total pressure in the stationary frame at inlet and in the
Vφ ,∞ − Ωrb rotating frame at outlet. This difference is largely independent
tan α = (7a) of flow rate. There is a significant increase in pressure loss as
Vz,b
the pre­swirl ratio is increased. Fig. 7a shows that the discharge
where: coefficient for the receiver holes increases as the ratio rp/rb
increases; Fig. 9 shows however that increasing rp/rb causes a
�b
m slight increase in pressure drop.
Vz,b = (7b)
ρ Ab
and

ADIABATIC EFFECTIVENESS
Vφ ,∞ − Ω rb = β ∞ − 1 Ω rb (7c)
The adiabatic effectiveness Θb,ad is defined as the non­
dimensional change in total temperature between the nozzles in
Using the definition of λT given in the nomenclature it the stationary frame and the receiver holes in the rotating frame:
follows that

7 Copyright © 2008 by ASME

 0.01
2.5 rp/rb=0.8 ­ Computations rp/rb=0.8 ­ Rotor
rp/rb=0.9 ­ Computations 0.008 rp/rb=0.9 ­ Rotor
rp/rb=1.0 ­ Computations rp/rb=1.0 ­ Rotor
2 rp/rb=0.8 ­ Karabay et al. rp/rb=0.8 ­ Stator
rp/rb=0.9 ­ Karabay
Eq. (11)et al. rp/rb=0.9 ­ Stator
0.006
rp/rb=1.0 ­ Karabay et al. rp/rb=1.0 ­ Stator
1.5

0.004
Θb,ad

Cm
0.002
0.5

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
βb βb
­0.5 ­0.002

­1 ­0.004
Fig. 10 Computed variation of adiabatic effectiveness Θb,ad Fig. 11 Computed values of moment coefficient CM on rotor and
with βb stator, λT = 0.24.

For turbine blade cooling the effectiveness should be as

c p (To,p − Tt,b ) high as possible as this ensures that the fluid reaching the blades
Θ b,ad = (9) has the lowest possible total temperature. In general,
1/ 2 Ω 2 rb2 configurations with low pre­swirl ratios require work to be
performed on the flow by the rotor to bring its tangential
A theoretical value for Θb,ad was derived by Karabay et al velocity to that of the receiver holes, and this work input raises
[15] for a cover­plate system using the First Law of the total temperature of the flow. Conversely, configurations
Thermodynamics. The equivalent theoretical expression derived with high pre­swirl ratios perform work on the rotor, thus
by Farzeneh­Gord et al [5] for the direct­transfer system reducing the total temperature.
considered here is: As shown in eq. (11), the relationship between the pre­swirl
ratio and the adiabatic effectiveness is approximately linear,
2 with the gradient dependent on rp/rb. As the radius of the inlet
⎛ rp ⎞ Ms
Θ b,ad = 2β p ⎜ ⎟ −1− (10) increases there is consequently a resulting increase in
⎜r ⎟ � Ωrb2
1/ 2m effectiveness.
⎝ b ⎠
A secondary effect adding to the improvement in
When expressed in terms of βb, the pre­swirl ratio based on effectiveness for high radius inlets is the reduction in moment
the receiver­hole radius, this relationship becomes on the stator. This is shown in Fig. 11; as the pre­swirl ratio is
increased, the moment on the stator increases and that on the
rotor decreases. The change in moment coefficient is slightly
rp Ms less for rp/rb = 1 than for the other two locations.
Θ b,ad = 2 β b −1− (11)
rb � Ωrb2
1/ 2 m

Fig. 10 shows the computed effectiveness plotted against the CONCLUSIONS

relationship in eq. (11). Note that a computed moment on the The effect of nozzle location on the fluid dynamics of a
stator is necessary to evaluate the relationship and hence only pre­swirl system has been studied computationally. The model,
discrete values are available. The agreement between the two is which has been validated previously using an experimental rig
excellent. As shown in the Appendix, in practice the increase of at the University of Bath, produces a flow structure
Θb,ad with increasing rp will be less than that shown in Fig. 10. representative of that found in gas turbine engines.

8 Copyright © 2008 by ASME

Computations were performed at Reφ = 106, 0.12 < λT < 0.36 [7] Yan, Y., Farzaneh­Gord, M., Lock, G., Wilson, M., and
and 0.5 < βb < 2.0. Owen, J. M., 2003, “Fluid dynamics of a pre­swirl rotor­
The main conclusions from the study are: stator system,” ASME J. Turbomach., 125, pp. 641–647.
1. Cd,b, the discharge coefficient for the receiver holes, is [8] Lewis, P., Wilson, M., Lock, G. D., and Owen, J., 2007.
maximized when the core flow is in synchronous rotation “Physical interpretation of flow and heat transfer in pre­
with the holes (β∞ = 1). swirl systems”, J. Engineering for Gas Turbines and
2. A simple model based on the effective receiver­hole area Power, 129, pp 769­777
can be used to estimate the reduction in Cd,b when β∞ ≠ 1. [9] Jarzombek, K, Benra, F.­K., H. Dohmen, J. and Schneider,
3. The maximum Cd,b is achieved at a value of βb that O., 2007, “CFD analysis of flow in high­radius pre­swirl
decreases as rp/rb increases, with a corresponding slight systems”, ASME paper GT2007­27404
increase in the pressure drop in the system. [10] Barth, T. J., and Jesperson, D. C., 1989, “The design and
4. The adiabatic effectiveness increases as rp/rb increases, and application of upwind schemes on unstructured meshes,”
computed values are in excellent agreement with the Proc. 27th AIAA Aerospace Sciences Meeting, Reno, NV,
theoretical analysis. January 9–12.
As shown in the Appendix, the variation of static pressure [11] Menter, F. R., 1994, “Two­equation eddy viscosity
in the core will reduce, but will not negate, the advantages of turbulence models for engineering applications,” AIAA J.,
locating the pre­swirl nozzles at as high a radius as practicable. 32, pp. 269–289.
It is also shown in the Appendix that the increase in [12] Wilcox, D. C., 1998, “Turbulence modelling for CFD”,
effectiveness as rp increases is caused solely by losses in the 2nd ed., DCW Industries, La Canada, CA.
nozzles and in the core. [13] Mirzaee, I., Gan, X., Wilson, M. and Owen, J. M., 1998,
“Heat transfer in a rotating cavity with a peripheral inflow
and outflow of cooling air”, ASME J. Turbomach, 120, pp
818­823
ACKNOWLEDGMENTS [14] Owen, J. M. and Rogers, R. H., 1989, “Flow and heat
The Engineering and Physical Sciences Research Council transfer in rotating­disc systems (vol. 1 Rotor­stator
(EPSRC) sponsored Paul Lewis' research at the University of systems)”, Research Studies Press, Taunton, UK
Bath through Doctoral Training Account EP/P500036/1. The [15] Karabay, H., Wilson, M., and Owen, J. M., 2001,
authors thank the reviewers for comments that have given rise “Predictions of effect of swirl on flow and heat transfer in
to improvements to the focus and completeness of the paper. a rotating cavity,” Int. J. Heat Fluid Flow, 22, pp. 143–
155.

REFERENCES
[1] Meierhofer, B., and Franklin, C. J., 1981, “An APPENDIX: EFFECT OF CORE SWIRL ON PRE­SWIRL
investigation of a preswirled cooling airflow to a turbine RATIO
disc by measuring the air temperature in the rotating If VT is the total velocity leaving the nozzles, then for
channels,” ASME Paper 81­GT­132. incompressible flow
[2] Wilson, M., Pilbrow, R., and Owen, J. M., 1997, “Flow
and heat transfer in a pre­swirl rotor­stator system,” 1/ 2
ASME J. Turbomach., 119, pp. 364–373. ⎡ 2 ( p0 − p p ) ⎤
VT = C d , p ⎢ ⎥ (A1)
[ 3] Geis, T., Dittmann, M., and Dullenkopf, K., 2004, ⎢⎣ ρ
⎥⎦
“Cooling air temperature reduction in a direct transfer
preswirl system,” J. Engineering for Gas Turbines and
where p0 is the total pressure upstream of the nozzles, pp is the
Power, 126, pp 809­815.
static pressure in the core at the radius of the nozzles, Cd,p is the
[4] Chew, J. W., Ciampoli, F., Hills, N. J., and Scanlon, T.,
discharge coefficient for the nozzles, and
2005, “Pre­swirled cooling air delivery system
performance,” ASME Paper GT2005­68323.
[5] Farzaneh­Gord, M., Wilson, M., and Owen, J. M., 2005, Vφ , p β
b Ωrb
VT = = (A2)
“Numerical and theoretical study of flow and heat transfer cos θ cos θ

in a pre­swirl rotor­stator system,” ASME Paper GT2005­

68135. θ being the nozzle angle to the tangential direction. Hence
[6] Dittmann, M., Geis, T., Schramm, V., Kim, S., and Wittig,
S., 2002, “Discharge coefficients of a preswirl system in 1/ 2
secondary air systems,” ASME J. Turbomach., 124, pp. Cd , p cos θ ⎧ 2 ( p0 − p p ) ⎫
β
b = ⎨ ⎬ (A3)
119–124. Ω rb ⎩ ρ
⎭

9 Copyright © 2008 by ASME

If β b,1 is the value of β b when rp/rb = 1, it follows that The results corresponding to Fig. 5 are shown in Table A1
for an assumed value of Cd,p = 0.9 and θ = 20º, which
1/ 2 corresponds to the nozzle angle in the experimental rig.
Cd , p cos θ ⎧2 ( p0 − pb ) ⎫
β b,1 = ⎨ ⎬ (A4)
Ω rb ⎩ ρ ⎭ Consider the ideal case where Cd,p = 1, θ = 0 and free­
vortex flow occurs in the core, such that A = 0 and
and hence
−2
Vφ ⎛r⎞
−1/ 2 = B⎜ ⎟ (A10)
β b ⎧⎪ Cd , p cos θ ( pb − p p) ⎫⎪
2 2 Ωr ⎝b⎠
= ⎨1− ⎬ (A5)
β b,1 ⎪ 1/ 2 ρ β b 2 Ω 2 rb 2 ⎪
⎩ ⎭ It follows that

Consider a combined vortex in the core, as shown in Fig. 5, −2

Vφ ,p ⎛ rp ⎞
where = B ⎜⎜ ⎟⎟ (A11)
Ωrp ⎝b⎠
−2
Vφ ⎛r⎞
= A + B ⎜ ⎟ (A6) and
Ωr ⎝b⎠
−1
It follows that Vφ ,p ⎛ rp rb ⎞
βb = = B ⎜⎜ 2 ⎟
⎟ (A12)
Ωrb ⎝ b ⎠
⎧⎪ −2 −4
dp Vφ 2 ⎛r⎞ ⎛ r ⎞ ⎫⎪
=ρ = ρΩ 2 r ⎨ A 2 + 2 AB⎜ ⎟ + B 2 ⎜ ⎟ ⎬ (A7)
dr r ⎪⎩ ⎝b⎠ ⎝ b ⎠ ⎭⎪ Hence

rp rb
Integrating from r = rp to r = rb B = βb (A13)
b2
⎧ ⎛ r2⎞ −2
⎛r ⎞⎫
⎪A 2 ⎜1− p ⎟ − 4AB⎜⎛ b ⎞⎟ ln⎜ p ⎟ ⎪
r and eq (A9) reduces to
⎪ ⎜ rb 2 ⎟ ⎝b⎠ ⎜r ⎟⎪
1 ⎪ ⎝ ⎠ ⎝ b ⎠⎪
p b − p p = ρΩ 2 rb 2 × ⎨ ⎬ (A8) β b rb
2 −4 ⎡ −2 ⎤ = (A14)
⎪ 2 ⎛ rb ⎞ ⎢⎛ rp ⎞ ⎪ β b,1 rp
⎪ + B ⎜ ⎟ ⎜⎜ ⎟⎟ −1⎥ ⎪
⎪⎩ ⎝ b ⎠ ⎢⎝ rb ⎠ ⎥
⎣ ⎦ ⎭⎪
Reference to eq (11) shows that for this ideal case (where
and from eq (A5)
β brp/rb is constant) the nozzle location has no effect on the
adiabatic effectiveness! However, Table A1 shows that for the
β b ⎧⎪ C d,p cos θ

2 2
real case, although the core swirl reduces the advantage, the
= ⎨1− ×

β b,1 ⎪ β b 2
effectiveness increases as rp increases. It can therefore be
⎩ concluded that the increase in effectiveness as rp increases is
−1/ 2
⎛ ⎛ r 2 ⎞ −2 −4 ⎡⎛ r −2 ⎤ ⎞⎫ caused solely by losses in the nozzles and in the core.
⎜ 2⎜ p ⎟ − 4AB ⎛⎜ rb ⎞⎟ ln⎛⎜ rp ⎞
⎟ + B 2 ⎛⎜ b ⎞⎟
r
⎢⎜ p
⎞
⎟ ⎟⎪
−1⎥ ⎟⎬
⎜ A ⎜ 1− 2 ⎟ ⎜ ⎟ ⎢⎜⎝ rb ⎟ ⎥ ⎟⎪
⎜ ⎝ rb ⎠ ⎝b⎠ ⎝ rb ⎠ ⎝b⎠ ⎠
⎝ ⎣ ⎦ ⎠⎭
(A9)

rp / rb 0.8 0.9
βb 0.40 0.80 1.20 1.60 0.45 0.90 1.35 1.80
A 0.17 0.32 0.42 0.43 0.31 0.35 0.46 0.49
B 0.13 0.21 0.32 0.48 0.11 0.28 0.43 0.62
β b / β b,1 1.12 1.09 1.08 1.07 1.08 1.05 1.04 1.04

Table A1 Effect of core swirl on β b / β b,1 according to eq (A9) with θ = 20 0 and Cd,p = 0.9.

10 Copyright © 2008 by ASME

Appendix 4

International Gas Turbine Congress 2007

Three-Dimensional Computations of Ingress in Gas Turbine Cooling

Systems
 Proceedings of the International Gas Turbine Congress 2007 Tokyo
December 3-7, 2007 IGTC2007 Tokyo TS-040
IGTC2007­ABS­10 

Three­Dimensional Computations of Ingress

in Gas Turbine Cooling Systems
Paul Lewis1 and Michael Wilson2
1
Department of Mechanical Engineering, University of Bath
2 
Department of Mechanical Engineering
University of Bath
Claverton Down, Bath BA2 3BA, United Kingdom
Phone: +44­1225­386325, FAX: +44­1225­386928, E­mail: m.wilson@bath.ac.uk

ABSTRACT  sc axial seal clearance [m]

This paper describes a computational study of ingress in a sim­ T, To static temperature, total temperature [K]
plified model of a gas­turbine rotor­stator wheel­space with an T non­dimensional static temperature (Eq. 6) 
axial clearance rim seal, with non­axisymmetric flow conditions v velocity [m/s]
created using a stator vane in an external mainstream. Steady­state vr average radial velocity in the seal (Eq. 4)
computational fluid dynamics (CFD) simulations are carried out
using the commercial CFD code CFX in order to investigate the x non­dimensional radius (=r/b)
effects of geometry and boundary condition assumptions on the y+ wall­distance Reynolds number
results, providing information for the design of simplified experi­ β swirl ratio (=vφ/Ωr)
mental apparatus. The computations are carried out for a rotational φb vane spacing angle [degrees]
Reynolds number of 2.5×106, such as might typically be used in η sealing effectiveness (Eq. 5)
experiments, and for non­dimensional values of mainstream and µ dynamic viscosity [kg/m/s]
sealing flow­rates selected to match some of the conditions that θ non­dimensional angle (=φ/φb)
might be encountered in engines. ρ density [kg/m3]
The computed results show that the amount of ingress into the � disc rotation rate [rad/s]
wheel­space depends upon the distance between the stator vane Subscripts
trailing edge and the rim seal. A recirculation region set up within r, φ, z radial, circumferential or axial component
the seal is responsible for transporting ingested fluid inwards into e external (mainstream), at mainstream inlet
the wheel­space at some circumferential locations and for sealing i internal (wheel­space)
the wheel­space from this ingress at others. s at sealing flow inlet
The sealing effectiveness of the rim seal is calculated from
computed levels of concentration of a tracer  scalar variable. The
concentration results illustrate the importance of three dimensional INTRODUCTION
effects, and computed heat transfer results show that frictional The internal air system in a gas­turbine engine is of great im­
heating due to the rotating disc needs to be considered in planning portance in providing cooling air to temperature­critical compo­
experiments to investigate ingress phenomena further. nents such as the turbine blades, and also pressurises the seals to
prevent ingestion of hot gases from the mainstream into the
wheel­spaces between rotating and stationary elements. Minimis­
NOMENCLATURE 
ing this ingestion, or ingress, is particularly important in the high
a wheel­space inner radius [m] pressure turbine stages immediately downstream of the combustion
b wheel­space outer radius [m] chamber, where the high temperature of ingested mainstream gas
C  concentration of scalar quantity can lead to fatigue and damage in the important region around the
CP pressure coefficient (Eq. 3) outer rim seal, including possibly the blade­cooling air  receiver
CW non­dimensional sealing flow rate (= m & s / µb) holes located near the periphery of the turbine disc.
The rotating flow between a rotating turbine disc (the rotor) and
CW,MIN minimum Cw required to prevent ingress (Eq.1)
an adjacent stationary casing (the stator) creates a radial pressure
G wheel­space gap ratio (=s/b) 
gradient that encourages ingestion (radially inward flow) of main­
Gc seal gap ratio (=sc/b) 
stream gas, as illustrated schematically in Fig. 1. The stationary
m& mass flow rate [kg/s]
vanes and rotating blades in the turbine annulus create
p static pressure [Pa]
non­axisymmetric, unsteady variations of pressure that also drives
r, φ, z radial, circumferential and axial coordinates hot gas into the wheel­space in regions of high external pressure.
Reφ rotational Reynolds number (Eq. 2a) The relative importance of the influence on ingress of the flow
Rez mainstream flow axial Reynolds number (Eq. 2b) inside the wheel­space, the stationary vanes and the rotating blades
s wheel­space rotor­stator axial separation [m] has been studied using simplified experimental rotating­disc rigs,
where greater control can be exerted and more detailed measure­
Copyright © 2007 by GTSJ ments can be made than is usually the case with instrumented
Manuscript Received on September 4, 2007 engine components. A summary of such research has been given by
Owen (2006).

­1­
 Bayley and Owen (1970) carried out experiments for so­called
“rotationally­induced” ingestion into an unconfined simple ro­
tor­stator system with an axial clearance rim­seal,  and using a
theoretical “orifice model” for the seal correlated the minimum
non­dimensional flow rate of sealing air, CW (= m& s / µb), required
to prevent ingress by the expression:

CW,MIN = ΦMIN × GC × Reφ , where ΦMIN = 0.61 (1)  r = b 

seal
(where GC is the non­dimensional axial width of the seal and Reφ is
the rotational Reynolds number based on the disc outer radius, b).
Graber et al. (1987)  made measurements of ingestion of carbon
dioxide into a rotor­stator wheel­space from a swirling external
mainstream flow for a number of different seal geometries. The
results were in reasonable agreement with Eq. (1), although ΦMIN
was found not to be constant but to decrease with increasing Reφ.
Phadke and Owen (1988a,b,c) found that, for “exter­ r = a 
nally­induced” ingestion due to a non­axisymmetric external flow,
CW,MIN was independent of Reφ and increased with increasing axial
Reynolds number, ReZ, for the external flow. Phadke and Owen and r

others have correlated both CW,MIN and η, the non­dimensional

sealing effectiveness based on local concentration of fluid from the z 

external flow, with a non­dimensional measure of the peak pressure

difference in the external flow.
A review of research carried out under engine­relevant main­ Fig. 1 Schematic of configuration studied computationally
stream conditions involving rotating blades as well  as stationary
vanes is given by Owen (2006). Wang et al (2007) discuss results
of unsteady computations and comparisons with experiments for A stator vane of generic geometry is included in the mainstream
different combinations of locations for the stator vanes, the axial annulus upstream of the seal, see Fig. 2, to provide circumferential
clearance seal and the rotor blades. The pressure field due to the variations of pressure and velocity. A 15o circumferential sector has
rotor blades was found to dominate ingestion for the closely­spaced been modelled computationally, representing the pitch between
stages considered. Roy et al (2007) computed velocity distributions each of twenty four stator vanes proposed for a future experimental
and measured sealing effectiveness for a radial clearance rim seal, rig. Two geometric configurations have been tested, having dif­
and found that large­scale unsteady flow structures could occur ferent axial spacings of 7.92mm and 15.42mm between the vane
within the wheel­space (other similar findings to this are also re­ trailing edge and the seal. These are denoted here as NEAR­VANE
ported by Owen, 2006). and FAR­VANE configurations respectively.
Correlations such as that given in Eq. (1) based on the results of The system has two inlets; the sealing air inlet at the inner radius
very much simplified experiments have been used by designers to of the wheel­space (at r = a, see Fig. 1) and the external mainstream
estimate the amount of sealing flow needed in engines. Better inlet upstream of the vane, at both of which uniform values for
theoretical models of ingress and improved versions of design velocity components and temperature are prescribed as described
correlations are likely to result from the further investigation of the below. An average static pressure is prescribed at the mainstream
individual contributions to ingress of the fluid dynamics of the outlet boundary. Cyclic symmetry and no­slip conditions are ap­
wheel­space, the non­axisymmetric external flow produced by the plied at other  boundaries as appropriate and all solid boundary
stationary vanes and the additional unsteady variations due to surfaces are assumed to be adiabatic.
rotating turbine blades. The present paper describes a computa­ An unstructured mesh has been used, with a blend of quadrilat­
tional study of the first two of these influences. A simplified ro­ eral elements near wall surfaces and a Delaunay triangulation in the
tor­stator system is modelled and a stator vane is used to produce a core away from them, rotated around the central axis. The mesh
representative pressure field in the external mainstream. Computa­ around the stator vane was generated using regular layers in the
tions are carried out, assuming steady flow in the absence of ro­ near wall region and an advancing front scheme in the core. Sensi­
tating blades, at conditions relevant to the experimental rigs that are tivity to mesh size was tested over a wide range, see Table 1.
used to provide understanding and detailed quantitative informa­
tion on some of the fundamental features of ingress. The results are
intended to be used to guide aspects of the design of new apparatus.

&e
m axial seal, clearance sc
COMPUTATIONAL DOMAIN AND MODEL 
The computational domain, illustrated in Fig. 1, comprises a
θ=0
rotor­stator wheel­space having an outer radius b = 0.216m and an
axial­clearance rotor­side rim seal leading to an outer annulus
representing the mainstream gas path through the turbine stage. The
annulus height is 10mm and the wheel­space gap ratio and seal gap z  15o
ratio are G = 0.07 and GC = 0.01 respectively. The radial offset to
the annulus boundary surface on the stator side was included as a
φ
result of preliminary computations that showed that the mainstream
flow could stagnate otherwise at the top of the seal on the rotor side, θ=1
leading to very low computed levels of ingress. This may be de­
sirable in practice, however experimental apparatus would be likely
to be designed to promote ingress at the chosen flow conditions for Fig. 2 Geometric arrangement for mainstream and stator vane
the purpose of making accurate and informative measurements. (cyclic symmetry imposed at boundaries at θ = 0 and θ = 1) 

­2­

 0.16
Cp
Table 1 Computational grid sizes studied
0.14

Elements

Geometry
Mesh Total Circumferential Axial 0.12
(Across Seal Gap)

Coarse 583,200 45 15
0.10

Regular 1,526,220 90 25

NEAR­VANE Fine 2,777,355 135 33 0.08

Very Fine 3,963,600 180 39

Coarse Mesh
0.06
Very Fine (90) 1,993,140 90 39 Regular Mesh
Fine Mesh NEAR­VANE
FAR­VANE Regular 1,579,140 90 25 0.04 Very Fine Mesh
Very Fine Mesh (90)

0.02 FAR­VANE Regular Mesh

As illustrated in red in Fig. 1, all of the wall boundary surfaces θ θ

are stationary with the exception of the rotor. Other computations 0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
were carried out for which the annulus inner boundary surface
attached to the rotor also rotated. In this case, it was found that Fig. 3 Computed circumferential static pressure distributions
Taylor vortices were set up in the mainstream flow downstream of in the mainstream radially outward of the seal
the seal. These vortices affected the computed flow structure in the
seal and initiated unsteady flow inside the wheel­space (related
possibly to structures described by Roy et al (2007) and Owen, vz,e = 196 m/s

2006). These destabilising effects are still being studied and are not Reφ = 2.5 x 106

considered in detail here, as the single blade pitch angular domain Rez = 2.7 x 106

shown in Fig. 2 was found to constrain the flow to remain steady

under the aasumption of cyclic symmetry. The stator vane used gives an average flow angle of around 24o to
The commercial code ANSYS­CFX Version 10 was used for the the circumferential in the mainstream in the region radially outward
computations. This finite volume algebraic multi­grid solver uses a of the seal. The values of � and CW were selected as likely test
pressure coupling method for the non­staggered mesh based on that conditions in planned future experiments, and the value of Rez used
of Rhie and Chow (1982). The advection scheme is second order is that which gives rise to a swirl ratio βe ≈ 1 in the mainstream at
accurate, based on the method of Barth and Jesperson (1989). In the seal location (matching qualitatively conditions in the experi­
addition to the RANS momentum and energy equations, a further ments by Graber et al, 1987). The sealing flow rate used is much
transport equation was solved for conservation of a non­interacting lower than the value CW,MIN =15,250 suggested by Eq. (1) for this
scalar. This allowed a tracer to be introduced at the external main­ configuration. The matching of the sealing flow rate to the rota­
stream inlet in order to calculate the amount of ingress and hence tional speed through the non­dimensional parameters Reφ and CW
sealing effectiveness. Gravitational buoyancy effects were ignored. allow the findings for fluid dynamics to be extrapolated with some
Normalised convergence levels below 10­5 were achieved. confidence to the engine situation, see Owen and Rogers, 1989.
The Baseline (BSL) turbulence model of Menter (1994) em­
ployed is a blended formulation of a k­ω model in the near wall FLUID DYNAMICS RESULTS
region and a k­ε model further from the wall. So­called “scaleable” The circumferentially varying pressure field in the external
wall functions allow near wall grid resolution of y+ ≈ 11. A similar mainstream produced by the stationary vane is illustrated in Fig. 3.
computational approach was taken by Wang et al (2007). The CFX The pressure coefficient, Cp, is based on pe, the computed external
code was validated by Lewis et al (2007) using the same turbulence mainstream static pressure at the half height of the annulus radially
model for a rotor­stator system with a superposed flow, and very outward of the seal, and p i , the spatially averaged static pressure
similar  methods and software have been used by, for example, on the stator inside the wheel­space at a non­dimensional radial
Jarzombek et al (2007) to study gas­turbine cooling systems and by location x = 0.95, and is defined as follows:
Sun et al (2006) to compute the unstable flows inside rotating
cavities. p e − pi
Cp = (3)
0.5ρΩ 2 b 2
GOVERNING PARAMETERS AND TEST CONDITIONS
The rotational Reynolds number for the wheel­space is:
The circumferential distribution for Cp for the NEAR­VANE
2 configuration shown in Fig. 3 shows a minimum at a
ρΩb non­dimensional circumferential location θ ≈ 0.07 and a distinct
Re φ = (2a)
µ peak at θ ≈ 0.61.
The circumferential variation of Cp decreases with increasing
and the mainstream annulus axial Reynolds number is: distance downstream from the vane trailing edge. With the seal in
the further downstream (FAR­VANE) position the peak magnitude
for Cp is approximately half that for the NEAR­VANE configura­
ρv z,e b tion, due to mixing and pressure recovery in the mainstream. (In an
Re z = (2b)
µ engine, changing the seal location relative to the vane trailing edge
in this way would increase the relative effect on ingress of the
The values of the main parameters used for the computations are: pressure distribution due to the rotating turbine blades.) The main­
stream pressure variations for the FAR­VANE configuration were
� = 848 rad/s ( ≈ 8,000 rpm) found to give rise to very low levels of ingress, and further results
CW = 1,600 are presented here only for the NEAR­VANE configuration.

­3­

 10 v
Vr r Coarse Mesh

8
Regular Mesh

6 Fine Mesh

Very Fine Mesh

4
Very Fine Mesh (90)
2
x ≈1.01
θθ
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

­2

­4
Plane 1 Plane 2 Plane 3 Plane 4

­6

­8

­10

Fig. 4 Computed circumferential variation of average

radial velocity within the seal

The influence of the mainstream pressure variations in driving

fluid into the wheel­space can be characterised using v r , a
mass­weighted average radial velocity across the seal, defined as:
Fig. 5 Secondary flow in the seal at the four circumferential
planes defined in Fig. 4

vr =
∫ v (ρv )dz
sc
r r
(4)
∫ (ρv )dz
sc
r
η = 1−
C − Cs
(5)
C e − Cs
The circumferential variation of v r at the seal half­height
(x_≈_1.01) is shown in Fig. 4. The distribution shows a peak where Cs and Ce are the prescribed values of concentration at the
negative value (indicating net flow radially inward) at sealing and mainstream inlets respectively. (If no ingestion occurs,
non­dimensional circumferential position θ ≈ 0.73, a location C = Cs inside the wheel­space and the sealing effectiveness is
shifted circumferentially by about one­eighth of the stator­vane unity.)
pitch in the direction of rotation of the disc from that (θ ≈ 0.61) Values for sealing effectiveness are shown in Fig. 6a and Fig. 6b
shown in Fig.3 for the corresponding maximum driving pressure. for near­wall solution points adjacent to the stator and rotor re­
This illustrates the effect of the tangential swirl in the mainstream spectively, point results having been averaged circumferentially to
flow on the flow within the seal. give these radial distributions. The results show significant grid
The velocity vectors shown in Fig. 5 illustrate the secondary sensitivity, particularly for the coarser grids. There are no similarly
flow (i.e. the flow in the axial­radial plane) in the seal at the four significant differences for the computed velocity field (see Fig. 4),
suggesting that the computed transport of the scalar variable is
circumferential planes indicated in Fig. 4. In Plane 1, where v r is a
more sensitive to the grid than are the associated velocities.
maximum, this radially outward flow is due to the boundary layer Fig. 6a shows that (on each of the different meshes tested) the
on the rotor. In Plane 2, where v r ≈ 0, the secondary flow velocity computed sealing effectiveness is approximately constant with
magnitude is small in the seal region, with some outward flow on radius near the stator. This behaviour is consistent with ingested
the rotor side and some flow drawn inwards from the mainstream mainstream flow flowing radially inward in the wheel­space within
onto the stator side of the seal. Ingress into the system is a maxi­ the boundary layer on the stator. The ingested fluid then migrates
mum at Plane 3, where a powerful recirculation is formed at the axially across the wheel­space towards the rotor, however there is
stator side of the seal. This recirculation transports the fluid in­ no entrainment of fresh fluid into the stator boundary layer to dilute
gested into the seal from the mainstream further radially inward and the concentration. (The spatially averaged computed flowfield
this fluid then flows towards the stator as it enters the wheel­space. within the wheel­space was found to be very similar to that in a
classical rotor­stator system with a superposed radial outflow,
In Plane 4, where again v r ≈ 0, the strong recirculation apparent at where for the value of gap ratio G considered here there are
Plane 3 is maintained, but now acts to seal the system. This is due to boundary layers on the rotor and stator separated by a rotating core
the smaller pressure differences between the wheel­space and the of fluid, see Owen and Rogers, 1989.)
mainstream influencing the flow at this location (θ ≈ 1), see Fig. 3. The radial variation of averaged effectiveness with radial loca­
Similar underlying flow structures to those described here are tion near the rotor is more significant than for the stator. At the
expected also to be found in engines, modified by additional un­ inner radius (r = a corresponding to x = 0.61 for the configuration
steady pressure variations due the rotating turbine blades. considered) where sealing flow enters the system, effectiveness
approaches unity. As radius increases, fluid is entrained into the
COMPUTED SEALING EFFECTIVENESS rotor boundary layer from the stator, increasing the concentration of
The sealing effectiveness of the system, η, is evaluated using the ingested fluid and thus reducing the effectiveness.
computed local concentration (mass­fraction) C of the A qualitative comparison is made in Fig. 6a between the com­
non­participating scalar “tracer” variable transported throughout puted values for sealing effectiveness near the stator and local
the system: values measured on the stator in experiments Roy et al (2007), for a

­4­

 1.4
CT
similar value of non­dimensional sealing flow­rate CW although at T
lower values of Reφ and Rez. The lower values of effectiveness 1.2
stator B

measured by Roy et al for x > 0.8 (the outer region of the

wheel­space closest to the seal) suggests that more ingress occurred A
in this experiment than in the computations described here. As 1

mentioned above, other computations were carried out for which rotor
there was greater non­uniformity in the mainstream flow down­
0.8 Rotor (A)
stream of the seal that gave rise to more complex flow inside the
wheel­space. The computed values of effectiveness in this case
were lower than those shown in Fig. 6a, suggesting that the greater 0.6 Stator (B)
mainstream non­uniformity in this case may have had some similar
effects to those due to the rotating blades used on the rig of Roy et
al. 0.4

1
η 0.2
η A
0.9
X
B
0
0.8 D
E 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00
C x
0.7
Fig. 7 Computed circumferentially averaged values of
0.6 non­dimensional temperature T on the rotor and stator

0.5 Coarse Mesh ­ A

Regular Mesh ­ B T − To,s

0.4
T= (6)
Fine Mesh ­ C
To,e − To,s
0.3

Very Fine Mesh ­ D

0.2 Two mechanisms contribute to the elevated temperatures com­
a) Very Fine Mesh (90) ­ E puted within the wheel­space; the ingress of higher temperature
0.1 fluid from the mainstream and frictional heating (windage) due to
Measurements of Roy et al (2007) X the rotating disc.
0
Fig. 7 shows that dimensionless temperatures reach values
0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95
x 1.00
greater than unity at both the rotor and stator surfaces, indicating
1
that for this situation the fluid in the wheel­space is heated above
η
A that of the external mainstream. This is caused by frictional heating,
0.9
η B
exaggerated by the use of adiabatic boundary conditions at solid
D surfaces and the low value of sealing flow rate used. Typically in
E
0.8 C engines (Te – Ts) ≈ 1000K and the windage heating may be up to
around 50K, however the values used in the computations are
0.7
characteristic of test conditions that might be used in simplified
0.6
experiments. Sealing effectiveness can be defined in terms of T
rather than local concentration as in Eq. (5), and the results shown
0.5 in Fig. 7 are a caution to researchers designing experiments to study
Coarse Mesh ­ A the thermal effects of ingress using modest differences in tem­
0.4
Regular Mesh ­ B
perature, in order for example to make heat transfer measurements
using thermochromic liquid crystal (TLC), as it is necessary to
0.3 Fine Mesh ­ C
account for both ingestion and windage. Such experiments are
0.2
Very Fine Mesh ­ D currently being devised by colleagues of the authors, based on
Very Fine Mesh (90) ­ E further development of the TLC techniques described by Lock et al
0.1
b) (2005). Computations such as those described here, as well as
x measurements of concentration, are likely to be needed in order to
0 interpret fully measurements of temperature and heat transfer.
0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95
x 1.00

CONCLUSIONS
Fig. 6 Computed averaged values of effectiveness: Three­dimensional steady turbulent flow computations of a
(a) near the stator, (b) near the rotor rotor­stator system and an external mainstream have been carried
out using the commercial CFD code CFX, at conditions typical of
those likely to be used in simplified experiments devised to monitor
COMPUTED TEMPERATURE DISTRIBUTIONS and measure the effects of ingress of hot fluid from the mainstream
The inlet total temperature of the wheel­space sealing flow was into the rotor­stator wheel­space.
set at To,s = 20oC for the computations and that of the mainstream The computed results show that a stator vane in the mainstream,
flow at inlet at To,e = 60oC. These values were based on the condi­ with its trailing edge sufficiently close to the rotor­stator axial seal,
tions in heat transfer experiments by Lewis et al (2007) for a dif­ produces a non­axisymmetric flow distribution sufficient to cause
ferent turbine cooling application. significant levels of ingress into the wheel­space, as deduced from
Fig. 7 shows computed, circumferentially averaged, profiles of sealing effectiveness values calculated using computed concentra­
non­dimensional static temperature T on the stator and rotor, where tions of a tracer scalar variable. Investigation of the fluid dynamics
T is defined here as: within the axial clearance seal shows that ingested fluid is trans­

­5­

ported radially inwards, and towards the stator boundary layer RANS investigations into buoyancy affected convection in a ro­
inside the wheel­space, through the action of a recirculating flow tating cavity with a central axial throughflow, J. Engineering for
established in the seal which also acts to seal the system locally Gas Turbines and Power, v. 129, n. 3, pp.318­325
from ingress at other circumferential locations, even when the Wang, C­Z, Johnson, B. V., De Jong, F. and Vashist, T. K., 2007.
superposed wheel­space sealing flow rate is low compared with the Comparison of flow characteristics in axial gap seals for close and
minimum expected theoretically to be required to prevent ingress. wide­spaced turbine stages, ASME paper GT2007­27909
Greater grid sensitivity was observed for computed results for
effectiveness compared with velocity distributions. Computations
of the thermal field suggest that identification of the thermal effects
of ingress in simplified experiments may be complicated by the
frictional heating (windage) of the fluid in the wheel­space due to
the rotating disc.

ACKNOWLEDGEMENTS
Mr Lewis is a PhD student supported by a studentship from the
Engineering Innovative Manufacturing Research Centre at Bath,
which is funded by the UK Engineering and Physical Sciences
Research Council. The present research into ingress in gas turbines
has been stimulated and supported by collaboration with Mitsubishi
Heavy Industries, Japan.

REFERENCES
Barth, T. J., and Jesperson, D. C., 1989. The design and application
of upwind schemes on unstructured meshes,. Proc. 27th AIAA
Aerospace Sciences Meeting.
Bayley, F. J. and Owen, J. 1970. The fluid dynamics of a shrouded
disk system with a radial outflow of coolant; ASME J. Eng. Power,
v. 92, pp 335­341
Graber, D. J., Daniels, W. A. and Johnson, B. V. 1987. Disk pump­
ing test, Final Report. Air Force Wright Aeronautical Laboratories,
Report No. AFWAL­TR­87­ 2050
Jarzombek, K, Benra, F.­K., H. Dohmen, J. and Schneider, O., 2007.
CFD analysis of flow in high­radius pre­swirl systems, ASME
paper GT2007­27404
Lewis, P., Wilson, M., Lock, G. D., and Owen, J., 2007. Physical
interpretation of flow and heat transfer in pre­swirl systems, J.
Engineering for Gas Turbines and Power, v. 129, n. 3, pp 769­777
Lock, G. D., Wilson, M. and Owen, J. M., 2005. Heat transfer
measurements using liquid crystals in a pre­swirl rotating­disc
system, J. Engineering for Gas Turbines and Power , v. 127, n. 2, pp
375­382
Menter, F. R., 1994. “Two­equation eddy viscosity turbulence
models for engineering applications”. AIAA­Journal, 8(32), pp.
269–289.
Owen, J. M., 2006. Modelling internal air systems in gas turbine
engines, Proc. 1st Int. Symp. Jet Propulsion & Power Eng., Kun­
ming, China, Sept 17­22
Owen, J. M., and Rogers, R. H., 1989, Flow and heat transfer in
rotating disc systems: Vol. 1, Rotor­Stator Systems, Research
Studies Press, Taunton, UK and Wiley, New York.
Phadke, U.P. and Owen, J.M. 1988a. Aerodynamic aspects of the
sealing of gas­turbine rotor­stator systems, Part 1: The behaviour of
simple shrouded rotating­disk systems in a quiescent environment,
Int J Heat Fluid Flow, v. 9, pp 98­105
Phadke, U.P. and Owen, J.M. 1988b. Aerodynamic aspects of the
sealing of gas­turbine rotor­stator systems, Part 2: The performance
of simple seals an a quasi­axisymmetric external flow, Int J Heat
Fluid Flow, v. 9, pp 106­112
Phadke, U.P. and Owen, J.M. 1988c. Aerodynamic aspects of the
sealing of gas­turbine rotor­stator systems, Part 3: The effect of
non­axisymmetric external flow on seal performance, Int J Heat
Fluid Flow, v. 9, pp 113­117
Rhie, C. M., and Chow, W. L., 1982. “A numerical study of the
turbulent flow past an isolated airfoil with trailing edge separation”.
AIAA Paper 82­0998.
Roy, R. P., Zhou, D. W., Ganesan, S., Wang, C­Z, Paolillo, R. E. and
Johnson, B. V., 2007. The flow field and main gas ingestion in a
rotor­stator cavity, ASME paper GT2007­27671
Sun, Z, Lindblad, K, Chew, J. W. and Young, C, 2006. LES and

­6­