Anda di halaman 1dari 227

VIBRATION OF AXIAL TURBOMACHINERY BLADES : MEASUREMENT

AND FLUID-STRUCTURE INTERACTIONS

A Dissertation

Submitted to the Graduate School


of the University of Notre Dame
in Partial Fulfillment of the Requirements
for the Degree of

Doctor of Philosophy

by

Paul Louis Mikrut,

Scott C. Morris, Director

Graduate Program in Aerospace and Mechanical Engineering


Notre Dame, Indiana
April 2012

c Copyright by

Paul Mikrut
2012
All Rights Reserved

VIBRATION OF AXIAL TURBOMACHINERY BLADES : MEASUREMENT


AND FLUID-STRUCTURE INTERACTIONS

Abstract
by
Paul Louis Mikrut
The focus of this dissertation is on turbomachinery blade vibration measurements and unsteady fluid-structure interactions. Vibration of turbomachinery
blades are critical to jet engine durability and performance. The combined high
natural frequency of the vibrations and long service life of modern jet engines
can result in high cycle fatigue. There are two main topics discussed in this
dissertation. The first topic of this dissertation is the investigation of unsteady
fluid-structure interactions an isolated compressor blade in transonic flow. This
was preferred as a simpler alternative to a cascade of blades. Note that the
boundary conditions of an single vibrating blade are much simpler than those for
a vibrating cascade, and so a more clear understanding of the fundamental interactions are provided with the simple setup. New insights were obtained regarding
aerodynamic damping and quasi-steady blade vibrations in transonic flow.
The second topic of this dissertation discusses the development and application of a novel blade vibration measurement technique. Accurate blade vibration
measurements are critical in product aero-mechanical design validation and can be
difficult to obtain. This measurement technique, termed Blade Image Velocimetry,
provides an alternative to the current measurement methods which is both easy

Paul Louis Mikrut


to implement and can have the potential to exceed the current accuracy of Blade
Tip Timing. The theory of measurement and uncertainty analysis and benchtop
validation measurements will be presented. This will be followed by the application of the measurement technique to a high speed axial compressor rotor. Tip
deflections as low as 8m were resolved by the measurement technique at a rotor
tip speed of 350 m/s.

To my wife, Andrea and my son, Edward.

ii

CONTENTS

FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vi

TABLES

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

ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . .

xi

SYMBOLS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xii

CHAPTER 1: INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . .
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 A review of modern turbomachinery aeroelasticity . . . . . . . . .
1.2.1 Theoretical studies . . . . . . . . . . . . . . . . . . . . . .
1.2.2 Numerical simulations . . . . . . . . . . . . . . . . . . . .
1.2.3 Experimental aeroelasticity of axial compressors . . . . . .
1.2.3.1 Experimental Methods . . . . . . . . . . . . . . . . . . .
1.2.3.2 Experimental Studies . . . . . . . . . . . . . . . . . . . .
1.2.4 Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Background and motivation for the investigation of an isolated compressor blade . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4 Background and motivation for the development of a novel blade
vibration measurement technique . . . . . . . . . . . . . . . . . .
1.5 Objectives of this dissertation . . . . . . . . . . . . . . . . . . . .

1
1
8
9
12
15
15
20
23

CHAPTER 2: EXPERIMENTAL SETUP AND DATA PROCESSING


2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Fluid-structure interactions of an isolated compressor blade . .
2.2.1 General setup . . . . . . . . . . . . . . . . . . . . . . .
2.2.2 PIV measurements . . . . . . . . . . . . . . . . . . . .
2.2.3 LDV measurements . . . . . . . . . . . . . . . . . . . .
2.3 Development and validation of BIV . . . . . . . . . . . . . . .

34
34
35
35
41
44
46

iii

.
.
.
.
.
.
.

.
.
.
.
.
.
.

24
28
32

2.3.1
2.3.2
2.3.3

Demonstration of simultaneous fluid and structure velocity


measurements . . . . . . . . . . . . . . . . . . . . . . . . .
Benchtop validation of measurement theory . . . . . . . .
Application of BIV to an axial compressor rotor . . . . . .

CHAPTER 3: PIV MEASUREMENTS OF STEADY FLOW ABOUT A


FLEXIBLE BLADE . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Theoretical approach for non-contact surface pressure measurements
3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

46
49
52

61
61
63
67
88

CHAPTER 4: NATURAL RESPONSE OF AN ISOLATED COMPRESSOR BLADE IN COMPRESSIBLE FLOW . . . . . . . . . . . . . . .


4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.1 Structural dynamics . . . . . . . . . . . . . . . . . . . . .
4.2.2 Unsteady thin airfoil theory . . . . . . . . . . . . . . . . .
4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.1 Time-series of blade natural response . . . . . . . . . . . .
4.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

92
92
93
93
96
102
111
125

CHAPTER 5: IMPULSE RESPONSE OF AN ISOLATED COMPRESSOR


BLADE IN COMPRESSIBLE FLOW . . . . . . . . . . . . . . . . . .
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2.1 Model of aerodynamic damping . . . . . . . . . . . . . . .
5.2.2 Estimation of the damping ratio from measurements . . .
5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.3.1 Variation in effective damping ratio for the first two modes
5.3.2 Time-resolved blade response . . . . . . . . . . . . . . . .
5.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

127
127
128
128
129
131
131
134
139

CHAPTER 6: DEVELOPMENT AND UNCERTAINTY ANALYSIS OF


BLADE IMAGE VELOCIMETRY . . . . . . . . . . . . . . . . . . . .
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.2 PIV/BIV implementation . . . . . . . . . . . . . . . . . . . . . .
6.2.1 Aeroelastic demonstration of simultaneous PIV/BIV . . .
6.2.2 Uncertainty in structural velocity . . . . . . . . . . . . . .
6.3 Determination of vibration modal amplitude . . . . . . . . . . . .
6.4 Uncertainty analysis . . . . . . . . . . . . . . . . . . . . . . . . .
6.4.1 Noise corruption of the correlation matrix . . . . . . . . .

142
142
143
143
148
149
154
154

iv

6.4.2 Bias error due to velocity discretization . . . . . . . . . . .


Experimental validation of uncertainty . . . . . . . . . . . . . . .
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

156
158
160

CHAPTER 7: APPLICATION OF BIV TO AN AXIAL COMPRESSOR .


7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2.1 Results, 2,886 RPM shaft speed with magnetic bearing excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2.2 Results, 13,531 RPM . . . . . . . . . . . . . . . . . . . . .
7.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

163
163
164

CHAPTER 8: CONCLUSIONS AND FUTURE WORK . . . . . . . . . .


8.1 Objectives of dissertation . . . . . . . . . . . . . . . . . . . . . . .
8.2 Investigation of fluid-structure interactions between an isolated compressor blade and a high-speed flow . . . . . . . . . . . . . . . . .
8.2.1 Discussion of results . . . . . . . . . . . . . . . . . . . . .
8.2.2 Future directions . . . . . . . . . . . . . . . . . . . . . . .
8.3 Blade vibration measurement technology . . . . . . . . . . . . . .
8.3.1 Discussion of results . . . . . . . . . . . . . . . . . . . . .
8.3.2 Future directions . . . . . . . . . . . . . . . . . . . . . . .

174
174

6.5
6.6

APPENDIX A: VARIATION IN BLADE DAMPING RATIO


CENT AIR AS A FUNCTION OF STATIC PRESSURE .
A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . .
A.2 Experimental setup . . . . . . . . . . . . . . . . . . .
A.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . .
A.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . .

174
175
179
181
181
183

QUIES. . . . .
. . . . .
. . . . .
. . . . .
. . . . .

186
186
186
190
190

APPENDIX B: FIRST 10 EIGEN-MODES OF THE ISOLATED COMPRESSOR BLADE . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

193

APPENDIX C: INTEGRAL
C.1 Introduction . . . .
C.2 Approach . . . . .
C.3 Results . . . . . . .
C.4 Conclusions . . . .

.
.
.
.
.

200
200
200
201
204

BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

205

TIME
. . . .
. . . .
. . . .
. . . .

SCALE
. . . . .
. . . . .
. . . . .
. . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

IN
. .
. .
. .
. .
. .

164
168
172

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

FIGURES

1.1

Sketch of a single stage axial compressor. . . . . . . . . . . . . . .

1.2

Campbell diagram for forced response of a rotor blade . . . . . . .

1.3

Axial compressor characteristic map with flutter regions

. . . . .

2.1

Sketch of the blow-down tunnel immediately upstream of the test


section. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

36

2.2

Schematic of test section used for aerodynamic damping investigation. 37

2.3

Scatter plot of time-resolved blade response to mechanical impulse


in quiescent air & ensemble standard deviation of the response . .

38

2.4

Cross-section of the blade . . . . . . . . . . . . . . . . . . . . . .

39

2.5

First two eigen-modes of the blade . . . . . . . . . . . . . . . . .

40

2.6

Graphical decomposition of measured blade surface velocity into


modal velocities and instantaneous envelopes . . . . . . . . . . . .

45

2.7

Sketch of a stationary blade and measurement points. . . . . . .

48

2.8

Schematic of the BIV benchtop validation experiment. . . . . . .

50

2.9

Sketch of the blade cross section geometry . . . . . . . . . . . . .

50

2.10 Shapes of the two structural modes of the test blade as computed
by FEM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

51

2.11 Sketch of a cantilevered rotor blade and tip velocity measurement


points. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53

2.12 Cross-section of rotor blade at 4 locations along span . . . . . . .

54

2.13 First three eigen-modes of the rotor blade . . . . . . . . . . . . .

55

2.14 Schematic of the experimental setup used to demonstrate BIV. . .

56

2.15 Sample image and estimated tip velocity for BIV applied to an axial
compressor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

57

Fluid velocity and vorticity at M = 0.16 and 0 = 8 . . . . . . .

68

3.1

vi

3.2

Phase-averaged velocity, vorticity and coefficient of pressure of fluid


at M = 0.16 and 0 = 8 . . . . . . . . . . . . . . . . . . . . . . .

70

3.3

Phase-averaged surface pressure coefficient at 0 = 8 and M = 0.16 71

3.4

Fluid velocity and vorticity at M = 0.55 and 0 = 8 . . . . . . .

73

3.5

Phase-averaged velocity, vorticity and coefficient of pressure of fluid


at M = 0.55 and 0 = 8 . . . . . . . . . . . . . . . . . . . . . . .

75

3.6

Phase-averaged surface pressure coefficient at 0 = 8 and M = 0.60 76

3.7

Fluid velocity and vorticity at M = 0.75 and 0 = 8 . . . . . . .

77

3.8

Phase-averaged velocity, vorticity and coefficient of pressure of fluid


at M = 0.75 and 0 = 8 . . . . . . . . . . . . . . . . . . . . . . .

79

Instantaneous fluid velocity and blade velocity for 0 = 8 and


M 0.75 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

82

3.10 Fluid velocity and vorticity at M = 0.12 and 0 = 13 . . . . . . .

83

3.11 Phase-averaged velocity, vorticity and coefficient of pressure of fluid


at M = 0.12 and 0 = 13 . . . . . . . . . . . . . . . . . . . . . .

84

3.9

3.12 Phase-averaged surface pressure coefficient at 0 = 13 and M = 0.12 85


3.13 Velocity and vorticity at M = 0.75 and 0 = 13 . . . . . . . . . .

87

3.14 Phase-averaged velocity, vorticity and coefficient of pressure of fluid


at M = 0.75 and 0 = 13 . . . . . . . . . . . . . . . . . . . . . .

90

3.15 Instantaneous fluid velocity and blade velocity for 0 = 13 and


M 0.75 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1

Autospectral density of blade surface velocity as a function of nozzle


exit Mach number at 0 = 8 . . . . . . . . . . . . . . . . . . . .

91
103

4.2

Autospectral density of blade surface velocity at 0 = 8

. . . . .

104

4.3

RMS modal velocity as a function of nozzle exit Mach number


at0 = 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

105

Autospectral density of blade surface velocity as a function of nozzle


exit Mach number at 0 = 13 . . . . . . . . . . . . . . . . . . . .

108

4.5

Autospectral density of blade surface velocity at 0 = 13 . . . . .

109

4.6

RMS modal velocity as a function of nozzle exit Mach number at


0 = 13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

111

4.7

Time series of instantaneous modal velocity at M = 0.65 . . . . .

112

4.8

Auto/cross correlation of instantaneous modal envelope at M = 0.65114

4.9

Time series of modal velocity for the first two modes at M = 0.75

4.4

vii

115

4.10 Auto/cross correlation of fluctuating modal velocity envelope at


M = 0.75 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

117

4.11 Time series of modal velocity for first two modes at M = 0.81 . .

118

4.12 Auto/cross correlation of fluctuating modal velocity envelope at


M = 0.81 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

120

4.13 Time series of modal velocity at M = 0.95 . . . . . . . . . . . . .

121

4.14 Auto/cross correlation of fluctuating modal velocity envelope at


M = 0.95 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

122

5.1

Effective damping ratio vs Mach number for 0 = 8

. . . . . . .

132

5.2

Effective damping ratio vs Mach number for 0 = 13 . . . . . . .

133

5.3

Ensemble-averaged change in damping ratio vs Mach number for


both angles of attack . . . . . . . . . . . . . . . . . . . . . . . . .

134

5.4

Time series of blade response to mechanical impulse at M = 0.65

136

5.5

Time series of blade response to mechanical impulse at M = 0.75

137

5.6

Time series of blade response to mechanical impulse at M = 0.81

138

5.7

Time series of blade response to mechanical impulse at M = 0.95

139

6.1

Image of seeded fluid and structure for combined BIV/PIV measurements & computed velocity and vorticity field . . . . . . . . .

144

6.2

Phase space behavior of the flat plate subject to uniform flow . .

146

6.3

Phase averaged flow around a blade undergoing aeroelastic oscillations147

6.4

Schematic of the BIV calibration experiment. . . . . . . . . . . .

148

6.5

Accuracy of the correlation algorithms in measuring structural velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

150

6.6

Error in amplitude estimate for a single mode as a function of (t)f 159

7.1

Image of the rotor blade tip and estimated velocity using the BIV
technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

165

Chordwise distribution of error in ensemble-averaged circumferential tip velocity for a shaft speed of 2,886 RPM . . . . . . . . . .

165

Chordwise distribution of RMS error in axial tip velocity for a shaft


speed of 2,886 RPM . . . . . . . . . . . . . . . . . . . . . . . . .

167

Phase-averaged axial velocity vs. reference time. shaft speed was


2,886 RPM and axial velocity was induced by magnetic bearings .

168

A comparison of RMS chord-normal fluctuating tip velocity as determined from POD and BIV analyses . . . . . . . . . . . . . . .

169

7.2
7.3
7.4
7.5

viii

7.6

A comparison of RMS chord-normal fluctuating tip velocity as determined from POD and BIV analyses . . . . . . . . . . . . . . .

171

A.1 Schematic of impulsive loading mechanism and general orientation


with respect to the test blade. . . . . . . . . . . . . . . . . . . . .

187

A.2 Geomtery of airfoil used in exploratory measurements. All dimensions in mm. (a) Isometric view of blade geometry. (b) Blade cross
section at 0 = 0o . indicates camber line, is the approximate
location of the elastic axis for the cross-section. . . . . . . . . . .

188

A.3 First two mode shapes of the blade. (a) Mode 1, f 1 = 1150 Hz.
(b) Mode 2, f 2 = 3820 Hz. . . . . . . . . . . . . . . . . . . . . . .

189

A.4 Equivalent linear damping ratio of the aluminum blade as a function


of static pressure.  - Mode 1, - Mode 2. . . . . . . . . . . . .

191

B.1 First eigen-mode of the blade. F0 = 349 Hz . . . . . . . . . . . . .

194

B.2 Second eigen-mode of the blade. F0 = 1, 540 Hz . . . . . . . . . .

194

B.3 Third eigen-mode of the blade. F0 = 1, 767 Hz . . . . . . . . . . .

195

B.4 Fourth eigen-mode of the blade. F0 = 3, 804 Hz . . . . . . . . . .

195

B.5 Fifth eigen-mode of the blade. F0 = 3, 985 Hz . . . . . . . . . . .

196

B.6 Sixth eigen-mode of the blade. F0 = 4, 990 Hz . . . . . . . . . . .

196

B.7 Seventh eigen-mode of the blade. F0 = 6, 976 Hz . . . . . . . . . .

197

B.8 Eight eigen-mode of the blade. F0 = 8, 570 Hz . . . . . . . . . . .

197

B.9 Ninth eigen-mode of the blade. F0 = 8, 649 Hz . . . . . . . . . . .

198

B.10 Tenth eigen-mode of the blade. F0 = 10, 084 Hz . . . . . . . . . .

198

C.1 Integral time-scale of the first two modes at 0 = 8 . . . . . . . .

201

C.2 Integral time-scale of the first two modes at 0 = 13 . . . . . . . .

202

C.3 Ensemble-averaged integral time-scale of the first two modes as a


function of Mach number. (a) mode 1; (b) mode 2. 0 = 8 ;
0 = 13 . Error bars denote ensemble RMS. . . . . . . . . . . . .

203

ix

TABLES

3.1
4.1

4.2

6.1

TEST MATRIX FOR PIV INVESTIGATION OF FLOW FIELD


ABOUT THE BLADE . . . . . . . . . . . . . . . . . . . . . . . .

62

NON-DIMENSIONAL CHARACTERISTICS OF THE FLUCTUATIONS IN THE MODAL VELOCITY AMPLITUDES OF THE


BLADE AT 0 = 8 . . . . . . . . . . . . . . . . . . . . . . . . . .

123

NON-DIMENSIONAL CHARACTERISTICS OF THE FLUCTUATIONS IN THE MODAL VELOCITY AMPLITUDES OF THE


BLADE AT 0 = 13 . . . . . . . . . . . . . . . . . . . . . . . . .

124

COMPARISON OF MODAL AMPLITUDE ESTIMATES FOR A


BEAM UNDER TWO COMPONENT HARMONIC FORCING. .

160

7.1

COMPARISON OF RMS AMPLITUDES OF THE POD EIGENMODES AND THE FEM EIGEN-MODES USING THE BIV TECHNIQUE FOR A SHAFT SPEED OF 13,531 RPM NEAR DESIGN
CONDITIONS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

7.2

COMPARISON OF RMS AMPLITUDES OF THE POD EIGENMODES AND THE FEM EIGEN-MODES USING THE BIV TECHNIQUE FOR A SHAFT SPEED OF 13,531 RPM OPERATING
AT STALLED CONDITIONS. . . . . . . . . . . . . . . . . . . . . 172

A.1 MODAL DAMPING RATIO OF THE BLADE AT ATMOSPHERIC


PRESSURE AND VACUUM . . . . . . . . . . . . . . . . . . . . 191
B.1 UNDAMPED NATURAL FREQUENCIES OF THE FIRST TEN
EIGEN-MODES OF THE BLADE AS COMPUTED BY FEA. .

199

ACKNOWLEDGMENTS

Few dissertations are performed by a single person. I personally owe a debit


of gratitude to many people, without whose help this work may have not been
completed. First, I would like to thank Mike Swadner, Joel Preston and Rodney
McClain for providing the expertise, guidance and support that allowed me to
perform the experiments. I would also like to thank the turbomachinery group at
Notre Dame, especially Josh Cameron, Matt Bennington and Mark Ross for their
support in testing the BIV technique on the transonic axial compressor rig.
I would also like to thank my advisor, Scott Morris, for all the support and
guidance he has provided over the years, and for trusting me with some very
expensive measurement equipment.
Finally, I would like to thank my family, especially my wife Andrea and my
parents Joe and Paula. Your patience and understanding was greatly appreciated.

xi

SYMBOLS

x, y, z

Cartesian coordinates (chordwise, spanwise and blade-normal)

x, r,

Cylindrical coordinates (axial, radial and circumferential)

r, C, N

Blade coordinates (radial, chord-wise, and chord-normal)

t time
f
?
IBPA

Vector of natural frequencies of the structure


Reduced frequency of vibration, f c/U
Inter-Blade Phase Angle

Blade chord

Blade vibration velocity

Zero-flow angle of attack

Fluctuating angle of attack

Fluctuating angular velocity

Density of the blade

Displacement of the blade

Velocity of the blade

Youngs modulus of the solid blade material

Mp

Modal mass of the blade for the pth mode

Kp

Modal stiffness of the blade for the pth mode

Cp

Modal damping of the blade for the pth mode

xii

Modal damping ratio of the blade for the pth mode

Fp

Modal force acting on the blade for the pth mode

Lp

Modal lift force acting on the blade for the pth mode

Tp

Modal moment acting on the blade for the pth mode

t?n

Width of the main peak of the autocorrelation of the nth mode.

t?1,2

Time lag for maximum correlation between modes 1 and 2.

Fluid velocity

Ratio of specific heats for the fluid

Density of the fluid

Pressure of the fluid

Mach number

C(k)

Theodorsen function, C(k)

K1 (k)
K1 (k)+K0 (k)

Kn

Modified bessel function of the second kind and of order n

H[]

Hilbert transform of argument

Diagonal element operator

<[]

Real component of argument

=[]

Imaginary component of argument

2
()

Variance of the argument ()

Matrix transpose

t Time between two PIV images


v Matrix of vz evaluated at discrete points in the x y plane
V

Vector of vz evaluated at discrete points at the blade tip

Matrix of the z component of the blade mode shapes


Matrix of evaluated at the blade tip

Vector of blade modal amplitudes

xiii

Ensemble of

Ensemble of V

Correlation matrix of V

Ensemble of additive noise corrupting V

L Left pseudo-inverse of the tip mode matrix ([T ]1 T )


R

Right pseudo-inverse of the tip mode matrix ( [T ] )

Bias error in using BIV due to the discrete approximation of V

Bias error in using BIV due to additive noise in V

Total Bias error in using BIV

A
R()
N

Area of the flow passage


Radial distance from the center of rotation to the argument
Rotor design shaft speed
Free-stream value

xiv

CHAPTER 1
INTRODUCTION

1.1 Introduction
Aeroelasticity is broadly defined as the interaction between a fluid and a solid
structure. According to Bisplinghoff et al. [11], ...the term aeroelasticity includes phenomena involving interactions among inertial, aerodynamic and elastic forces. This is a multidisciplinary field of engineering which includes solid
mechanics, structural dynamics and unsteady aerodynamics. The aeroelastic response of structures is widely studied in many fields such as Aerospace and Civil
engineering. In the field of Aerospace engineering, there exist three major areas of
aeroelasticity. These include: wing response of aircraft, helicopter rotor response,
and turbomachinery blade response. The focus of this dissertation will be on
dynamic aeroelasticity1 in the context of axial turbomachinery.
Axial turbomachinery can be divided into two types, depending upon whether
the rotor is supplying energy to the gas (compressor, fan) or if the gas is supplying
energy to the rotor (turbine). The discussion will be limited to axial compressors,
where the rotor is supplying energy to the gas. A simplified schematic of the major
components constituting a single stage axial compressor are shown in figure 1.1.
The gas progresses through the machine along the axial direction (x-ordinate).
1

The terms dynamic aeroelasticity and static aeroelasticity are used in reference to the
time-response of the solid structure to the aerodynamic loads.

Figure 1.1. Sketch of a single stage axial compressor. (a) Sketch of an


axial compressor cross-section. (b) Cascade representation of an axial
compressor with potential sources of aeroelastic excitation.

The gas passes through the inlet guide vanes, which set the incidence angle (angle
of attack) of the rotor blades. Energy is transferred to the flow as it passes
through the rotor blades. The gas is discharged from the rotor to another set
of stators (not pictured) which further condition the flow. Multiple stages of
rotor-stator pairs can exist in a given machine. Lakshminarayana[42] provides a
comprehensive discussion of the current state of knowledge regarding fundamental
turbomachinery flow physics.
The aerodynamic and structural design of compressor blades can result in conflicting requirements. Modern blade designs specify thin blades with large aerodynamic loads. This is done to increase the power-to-weight ratio of the engine,
allowing for more agile combat aircraft or higher payload capacity for civilian
transport and cargo aircraft. These designs can result in flexible blades with
high mechanical stresses. Unsteady aerodynamic loads can induce large stresses
in these blades, and as a result these designs can be susceptible to mechanical
fatigue failure [50]. A detailed knowledge of the aeroelastic interactions within
an axial compressor are therefore essential for accurate estimations of machine
durability.
There are many mechanisms which can result in unsteady blade loads. Figure 1.1(b) summarized a few sources which are found in modern axial compressors. The unsteady loads can be divided based on the fluid-structure coupling.
The categories are weak-coupled and strong-coupled. Weak-coupled loads are
characterized by little to no feedback between the fluid which is inducing the load
and the response of the structure. In contrast, strong-coupled aerodynamic loads
implicitly depend upon the vibration of the blade.
An example of an weak-coupled blade load is the interaction between the rotor

blades and the wakes from the stators. The change in the fluid forces as a rotor
blade translates through a stator wake can be independent of the vibration of the
rotor. The stator wakes cause an unsteady loading of the rotors whose frequency
is proportional to the shaft speed of the machine and the number of stators (the
number of excitations per revolution, N , is referred to as the N th engine order).
Strong vibrations are expected to occur when the loading frequency coincides with
a resonant frequency of a rotor blade. The relationship between the forcing and
resonant blade frequencies as a function of compressor shaft speed are visualized
using a Campbell diagram. An example of a typical Campbell diagram is shown
in figure 1.2. The blade vibrations which occur at these intersections are termed
synchronous vibrations, because they are synchronous with some multiple of
rotor shaft speed.
The prediction of blade vibration due to weak-coupled loads can be straightforward, precisely because the aerodynamic loads can be approximated without
detailed knowledge of the blade vibration. The ease of prediction and intuitive
understanding of the source of these loads allow engineers to modify the compressor design to avoid large blade vibrations (and thus extend the blade fatigue life).
Referring back to the stator-wake rotor blade interaction mentioned previously,
an approach to mitigate aeroelastic resonance is to change the number of stators.
This changes the frequency of aerodynamic loading the rotor blades encounter,
thus avoiding a potentially dangerous blade resonance condition.
Strong-coupled loads result in a system where the aerodynamic excitation is
implicitly dependent upon the motion of the structure. The conditions which
result in this coupling depend on a variety of factors, such as fluid properties,
blade material, flow velocity and blade geometry. An example is unsteady shock

Figure 1.2. Campbell diagram for forced response of a rotor blade from
34 stators. From Japikse and Baines [38].

motion, commonly found in high-speed axial compressors. The motion of a blade


(relative to its neighbors) alters the geometry of the flow passage between the
blades. Changes in geometry alter the location of the passage shock and the
aerodynamic load imposed on the blade. The change in load causes the blade to
vibrate and thus a feedback loop between the structure and the fluid is established.
The blade response due to strong-coupled aerodynamic loads can be further
characterized by the net work done by the aerodynamic load on the blade over one
period of harmonic structural vibration. The metric used in this work will be the
aerodynamic damping, which, when added to the structural damping present
in the blade, describes the component of force which is out-of-phase with the
blade deflection. It will be assumed that the structural damping of a given blade
is positive definite. A stable system results when the sum of the aerodynamic
damping and structural damping is positive definite, such that the amplitude
of blade vibration decays with time. Neutrally-stable systems result when the
aerodynamic damping is equal in magnitude and opposite in sign to the structural
damping such that the total damping is zero. Here, the vibration amplitude
remains constant with time. Unstable systems occur when the total damping is
less than zero, resulting in an exponentially growing vibration amplitude.
The conditions which result in an unstable aeroelastic system are termed flutter. Compressor operating conditions which result in flutter are avoided because
the strong-coupled aerodynamic loads result in large structural vibration amplitudes (and associated blade mechanical stresses).
Figure 1.3 presents a compressor operating map, which shows the fluid massflow rate through the stage on the abscissa and the increase in static pressure
between the outlet and inlet of the stage on the ordinate. The maximum pressure

Figure 1.3. Axial compressor characteristic map with flutter regions. I.


subsonic/transonic stall flutter, Ia. system mode instability, II. choke
stall flutter, III. low incidence supersonic flutter, IV. high incidence
supersonic flutter, V. supersonic bending flutter. From Marshal and
Imregun [48].

that the stage can sustain at a fixed mass flow has an upper bound described
by the stall line, whereas the lower limit is described by the surge line. The
operating line describes the locus of flow conditions for which the stage was
designed. It can be seen that there are several distinct regions where flutter is
expected. Note that each flutter region can be described based upon the physics
of the strong-coupled aerodynamic loading. For example, stall-flutter is associated
with an unsteady flow separation over the blade, whereas low-incidence supersonic
flutter is associated with moving shock-waves in between adjacent blades.
Aerodynamic damping can be one of the largest determinants of turbomachine
durability. Modern blade designs have very low structural damping, and so aerodynamic damping can account for a large portion of the total blade damping. A
decrease in aerodynamic damping can result in an increase in vibration amplitude, thus decreasing the fatigue life of the blade. Therefore, it is desirable to
understand the physical causes of aerodynamic damping.
A review of the current work regarding axial-turbomachinery aeroelastic studies will be presented in the following section. This will include a description of
the current knowledge regarding aeroelastic behavior (from both experiment and
simulation) as well as a review of the experimental methods. Following this section will be a description of two research programs that were initiated to address
some deficiencies in both the experimental methods and theory describing the
aerodynamic damping of axial compressor blades.

1.2 A review of modern turbomachinery aeroelasticity


There are three main research areas in turbomachinery aeroelasticity. They
are theoretical, numerical simulations and experimental measurements. Although

this study will be focused on experimental aeroelasticity, it is useful to review the


state of the art with respect to both theory and numerical simulations. Theoretical aeroelastic studies will be distinguished from numerical simulations by their
use of relatively few analytic functions to describe the flow field. This is in contrast to numerical simulations where the entire domain must be discretized and a
numerical solution found at every point.

1.2.1 Theoretical studies


Theoretical aeroelastic studies represent the most fundamental analysis available for routine use. There are many necessary simplifying assumptions, but these
analyses can yield predictive trends with minimal computational effort. The general approach will be discussed in this section along with a few examples and a
discussion of their limitations.
There are two important aspects to a theoretical consideration of axial compressor aeroelasticity: the unsteady aerodynamics and the structural dynamics.
The unsteady aerodynamics of an axial compressor are typically approximated by
considering a cascade of airfoils, as shown in figure 1.1(b). The unsteady aerodynamic forces and moments are calculated by decomposing the flow field into
a steady (with respect to time) component and an unsteady component. The
analysis is simplified by assuming that the effect of the flow unsteadiness is small
and thus linearizable. The blades are assumed to be vibrating with an identical
motion (e.g. plunging motion) and a harmonic time dependence. However, adjacent blades can be vibrating out of phase. This phase difference is known as the
interblade phase angle (IBPA). Solutions to the unsteady flow field are formulated
within the framework of unsteady potential flow theory. The unsteady aerody-

namics are handled by enforcing the Kutta condition and allowing vorticity to be
shed from the airfoil trailing edge. This is done to account for changes in airfoil
lift associated with the unsteady flow. The blade can be approximated by a flat
plate, however more complex shapes can be accounted for by yet smaller perturbations about this initial unsteady flow (cf. Atassi and Akai [6], Akai and Atassi
[2]). Chapter 8 of Dowell et al. [21] and chapters 2-7 of the AGARD manual [52]
provide a detailed account of theoretical turbomachinery aeroelasticity models.
The structure is typically assumed to be constructed from a linear-elastic solid
with homogeneous, isotropic material properties. The structural dynamics are
usually found using a Rayleigh-Ritz approach. In general, the blades are assumed
to have negligible structural damping, such that the stability of the system depends upon the unsteady aerodynamics. Chapters 2-4 of Bisplinghoff et al [11]
present a comprehensive review of these approaches for single wings and blades
while chapters 13-15 of the AGARD manual [53] give an advanced discussion of
theoretical structural dynamics in the context of axial turbomachinery. The coupling between the unsteady aerodynamics and structure dynamics is accomplished
by LaGranges equation. The generalized coordinates are typically associated with
the structure, whereas the generalized forces/moments are associated with the unsteady aerodynamics. Chapter 19 of the AGARD manual [53] provides a detailed
discussion regarding the formulation of the aeroelastic equations of motion for
axial turbomachinery models of varying sophistication.
One of the main objectives of a theoretical analysis of a given compressor design
is to assess the stability of the aeroelastic system. This can be done by assuming
a set of blade vibration mode shapes (as determined from the Rayleigh-Ritz approach above). The effect of forced excitation is neglected, and an eigen-analysis

10

is performed. The eigen-values of the system correspond with the frequency and
total damping of a given vibration mode, and typically vary with mode shape and
fluid properties (density, pressure, velocity, Mach number, etc.). It is useful to
examine the factors which tend to reduce blade aerodynamic damping and lead
to flutter.
The reduced frequency, defined as

f c
,
U

(1.1)

is the ratio between the period of time required for a fluid particle to advect over
1/2 of the blade chord to the period of blade vibration. Here f is the frequency of
vibration (in Hz), c the chord of the blade and U is the free-stream fluid velocity.
The reduced frequency plays an important role in determining the stability of the
compressor blade vibrations. Flutter tends to occur at low reduced frequencies
(O(0.1)). Within the range of 102 ? 1, the unsteady aerodynamic loads
have a component which is in phase with the blade vibration velocity. As a result,
there is a significant energy exchange from the flow to the structure within this
reduced frequency range.
The interblade phase angle (IBPA) can also influence system stability. Recall
that the IBPA describes the relative phase difference in vibration between adjacent
blades. The flow disturbance associated with the adjacent blade motion induces an
unsteady load with a similar phase lag. The IBPA can vary from 0 to 2 radians in
discrete angles only (due to the discrete number of blades). As a result, for every
IBPA for which the induced loads contribute to the stability of the blade motion
there is a corresponding IBPA where the loads are destabilizing. In practice, it is
difficult to specify the IBPA of a rotor a priori. Therefore, the total damping for

11

the individual blade must be sufficiently large to accommodate the induced loads
due to adjacent blades at the least stable IBPA (cf. Chapter 19 of [53]).
There is a fundamental dependence on the type of structural motion: plunging
motions (translation normal to the blade chord) are typically more stable in low
speed flow than pitching (rotation of the blade) motions. That having been said,
the effect of a non-zero static imbalance2 can either enhance or reduce system
stability[2]. Finally, there are fundamental changes in the aeroelastic stability of
the cascade as the Mach number increases[62], [27]. Specifically, the presence and
motion of shock waves on the surface of the blade can be destabilizing for blade
motion which would otherwise be stable for subsonic flow.
A disadvantage of theoretical aeroelastic studies is the linearization of the fluid
dynamics. This approach is incapable of capturing strong nonlinearities such as
boundary layer separation (including airfoil stall and shock-induced). Numerical
simulations have been developed over the last four decades to address these shortcomings. A summary of the general approach as well as a few studies will be
presented in the next section.

1.2.2 Numerical simulations


Numerical simulations, where the aeroelastic system is discretized in both
space and time, have constituted a large area of research over the last few decades.
This field of study has advantages over theoretical studies because its implementation requires fewer restrictive assumptions. A summary of a few approaches used
in numerical simulations will be given along with some general trends. This section
will conclude by presenting some of the known issues regarding these simulations.
Similar to theoretical approaches, numerical simulations must address the un2

The chordwise separation between the center of gravity and the elastic axis

12

steady aerodynamics and structural dynamics. In addition, the discrete meshes


associated with the fluid and structure must be coupled appropriately. The formulation of this coupling may not be trivial [22]. The unsteady aerodynamics are
typically simulated using a finite volume formulation and either an Euler or Unsteady Reynolds-Averaged Navier-Stokes (URANS) approach. It should be noted
that Large-Eddy Simulation (LES) studies have been performed [24], but the computational burden currently restricts their routine use to academic studies. The
flow is typically split into a nonlinear steady component and a linearized unsteady
component [48], however fully nonlinear analyses are becoming more common.
The structural dynamics can be simulated using a Finite Element Method.
The incorporation of the structural dynamics into the aeroelastic system is accomplished by considering a small set of free-vibration modes with an assumed
harmonic time-dependence. The fluid mesh is typically deformed using an algebraic dependence to account for blade motion. There are two approaches used
to address the time-evolution of the spatially discretized aeroelastic system. One
approach is to discretize the system in time and integrate the equations of motion. This approach is the most general solution method, but requires that a large
amount of data be stored. Another approach assumes that the flow variables have
a harmonic dependence in time but are allowed to vary in space [18]. Using this
approach, the flow variables can be approximated by a spatially varying fourier
series in time, thus substantially reducing the computational effort [31].
The use of numerical simulations has yielded insight regarding the unsteady
loads associated with non-linear aerodynamic phenomena, which is a direct result
of the nonlinear approach [31], [56]. The interaction between shocks and boundary
layers is an example of a non-linear, strong-coupled aerodynamic load. Shocks are

13

generally destabilizing for blades undergoing bending oscillations, but tend to be


stabilizing for torsional motion. However, shock induced flow separation tends
to destabilize the torsional oscillations [61], [59], [31]. Thus, shock-induced flow
separation can have differing effects on aerodynamic damping, depending on the
mode shape, strength of the shock and the flow separation.
The periodic stalling of airfoils in a cascade is another example of a non-linear
weak-coupled aerodynamic load. Periodic stalling can be due to rotating stall (cf.
Hoying et al. [36] and Cameron [16]), or blockage in the flow field upstream of the
rotor. This type of aerodynamic load can result in quasi-periodic blade vibrations
[1].
Although numerical simulations offer a substantial amount of insight regarding the physics of non-linear aerodynamic loads, there are still many areas which
require improvement. The single largest issue is the accurate simulation of viscous
effects. The unsteady aerodynamics are dependent upon accurately capturing the
effect of turbulence and boundary layer separation, yet even steady flow simulations which are routinely used for aerodynamic studies have difficulty in predicting
these phenomena [30], [59]. Consequently, simulations are performed as a postprocessing step as a supplement to experimental measurements [56]. This approach
allows for additional insight into the flow physics which cause the unsteady aerodynamic loads, but may be of limited use as a predictive tool for conditions which
differ significantly from those observed in the experiments.
At the present time, it is unfeasible to routinely execute simulations which can
completely resolve the viscous, unsteady flow typically found in axial compressors.
Consequently, there is a strong demand for experimental characterization of the
nonlinear aspects of the axial compressor aeroelastic system.

14

1.2.3 Experimental aeroelasticity of axial compressors


Experimental studies of axial compressor aeroelasticity can yield much insight
of the physics without the need for simplifying assumptions. However, successful
implementation currently requires large investments in both time and equipment.
This high cost has been the primary motivation for the development of both
theory and numerical simulations. It should be noted that the current limitations
of the previous two approaches implies that the successful development of an
axial compressor may require multiple experimental investigations to identify and
suppress potentially dangerous aeroelastic resonances which were not previously
predicted [58].
There are two main areas of research in experimental axial compressor aeroelasticity. These are the development of new measurement technology and the experimental investigations. The need for new measurement technology is motivated
by the complex aeromechanical environment in which modern axial compressors
operate. Experimental investigations are critical to aeromechanical design validation. These studies can reveal aeroelastic excitation which are not predicted using
theoretical or numerical models. The remainder of this literature review will focus
on recent developments in these two areas of research.

1.2.3.1 Experimental Methods


The complex aeromechanical environment which characterizes modern axial
compressors necessitates specialized measurement technology. Specifically, measurements must characterize the dynamics of a system where the structure has
both a low mass and low internal damping, rotating at high speed and is operating in a highly unsteady aerodynamic environment. There are two aspects

15

of this system which need to be characterized by measurements. They are the


unsteady aerodynamics and the structural dynamics. The technology associated
with both structural dynamics measurement and unsteady flow can be classified
based upon the method by which measurements are made. Two categories will
be distinguished: direct and non-contact. Direct measurement technology will be
considered as technology where the sensor is in direct contact with the system to
be measured. Hot wires/films, pressure transducers, strain gages and accelerometers are examples of direct measurement technology. Particle Image Velocimetry
(PIV), Laser Doppler Velocimetry (LDV) and Blade Tip Timing (BTT) are examples of non-contact measurement technology. Although there are many measurement technologies which have been applied to axial compressors, this review
will only focus on four. They are surface-mounted pressure transducers, strain
gages, Blade Tip Timing (BTT) and Particle Image Velocimetry (PIV). The first
three represent the most common measurement techniques used in experimental
compressor aeroelasticity. The last technique, PIV, is included because it has
specific abilities which make it an ideal measurement system for aeroelastic studies. A brief discussion of these measurement techniques in the context of axial
compressor aeroelastic measurements will be the topic of this section.
The measurement technology for unsteady flow has been developed extensively. Sieverding et al. [57] presents a review of conventional direct measurement
technology while chapter 14 of [60] provide a detailed review of both direct and
non-contact flow measurement technology. In the context of experimental aeroelastic studies, the most popular unsteady flow measurement technology is high
frequency surface mounted pressure transducers (cf. chapter 9 of [52] and chapter
20 of [53], Gill et al. [26], Buffum et al. [14], Kobayashi [41] and Belz and Hen-

16

nings [9]). The sensors are characterized by their small size and high frequency
response. They are typically recessed into the surface of the compressor blades
so as to mimic the original aerodynamics. The advantage of this approach is
that a direct measurement of unsteady pressure distribution can be obtained in
a time-resolved manner. However, there are a number of disadvantages. Spatial
resolution of the unsteady blade surface pressure is limited by the physical size of
the pressure transducer/pressure taps. The information available only characterizes the behavior of the flow in the vicinity of the blade surface. Although this
is useful in predicting aeroelastic stability, it may not yield much insight into the
source of the surface pressure fluctuations. The application of surface mounted
pressure transducers to rotating structures requires that both power and data
be transmitted between a stationary data acquisition system and the rotating
sensors. Traditionally, this is accomplished through slip-rings, although wireless
approaches are becoming more popular. Finally, the installation of these sensors
on a flexible blade can alter the structural dynamics. This can influence the measured unsteady pressure distribution, especially for self-excited blade vibration.
Particle Image Velocimetry (PIV) is a minimally-intrusive measurement technique that can be used to estimate fluid velocity [54]. The technique utilizes a
specialized CCD camera and a laser to estimate the displacement of particles dispersed in the fluid flow. The displacement of the particles between two images
acquired by the CCD camera divided by the time lag between the first and second image yields an estimate of the flow velocity. The spatial extent over which
fluid velocity estimates are obtained and the spatial resolution, depend upon the
resolution of the CCD camera and associated lens. This technique has been used
to successfully characterize the quasi-steady axial compressor flow environment

17

(cf. Wernet [64], Gorrell and Copenhaver [29] and chapter 14 of [60]), primarily
because of its ability to acquire a large amount of flow velocity information in a
non-intrusive manner with minimal equipment (in most cases the only sensor to
be calibrated is the CCD camera). Despite these advantages PIV has not been
widely integrated into axial compressor aeroelastic studies due to challenges associated with optical access and high equipment costs. Chapter six will present a
method which allows for the acquisition of simultaneous fluid and structure velocity estimation using a conventional single camera PIV system. Chapter three will
also use PIV to estimate the fluid static pressure due to a vibrating compressor
blade.
There are few techniques available for the characterization of the structural
dynamics of axial compressors. The review article of Al-Bedoor [3] provides a
survey of the current technology applied to blade vibration measurement. The
most common direct measurement technology is the strain gage. The sensors
can be recessed to preserve the aerodynamic shape of the blade. Few sensors
are necessary to deduce the entire motion of the blade. This technology remains
attractive because it represents the most reliable approach to obtaining accurate,
time-resolved measurements of the blade motion. The drawbacks of this approach
are similar to those found with the surface pressure measurements. Specifically,
the transmission of power and data between the rotating sensor and stationary
data acquisition system is not trivial. In addition, the machining required to
recess the gages to be flush with the surface of the blade can change the structural
dynamics.
Blade Tip Timing (BTT) is a non-contact structural vibration measurement
technology which has gained popularity over the last 40 years. Heath and Imregun

18

[33] and Lawson and Ivey [43] provide a detailed description of this measurement
technique. The concept of measurement will be summarized briefly. A set of
proximity sensors are installed into the outer casing in a circumferential pattern.
The proximity sensors can either be capacitive [43] or optical [66]. The sensors
are installed such that they detect when a blade tip passes by the sensor location.
The average time required for a blade tip to move from one sensor location to
another is known by the shaft speed of the rotor, the radius of the rotor at the tip
and the circumferential distance between the two sensors. Any deviations from
this average time indicates blade motion with respect to the hub. The advantage
of this method over conventional strain gages is that the rotor does not need to
be modified and installation only requires a few holes to be drilled in the outer
casing. The output from this measurement system is a time-series of blade tip
displacement. The sampling frequency of the method depends on the shaft speed
of the compressor and the circumferential separation of the sensors. The sampling
frequency is typically much lower than the frequency of blade vibration, and as
a result modes are distinguished by observing their aliased frequencies [66]. One
unique issue with this measurement system is the difficulty in obtaining amplitude
and frequency estimates when the blade vibration frequency is an integer multiple
of the shaft speed (a condition known as synchronous resonance). Data processing
algorithms based on linear regression principles have been proposed to overcome
this limitation [34], [25]. The accuracy of the vibration amplitude measurement
can be comparable to strain-gage measurements under controlled circumstances
[43]. However, the presence of noise as well as the assumptions used in data
processing can reduce the accuracy substantially [17], [43].

19

1.2.3.2 Experimental Studies


The remainder of this literature review will focus on several case-studies of axial
compressor aeroelastic stability. These case studies will elucidate flow mechanisms
which contribute to the overall stability of the blade vibrations. The studies were
performed with the objective of determining the flow mechanisms which can lead
to blade flutter. Knowledge of the interaction between these mechanisms and the
blades will help yield insight into the flow physics associated with aerodynamic
damping.
An example of a complete aeroelastic investigation of an axial compressor rotor
operating at high Mach number and high incidence was presented by Stargardter
in chapter 20 of [53]. This type of flutter occurs in regions I and Ia of figure
1.3. The following observations can be made regarding this particular form of
flutter. Blade flutter manifests itself as a high amplitude oscillatory vibration
whose frequency is usually not an integer multiple of the rotor shaft speed. The
blades are typically excited in the first torsion mode (pitching of the blade crosssection). Flutter is typically preceded by stall-induced aeroelastic oscillations.
These stall-induced oscillations, commonly known as buffeting, have a random
amplitude and excite the first bending mode of the blades. The distribution of
aerodynamic forcing is concentrated near the leading edge on the suction side of
the blades. It was observed that this type of flutter required that supersonic flow
exist over some portion of the blade, suggesting that shocks which oscillate in
both strength and chordwise position are present during flutter. Finally, it was
observed that flutter occurred only for a narrow range of reduced frequencies.
Kobayashi [41] studied the aeroelastic response of an annular cascade in transonic flow. The blades were forced to oscillate in first torsion about the mid-chord.

20

This investigation was concerned with the effects of reduced frequency, inlet Mach
number and interblade phase angle on the unsteady aerodynamic forces observed
on the blades. In contrast to the study by Stargardter, Kobayashi was concerned
with the physics of supersonic unstalled torsional flutter, a condition which is
characterized by high operating speed and low compressor backpressure (region
III of figure 1.3). A common feature of this system is the presence of a shock
whose chordwise position along the suction side of the blade oscillates in time.
The mean location of this oscillating shock was aft of the torsional axis of rotation. Measurements of unsteady surface pressure indicated that the suction side
downstream of this axis and the pressure side upstream of this axis were responsible for the unsteady forcing, whereas the rest of the airfoil surface had a damping
effect on the vibration. Changes in the reduced frequency of the airfoil oscillations
significantly affected the aerodynamic forcing on the suction surface. Increasing
the reduced frequency increased the stability of the aeroelastic system. This was
attributed to a decrease in the magnitude of the shock oscillations associated with
high frequency pitching oscillations.
Buffum et al. [14] also studied the stability of torsional oscillations of a cascade.
However, the inlet Mach number was much lower than the previous two studies
(M = 0.5). The effect of mean incidence angle were investigated for a constant
interblade phase angle of 180o . There was a region of separated flow extending
from the leading edge to about 30% chord along the suction side of the blades.
This flow separation resulted in strong aerodynamic forcing of the blades. The
largest contribution to the unstable forcing occurred within the separation region
near the leading edge. In contrast, a stabilizing forcing was observed in the region
where the flow reattached. The effect of increasing the reduced frequency was to

21

magnify the forcing. In general, this caused a decrease in cascade stability. This
study indicated that the unsteady flow associated with the initial separation point
can have a strong destabilizing effect.
Srinivasan [58] presented a review of many turbomachinery aeroelasticity case
studies. Similar to the work of Stargardter, the emphasis was on the complete
aeroelastic system in the context of industrial design. There are a few mechanisms
associated with the structural dynamics which may contribute to the self-limiting
behavior of compressor blades. The first was the disk mistuning, whereby the
frequencies of vibration of adjacent blades for a vibration mode are not identical.
If done correctly, the intentional mistuning of a rotor can increase the stability
of the system. The overall effect is to decrease the influence of the interblade
phase angle on the aeroelastic stability of the blade vibrations. It should be
noted that optimal mistuning will yield a rotor whose stability is governed by the
stability of the individual blades. However, mistuning can result in inordinately
large vibrations due to forced response. External sources of damping can be an
important source of energy dissipation for optimally mistuned rotors under forced
excitation.
Belz and Hennings [9] studied the effect of reduced frequency on the stability of
torsional oscillations in a transonic annular cascade. The investigation considered
two flow conditions which resulted in substantially different shock structures. A
transonic reference case had a passage shock which extended from the leading
edge of the pressure side to between 25% and 35% chord on the suction side. The
flutter case, in which the inlet Mach number was increased slightly, resulted in a
passage shock at 25%-35% chord on the pressure side and 65%-85% chord on the
suction side. In both cases, the shocks were observed to be both stabilizing and

22

destabilizing, depending on the interblade phase angle and the surface of the airfoil
on which they acted. The shock had a stabilizing influence on the pressure side but
a slight destabilizing influence on the suction side for the transonic reference case.
Note that the blade vibrations at these conditions were stable. In contrast, the
intersection of the shock on the suction side for the flutter case was consistently
responsible for the destabilization of the cascade. This configuration was most
unstable when the interblade phase angle was around 90 .

1.2.4 Review
The current design trends for axial compressors can result in blade designs
which experience flow-induced vibrations. These vibrations can lead to a reduction in the fatigue life of the blades, which can shorten the service life of the
engine. The aerodynamic loads associated with these vibrations can be classified
as either weak-coupled or strong-coupled with respect to the blade motion. Weakcoupled aerodynamic loads are not significantly influenced by the blade motion,
and potential vibration issues can be addressed relatively easily. Strong-coupled
aerodynamic loads implicitly depend on the motion of the blade, and can either
attenuate or amplify blade vibrations. The rate at which the blade vibration
amplitude grows or decays is described by the total damping, which is a combination of the structural damping and the aerodynamic damping associated with
the strong-coupled aerodynamic loads. Modern axial compressor rotors tend to
have very little structural damping, so aerodynamic damping can constitute most
of the total blade damping. Changes in aerodynamic damping can profoundly
influence the blade structural dynamics, which in turn effects the durability of the
compressor.

23

The literature review presented here demonstrates that although the current
state of knowledge regarding axial compressor aeroelasticity has improved dramatically over the last 70 years, the current tools used for predicting and measuring
the unsteady aerodynamic loads and resulting blade response in axial turbomachinery can be improved. The focus of the work described in this dissertation
was on the improvement of blade vibration measurements, and the investigation
of the fundamental fluid-structure interactions for an axial compressor blade. A
brief introduction to the experiments that were conducted, as well as a review of
the relevant literature associated with each will be presented.

1.3 Background and motivation for the investigation of an isolated compressor


blade
The current design trend of axial compressors can result in blades that have
large aerodynamic loads and are structurally very flexible. Consequently, the
blades can fail prematurely due to fatigue from vibration. There are three actions
which can be taken to avoid fatigue failure : decrease or eliminate the unsteady
load responsible for the vibrations, modify the geometry of the blade to reduce
the cyclic stresses, or increase the damping of the blade. Current design practices
greatly reduce the magnitude of forced excitation rotor blades experience. However, not all sources of forcing can be eliminated while allowing the stage to meet
performance requirements. Furthermore, alterations in blade geometry to reduce
cyclic vibratory stresses can negatively impact compressor aerodynamic performance or weight. This can adversely effect vehicle performance or fuel economy.
The damping of a structure is a measure of the rate at which vibration energy is
dissipated. Ideally, blades would be designed with very high damping such that

24

most of the energy which is being transferred to the blade from the forcing is
dissipated as heat rather than used to nucleate and grow cracks in the structure.
The sources of the unsteady aerodynamic loads acting on the blade can be
characterized as either external or self-induced. External sources, such as wakes
from stators upstream of the rotor blades, are well understood and their effects
can be predicted using fairly simple models. Self-induced sources are typically
characterized by a feedback between the flexible blade and the unsteady aerodynamics. As a result, these sources are more difficult to model and their effects are
not as well understood. The response of the blade due to these sources will be
termed the natural response. The fatigue life of the blades can be dependent
upon the natural response. Therefore, it is important to characterize the natural
response and understand the unsteady forces associated with it.
A characteristic of modern axial compressor design are blades with extremely
low structural damping. Typical compressor blades can have a structural damping
ratio well under 1% of the critical damping ratio [53], [58], [40]. In contrast, blade
damping due to unsteady aerodynamic loads (aerodynamic damping) can vary by
a few percent. As a result, aerodynamic damping can have a large contribution
to total blade damping, and thus can play an important role in the fatigue life of
the rotor.
Aerodynamic damping results from unsteady fluid-structure interactions where
the induced aerodynamic loads are strongly coupled to the blade vibration. These
loads and the resulting aerodynamic damping can vary with blade geometry, local
flow conditions and vibration mode shape. The methods used for predicting these
unsteady loads can be categorized based upon the computational approach. The
methods are theoretical calculations and numerical simulations.

25

Theoretical calculations simplify the equations of motion by a-priori assumptions regarding the dynamics of the fluid and blade. The result is a description of
the unsteady aerodynamic loads using few equations. The review article by Atassi
[5] presents a summary of modern aeroelastic theory. The general approach can be
described as follows. The fluid dynamics are linearized about the mean flow. The
unsteady aerodynamic loads are estimated by assuming harmonic blade vibration.
The flow is assumed to be well-behaved (no boundary layer separations), inviscid
and that the Kutta condition is satisfied (flow stagnation point is at the apex of
trailing edge). The aerodynamic damping is estimated from the phase difference
between the unsteady load and the blade displacement, as well as the mechanical
properties of the blade and the magnitude of the unsteady forces.
Numerical simulations have gained in popularity over the last 30 years due to
the proliferation of cheap and powerful computers. This approach discretizes the
problem and numerically solves the equations of motion that describe the system.
Increasing computational power has allowed for sophisticated simulations which
do not require as many simplifying assumptions to describe the unsteady fluidstructure interaction. As a result, numerical simulations can predict the unsteady
loads for flows which have strong non-linearities, such as shocks and boundary
layer separations. Note however that given the constraints of finite computation
time and processing power, only approximate solutions can be obtained on a routine basis. Guidance from theory, experiment or experience must be incorporated
with these simulations to provide a complete understanding of the interactions.
A set of experiments were designed to explore the fundamental fluid-structure
interactions of a typical compressor blade at high incidence angle. The experiments were performed using an isolated, cantilevered blade with an end-wall

26

exposed to a two sided compressible jet. Note that real compressors have adjacent
blades which influence the flow field. The objective for this set of experiments was
to investigate the fundamental physics, and so an isolated compressor blade offered very simple boundary conditions which allowed for a direct interpretation
of the results. A major obstacle in the prediction of aerodynamic damping is the
accurate description of these fundamental mechanisms, especially for transonic
flows. The motivation of this work was to provide further understanding of the
dynamics of an isolated compressor blade such that these fundamental models
may be refined.
A key feature of this experimental setup was the cantilevered blade, which
more closely reproduced the structural dynamics of a typical rotor blade. In
contrast, other studies such as Kobayashi [41], Belz and Hennings [9] and Buffem
et al. [14] prescribed a known motion profile to an airfoil. The prescribed motion
profile, while useful in determining the fundamental physics of a vibrating airfoil,
can greatly suppress the feedback which is critical to the blades natural response.
A more accurate estimate of the dynamics of a flexible rotor blade was obtained
by using a cantilevered blade that was allowed to vibrate freely.
The aerodynamic damping for this series of experiments was directly evaluated
by observation of the blade response to a mechanical impulse. The experiment was
set up in such a way as to allow for the mechanical impulse to be applied without
significant distortion of the flow field. Previous studies relied upon phase differences between the unsteady aerodynamic forces and the prescribed airfoil motion
to deduce the aerodynamic damping. The author knows of no prior study which
has utilized the impulse-response approach to characterize aerodynamic damping.
Consequently, the results of these experiments will provide a new perspective on

27

the aeroelastic behavior of the blade.


The investigation of the fundamental fluid-structure interactions of an isolated
compressor blade will be presented in chapters 2-5. Chapter 2 discusses the experimental setup used. Measurements of the flow using PIV will be presented
in chapter 3. An analysis of the flow field at the mid-span of the cantilevered
blade will be presented for incompressible, subsonic compressible and transonic
flow. Additionally, estimates of the fluid static pressure deduced from the measured flow field will be presented. Chapter 4 describes observations of the blade
vibrations due only to the high speed jet. A comparison of the measured blade
vibration against predictions from a linear aeroelastic theory will be presented. A
detailed analysis of the unsteady blade vibration will also be presented. Chapter
6 will present results of a novel approach to characterize the transient response.
Specifically, a mechanical impulse was applied to the blade by way of a mechatronic hammer and the structure response was observed. The transient response
was used to estimate the aerodynamic contribution to the damping of the first two
modes of the blade.

1.4 Background and motivation for the development of a novel blade vibration
measurement technique
Modern turbomachinery operate in a complex aeromechanical environment
that leads to unique challenges for blade vibration measurements. These measurements are intended to characterize the dynamics of a system where the structure
has low mass, low internal damping, and is rotating at high speed in an unsteady
aerodynamic environment. Traditional measurement techniques either require a
large investment in experiment design and implementation (strain gages) or yield

28

ambiguous results. A novel measurement technique which utilizes a digital camera and laser system, similar to the equipment used for PIV measurements, to
deduce blade vibration from observations of the tip motion. A brief review of
other non-contact structural vibration measurement techniques will be presented.
There are several other non-contact structural vibration measurement techniques which have been applied to non-rotating structures. Laser Doppler Vibrometry (LDV) [8], [19] and fiber optic interferometry [20], [4] do not influence
the dynamics of the system, and can be scanned in order to obtain the spatial
information that is required to estimate the amplitude of various eigenmodes of
the vibration. These methods are not well suited for measurements on rotating
structures given that they yield information at a small number of spatial locations only when there is a clear line of sight between the sensor and the structure.
The accurate determination of three-dimensional structural motion using these
data can be difficult. Harris and Piersol [32] provide a review of various vibration
measurement techniques.
Digital Image Correlation (DIC) is a general technique that utilizes two successive images of a structure or flow field in order to estimate displacement. The
method often partitions the image into interrogation zones for which a spatial
correlation algorithm is applied. This results in a two-component displacement
vector in the plane of the image at each interrogation zone. The displacement is
related to the velocity field by the time difference between the images [54]. The
implementation of DIC for both static and dynamic strain measurements has been
reviewed by Hild and Roux [35].
Photogrammetry is another non-contact measurement technique which estimates structural displacement from images. The similar technique, Videogram-

29

metry, is the Photogrammetric technique applied to a time series of images [51].


Photogrammetry is often used to reconstruct three dimensional motion from twodimensional images. The technique measures the motion of a few markers whose
locations with respect to the camera are known. The mapping between the
measured two-dimensional displacement (seen by the camera) and the threedimensional model deformation is determined by calibration [15]. The fundamentals of Photogrammetry have been reviewed by Liu et al. [46], while Pappa
et al. [51] describe some practical examples related to aerospace applications.
A novel, non-contact blade vibration measurement technique has been developed as an alternative to both BTT and strain gages. This technique uses equipment common to conventional Particle Image Velocimetry (PIV) systems, and
is capable of simultaneous fluid and structural velocity measurements. A CCD
camera measures the velocity of a cantilevered blade at the tip, and can estimate
vibration amplitudes on stationary and rotating structures. The simultaneous acquisition of fluid velocity can provide information about the aerodynamic forcing
and aerodynamic damping of the blade vibratory motion. Since the technique
measures blade velocity using digital images, the technique is referred to as Blade
Image Velocimetry (BIV).
Previous measurements have been made that obtain simultaneous acquisition
of fluid and structural velocity using a correlation based approach. For example,
Breuer et al. [13] used a standard PIV system to obtain fluid and structural
displacement on a silicon microturbine. The extremely small size precluded the
use of conventional vibration measurements. The structural motion was obtained
by tracking the the intersection of the PIV laser light sheet with the surface of the
blade. DIC algorithms were used to estimate the degree of rotor whirl from these

30

data. Gomes and Lienhart [28] also utilized PIV equipment to measure structural
velocity. Their objective was to obtain a set of high fidelity measurements for
aeroelastic code validation. The structural velocity measurement was similar to
that employed by Breuer et al. In both cases, the features used in the correlation
algorithms were generated from the intersection of the laser light sheet and a solid
surface.
The main difference between the non-contact vibration techniques discussed
and the present BIV measurements is the application of a small paint dots to
the structure in order to mimic the seeding images that are typically found in
conventional PIV measurements. This allowed standard correlation algorithms to
be employed in order to determine the structural velocity. There are a number
of advantages to this approach. First, the BIV provides a spatial distribution
of structural and fluid velocity simultaneously (in a plane). Secondly, the PIV
algorithms are convenient to use and provide an extremely accurate estimate of
the true structural velocity. These measurements were acquired at a relatively
low temporal frequency, and so the technique is fundamentally different from point
measurements of strain or vibration velocity where high frequency time series data
are recorded. However, the measured spatial distribution of structural position
and velocity allows for the determination of the instantaneous amplitude of the
vibration modes when a priori knowledge of the the three-dimensional eigen-mode
shapes is available.
BIV and BTT both use velocity measurements at the tip to infer blade vibration. However, there are several important differences between BIV and BTT.
First, estimates of the blade tip velocity are acquired along the entire chord at the
blade tip. This allows for the discrimination of different modes of vibration based

31

upon the velocity distribution at the tip and thus avoids the need to rely upon the
aliased frequencies of vibration. The high spatial resolution afforded by BIV also
provides more estimates of the blade tip velocity. These estimates are statistically
independent from one another and can be used to minimize the influence of noise
on the velocity estimates.
A description of the experimental setup used to validate the BIV technique will
be presented in Chapter 2. Chapter 6 will focus on the development of the BIV
technique through several bench-top experiments. A discussion of the theory and
uncertainty analysis of the technique will be provided. Chapter 7 will discuss the
application of BIV to a high speed axial compressor rotor. A second validation will
be presented to verify the techniques applicability to rotating structures. This
will be followed by a discussion of the measured blade vibration at two compressor
operating conditions near the design shaft speed of 15,000 RPM.

1.5 Objectives of this dissertation


The objectives of this dissertation were two fold. The first was to explore
the fluid-structure interactions of a compressor blade under self-excited forcing.
The second was to develop a novel, non-contact blade vibration measurement
technique.
Two responses were investigated in the exploration of the fluid-structure interactions of a compressor blade. The first, termed the natural response can be
described as the response of the blade due to the unsteady aerodynamic forces that
result from a uniform flow. The second response that was investigated was the response of the blade due to a mechanical impulse. This response can be important
for evaluating the stability of the blade vibrations under different flow conditions.

32

These fluid-structure interactions can difficult to predict for transonic conditions


and so these measurements can be used to both improve understanding of the
physics and provide validation data for either CFD codes or theoretical models.
Furthermore, it should be noted that the approached used here for investigation
of blade response to a mechanical impulse is, to the authors knowledge, unique
among current aeroelastic literature.
The second objective of this dissertation was to develop an accurate, noncontact blade vibration measurement technique. The literature review revealed
that blade vibration measurements are typically acquired using one of two approaches : strain gages and Blade Tip Timing. Both techniques have unique
challenges that can be difficult to address, and so there exist an opportunity for
improvement in the field of blade vibration measurement technology. Blade Image Velocimetry is a novel approach to blade vibration measurements which is
non-contact, can be easy to implement and can provide accurate estimates of the
blade vibration amplitudes.

33

CHAPTER 2
EXPERIMENTAL SETUP AND DATA PROCESSING

2.1 Introduction
This chapter describes the experimental setups and data processing approaches
used in this dissertation. The chapter is divided into two sections. The first
section discusses the equipment and data processing algorithms used to investigate
the fluid-structure interactions of an isolated compressor blade in a high speed
jet. Two sets of experiments were performed. The first used Particle Image
Velocimetry to quantify the flow about the blade at mid-span and at several
Mach numbers and angles of attack. The second set of experiments used a Laser
Doppler Vibrometer to acquire time-resolved blade vibration measurements.
The second section discusses the experimental setup and data processing used
in the development and validation of the BIV technique. Three sets of experiments
were performed. The first experiment used a PIV system to acquire simultaneous
fluid and structure velocity measurements to motivate a unique capability of the
BIV technique. The second experiment was a bench-top validation of the measurement technique. The third experiment applied BIV to the rotor blades of a
single stage axial compressor.

34

2.2 Fluid-structure interactions of an isolated compressor blade


This section describes the equipment and data processing algorithms used to
investigate the fluid-structure interactions of an isolated compressor blade exposed
to a high speed jet. Two experiments were performed. The equipment common to
both experiments will be presented first. A discussion of the equipment and data
processing algorithms which were unique to each experiment will then follow.

2.2.1 General setup


The experiments were conducted using the Notre Dame Compressible Jet Shear
Layer facility. This facility is a blow-down tunnel powered by a 3 stage, 550 psig
piston-type compressor. The compressed air passes through a drier which reduces
the dew point to 73 C and is stored in a 7.8m3 insulated tank. Figure 2.1 shows
a sketch of the blowdown facility immediately upstream of the test section. The
mass-flow through the tunnel was controlled by an pneumatic valve. The valve
operated using proportional feedback to set the static pressure of the plenum
downstream of the second acoustic baffle. Honeycomb and a series of 6 screens
were used to condition the flow before exhausting through the nozzle and into
the test section. Measurements of static pressure and temperature were taken
upstream of the nozzle (shown in brown). Seeding fluid (for PIV measurements)
was introduced in the plenum, upstream of the honeycomb. A detailed description
of the facility can be found in [63].
The test section consisted of a nozzle exhausting into a two-sided test section.
Figure 2.2 (a) shows a schematic of the geometry of the test section. The nozzle
exit was 104 52 mm and had a contraction ratio of 10:1. Static pressure measurements at the nozzle exit were obtained to estimate gas exit velocity and Mach

35

Figure 2.1. Sketch of the blow-down tunnel immediately upstream of the


test section.

number. The test section had two walls. The first consisted of an aluminum plate
bolted to a rigid structure and acted as a stiff base. The blade (shown in green)
was bolted to this wall as a cantilever. The second wall was a clear acrylic plate
with a nominal thickness of 12.7 mm. The plate was installed to reduce finite-span
effects such that the flow over the blade was nominally 2D. The plate was secured
to a fixture extending from the nozzle exit (not shown) and to the aluminum plate
using struts located downstream of the blade.
A contoured hole was cut in the plate to allow for a reduced cross-section
of the blade to extend out beyond the test section. This portion of the blade,
referred to as the handle, was used to apply a mechanical impulse to the blade.
A mechatronic hammer was used to apply a repeatable mechanical impulse to the
blade handle. Figure 2.2 (b) shows the mechatronic hammer installed in the test
36

(a)

(b)

Figure 2.2. Schematic of test section used for aerodynamic damping


investigation.

section. This configuration was used because it eliminated disturbances to the


flow in the test section due to the motion of the hammer.
The excitation provided by the hammer can be described as follows. The tip
of the hammer deflected the blade handle by some known amount. The handle
was then released rapidly, causing the blade to vibrate. Figure 2.3 shows the
time resolved blade velocity at a fixed position on the blade handle after the
mechatronic hammer hit the blade in quiescent air. Twenty nine responses are
superposed in this figure. The ensemble RMS of these twenty-nine responses is
also shown as a function of time. The mechanical impulse applied to the blade
resulted in vibrations whose amplitude decayed with time. It can be seen that most
of the vibration was at a single frequency; this frequency corresponded to the first
natural mode of the blade. Other, smaller amplitude vibrations were present in the
first few cycles after the impulse; their presence was most noticeable at the local
maxima and minima of the main vibration. This response was typical of a beam

37

0.3

Velocity [m/s]

0.2
0.1
0
0.1
0.2
0

10

10

15

20

25

15

20

25

Ensemble RMS [m/s]

0.01
0.008
0.006
0.004
0.002
0

Time [ms]

Figure 2.3. Scatter plot of time-resolved blade response to mechanical


impulse in quiescent air (upper), and ensemble standard deviation of the
response (lower). 29 responses shown.

impulse-response with non-zero damping. The ensemble-RMS vibration revealed


that the response was remarkably repeatable. The highest observed RMS occured
near the maximum observed velocity, coinciding with the beginning of the impulseresponse. The RMS at this point was observed to be 2.5% of the peak velocity,
confirming that the mechatronic hammer produced repeatable blade mechanical
loads.
The blade was constructed as three separate pieces composed of low-carbon
steel. A constant cross-section beam was clamped between two blocks. The blocks
were then bolted to the aluminum end-wall. The beam cross-section is shown
in figure 2.4. The cross-section was based upon a NACA 65 series airfoil with
a 3% thickness to chord ratio and a circular arc camber line. The angle of the

38

10
8
6

Z [mm]

4
2
0
2
4
6
8
0

10

15

20

25

30

35

X [mm]

Figure 2.4. Cross-section of the blade. Blade surface; Camber


line; Center of gravity.

camber line at the leading and trailing edges relative to the x-axis were 4 and 4
respectively. The airfoil geometry was consistent with a subsonic, low pressure,
low solidity compressor. The cross-section shown in figure 2.4 was held constant
over a 54 mm span. The handle section was formed by extending a reduced airfoil
profile, corresponding to between 25% and 50% chord, an additional 25 mm.
The first two eigen-modes of vibration are shown in figure 2.5. The eigenmodes were calculated using ANSYS 12.0. It can be seen that the first eigenmode corresponds with a bending-type vibration, whereas the second mode wa a
torsion-type vibration. Note also that the frequency of the second eigen-mode was
higher than the first eigen-mode by a factor of 4.4. The mode-shapes and natural
frequencies of the first ten eigen-modes can be found in appendix 1. The first two
modes will be referred to as bending and torsion respectively.
The instrumentation used in this series of experiments consisted of the following. A pressure transducer and k-type thermocouple were installed downstream of
the screens shown in figure 2.1. The pressure transducer was installed to measure

39

(a)

(b)
||

0.3

Z [mm]

10

0.25

0.2

10
0.15

50
40
30

0.1

20
10
20

0
10

Y [mm]

0
20

20

X [mm]

0.05

Figure 2.5. First two eigen-modes of the blade. (a) Mode 1, f1 = 349
Hz; (b) Mode 2, f2 = 1540 Hz.

the differential static pressure between the nozzle inlet and exit, whereas the thermocouple provided a temperature measurement at the nozzle inlet. Measurements
of atmospheric pressure were obtained from a mercury barometer. The nozzle exit
Mach number, velocity and temperature were calculated by assuming isentropic
flow and the relations

"
A1
M1 1 +
=
A2
M2 1 +

1
M22
2
1
M12
2

+1
# 2(1)

(2.1)

and
"
1+
P1
=
P2
1+

1
M22
2
1
M12
2

# 1

(2.2)

These relations provide a unique estimate of the nozzle exit Mach number when
the upstream Mach number is less than 1.
Blade vibration was measured using a commercial Laser Doppler Vibrometer
(LDV) with a typical accuracy of 20 m/s. The beam was focused on the handle
40

section of the blade. The blade vibration was quantified by estimating the modal
amplitude. The instantaneous velocity of the blade surface was filtered to isolate
the mode of interest, and then a conversion factor based upon the mode shape
was applied to obtain the modal amplitude from the surface velocity.

2.2.2 PIV measurements


The fluid velocity was measured using a two-camera PIV system. The cameras
were focused on the suction and pressure sides of the blade respectively. The flow
was seeded using a commercial Laskin nozzle and The 3 mm thick light sheet was
positioned at 65% span. This location was selected based on surface-streaking
measurements, which revealed it to be the furthest out-board location for which
there was little spanwise-curvature of the surface streak-lines. A mirror was used to
eliminate shadows in the cameras field of view. The PIV system was synchronized
to the mechatronic hammer to allow for phase-averaged measurements of the fluid
response when a mechanical impulse was applied to the blade. Portions of the
acrylic end-wall were either tinted or completely masked off to reduce the intensity
of the laser light reflected off the blade surface.
The PIV vector calculation was performed in two parts based on proximity to
the blade surface. A far-field vector calculation was performed to estimate the fluid
velocity at distances greater than 5 mm from the blade surface. A separate, nearfield vector calculation was also performed. The two approaches were necessary
due to the bright background reflections present near the blade surface.
The far-field vector calculation consisted of a sliding background subtraction
which removed large scale brightness fluctuations. A particle intensity normalization was then performed to equalize the brightness of all the visible particles in the

41

image. A multi-pass vector calculation was performed using progressively smaller


interrogation windows. The initial window size was 64 64 pixels, overlapped
by 50% of the window size and was weighted using an axisymmetric Gaussian
weighting function. The final two passes used 32 32 pixel interrogation windows
with similar weighting and overlap. Vector validation consisted of the following
operations. Vectors which had correlation peak ratio which were less than 3 were
discarded. A velocity threshold which limited the fluid velocity results to between
20% and 200% of the expected free-stream velocity in the streamwise direction,
and 60% of free-stream in the chord-normal direction. Finally, the vectors which
deviated by more than 3 standard deviations from their immediate neighbors were
discarded.
The near-field calculation was similar to the far field calculation. The image
preprocessing consisted of a sliding background subtraction was applied, followed
by a particle intensity normalization. Then, the image from the second frame
was subtracted from the first and vice-versa. All pixels whose intensity was negative was then set to zero. The main effect of these additional operations was
to remove all stationary features from the images. Near the blade, a majority of
the stationary features were associated with background reflections, and so this
image pre-processing effectively isolated the seed particles from the background.
A multi-pass vector calculation was also used, but started with 256 256 pixel
interrogation windows and finished with 32 32 pixel windows. The overlap and
weighting function was identical to that used for the far field calculation. Vector
validation consisted of the following operations. First, vectors which had a correlation peak ratio less than 2 were discarded. The velocity threshold ranged from
10% to 200% of the freestream along the streamwise direction and to 60% along

42

the chord-normal direction. The vectors which deviated by more than 3 standard
deviations from their immediate neighbors were discarded.
The four vector fields, two for each camera, were then stitched together to
form a composite fluid velocity field. The velocity field had a spatial resolution of
approximately 4 vectors per mm and a field of view of 8080 mm. The accuracy of
the velocity measurements was approximately 1% of the freestream fluid velocity.
The PIV system was synchronized to the impact hammer and measurements
were phase-averaged to re-construct the first bending response. This measurement approach was used because it provided a well-defined reference point. This
reference point allowed for the investigation of the flow field as it evolved with
the unsteady vibration of the blade, and may provide insight into the dynamic
fluid-structure interactions.
The bending response was isolated by aligning the tip of the impactor with the
mid-chord line of the blade. Note that the mid-chord location was near a nodal
line for the torsion eigen-mode, and so it was expected that the response would be
primarily due to bending. Approximately 1500 vector fields were phase averaged
about the first 1.5 cycles of vibration after the mechanical impulse was applied.
Twelve evenly spaced points were used for the subsonic tests cases and 24 points
were used for the transonic tests cases. The number of points was increased for
the transonic test cases because of the large response observed for first torsion at
these conditions. The resulting time series could be used to estimate the unsteady
lift and identify key features associated with the impulse response.

43

2.2.3 LDV measurements


The modal velocity and modal damping ratio were calculated from the LDV
measurements. A brief description of the method by which the modal velocity
and damping ratio were calculated will be presented. First, a continuous wavelet
transform of the impulse response was performed using Meyer wavelets and scale
factors corresponding to the natural frequencies of the first two modes to obtain the modal velocity of the pth mode, p . This transformation is described by
equation 2.4

1
p (t) =

p (xm , ym ) ap


v(xm , ym , )

t
ap

ap =

d,

(2.3)

fc fs
fp

(2.4)

where v(xm , ym , t) is the measured surface velocity at position xm , ym and time t,


p is the mode shape for the pth mode, ap is the scale factor, xm and ym describe
the chord-wise and span-wise location of the LDV measurement, is a dummy
variable of integration, fc is the center frequency of the wavelet, fs is the sampling

frequency of the signal and tb
is the wavelet function. The center frequency
a
of the wavelet was estimated from Matlabs scal2frq function, and represents
the approximate frequency of the wavelets main oscillations. This transformation
produced a narrow-band signal.
Matlabs Hilbert function was used to provide a complex representation of
the modal velocity, p (t). This operation shall be denoted by the operator H[].
The function calculates the Hilbert transform of a signal and adds it to the original
signal. The advantages of working with the complex representation of the signal

44

instead of the real valued signal is that accurate estimates of the instantaneous
amplitude and phase may be obtained.
Figure 2.6 presents a graphical summary of how the measured blade surface was
decomposed into modal velocities to estimate the modal damping ratio.

v [m/s]

0.5
0
0.5

1 [m/s]

10

15

20

25

30

35

40

10

15

20

25

30

35

40

10

15

20

25

30

35

40

0.5
0
0.5

2 [m/s]

0
0.5
0
0.5
0

Time [ms]

Figure 2.6. Graphical decomposition of measured blade surface velocity,


v () into modal velocities () and instantaneous envelopes,
|H [p ] | ( ). Response shown was acquired for quiescent air.

The magnitude of H [p (t)] provides a good estimate of the variation in the signals
envelope with time. The frequency of the signal may be estimated from the

45

derivative of the unwrapped phase angle of the signal,


d
tan1
2f
dt

= [H [p (t)]]
< [H [p (t)]]


(2.5)

where =[] and <[] select the real and imaginary component of the signal. The
modal damping ratio may be estimated by

2f

d
log (|H [p (t)] |)
dt

(2.6)

If the magnitude of the modal velocity had an exponential decay rate, then
equations 2.5 2.6 should be constant. This assumption was used for the estimate of all damping ratios reported. Note that a linear curve fit was applied to
tan1 (= [H [p (t)]] /< [H [p (t)]]) and log (|H [p (t)] |) to provide accurate estimates
of the natural frequencies and damping ratios.

2.3 Development and validation of BIV


This section describes the equipment and data processing algorithms used to
validate the BIV measurement technique. Three experiments were performed. All
three experiments used similar equipment in different configurations. Additionally,
the data processing algorithms varied significantly between all three experiments,
due to the differences in the objectives. Therefore, a section will be devoted to
the description of each experiment and its subsequent data processing approach.

2.3.1 Demonstration of simultaneous fluid and structure velocity measurements


The simultaneous measurement of fluid and structure velocity was demonstrated using a standard single camera PIV system [54]. A schematic of the setup

46

is shown in Figure 2.7. The model shown is a cantilevered beam which nominally
represents a generic axial compressor blade geometry. The 6061-T6 aluminum
plate had a thickness of 0.50 mm, a chord of 78 mm, and a span of 189 mm. The
plate was positioned in the test section of an open jet wind tunnel at a nominal angle of 20o relative to the incoming flow. The free-stream fluid velocity was 12 m/s.
This resulted in large amplitude oscillations of the plate. The frequency of oscillation was obtained from a conventional Laser Doppler Vibrometer (LDV) system,
and was found to be 53 Hz. The reduced frequency, defined by ? = f c/(U ),
was 1.08 where c is the chord length, f is the frequency of oscillation, and U is
the free stream velocity. The CCD camera field of view was 84 84 mm resulting in a spatial resolution of 24.3 pixels/mm. The camera was calibrated using a
standard PIV calibration plate using a 3rd order polynomial fit with an RMS of
0.08 pixel.
The coordinate system is defined by the x, y and z vector directions that are
oriented in the chord-wise, span-wise and blade normal directions, respectively. A
dual-cavity Nd:YAG laser light source was focused into a 1 mm thick light sheet
in order to illuminate the fluid plane of interest. The sheet was oriented at a
small angle relative to the tip plane such that the tip was illuminated. Note that
if the camera can not focus on the fluid plane of interest and the blade tip at the
same time, a two-camera system would need to be employed. A random pattern
of matte white paint dots were applied to the tip of the cantilevered beam.
This pattern was designed to mimic that of a seeded flow typical of conventional
PIV so that the same vector calculation algorithms could be employed without
modification. The CCD camera was synchronized to the laser trigger to obtain two
consecutive images of the vibrating blade and seed particles in the fluid. Image

47

pairs were acquired at 1 Hz.

Figure 2.7. Sketch of a stationary blade and measurement points.

All images were processed using LaVisions DaVis software suite. The images
were first filtered using a Sobel edge detection filter. Local intensity fluctuations were minimized using a sliding background subtraction algorithm. The BIV
calculation consisted of a three iteration multipass approach using progressively
smaller square windows. The interrogation windows used on the first pass were
64 64 pixels in size weighted with an axisymmetric Gaussian weighting function
(the weighting function had no preferential direction with respect to the image).
The overlap between interrogation windows was 50%. The last two passes utilized 32 32 pixel windows with an axisymmetric Gaussian weighting function
and a Whittaker image reconstruction on the final pass. This resulted in a vector
resolution of 1.5 vectors/mm.
Vector validation allowed a maximum vector velocity that was set to 3 times

48

greater than the maximum anticipated tip velocity and a minimum correlation
peak ratio (for a given window, the ratio of the highest correlation peak had to
be at least 1.3 times greater than the second highest correlation peak). Lastly, a
median filter (velocity vectors which deviate by more than 2 standard deviations
of their neighbors were rejected) and a single smoothing operation were applied.
For a brief introduction to these processing operations, the reader is referred to
chapters 5 and 6 in Raffel et al. [54].

2.3.2 Benchtop validation of measurement theory


The bench-top validation of the BIV measurement technique was performed
using a vibrating cantilevered beam. Harmonic forcing in the form of acoustic
pressure was applied to the beam using a 100W speaker. The separation between
the cone of the speaker and the surface of the beam (zdirection) was 2.5 mm. The
forcing was applied at the resonant frequencies. A schematic of the experimental
setup is shown in figure 2.8.
The cross-section of the beam is shown in figure 2.9. The cross section was
constant over the 115.6 mm span. The beam was made from 6061T6 aluminum
bar stock. The second and third natural modes are shown in Figure 2.10. These
were computed using FEA software (ProE Wildfire 2.0). The frequencies of the
modes showin in figure 2.10 are 782.4 and 959.4 Hz respectively. The natural
modes and frequencies of vibration f of the blade were experimentally verified
using a scanning LDV and harmonic forcing at the resonance frequencies using
the setup shown in figure 2.8.
The blade tip motion was recorded using the same PIV system described in
section 2.3.1. Note that similar image processing and vector validation algorithms

49

Figure 2.8. Schematic of the BIV benchtop validation experiment.

12.7
3o
3.17
LDV measurement point
50.8
z
x

Figure 2.9. Sketch of the blade cross section geometry (not to scale). All
dimensions in mm. The cross section was constant across the 115.6 mm
span.

were used. One thousand image pairs (N = 1000) were acquired.


A Laser Doppler Vibrometer (LDV) was used as a reference sensor that was
assumed to provide the true velocity at one point on the blade. The uncertainty in

50

Normalized Velocity

(a)
1
0.5
0
0.5
1
0
20
40
60
50

80
100

y [mm]

x [mm]

Normalized Velocity

(b)
1
0.5
0
0.5
1
0
20
40
60
50

80

y [mm]

100
0

x [mm]

Figure 2.10. Shapes of the two structural modes of the test blade as
computed by FEM. Shapes are exaggerated for clarity. (a) First torsion
mode of the model, [f ]1 = 782.4 Hz; (b) Second bending mode of the
model, [f ]2 = 959.4 Hz.

the LDV velocity measurement was 20 m/s. The single point LDV was positioned
near the blade tip.
The accuracy of the structural velocity measurements was evaluated experimentally. Velocity measurements obtained from BIV were spatially averaged in
the vicinity of the LDV measurement point. The LDV measurements were time
averaged over the 200s required to obtain two BIV images. Only the first mode

51

was excited for this test.


The measurement theory and uncertainty analysis were validated by considering the amplitude of blade vibration. Time series data were acquired using the
LDV for 60 seconds before and after BIV data acquisition. The autospectral density of the LDV data was computed and the mean-squared amplitude of the pth
h i
mode, 2 , was determined by integrating around the frequency band of interp

est. The experimentally measured error in the estimation of modal amplitude was
defined by
r h i
q
2 2 2 [ 2 ]p
p
r h i
,
[E ]p
2
2

(2.7)

r h i
where 2 2 is the amplitude of the pth mode as measured by the LDV and
p
q
2 [ 2 ]p the amplitude of the pth mode as measured by the BIV technique.
The beam was harmonically forced only at first torsion at two different amplitudes to demonstrate the interaction of the two types of error. The time between
the two images, denoted by t, was varied among different tests to vary the
influence of the random and bias errors on the measured amplitude of vibration.

2.3.3 Application of BIV to an axial compressor rotor


Figure 2.11 shows a sketch of the compressor rotor stage used for this series
of tests. Note the presence of two sets of coordinate axes. The first, shown in
black, is the cylindrical coordinate system which is used to describe the threedimensional rotor blade system. The second coordinate system, shown in red, is
oriented along the blade stagger angle and describes the blade tip motion in terms

52

Figure 2.11. Sketch of a cantilevered rotor blade and tip velocity


measurement points.

of the chordwise (C) and chord-normal (N ) axes.


The BIV technique was applied to the rotor of a high-speed, single stage
axial compressor. The twenty blade rotor was an integrated bladed-disk (blisk)
design with tip and hub radii of 228 mm and 177 mm respectively. The drive shaft
was supported using magnetic bearings, which allowed for unsteady actuation of
the rotor along the axial and radial axes. A detailed description of the compressor
can be found in Cameron et al [16].
The compressor aerodynamic design was representative of a modern transonic axial compressor stage. A sample of the 3D blade geometry is shown in
Figure 2.12. Here, four cross-sections at varying span are shown. The compressor
was originally designed to study stall, an unsteady operating condition associated
with low air mass flow. Compressor stall can result in large unsteady aerodynamic
loads. The mechanical design required infinite fatigue life under stalled conditions,
which resulted in stiff blades. The maximum thickness to chord ratio varied from
9% near the hub to 3% at the tip. This design was ideal for determining the noise

53

Figure 2.12. Cross-section of rotor blade at 4 locations along span.


tip, 75% span, 50% span, 25% span.

limits of the BIV technique on a high-speed rotor since at design conditions the
blades could be considered nearly rigid.
The first three eigen-modes of the blades and their associated frequencies
are shown in the left column of figure 2.13. The tip mode shapes, , are also
shown in the right column of figure 2.13. The eigen-modes were calculated using
ANSYS 12.0. Note that the complete eigen-modes are highly three-dimensional,
which results in a set of nearly orthogonal tip modes.
A schematic of the experimental setup is shown in figure 2.14. The BIV
system utilized hardware common to a typical, two-camera, Particle Image Velocimetry (PIV) system [54]. The cameras (FlowMaster 2S, 2048 x 2048 pixel
CCD) were focused on rotor tip through a casing window. The field of view of

54

(a)
0
10

RT [mm]

20
30
40
50
60
70
80
0

10

20

30

40

30

40

30

40

X [mm]

(b)
0
10

RT [mm]

20
30
40
50
60
70
80
0

10

20

X [mm]

(c)
0
10

RT [mm]

20
30
40
50
60
70
80
0

10

20

X [mm]

Figure 2.13. First three eigen-modes of the rotor blade. (a) [f ]1 = 2, 148
Hz; (b) [f ]2 = 2, 698 Hz; (c) [f ]3 = 4, 025 Hz.

55

Figure 2.14. Schematic of the experimental setup used to demonstrate


BIV.

the cameras was 65 120 mm with a spatial resolution of 29.5m/pixel. The field
of view for both cameras overlapped by 100 pixels and the images were used
to form a composite view of the blade tip. The cameras were calibrated using
a printed calibration plate which conformed to the inner casing window. The
spatial calibration mapped the camera images to a uniform grid in the axial and
circumferential directions. A 3rd order polynomial fit was used to calibrate the
cameras.
The cameras were operated in double-frame double-exposure mode, where
a pair of images were acquired with the blade illuminated once per image. This
recording approach is typically used in PIV to provide an unambiguous estimate
of particle direction and velocity. Single-frame, double exposure (SFDE) was also
used successfully since the mean blade motion eliminated directional ambiguity.
A detailed description of these recording approaches can be found in Raffel et al.
[54].
The blade tip was illuminated using a dual cavity Nd:YAG PIV laser (New

56

(b)
0

10

10

20

20

30

30
RT [mm]

RT [mm]

(a)

40

40

50

50

60

60

70

70

80

80
0

10

20

30

X [mm]

10
20
X [mm]

30

Figure 2.15. (a) Sample image of a blade tip used for BIV measurements.
(b) Sample tip velocity estimated from the BIV technique.

Wave SOLO 120). The beams were expanded using a spherical lens and an optical diffuser such that illumination was uniform. The laser and cameras were
synchronized to a once-per-revolution (OPR) sensor.
A surface treatment was applied to the blade tips to facilitate velocity estimation. This surface treatment consisted of a random pattern of matte white
dots on the matte-black blade tips (see figure 7.1(a)). The diameter of the
dots was between 50 and 150 m and were designed to mimic a seeded flow
commonly found in PIV. This allowed the use of standard PIV vector calculation
algorithms to determine blade tip velocity.
All images were processed using LaVisions DaVis software suite. The images from each 4 megapixel camera were warped according to the cameras spatial
calibration and stitched together to form a set of 8 megapixel composite images.

57

The images were enhanced to reduce background reflections. The intensity of the
image was multiplied by 10 and a sliding background subtraction algorithm was
then applied. A bias of -50 counts was applied to the intensity of each pixel. Any
pixel whose intensity was less than zero was then set to zero.
The vector calculation consisted of a three iteration multipass approach
using progressively smaller square interrogation windows. All passes utilized an
axisymmetric Gaussian window weighting function and the interrogation windows
were overlapped by 50% of the window size. The first pass used 128 128 pixel
interrogation windows. An initial guess of the blade velocity, specified in equivalent pixel displacement, was provided to improve correlation accuracy. The last
two passes used 64 64 pixel interrogation windows, overlapped by 50% of the
window size and weighted using an axisymmetric Gaussian weighting function.
Vector validation consisted of the following criteria. Vectors which deviated by
more than 1% of the theoretical average tip velocity were rejected, as well as
those whose peak correlation value was less than 1.75 times larger than the next
highest peak. Vectors which deviated by more than 2 standard deviations from
their immediate neighbors were also removed and replaced with the local median
velocity. Groups of 4 vectors or less were removed entirely. Lastly, the vector field
was then spatially averaged about a set of points uniformly distributed along the
blade camber line. Twenty-one points along the camber-line were used, resulting
in a spatial resolution of about 4 mm, see figure 7.1(b).
An a-posteriori calibration was applied to the vector field. This consisted of
a correction of vector magnitude and direction. The correction set the difference
between the ensemble-averaged un-corrected BIV velocity field and the known
mean tip velocity to zero. The mean tip velocity was found from the shaft speed

58

and tip radius. This calibration was applied to the data in order to account for
small scale optical distortions in the casing window and misalignment between the
two cameras.
Two rotor shaft speeds were investigated using the BIV measurement technique. The first was at low shaft speed (2,886 RPM), and was used to validate
the measurement technique. The second was at high shaft speed (13,531 RPM).
The first set of experiments were conducted at low shaft speed and were
performed to validate the accuracy of the measurement system against a known
rotor tip velocity. The rotor shaft speed was held constant at 2,886 RPM, corresponding to a blade tip velocity of 65.5 m/s. The magnetic bearings were used
to oscillate the rotor axially. The axial oscillation was synchronous to the shaft
rotation with a frequency of 96.2 Hz and an amplitude of 49 mm/s. The frequency
of the axial oscillation was well below the natural frequencies of the blades and
so the motion of the tip due to blade vibration was assumed to be negligible.
Three blades were analyzed, corresponding to different points within a cycle of
oscillation. Blade number 5 was used as the reference case for the a-posteriori
calibration. Five-hundred realizations of the tip velocity were acquired for each
blade.
The second set of experiments were conducted at a shaft speed of 13,531
RPM, corresponding to a blade tip velocity of 322 m/s. The magnetic bearings
maintained the axial position of the shaft to 30m, and controlled the whirl
orbits to within 6 mum. BIV data were acquired at two operating points. The
first was at the aerodynamic design point of the compressor. The second was a
low air mass flow condition where the compressor was operating in deep stall.
Large unsteady aerodynamic loads were expected at this flow condition and the

59

BIV technique was used to determine if these unsteady loads would induce a
detectable level of blade vibration. A single blade was investigated and an aposteriori calibration based on the ensemble-averaged tip velocity was applied
at each operating condition. One thousand realizations of the tip velocity were
acquired for each operating point.

60

CHAPTER 3
PIV MEASUREMENTS OF STEADY FLOW ABOUT A FLEXIBLE BLADE

3.1 Introduction
This chapter presents PIV measurements of the flow around the flexible cantilevered compressor blade. Experimental measurements of the flow about the flexible compressor blade will provide perspective for the results presented in chapters
4 and 5.
PIV measurements of the flow were obtained at 65% span for several combinations of blade angle of attack and Mach number. The measurements were
synchronized to the impact hammer described in chapter 2 so that the phaseaveraged fluid response could be evaluated.
Five different combinations of Mach number and blade angle of attack were
investigated using PIV. The combinations of angle of attack and Mach number
are given in table 3.1. The Mach numbers investigated were representative of
three important types of flow fields: incompressible, subsonic compressible, and
transonic.
Data were acquired at 0 = 8 and for incompressible(M = 0.16), highsubsonic (M = 0.60) and transonic flow (M = 0.75). Additionally, data were
acquired at 0 = 13 and for incompressible (M = 0.16) and transonic flow
(M = 0.75). The Mach numbers and angles of attack investigated were selected
61

TABLE 3.1
TEST MATRIX FOR PIV INVESTIGATION OF FLOW FIELD
ABOUT THE BLADE

0 [degrees]

LE incidence angle [degrees]

f 1 c
U

f 2 c
U

0.16

0.71

3.14

0.55

0.22

0.99

0.75

0.16

0.70

13

0.15

0.99

4.36

13

0.75

0.16

0.70

to represent several noteworthy flow conditions. Incompressible and compressible subsonic flow can be accurately described using linear aeroelastic theories.
Transonic flow can be challenging to predict using current theoretical or numerical approaches. Consequently, measurements of the flow at transonic conditions
can provide additional insight that are not available to theoretical or numerical
approaches.
The angles of attack investigated were selected to represent two specific conditions. Note that axial compressor blades use incidence angle in lieu of angle
of attack to account for differences between the camber-line angle at the leading
edge and the chord-line of the blade. The 0 = 8 was selected as an equivalent
low incidence condition. Such a condition might be found near the working line
of the compressor performance map presented in figure 1.3. The 0 = 13 was
selected as a high incidence condition that may be found near the surge line
shown in figure 1.3.

62

The chapter is divided into the following sections. First, a description of


the theory by which the static pressure may be obtained from phase-averaged,
2-component planar velocity fields will be described. A discussion of the fluid
velocity, vorticity and pressure fields obtained at the 5 combinations of angle of
attack and Mach number will be presented. The chapter will conclude with a brief
summary of the observations and conclusions.

3.2 Theoretical approach for non-contact surface pressure measurements


Measurement of unsteady fluid pressure acting on the blades offer a means to
directly evaluate the validity of theoretical and numerical models, as well as the
ability to identify key fluid-dynamic phenomenon associated with various aerodynamic loads. Surface mounted pressure transducers are an approach which can
directly measure these loads, but require modification of the blade. However, there
are certain applications (such as thin blades that are commonly found in turbomachinery) where surface mounted pressure transducers are impractical. An ideal
approach would estimate the unsteady pressure using a non-intrusive, non-contact
approach.
An approach which will be used in this investigation estimates the pressure
field about the airfoil from an unsteady, 2-component planar velocity field in the
following manner. Given a time series of velocity measurements, the local fluid
material acceleration may be estimated by virtual particle tracing. The material
acceleration can be used to estimate the pressure gradient field. Spatial integration
of this gradient field from a known reference location, such as the free-stream, can
provide an estimate of the static pressure.
The virtual particle tracing technique was pioneered by Jakobsen et al. [37]

63

and can be described as follows. An infinitely small volume of fluid, henceforth


referred to as a virtual particle, is defined at a specific time and location. The
instantaneous velocity of the center of mass of the particle is assumed to be equal
to the local velocity at that specific time and location. The objective is to determine the particles velocity and location after one time-step. The change in
velocity after the time step is a good approximation to the material acceleration
in Lagrangian specification of the flow field.
The change in velocity of the virtual particle was estimated by an iterative
procedure. The displacement of the particle for the time step was calculated
based on the average velocity (which was assumed to be the initial velocity at the
first iteration). The particle velocity at the end of the time-step was calculated
from the flow field by interpolation in both space and time. The average velocity
over the time step was then calculated and the process repeated. The time step
was limited such that the maximum particle displacement was less than twice
the spatial resolution of the PIV vector fields. Five iterations were found to be
sufficient. The material acceleration was calculated from the change in velocity
and was specified at the mid-point of the particles trajectory. The irregularly
spaced acceleration field was then mapped back onto a regular spatial grid by
interpolation.
Assuming inviscid and isentropic flow, the product of the fluid density and
the material acceleration is equal to the pressure gradient. Integration of this
gradient field from a reference location where the pressure is known allows for the
estimation of the static pressure field.
The conservation of momentum of an inviscid compressible flow are described
by the Euler equations,

64

f )
+ (f U) = 0
t
(f U)
+ (f UU) = P
t

(3.1)
(3.2)
(3.3)

where f is the density of the fluid, U is a vector describing the fluid velocity, P
is the pressure, and is the spatial derivative operator (

).
z

The

Euler equations state that the product of a component of the material acceleration
with the fluid density is equal to the sum of the pressure gradient along the given
direction. If the flow can be assumed to be isentropic, the fluid density and
pressure can be related. A scaled pressure can be defined as
1/

P?

P (1)/
P
.
1 f

(3.4)

Substituting equations 3.1 and 3.4 into equation 3.3 yields


U
+ U U = P ? .
t

(3.5)

Equation 3.5 relates the gradient of the scaled pressure field to the material
acceleration of an isentropic flow with negligible viscous effects.
The pressure field was calculated by spatial integration of the acceleration field
from a location where the pressure was known. Note that the pressure field, being
a scalar, should be independent of the path used for the integration. Different
path integrals from the same reference point to the same location should yield
identical results. This can be used to reduce the random error in the pressure
field estimate by averaging the pressure over many different path integrals. A
65

multi-path integration scheme was implemented to take advantage of this feature.


The virtual boundary integration scheme from Liu and Katz [47] was used to
provide an optimized set of path integrals for the estimation of the pressure field.
The pressure field was used to estimate the blade surface pressure distribution.
A 3rd order polynomial interpolation of the pressure field along lines normal to the
blade surface was used to deduce the pressure at the blade surface. One-hundred
points along each side (pressure and suction) were used.
The surface pressure distribution was compared against a theoretical distribution calculated using a compressible panel method code. The panel method
only considered two-dimensional flow over the blade cross section and could be
described as follows. A set of N irrotational vortices and sources were distributed
evenly along the blade camber line. Recall from potential flow theory that the
velocity field due to an irrotational vortex and source can be expressed as

x0 )
i(z z0 ) k(x
,
2 ((z z0 )2 + (x x0 )2 )
x0 ) + k(z
z0 )
k(x
,
US (x, z) = S
2 ((z z0 )2 + (x x0 )2 )

U (x, z) =

(3.6)
(3.7)

where and S are the strength of the irrotational vortex and source respectively,
i and k are unit vectors aligned to the x and z axes respectively and x0 , z0
describes the location of the source and/or vortex. A set of 2N control points
were distributed along the blade profile at the same chord-wise locations as the
source and irrotational vortices. The control points were used to enforce an impermeability boundary condition at the blade surface. The Kutta condition was
enforced by requiring that the velocity magnitude at the two control points at

66

the blade trailing edge had equal velocity magnitude. Note that this resulted in
2N +1 equations for the 2N unknowns (the strength of the sources and vortices,
and S). A least-squares solution was obtained for the source and vortex strengths.
Compressibility was accounted for by a Prandtl-Glauert transformation.

3.3 Results
The average fluid velocity, RMS velocity and average vorticity at 65% span are
shown in figure 3.1 for a free stream Mach number of M = 0.16 and 0 = 8 . These
data represent the in-plane (x z axes) time-averaged and RMS fluid velocity
and out-of plane (yaxis) time-average vorticity. The fluid velocity statistics
were derived from 1500 PIV vector fields acquired between 0.5 and 1.5 bending
vibration cycles after the impact hammer hit the blade. The vorticity field was
calculated from the time-averaged velocity field. The average and RMS velocity
fields were normalized by the 57 m/s free-stream velocity. The fluid vorticity was
normalized by the factor |U |/2b, where 2b is the chord length of the blade. The
coordinate system used in the plots was aligned to the blade coordinates and so
the free-stream velocity is inclined relative to the xaxis by approximately 8 and
proceeds along both positive x and z axes.
The average fluid velocity and vorticity fields indicated that the flow was attached over the first 80% of the chord. The wake region, denoted by the velocity
magnitude deficit (figure 3.1 (a)) was approximately 5 mm in chord-normal extent
and centered about the blade trailing edge. The average vorticity field indicated
that outside of the wake, the flow was very nearly irrotational. The wake centerline, denoted by a region of zero vorticity, extended away from the trailing
edge nearly parallel with the x-axis. Note that the trailing edge of the blade was

67

(a)
|U|/|U |
1.75
1.5

1.25

z [mm]

10

0.75

10

0.5

15
0

10

20

30

40

50

60

0.25

x [mm]

(b)
U /|U |
0.4
10
0.3

z [mm]

5
0

0.2

5
0.1

10
15
0

10

20

30

40

50

60

x [mm]

(c)
y 2b/|U |
8
6

10

z [mm]

2
0

2
4

10

6
15
0

10

20

30

40

50

60

x [mm]

Figure 3.1. Fluid velocity and vorticity at M = 0.16 and 0 = 8 . (a)


average velocity, (b) RMS velocity, (c) average fluid vorticity.

68

angled 10 downward relative to the x-axis, and so it can be observed that the
wake center-line deviated from the blade trailing edge angle. The fluid velocity
vectors near the trailing edge were nearly parallel with the blade surface. There
was no direct evidence of flow separation. However, the flow field near the trailing
edge suggests a substantial flow deceleration which could result in boundary layer
separation.
The RMS velocity is shown in figure 3.1 (b). The typical RMS in the freestream was 0.03U , whereas the RMS velocity in the wake was as high as
0.2U . A correlation of the fluctuations in the velocity field far away from the
blade indicated that the free-stream fluctuations could be attributed to tunnel
velocity fluctuations and noise present in the PIV measurement. The tunnel velocity fluctuations were approximately 1% of free-stream, and the error in the PIV
measurements were approximately 1% 2%.
The phase-averaged fluid response around the blade at 0 = 8 and M =
0.16 is shown in figure 3.1. The figure shows three plots, corresponding to the
phase-averaged normalized velocity magnitude (figure 3.2 (a)), the out-of-plane
normalized vorticity (figure 3.2 (b)), and the calculated static pressure field (figure
3.2 (c)). The fields represent the flow response 0.5 cycles (with respect to the
bending mode) after the mechanical impulse was applied to the blade. The static
pressure is presented in terms of the non-dimensional pressure coefficient, Cp .
It can be observed that the phase-averaged velocity and vorticity fields were
similar to the time-averaged fields. This indicated that the unsteady flow induced
by the vibrating blade was small relative to the steady flow. The vorticity and
velocity fields also indicated that, in the phase-averaged sense, there were no large
scale boundary-layer separations. The pressure field (figure 3.2 (c)) indicated that

69

(a)
|U|/|U |
1.75
10

1.5

z [mm]

1.25

0.75

10

0.5

15
0

10

20

30

40

50

60

0.25

x [mm]

(b)
y 2b/|U |
8
6

10

z [mm]

2
0

0
2

10

6
15
0

10

20

30

40

50

60

x [mm]

(c)
Cp
1
10

0.5

z [mm]

0.5

10

1.5

15
10

10

20

30

40

50

60

x [mm]

Figure 3.2. Phase-averaged velocity, vorticity and coefficient of pressure


of fluid at M = 0.16 and 0 = 8 . The phase shown was at the first
maximum in bending vibration velocity after a impulse was applied.
fluid velocity; 2U blade velocity.

70

the net lift on the airfoil was positive, and that the pressure gradients above the
suction side were moderate. This was consistent with the velocity magnitude
results, which indicated a gradual deceleration from the peak velocity at 2%
chord. Additionally, the pressure field on the pressure side of the blade indicated
a higher flow acceleration from the stagnation point.

1.5

Cp

0.5

0.5

10

20

30

40

50

60

70

80

90

100

% chord

Figure 3.3. Phase-averaged surface pressure coefficient at 0 = 8 and


M = 0.16. Steady 2D panel method; from PIV data 1/2 cycle
after impulse; 3/4 cycle after impulse; 1 cycle after impulse; 1
1/4 cycle after impulse.

Figure 3.3 presents a comparison of the measured phase-averaged surface pressure distribution and the 2D steady panel method calculation. The surface pressure derived from the PIV data for the four phases shown had a similar chord-wise
71

distribution as the panel-method prediction. The PIV data and the panel method
show fair agreement along the first 80% of the pressure side of the blade. A slight
over-prediction of the surface pressure (resulting in a high Cp ) along the suction
side between 15% and 75% chord was observed. Note that this over-prediction
was found for all phase-averaged data, but varied with blade velocity. A portion of
this bias could be attributed to errors in both the static pressure field calculation
and errors in extrapolation to the blade surface. In contrast to the suction side,
excellent agreement was obtained for the the pressure side of the blade over the
first 80% of the chord.
A bias in surface pressure aft of 80% chord was observed. Here, the PIV data
indicated a lower pressure at the trailing edge than was predicted using the panel
method. The lower pressure coefficient was observed on both suction and pressure
side surfaces, and was present for all phase-averaged data. The source of this bias
was most likely a boundary layer separation. The asymmetry of the vorticity field
within the wake and the large chord-normal extent suggested that a minor flow
separation may have been present. This was not surprising given that the angle
of the suction side trailing edge was approximately 12 relative to the free-stream
velocity and that a significant adverse pressure gradient was predicted by the panel
method over the last 50% of chord.
The time-averaged velocity, vorticity and RMS velocity are shown in figure 3.4
for M = 0.55 and 0 = 8 . Here, the velocity magnitude was normalized by the
180 m/s free-stream velocity. The average velocity was qualitatively similar to the
observations at M = 0.16. The normalized velocity over the suction side of the
blade was higher at M = 0.55 compared to M = 0.16. Additionally, the velocity
deficit in the wake was smaller at M = 0.55 when compared against M = 0.16.

72

(a)
|U|/|U |
1.75
1.5

1.25

z [mm]

10

0.75

10

0.5

15
0

10

20

30

40

50

60

0.25

x [mm]

(b)
U /|U |
0.4
10
0.3

z [mm]

5
0

0.2

5
0.1
10
15
0

10

20

30

40

50

60

x [mm]

(c)
y 2b/|U |
8
6

10

z [mm]

5
2
0

2
4

10
6
15
0

10

20

30

40

50

60

x [mm]

Figure 3.4. Fluid velocity and vorticity at M = 0.55 and 0 = 8 . (a)


average velocity, (b) RMS velocity, (c) average fluid vorticity.

73

Specifically, the velocity magnitude of the wake for M = 0.16 was less than half of
the free-stream velocity, however at M = 0.55 the magnitude was nearly 75% of
free-stream. A comparison of the RMS velocity fields at M = 0.16 and M = 0.55
(figures 3.4 (b) and 3.1 (b)) revealed an increase in the RMS velocity in the wake
and near the suction side of the blade. The largest increase in RMS velocity was
observed on the suction side of the blade aft of mid-chord and in the wake near
the trailing edge.
The phase-averaged fluid response at 0 = 8 and at M = 0.55, 1/2 cycle
after the mechanical impulse was applied to the blade is shown in figure 3.5. The
phase-averaged flow field was qualitatively similar to the time-averaged flow field,
and so it may be concluded that the unsteadiness at these flow conditions due to
blade vibration was also much smaller than the steady flow. Furthermore, it can
be observed that at this phase the wake velocity magnitude decreased downstream
of the trailing edge of the blade. This decrease was observed up to 10 mm aft of
the trailing edge, beyond which the magnitude was approximately constant.
The vorticity field, shown in figure 3.5 (b), was qualitatively similar between
M = 0.16 and M = 0.55. Note however that the normalized wake vorticity
magnitude was smaller by a factor of 2, and was coincidental with the increase in
wake velocity magnitude observed in figure 3.5 (a).
The phase-averaged pressure field is shown in figure 3.5 (c). The following
differences can be noted between the M = 0.16 andM = 0.55 results. First,
the pressure over the suction side was significantly lower at M = 0.55 than at
M = 0.16. The minimum Cp was also lower due to the large flow acceleration
near the suction side leading edge. Larger pressure gradients were observed over
the suction side surface, which agree well with the larger velocity gradients.

74

(a)
|U|/|U |
1.75
1.5

1.25

z [mm]

10

0.75

10

0.5

15
0

10

20

30

40

50

60

0.25

x [mm]

(b)
y 2b/|U |
8
6

10

z [mm]

2
0

2
4

10

6
15
0

10

20

30

40

50

60

x [mm]

(c)
Cp
1

z [mm]

10

0.5

0.5

10

1.5

15
10

10

20

30

40

50

60

x [mm]

Figure 3.5. Phase-averaged velocity, vorticity and coefficient of pressure


of fluid at M = 0.55 and 0 = 8 . The phase shown was at the first
maximum in bending vibration velocity after a impulse was applied.
fluid velocity; 2U blade velocity.

75

1.5

Cp

0.5

0.5

10

20

30

40

50

60

70

80

90

100

% chord

Figure 3.6. Phase-averaged surface pressure coefficient at 0 = 8 and


M = 0.60. Steady 2D panel method; from PIV data 1/2 cycle
after impulse; 3/4 cycle after impulse; 1 cycle after impulse; 1
1/4 cycle after impulse.

Figure 3.6 shows the surface pressure coefficient of the blade calculated from
the panel method previously described and the estimates from the phase-averaged
PIV data 1/2 cycle after impact. Good agreement was obtained between the
steady panel method and the measured pressure distribution over the first 80% of
the chord. A negative bias was observed near the trailing edge of the blade for all
phase points investigated. This was consistent with the M = 0.16 results.
The time-averaged velocity, vorticity and RMS velocity fields for 0 = 8 and
M = 0.75 are shown in figure 3.7. The velocity magnitude was normalized by the
242 m/s free-stream velocity. A black isocontour denotes the region of supersonic
flow over the blade. Note that despite the presence of supersonic flow over the
suction side of the blade, no discernable shock was identified. The aerodynamic

76

(a)
|U|/|U |
1.75
1.5

1.25

z [mm]

10

0.75

10

0.5

15
0

10

20

30

40

50

60

0.25

x [mm]

(b)
U /|U |
0.4
10
0.3

z [mm]

5
0

0.2

5
0.1
10
15
0

10

20

30

40

50

60

x [mm]

(c)
y 2b/|U |
8
6

10

z [mm]

5
2
0

2
4

10
6
15
0

10

20

30

40

50

60

x [mm]

Figure 3.7. Fluid velocity and vorticity at M = 0.75 and 0 = 8 . (a)


average velocity, (b) RMS velocity, (c) average fluid vorticity.

77

design of the blade was not optimized for transonic flow, and so a shock was
expected for transonic operating conditions. The large velocity gradients upstream
of the supersonic pocket indicated that the seed particles tracked the fluid velocity
with reasonable accuracy, and so particle lag (cf. Raffel et al. [54]) could not
have been entirely responsible for the absence of a shockwave. Consequently, the
PIV measurements of the flow at these conditions are included as a qualitative
description of the flow.
The RMS velocity was significantly higher at M = 0.75 than at either M =
0.55 or M = 0.16. The isocontours of the time-averaged blade velocity over the
suction side had a different shape at M = 0.75 when compared to the M = 0.55
results. The flow at M = 0.75 resulted in the highest normalized velocity over the
blade. The wake thickness nearly tripled over the 23 mm distance downstream
of the trailing edge. Nearly all of this expansion occurred on the upper side of
the wake (associated with negative vorticity). The minimum normalized velocity
within the core of the wake was approximately 80% of free-stream. For reference,
the minimum wake velocity for M = 0.55 and M = 0.16 were 75% and 40% of freestream respectively. Finally, substantial increases in RMS velocity were observed
in the wake region near the blade trailing edge and near the blade leading edge
on the suction side.
The phase-averaged fluid response at 0 = 8 and M = 0.75 is shown in figure
3.8. It is immediately obvious from the velocity magnitude field (figure 3.8 (a))
that the phase-averaged measurements did not indicate a shock bounding the
downstream side of the supersonic flow region. Thus, it should be noted that the
phase-averaged results were also a qualitative description of the flow.
The wake region differed substantially from the low speed results. Specifically,

78

(a)
|U|/|U |
1.75
1.5

1.25

z [mm]

10

0.75

10

0.5

15
0

10

20

30

40

50

60

0.25

x [mm]

(b)
y 2b/|U |
8
6

10

4
5

z [mm]

2
0

2
4

10
6
15
0

10

20

30

40

50

60

x [mm]

(c)
Cp
1

z [mm]

10

0.5

0.5

10

1.5

15
10

10

20

30

40

50

60

x [mm]

Figure 3.8. Phase-averaged velocity, vorticity and coefficient of pressure


of fluid at M = 0.75 and 0 = 8 . The phase shown was at the first
maximum in bending vibration velocity after a impulse was applied.
fluid velocity; 2U blade velocity.

79

the thickness was much larger 1/3 of a chord-length aft of the blade trailing
edge. This region was also characterized by a disruption of the vorticity from the
suction side of the blade. Recall that at low speed (figures 3.2 (b) and 3.5 (b))
the wake vorticity magnitude was slightly asymmetric about the wake centerline
and the wake was well defined by a region of low velocity magnitude and high
vorticity magnitude. Here, the lower side of the wake associated with positive
vorticity could be identified easily and was similar to the M = 0.16 and M = 0.55
observations. The upper side of the wake associated with negative vorticity had a
larger spatial distribution at M = 0.75 than was observed at lower tunnel speeds.
The static pressure field at 0 = 8 and M = 0.75 is shown in figure 3.8 (c).
A comparison of figure 3.8 (c) to the M = 0.55 results (figure 3.5 (c)) revealed
that the favorable pressure gradient observed on the pressure side of the blade was
proportionally smaller at M = 0.75 than at M = 0.55. Furthermore, the static
pressure over the suction side near the blade leading edge had a larger gradient
at M = 0.75 than at M = 0.55. This indicated a proportionally larger flow
acceleration for transonic flow conditions than at subsonic conditions.
Two instantaneous fluid velocity fields for the blade at 0 = 8 and M = 0.75
are shown in figure 3.9. Here, the blade was translating in the negative z direction.
The instantaneous fields are included as a reference to provide insight into the
anomalous results. Note that both vector fields were acquired t 1/(2f 1 ) after
the impulse had been applied, and were less than 12 out of phase (with respect to
the bending mode) with each other. It can be seen that the region of supersonic
flow above the suction side of the blade was roughly similar in size and shape
to that observed in the phase averaged results. A gradual deceleration was also
observed near the downstream edge of the supersonic pocket, which implied a

80

shock-free deceleration. This indicated that the gradual flow deceleration observed
in the phase-averaged data was not an artifact of phase-averaging a moving shock,
but rather was a reflection of the instantaneous flow measurements.
Another noteworthy feature of the instantaneous flow fields is the presence of
pockets of both high and low momentum flow near the trailing edge and wake.
The pockets were transonic and bound by steep velocity gradients. The presence
of these pockets indicated that the flow was highly unsteady at these conditions
and may have had a significant out-of-plane velocity component.
The time-averaged fluid velocity and vorticity, and the RMS of the in-plane
velocity for 0 = 13 and M = 0.12 are shown in figure 3.10. The velocity
magnitude was normalized by the 41 m/s free-stream fluid velocity. The peak
velocity over the suction side surface was higher for 0 = 13 than the 0 = 8
data. There were some minor differences between the incompressible flow fields at
0 = 8 and 0 = 13 . Specifically, the increase in angle of attack resulted in an
increase of the velocity magnitude within the wake and a reduction in the velocity
gradients bounding the wake.
The phase-averaged fluid velocity magnitude, vorticity and static pressure for
0 = 13 and M = 0.12 are shown in figure 3.11. The peak velocity over the
suction side surface was higher for 0 = 13 than the 0 = 8 data. The phaseaveraged flow field was similar to the time-averaged field, indicating that the
unsteady flow induced by blade vibration was small.
The vorticity field was also qualitatively similar to that observed at 0 = 8
and M = 0.16. A slight decrease in the wake vorticity magnitude was observed at
0 = 13 when compared with the 0 = 8 data. The similarities between both
the velocity and vorticity fields suggest that the change in angle of attack had a

81

(a)
Velocity [m/s]
30

450

25

400

20
350
15
300

Y [mm]

10
5

250

0
200
5
150
10
100

15
20
30

20

10

0
X [mm]

10

20

30

(b)
Velocity [m/s]
30

450

25

400

20
350
15
300

Y [mm]

10
5

250

0
200
5
150
10
100

15
20
30

20

10

0
X [mm]

10

20

30

Figure 3.9. Instantaneous fluid velocity and blade velocity for 0 = 8


and M 0.75. Both images were acquired approximately 1/2 cycle after
the mechanical impulse was applied. fluid velocity; 50 blade
velocity; Sonic line.

minor influence on the flow field far from the blade surface.
The phase-averaged pressure field (figure 3.11 (c)) was also qualitatively similar
82

(a)
|U|/|U |
1.75
1.5

1.25

z [mm]

10

0.75

10

0.5

15

10

20

30

40

50

60

0.25

x [mm]

(b)
U /|U |
0.4
10
0.3

z [mm]

5
0

0.2

5
0.1
10
15

10

20

30

40

50

60

x [mm]

(c)
y 2b/|U |
8
6

10

z [mm]

0
2

4
10
15

6
0

10

20

30

40

50

60

x [mm]

Figure 3.10. Fluid velocity and vorticity at M = 0.12 and 0 = 13 . (a)


average velocity, (b) RMS velocity, (c) average fluid vorticity.

83

(a)
|U|/|U |
1.75

10

1.5

1.25

z [mm]

15

0.75

10

0.5

10

10

20

30

40

0.25

50

x [mm]

(b)
y 2b/|U |
8

15

10

z [mm]

4
5

0
2

4
10

10

10

20

30

40

50

x [mm]

(c)

z [mm]

Cp
15

10

0.5
0

0.5
0
1
5
1.5
10
10

10

20

30

40

50

x [mm]

Figure 3.11. Phase-averaged velocity, vorticity and coefficient of pressure


of fluid at M = 0.12 and 0 = 13 . The phase shown was at the first
maximum in bending vibration velocity after a impulse was applied.
fluid velocity; 2U blade velocity.

84

to the 0 = 8 results. Note that the suction side pressure distribution at 0 =


13 had a lower minimum pressure coefficient than was observed at 0 = 8 .
Additionally, a larger favorable pressure gradient ( P
> 0) was also observed over
x
the pressure side of the blade.

1.5

Cp

0.5

0.5

10

20

30

40

50

60

70

80

90

100

% chord

Figure 3.12. Phase-averaged surface pressure coefficient at 0 = 13 and


M = 0.12. Steady 2D panel method; from PIV data 1/2 cycle
after impulse; 3/4 cycle after impulse; 1 cycle after impulse; 1
1/4 cycle after impulse.

The surface pressure for the blade at 0 = 13 and M = 0.12 is shown in figure
3.12. The surface pressure estimate deviated significantly from the theoretical
profile obtained from the 2D panel method. First, the PIV results indicated a
higher favorable pressure gradient on the pressure side of the blade. Second,
85

the location of minimum pressure varied, with the panel method predicting a
minimum at 1% of chord and the PIV estimate indicating the minimum at 10%
chord. Third, the adverse pressure gradient near the trailing edge of the blade
was much steeper for the theory when compared to the measurements.
The fluctuations in surface pressure were similar in magnitude to those observed at 0 = 8 and M = 0.16, suggesting that the unsteady flow was small
relative to the mean flow. In addition, the same bias in pressure near the trailing
edge on the suction surface was observed. The chordwise extent of the bias was
similar between 0 = 8 and 0 = 13 .
The time-averaged fluid velocity and vorticity, as well as the RMS velocity
of the flow are presented in figure 3.13. The velocity magnitude was normalized
by the 245 m/s free-stream velocity, and a black isocontour is shown to denote
the region of supersonic flow. Substantial differences may be observed between
these results and the fields presented for M = 0.12 and 0 = 13 (figure 3.10).
However, these results were qualitatively similar to the observations at M = 0.75
and 0 = 8 (cf. figure 3.7). A shock-free deceleration of the flow over the
suction side was observed, and so these results should be viewed as a qualitative
description of the flow. The key changes that an increase in angle of attack had
on the flow field were concentrated in the wake region. Specifically, at 0 = 13
the wake thickness near the blade trailing edge was about half of the thickness
observed at 0 = 8 . The thin wake was also associated with an increase in
negative fluid vorticity.
The phase-averaged fluid velocity, vorticity and static pressure fields for the
blade at 0 = 13 and M = 0.75 are shown in figure 3.14. A gradual deceleration
from supersonic to subsonic flow was observed over the suction side of the blade;

86

(a)
|U|/|U |
1.75
1.5

1.25

z [mm]

10

0.75

10

0.5

15

10

20

30

40

50

60

0.25

x [mm]

(b)
U /|U |
0.4
10
0.3

z [mm]

5
0

0.2

5
0.1
10
15

10

20

30

40

50

60

x [mm]

(c)
y 2b/|U |
8
6

10

z [mm]

0
2

4
10
15

6
0

10

20

30

40

50

60

x [mm]

Figure 3.13. Velocity and vorticity at M = 0.75 and 0 = 13 . (a)


average velocity, (b) RMS velocity, (c) average fluid vorticity.

87

consistent with the transonic flow observed at 0 = 8 . In addition, a similar wake


structure was also observed for the phase-averaged and time-averaged results.
The following observations may be made from the pressure field shown in figure
3.14 (c). First, the pressure side of the blade had a strong favorable pressure
gradient that was consistent with the large flow acceleration from the leading
edge stagnation point. A large favorable pressure gradient was also observed on
the suction side near the leading edge which also was consistent with the observed
flow acceleration. A comparison of figures 3.8 (c) and 3.14 (c) revealed a similar
spatial distribution of pressure coefficient on the pressure side of the blade.
Two instantaneous fluid velocity fields for the blade at 0 = 13 are shown in
figure 3.15. Note that pockets of high momentum flow were observed in the wake
region. This was consistent with the observations for 0 = 8 and M = 0.75, and
suggested that the flows were similar.

3.4 Conclusions
The PIV results suggest the following characteristics regarding the flow field.
First, at 0 = 8 and for subsonic flow the flow at mid-span was attached over
most of the chord. The phase-averaged surface pressure was in reasonable agreement with the predicted pressure from a 2D panel method code. The unsteady
component of the flow field due to the mechanical impulse applied to the blade
was small relative to the mean flow. This suggested that the linearization of the
fluid dynamics about the steady flow was appropriate for this experiment. The
RMS velocity fluctuations increased with increasing Mach number. Increasing
the angle of attack from 0 = 8 to 0 = 13 at incompressible flow conditions
resulted in noticeable differences in the blade surface pressure distribution.

88

The flow field at transonic conditions was highly unsteady, and most of the
unsteadiness was concentrated in the wake and near the suction side of the blade.
Pockets of high and low momentum flow were observed in the instantaneous vector fields for both angle of attacks observed. These pockets were found in the
wake region and were clearly discernable from the surrounding flow by high velocity gradients. Furthermore, an anomalous supersonic shock-free deceleration
was observed over the suction side of the blade at both angle of attacks. This deceleration was observed for the time-averaged, phase-averaged and instantaneous
velocity fields.

89

(a)
|U|/|U |
1.75
1.5

1.25

z [mm]

10

0.75

10

0.5

15
0

10

20

30

40

50

60

0.25

x [mm]

(b)
y 2b/|U |
8
6

10

z [mm]

2
4

10

15
0

10

20

30

40

50

60

x [mm]

(c)

Cp
1
0.5

z [mm]

10

0.5

1.5
10
10

10

20

30

40

50

60

x [mm]

Figure 3.14. Phase-averaged velocity, vorticity and coefficient of pressure


of fluid at M = 0.75 and 0 = 13 . The phase shown was at the first
maximum in bending vibration velocity after a impulse was applied.
fluid velocity; 2U blade velocity.

90

(a)
Velocity [m/s]
500

25

450

20

400

15

350

10

300

250

200

150

10

100

15

50

Y [mm]

30

20
30

20

10

0
X [mm]

10

20

30

(b)

Y [mm]

Velocity [m/s]
30

500

25

450

20

400

15

350

10

300

250

200

150

10

100

15

50

20
30

20

10

0
X [mm]

10

20

30

Figure 3.15. Instantaneous fluid velocity and blade velocity for 0 = 13


and M 0.75. fluid velocity; 50 blade velocity.

91

CHAPTER 4
NATURAL RESPONSE OF AN ISOLATED COMPRESSOR BLADE IN
COMPRESSIBLE FLOW

4.1 Introduction
The sources of the unsteady aerodynamic loads acting on the blade can be
characterized as either external or self-induced. External sources, such as wakes
from stators upstream of the rotor blades, are well understood and their effects
can be predicted using fairly simple models. Self-induced sources are typically
characterized by a feedback between the flexible blade and the unsteady aerodynamics. As a result, these sources are more difficult to model and their effects are
not as well understood. The response of the blade due to these sources will be
termed the natural response. The fatigue life of the blades can depend on the
natural response. Therefore, it is important to characterize the natural response
and understand the unsteady forces associated with it.
The PIV results indicated that the steady flow about the blade at the subsonic
Mach numbers investigated could be modeled with reasonable accuracy using a
linear, two-dimensional aerodynamic theory. In addition, transonic conditions
resulted in a highly unsteady flow field. The current chapter will discuss the timeresolved blade natural response. The PIV measurements from chapter 3 will be
used to provide perspective for these measurements.

92

A review of a theoretical model describing the unsteady structural dynamics


will be presented. This will be followed by a description of the model used to
predict unsteady aerodynamic lift and moment. These models will be incorporated
into the analysis of blade vibration measurements presented in the results section.
Conclusions will then be provided to summarize the insights gained from these
experiments.

4.2 Theory
This section will discuss the theoretical models used to describe the aeroelastic behavior of the cantilevered blade in a high speed flow. A discussion of the
structural dynamics will first be provided, followed by a brief discussion of the
unsteady aerodynamic model used. The section will conclude with an analysis of
the fluid-dynamic equations of motion used to estimate the unsteady aerodynamic
loads from measurements of fluid velocity.

4.2.1 Structural dynamics


It will be assumed that the blade is composed of a linear-elastic solid and that
the dynamics can be approximated by a thin plate that deflects in the chordnormal direction, z. The unsteady deflection of the blade can be described by


2w
2w
D
+ 2
+ ...
x2
y


2
2w
+2(1 )
D
+ ...
xy
xy
  2

2
w
2w
+ 2 D
+ 2
+ Fz (x, y, t).
y
y 2
x

2w
s 2 =
t

2
x2

93

(4.1)

Here, s is the density of the blade, is Poissons ratio for the given blade material,
Fz (x, y, t) is the external force applied to the blade surface and D describes the
stiffness of the blade. The boundary conditions are zero displacement and strain
along the clamped edge at the root (y = ys ), and zero net force and bending
moment along the free edges of the blade. These can be expressed as

w
= 0 y = ys x,
y
2w
2w
+

=
x2
y 2





2w
2w
2w

D 2 + 2
D
= 0 x = 0, xc y,
x
x
y
y
xy
2w
2w
+ 2 =
y 2
x




2
2
2

w
w
w

D 2 + 2
D(x, y)
= 0 x = 0, xc y,
y
x
x
x
xy
w=

(4.2)

(4.3)

(4.4)
(4.5)

A solution of equation 4.1 can be obtained by expressing the displacement as a sum


of orthonormal mode-shapes, (x, y) multiplied by their respective amplitudes,
Zp (t) such that

w(x, y, t) =

P
X

p (x, y)Zp (t).

(4.6)

p=1

p (x, y) and Zp (t) can be estimated from the equation of motion assuming Fz = 0
using a separation of variables approach. The solution of p for a given set of
boundary conditions is typically performed numerically for all but the most simple
blade geometries. Once are known, equation 4.1 can be re-cast as an ODE for

94

Zp (t),

p s

d2 Zp
+ Kp Zp = 0
dt2

(4.7)

where Kp is a separation constant which depends on the blade stiffness and mode
shape p . Assuming solutions of the form Zp = Z p ei2fp t yields

Kp = (2fp )2 s p .

(4.8)

Note that fp is the natural frequency of the pth mode. Considerable simplification
of equation 4.1 can be obtained by taking advantage of the orthogonality of .
Specifically, using equations 4.6 and 4.8, multiplying by q and integrating over
the planform of the blade yields

Mp


d2 Zp
2
+ (2fp ) Zp = Fp
dt2

(4.9)

where
Z Z

2p s dxdy

(4.10)

Fz (x, y, t)p dxdy

(4.11)

Mp
A

is the modal mass, and


Z Z
Fp
A

is the modal force for the pth mode.


All structures have some amount of damping. The damping of the blade in
quiescent air at standard conditions will be referred to as intrinsic damping.
It will be assumed that the damping force is small compared with the mass and

95

stiffness of the blade, such that the changes in are negligible. This was a
reasonable simplification for the blade under investigation. The modal damping
will be modeled as a viscous damping force

Fz (x, y, t) C(x, y)

w
t

where C(x, y) describes the spatial distribution of the damping coefficient. Equation 4.9 is modified as

d2 Zp
dZp
+ Cp
+ Mp (2fp )2 Zp = Fp
2
dt
dt

 2
d Zp
dZp
2
Mp
+ 2(2fp )p
+ (2fp ) Zp = Fp
dt2
dt
Mp

(4.12)

where

Z Z

C(x, y)2p dxdy

Cp

(4.13)

1 Cp
4fp Mp

(4.14)

are the modal damping force and damping ratio respectively.

4.2.2 Unsteady thin airfoil theory


The modal force is derived from the aerodynamic forces acting on the blade.
In general, these aerodynamic forces are dependent upon the blade geometry and
motion, and so the equations describing the fluid and structural dynamics are
coupled. Solutions to these equations can be challenging for practical turboma-

96

chinery stage designs because the flow is usually three-dimensional, compressible


and unsteady. Approximate solutions, obtained from the linearized equations of
motion, are easier to implement and can offer insight into the physics associated
with the unsteady fluid loads.
The general aeroelastic problem will be simplified using a set of assumptions
consistent with thin airfoil theory. These assumptions are inviscid, irrotational,
isentropic flow. Additionally, it will be assumed that the fluctuating velocities and
pressures are much smaller than their mean values. The airfoil thickness will be
assumed to be negligibly small and the cross-section of the blade remains rigid.
Finally, it will be assumed that streamline curvature in the yaxis is negligible
such that an aerodynamic strip theory approximation can be used.
The assumptions allow for the estimation of the three-dimensional unsteady
aerodynamic forces by dividing the blade into a series of two-dimensional segments
along the span. The effect that the unsteady aerodynamic pressures have on the
structural vibration of the blade will be modeled as a chord-normal force (lift) and
twisting moment. In a similar manner, the eigen-mode shapes of the blade will
also be decomposed into either a chord-normal deflection (henceforth referred to
as bending) or twist about some axis (referred to as torsion). This decomposition
facilitated the use of a linearized unsteady thin airfoil theory, which expressed the
unsteady lift and moment as a function of bending and torsion type motion of the
blade.
The delineation between torsion and bending modes was based on the orientation of the nodal lines of the mode shape. Bending modes had nodal lines
aligned to the chord-wise direction (xaxis), whereas torsion modes had nodal
lines aligned to the span-wise direction. The delineation between the two can be

97

quantified by performing a linear curve fit of the mode shape along the xaxis at
constant span,

p (x, y) (y)(x a(y)) + h(y),

(4.15)

where a is the distance between the center of rotation and the mid-chord of the
blade.
Mode shapes where classified as bending when |h(y)| >> |(y)|, and as torsion
when |(y)| >> |h(y)|. The delineation was straightforward for the first two
eigen-modes of the blade which were shown in figure 2.5. Here, the first eigenmode shape was clearly a bending type mode, whereas the second eigen-mode was
a torsion mode. Note that the axis of rotation for the torsion mode varied along
the span.
The delineation between bending and torsion modes resulted in considerable
simplification of the equations of motion,


dZp (t)
dZp (t)2
2
+ 2(2f p )p
+ (2fp ) Zp (t) = Lp
Mp
dt2
dt


dZq (t)
dZq (t)2
2
Iq
+ 2(2f q )q
+ (2f q ) Zq (t) = Tp
dt2
dt


(4.16)
(4.17)

where Zp (t) is the amplitude of the bending motion of the pth mode, Zq (t) is the
amplitude of the q th torsion mode, Lp is the modal lift force, Tq is the modal
moment, and Iq describes the moment of inertia of the blade cross section with
respect to the torsion motion. The subscripts were included to keep the form of
the equations as general as possible. A solution to the system of equations can
now be obtained by calculating the unsteady lift and moment due to the motion

98

of the airfoil.
The estimate of the unsteady lift and moment acting on the blade was calculated using a closed form approximate solution for the unsteady response of
an airfoil in two-dimensional compressible flow. The formulation of the modal lift
and moment were taken from Lin and Iliff [45]. This formulation was derived from
a power-series type solution from the Laplace transformed Possio equation, first
presented by Balakrishnan [7]. The Possio equation relates how the unsteady surface pressure on a vibrating thin flat plate varies with the chord-wise distribution
of plate velocity in compressible flow. The flow is assumed to be attached over
the entire plate and leaves smoothly off the trailing edge (the Kutta-Joukowski
condition). Finally, it is assumed that flow unsteadiness vanishes far away from
the blade. The power series solution presented has an accuracy of M 2 log(M ) and
is considered to be accurate forM < 0.7 [45]. Note that the relations presented
here assume subsonic flow, and so the model was expected to be inaccurate at
transonic flow conditions.
The lift and moment coefficients due to bending and torsion vibration of a flat
plate are given by [45]

99

Lh (M, k) k + 2kC(k) + M log(M )


k4
3
2
2
+ 2k C(k) + 2k C(k) ,(4.18)
2

L (M, k) k a(y)k 2 + C(k) [2 + (1 2a(y))k]


 3


 
k
k4
1
2
2
+M log(M )
a(y)C(k) 2k +
2a(y) k 3
2
2
2

+ +C(k)2 2k + [1 2a(y)] k 2 ,
(4.19)
Th (M, k) a(y)k 2 + (1 + 2a(y)) kC(k)

+M 2 log(M ) (1 + 2a(y)) k 2 C(k)2



k4
1
3
+ 2a(y) k C(k) + a(y) ,
+
2
2




1
1
2
T (M, k)
a(y)
+ a(y) k 2
k
2
8


 
1
2
2a(y) k
+C(k) 1 + 2a(y) +
2

a(y) 3 a(y)2 4
2
k
k
+M log(M )
2
2
 


1
2
2
2a(y) k 2
+C(k) (1 + 2a(y)) k +
2



1
2
2 3
2a(y) k 2a(y) k
+ C(k)
.
2

(4.20)

(4.21)

(4.22)

Here, k is the complex reduced frequency (k icf /U ) and C(k) is the Theodorsen
function, given by

C(k)

K1 (k)
,
K1 (k) + K0 (k)

(4.23)

where Kn are modified bessel functions of the second kind and of order n. Note
that the dependence of a on span was omitted for clarity. The lift and moment
coefficients are related to the lift and moment by

100

"

#
()
h(y)

p
2

b Lh (M, k)
+ L (y, M, k)(y) q () , (4.24)
L()
= U
b
"
#
()
h(y)

p
2
Th (M, k)
+ T (M, k)(y) q () .
(4.25)
T(y, ) = U
b
Here, is the complex frequency of vibration ( i2f ), b is the half-chord of the
denotes the Laplace transform
airfoil, k is the complex reduced frequency and []
of the argument with respect to time.
Note that the vibration amplitude can be factored out of both the aerodynamic forcing and structural dynamics equations. Consequently, this formulation
can only yield estimates of the stability of blade vibration at a given set of flow
conditions.
The following approach was used to provide a theoretical benchmark against
which the natural response of the blade may be compared. The measured modal
amplitudes of the blade were used with equations 4.24 and 4.25 to obtain the expected modal forces. These modal forces were used with equations 4.16 and 4.17
to predict the expected vibration amplitude. The damping for each mode was
incorporated into the theoretical model using the impulse response in quiescent
air (cf. figure 2.3). A discussion of the approach used to estimate the damping
ratio is provided in chapter 5.
A comparison of the measured amplitude and the prediction described above
can provide insight into the fluid-structure interactions. For example, if the theoretical and measured modal velocities are equal in magnitude, it may be concluded
that the model of the aerodynamic forces describe the unsteady fluid-structure interaction well. If the theory over-predicts the measured velocity, it may indicate

101

that additional damping was present. In contrast, if the theory under-predicts the
measured velocity, it may be concluded that additional forcing was present.

4.3 Results
The autospectral density of the blade surface velocity for 0 = 8 is shown
in figure 4.1. A minimum of 30 averages per Mach number with a Mach number
resolution of 0.025 and a frequency resolution of 12.5 Hz were used to generate
these plots.
Most of the observed vibration were due to excitation of the first two modes.
Multiple distinct peaks in the spectra were observed. These peaks correspond to
the natural modes of vibration (see Appendix A), suggesting that some degree
of excitation of these high frequency modes was present. At high tunnel speeds
(0.8 M 0.95) the magnitude of vibration for the first two modes were 4
orders of magnitude larger than the other high frequency modes. The broad-band
vibration not associated with a blade mode was also small over a large range
of tunnel speeds. It can be observed that at frequencies greater than 2kHz the
broadband response increased with increasing Mach number. Note that the broadband response was at least 6 orders of magnitude smaller than the first two blade
modes. Consequently, the natural response of the blade at this angle of attack
could be approximated by considering only the first two modes of vibration.
Figure 4.1 (b) shows the autospectra of blade vibration velocity at frequencies
less than 2, 000 Hz. There are a number of noteworthy features. The peaks
associated with the first two vibration modes had a slight increase in frequency
with Mach number. This was due to the steady aerodynamic loads generated by
the blade. The stress induced by the generation of lift and moment had the effect

102

(a)
Log () [m2/s2/Hz]
10

2
3

0.9

0.8

Mach number

0.7
6

0.6

0.5

0.4

0.3

10
11

0.2

12

0.1
13

5000
10000
Frequency [Hz]

15000

(b)
Log10() [m2/s2/Hz]

2
3

0.9

0.8

Mach number

0.7
6

0.6

0.5

0.4

0.3

10
11

0.2

12

0.1
13

500

1000
1500
Frequency [Hz]

2000

Figure 4.1. Autospectral density of blade surface velocity as a function


of nozzle exit Mach number at 0 = 8 . (a) Spectra for frequencies up to
15 kHz; (b) Spectra for frequencies up to 2 kHz.

of increasing the modal stiffness of the blade, which resulted in an increase in the
natural frequency. The increase in natural frequency was approximately 6.5% for

103

mode 1 and 2% for mode 2.

Log10() [m2 /s2 /Hz]

9
0

500

1000

1500

2000

2500

3000

Frequency [Hz]

Figure 4.2. Autospectral density of blade surface velocity at 0 = 8 .


M = 0.57; M = 0.74; M = 0.80; M = 0.95.

The spectra also show that the frequency of blade vibrations was quantitatively
different at tunnel speeds near M = 0.8. Specifically, excitation near 840 and 1,200
Hz (corresponding to ? = 0.35 and 0.5 respectively) was observed only around
this region; increasing or decreasing the tunnel speed eliminates this excitation.
Figure 4.2 shows the autospectral density of the blade vibration as a function
of frequency for four Mach numbers. It can be seen that the local maxima at
M = 0.80 were relatively broad and were only present and were only observed
between 0.74 M 0.90. Note that the critical Mach number for this angle of

104

attack was 0.62, as calculated from the two-dimensional panel method described
in chapter 3. Thus, these vibrations occurred when the flow was transonic.

0.3

[m/s]

0.25

0.2

0.15

0.1

0.05

0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Mach number

Figure 4.3. RMS modal velocity as a function of nozzle exit Mach


number at0 = 8 . Mode 1 velocity; Theoretical mode 1 velocity;
Ensemble-averaged mode 1 velocity; mode 2 velocity; Theoretical
velocity for mode 2; Ensemble-averaged mode 2 velocity.

The RMS modal velocity for the first two modes as a function of Mach number are shown in figure 4.3. The RMS velocity was calculated as follows. The
measured blade vibration velocity was band-pass filtered to isolate the modes of
interest. A conversion factor was applied to account for the mode shape at the
measurement location. Then, the RMS was calculated over a 0.68 second time

105

segment. The segment duration represented approximately steady state conditions of the blowdown tunnel while still allowing for convergence of the statistics
of the blade vibration. The resulting collection of RMS amplitudes were then
averaged based on Mach number using the same averaging procedure as was used
to estimate the autospectral density.
There are several noteworthy features shown in figure 4.3. The following may
be observed for the variation of the RMS velocity for the first mode (bending). The
average RMS velocity increased with Mach number up to M 0.7. The RMS
velocity was nearly independent with respect to Mach number between 0.70
M > 0.90. There exist a narrow band of Mach numbers (0.90 < M < 1) where
the RMS velocity was at a maximum. The local maxima had a corresponding
reduced frequency of ? 0.134. Additionally, the scatter in the individual RMS
was also largest at these conditions which implied that the vibration was very
unsteady.
The solid lines shown in figure 4.3 show the theoretical benchmark of blade
vibration using the measured RMS modal velocity. The theoretical model underpredicted the measured modal velocity at low Mach number for the first mode.
This suggested that additional fluid forcing was present. In contrast, the theoretical model over-predicted the modal velocity for the first mode at high subsonic
conditions, suggesting that there was an additional source of damping influencing
the system which was not accounted for.
The second mode (torsion) RMS modal velocity also increased with increasing
Mach number until M 0.75. A local maxima was observed at M 0.75 with a
corresponding reduced frequency of ? = 0.752. A gradual decrease in the RMS
modal velocity with increasing Mach number was observed at M > 0.75. The local

106

maxima for the torsion mode also had high scatter in the individual observations.
The theoretical model under-predicted the modal velocity over the entire range
of the models validity (M 0.7). This indicated that there may have been an
additional source of excitation for the torsion mode.
Figure 4.4 shows the autospectral density of the blade surface vibration as a
function of both Mach number and frequency at an angle of attack of 13 . Note
that at this incidence angle, the critical Mach number based on a steady 2D panel
method was found to be M 0.50. Although the spectra was qualitatively similar
to that observed for the 8 angle of attack, there are several important differences.
First, the high frequency broadband vibration was reduced over most of the Mach
numbers investigated. The third and fourth eigen-modes (with frequencies of 1740
Hz and 3800 Hz, see appendix 1) had a higher vibration amplitude when the blade
angle of attack was increased to 13 . The frequency shift for the first two eigenmodes was not as large as was observed for the 0 = 8 results. The shift in
natural frequency was 0.5% for the first mode and 3% for the second mode. The
decrease in frequency shift could be attributed to a smaller variation in mean
aerodynamic loading as a function of Mach number. This could be attributed to
the presence of flow separation at high angle of attack, which would effectively
reduce the steady aerodynamic lift and moment.
A few autospectral density curves of the measured blade surface velocity are
shown in figure 4.5 for 0 = 13 . Note the presence of two peaks near 840
Hz and 1200 Hz when M = 0.74 and 0.80 respectively. An increase in angle
of attack resulted in an increase of the magnitude of the peaks and reduction in
their subsequent width, though the effect was most noticeable for the 840 Hz peak.
However, more important was the fact that the amplitude, frequency and Mach

107

(a)
Log10() [m2/s2/Hz]

2
3

0.9

0.8

Mach number

0.7
6

0.6

0.5

0.4

0.3

10
11

0.2

12

0.1
13

5000
10000
Frequency [Hz]

15000

(b)
Log10() [m2/s2/Hz]

2
3

0.9

0.8

Mach number

0.7
6

0.6

0.5

0.4

0.3

10
11

0.2

12

0.1
13

500

1000
Frequency [Hz]

1500

2000

Figure 4.4. Autospectral density of blade surface velocity as a function


of nozzle exit Mach number at 0 = 13 . (a) Spectra for frequencies up
to 15 kHz; (b) Spectra for frequencies up to 2 kHz.

number at which these peaks occurred was nearly identical to the low incidence
case. The insensitivity to angle of attack indicated that the mechanism responsible
for these peaks were not significantly influenced by the steady flow field induced

108

Log10 () [m2 /s2 /Hz]

3
4
5
6
7
8
9
0

500

1000

1500

2000

2500

3000

Frequency [Hz]

Figure 4.5. Autospectral density of blade surface velocity at 0 = 13 .


M = 0.57; M = 0.74; M = 0.80; M = 0.95.

by the blade. The narrow Mach number range over which these peaks occurred
suggests that the source may be associated with the test facility rather than the
blade. Note that the peaks for the two humps were 2 orders of magnitude smaller
than the first two modes, and so represent a minor contribution to the overall
blade vibration.
The RMS amplitude of vibration for the first two modes as a function of Mach
number are shown in figure 4.6 for an angle of attack of 0 = 13 . Also included
are the theoretical benchmarks for the RMS modal velocity. The bending modal
velocity was similar in magnitude between the two angle of attacks over most
of the tunnel speed range. However, no distinct local maxima was observed at
M = 0.95 for 0 = 13 . It may be concluded that the bending response of the
blade at subsonic flow conditions was not very sensitive to changes in 0 between

109

0 = 8 and 13 . In contrast, the response was very sensitive to changes in 0


between 8 and 13 for transonic conditions.
The theoretical model previously developed over-predicted the high incidence
bending response beyond M = 0.40. Recall that a similar over-prediction was
observed at 0 = 8 at tunnel speeds beyond M 0.60. Thus, similar to the low
incidence results it can be concluded that an additional source of damping was
present at 0 = 13 and M > 0.40.
The RMS modal velocity for the torsion mode at 0 = 13 was significantly
different from that observed at 0 = 8 . Whereas the RMS modal velocity below
M = 0.50 was similar between the two incidence angles, two local maxima were
observed at M = 0.65 and 0.80 at 0 = 13 compared with a single maxima at
M 0.75 and 0 = 8 . The local maxima at 0 = 13 were discernable due to
their large amplitudes. In contrast, the increase in RMS modal velocity at the
local maxima for 0 = 8 and M = 0.75 was not as large. The reduced frequency
of the two local maxima at 0 = 13 were ? = 0.798 and 0.648. Additionally,
as the Mach number was increased beyond M = 0.90, the modal velocity was
observed to increase with Mach number. Finally, it should be noted that the
overall vibration amplitude increased when the blade angle of attack increased.
The scatter in the individual RMS amplitudes shown in figures 4.3 and 4.6
indicate that the vibration amplitude was very unsteady at high tunnel speeds.
An analysis of the time-resolved modal amplitudes can yield additional insight
into the source of the large scatter in the RMS amplitude estimates.

110

0.3

[m/s]

0.25

0.2

0.15

0.1

0.05

0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Mach number

Figure 4.6. RMS modal velocity as a function of nozzle exit Mach


number at 0 = 13 . Mode 1 velocity; Theoretical mode 1 velocity;
Ensemble-averaged mode 1 velocity; mode 2 velocity;
Theoretical velocity for mode 2; Ensemble-averaged mode 2
velocity.

4.3.1 Time-series of blade natural response


Figures 4.7 - 4.13 show time segments of the modal velocity at four Mach
numbers and at 0 = 8 and 13 . The modal velocities were calculated by filtering
the measured surface velocity using a Meyer wavelet filter and then applying a
correction based on the mode-shape. The Mach numbers were selected to coincide
with the local maxima in the RMS modal velocities. Only the first two blade
eigen-modes are shown because they represented a majority of the observed blade
vibration.
A time-series of the modal amplitudes at M = 0.65 are shown in in figure
4.7. This tunnel speed corresponded with a local maxima in the torsion mode

111

for 0 = 13 . The time-resolved modal velocity for 0 = 8 can be described


as follows. The bending response was composed of long periods of near-constant
vibration amplitude separated by brief periods of nearly zero vibration amplitude.
This vibration pattern will be referred to as a burst. A similar burst-type
pattern was also observed for the torsion response.

(a)

(b)
1

1 [m/s]

1 [m/s]

1
0
1
0

100

200

300

400

1
0

500

100

200

300

400

500

100

200

300

400

500

2 [m/s]

2 [m/s]

1
0
1
0

100

200

300

400

0
1
0

500

Time [ms]

Time [ms]

Figure 4.7. Time series of instantaneous modal velocity at M = 0.65.


(a) 0 = 8 ; (b) 0 = 13 .

The time-series of modal velocities for the first two modes at M = 0.65 and
0 = 13 are shown in figure 4.7 (b). The variation in bending modal velocity was
consistent with the response observed at 0 = 8 . However, the torsion modal
velocity was much larger in magnitude at 0 = 13 when compared to the 0 = 8
results.
The variation in the envelope of the modal velocity for both modes was also

112

investigated. This was done to determine if there were any significant patterns
of the bursts of vibration observed in the time-series modal velocity data. The
envelope of the modal velocity, A(t), was defined as
h

2 i1/2
A(t) 2p (t) + H p (t)
,

(4.26)

where H is the Hilbert transform of the argument [10].


Figure 4.8 shows the auto and cross-correlation coefficients of the fluctuations
in the envelope of the modal velocity for both bending and torsion at M = 0.65
and at both angles of attack investigated. Note that the time lags are normalized
by C/U .
The autocorrelation of the fluctuations in modal velocity envelope for the first
two vibration modes at 0 = 8 (shown in figure 4.8) had significant correlation at
long non-dimensional time. Furthermore, the local maxima of the autocorrelation
function for both modes occurred at somewhat regular intervals. The characteristic time scales over which the local maxima occurred was f1? 12.5 Hz for the
first mode and f2? 10.3 Hz for the second mode. The first zero crossing, denoted by t?n , represented a characteristic de-correlation time scale for the envelope
fluctuations. This time scale was t?1 160 for bending and t?2 125 for torsion.
A cross-correlation of the bending and torsion modal velocity envelope fluctuations is also shown in figure 4.8 (a). It can be seen that the fluctuations were
correlated at 0 = 8 and M = 0.65. Furthermore, moderate correlation was observed at very long time lags (t? > 500). The cross-correlation had repeated local
maxima, and the characteristic time scales of these maxima were 10.3 and 12.5
Hz. Recall that the bending and torsion modes shapes are orthogonal and their
natural frequencies differ by 1,200 Hz. Consequently, modal coupling through the

113

(a)
1

Correlation Coefficient

0.75

0.5

0.25

0.25

0.5
4000

3000

2000

1000

1000

2000

3000

4000

2000

3000

4000

t? = t0 U /C

(b)
1

Correlation Coefficient

0.75

0.5

0.25

0.25

0.5
4000

3000

2000

1000

1000

t? = t0 U /C

Figure 4.8. Auto/cross correlation of instantaneous modal envelope at


M = 0.65. (a) 0 = 8 ; (b) 0 = 13 ; Mode 1; Mode 2;
Cross-correlation between modes 1 and 2.

blade structure itself was very unlikely. Therefore, the cross-correlations observed
may indicate coupling through the unsteady fluid-dynamics interacting with the
blade.
The auto- and cross-correlation coefficients for the modal velocity envelope

114

fluctuations at 0 = 13 and M = 0.65 are also shown in figures 4.8 (a) and
(b) respectively. An increase in the characteristic de-correlation time scales was
observed, with t?1 335 and t?2 220 respectively. The characteristic time-scales
over which the local maxima in the autocorrelation coefficients occurred were 3.7
Hz and 10.4 Hz for the first and second modes respectively. Significant correlation was also observed in the cross-correlation between the envelope fluctuations
for the first and second modes at t? > 2, 000.

(a)

(b)
1

1 [m/s]

1 [m/s]

1
0
1
0

100

200

300

400

1
0

500

100

200

300

400

500

100

200

300

400

500

2 [m/s]

2 [m/s]

1
0
1
0

100

200

300

400

0
1
0

500

Time [ms]

Time [ms]

Figure 4.9. Time series of modal velocity for the first two modes at
M = 0.75. (a) 0 = 8 ; (b) 0 = 13 .

A time-series of the modal velocity for both bending and torsion modes at
M = 0.75 is shown in figure 4.9. The modal velocity for the bending mode for
both angles of attack investigated appear qualitatively similar to the response
observed at M = 0.65. Recall that for both 0 = 8 and 13 the RMS modal

115

velocity for bending had little variation between 0.65 M 0.75. Figure 4.9
indicated that the time-resolved bending modal velocity was also similar over this
speed range.
The torsion modal velocity at M = 0.75 and 0 = 8 was also similar to the
response observed at M = 0.65. A local maxima of the torsion response was
observed at M = 0.75. The similarities in the dynamic response of the torsion
mode between these Mach numbers implied a similarity in the dynamics of the
aerodynamic forcing.
Significant differences were observed for the torsion modal velocity at M = 0.75
and 0 = 13 when compared against the M = 0.65 and 0 = 13 observations.
Specifically, the long duration, high amplitude bursts of vibration were eliminated
at M = 0.75. The vibration could be described as combination of steady, low
amplitude vibration with many short-duration, high amplitude pulses. Recall
that at M = 0.75 and 0 = 13 , the RMS torsion modal velocity was between local
maximas, and so it can be concluded that the reduction in RMS modal velocity
was due to an increase in vibration intermittency rather than a decrease in peak
vibration amplitude.
The auto- and cross-correlation coefficients for the modal velocity envelope
fluctuations are shown in figure 4.10. The variation in correlation coefficient with
time-lag was similar to the results presented for M = 0.65 at 0 = 8 . Furthermore, moderate cross-correlation was observed between both modes over a
long time-scale. Note however that there were some noticeable differences in the
observed autocorrelation for the blade response when the tunnel speed was increased from M = 0.65 to M = 0.75. A decrease in t?2 was observed at M = 0.75.
Specifically, t?2 = 122 at M = 0.75, compared with t?2 = 166 at M = 0.65.

116

(a)
1

Correlation Coefficient

0.75

0.5

0.25

0.25

0.5
4000

3000

2000

1000

1000

2000

3000

4000

2000

3000

4000

t? = t0 U /C

(b)
1

Correlation Coefficient

0.75

0.5

0.25

0.25

0.5
4000

3000

2000

1000

1000

t? = t0 U /C

Figure 4.10. Auto/cross correlation of fluctuating modal velocity


envelope at M = 0.75. (a) 0 = 8 ; (b) 0 = 13 ; Mode 1; Mode 2;
Cross-correlation between modes 1 and 2.

The effect of increasing the tunnel speed from M = 0.65 to M = 0.75 also had
an effect on the auto- and cross-correlation coefficients at 0 = 13 . Specifically,
lower correlation coefficients were observed for both auto- and cross-correlation
and the coefficients fluctuated with the characteristic time lag over a much shorter

117

period. This implied that the fluctuations in the envelopes were probably random.

(a)

(b)
1

1 [m/s]

1 [m/s]

1
0
1
0

100

200

300

400

1
0

500

100

200

300

400

500

100

200

300

400

500

2 [m/s]

2 [m/s]

0
1
0

100

200

300

400

0
1
0

500

Time [ms]

Time [ms]

Figure 4.11. Time series of modal velocity for first two modes at
M = 0.81. (a) 0 = 8 ; (b) 0 = 13 .

A time-series of the blade modal velocity for the first two modes at M = 0.81
is shown in figure 4.11. The following may be observed for the response at 0 =
8 . The bending response was qualitatively similar to the responses observed at
M = 0.75 and M = 0.65. The torsion response was composed of both burst and
pulse type patterns. Note that the peak amplitude of the pulses for the torsion
response were similar between both observations at M = 0.81 and M = 0.75.
The response of the bending mode at 0 = 13 was also qualitatively similar
to the response observed at slower tunnel speeds at the same angle of attack.
There were significant differences in the response of the torsion mode between
M = 0.81 and M = 0.75. The pulses in modal velocity amplitude were much

118

larger at M = 0.81 than at M = 0.75. A low-amplitude burst type pattern was


also observed.
It was shown in figure 4.6 that local maxima in the torsion mode occurred at
M = 0.65 and M = 0.81. A comparison of figures 4.7 (b) and 4.11 (b) indicated
that the torsion mode response at these two points had significantly different
dynamic behavior. Specifically, the time-scales over which the modal velocity
envelope fluctuated were much longer at M = 0.65 compared to M = 0.81.
The auto- and cross-correlations of the modal velocity envelope fluctuations
at M = 0.81 are shown in figure 4.12. The following may be observed for 0 = 8
(figure 4.12 (a)). The characteristic de-correlation time scale for the bending mode
increased at M = 0.81 to t?1 = 360 . In contrast, a decrease in the time-scale for the
torsion mode was observed when the tunnel speed was increased from M = 0.65
to M = 0.81.
The auto- and cross-correlation coefficients for the modal envelope fluctuations
at 0 = 13 are shown in figure 4.12 (b). Qualitatively, the variation in autocorrelation for the bending and torsion modes at M = 0.81 and 0 = 13 were similar
to those observed at M = 0.75 and 0 = 13 . A reduction was observed for t?2
when the tunnel speed was increased from M = 0.75 to M = 0.81. In contrast, t?1
increased as tunnel speed was increased from M = 0.75 to M = 0.81.
The cross-correlation between bending and torsion modal amplitude fluctuations at M = 0.81 and 0 = 13 indicated moderate correlation between the two
modes at time-scales of t? 1000. There was a region of time lag between
1400 t? 800 where the cross-correlation coefficient was nearly constant
and of moderate amplitude.
The modal velocity of the blade at M = 0.95 is shown in figure 4.13. The

119

(a)
1

Correlation Coefficient

0.75

0.5

0.25

0.25

0.5
4000

3000

2000

1000

1000

2000

3000

4000

2000

3000

4000

t? = t0 U /C

(b)
1

Correlation Coefficient

0.75

0.5

0.25

0.25

0.5
4000

3000

2000

1000

1000

t? = t0 U /C

Figure 4.12. Auto/cross correlation of fluctuating modal velocity


envelope at M = 0.81. (a) 0 = 8 ; (b) 0 = 13 , autocorrelation of
mode 1; autocorrelation of mode 2; cross-correlation of modes 1
and 2.

RMS modal velocity at this Mach number had a local maxima at 0 = 8 . The
bending mode response at 0 = 8 (figure 4.13 (a)) can be characterized by long
bursts of vibration. The torsion response could be characterized by a series of

120

(a)

(b)
1

1 [m/s]

1 [m/s]

1
0
1
0

100

200

300

400

1
0

500

100

200

300

400

500

100

200

300

400

500

2 [m/s]

2 [m/s]

1
0
1
0

100

200

300

400

0
1
0

500

Time [ms]

Time [ms]

Figure 4.13. Time series of modal velocity at M = 0.95. (a) 0 = 8 ; (b)


0 = 13 .

pulses superposed onto low amplitude bursts of vibration. These results suggest
that the local maxima in RMS bending modal velocity could be attributed to
a reduction in the intermittency of the vibration and an increase in the peak
vibration amplitude. In contrast, the reduction in RMS modal velocity for the
torsion mode could be attributed to an decrease in the peak vibration amplitude.
The bending modal velocity at 0 = 13 and M = 0.95 (figure 4.13 (b))
was qualitatively similar to the response observed at 0 = 13 and M = 0.81.
Specifically, at both Mach numbers the bending vibration could be described as a
series of low amplitude pulses.
A large reduction in the peak amplitude was observed for the torsion mode
when the tunnel speed was increased from M = 0.81 to M = 0.95 at 0 = 13 .
This was also accompanied by a decrease in the vibration intermittency. Note
that the RMS modal velocity for the torsion mode was nearly constant between
0.81 M 0.95 at 0 = 13 . This was interesting because these two different
vibration patterns resulted in similar RMS vibration levels.
121

(a)
1

Correlation Coefficient

0.75

0.5

0.25

0.25

0.5
4000

3000

2000

1000

1000

2000

3000

4000

2000

3000

4000

t? = t0 U /C

(b)
1

Correlation Coefficient

0.75

0.5

0.25

0.25

0.5
4000

3000

2000

1000

1000

t? = t0 U /C

Figure 4.14. Auto/cross correlation of fluctuating modal velocity


envelope at M = 0.95. (a) 0 = 8 ; (b) 0 = 13 , autocorrelation of
mode 1; autocorrelation of mode 2; cross-correlation of modes 1
and 2.

The auto- and cross-correlation coefficients for modal velocity envelope fluctuations of the first two modes of the blade at M = 0.95 are shown in figure 4.14.
The average t? between successive zero crossings of the autocorrelation coefficients

122

were much larger at M = 0.95 than at any of the slower tunnel speeds presented
previously. This feature was observed for both modes, but was most noticeable
for bending. A similar pattern was also observed for the cross-correlation of both
modes. Finally, the characteristic de-correlation time for both modes were also
very large at t?1 = 970 and t?2 = 356.
The auto- and cross-correlation coefficients for 0 = 13 and M = 0.95 are
shown in figure 4.14 (b). The effect of increasing the tunnel speed from M = 0.81
to M = 0.95 effectively reduced the average t? between zero crossings for the
autocorrelation coefficients. This was most pronounced for the bending mode.
The local maxima in the autocorrelation coefficients at large t? was smaller at
M = 0.95 when compared to the M = 0.81 results.

TABLE 4.1
NON-DIMENSIONAL CHARACTERISTICS OF THE
FLUCTUATIONS IN THE MODAL VELOCITY AMPLITUDES OF
THE BLADE AT 0 = 8 .

Mach number

t?1

t?2

U
f 1c

U
f 2c

0.65

160 125

17.5

3.9

0.75

122 131

20.3

4.5

0.81

360 123

21.9

4.9

0.95

970 356

25.7

5.7

123

TABLE 4.2
NON-DIMENSIONAL CHARACTERISTICS OF THE
FLUCTUATIONS IN THE MODAL VELOCITY AMPLITUDES OF
THE BLADE AT 0 = 13 .

Mach number

t?1

t?2

U
f 1c

U
f 2c

0.65

335 220

17.5

3.9

0.75

282 173

20.3

4.5

0.81

229

95

21.9

4.9

0.95

131

79

25.7

5.7

Tables 4.1 and 4.2 present the measured non-dimensional de-correlation time
scales (t?1 , t?2 ) and the reduced period (1/ ? ) for the first two blade vibration modes.
It may be observed that t?1 tended to increase with increasing Mach number at
0 = 8 . The opposite trend was observed when the angle of attack was increased
to 0 = 13 . There was little variation in t?2 with Mach number until M = 0.95
when the blade angle of attack was 0 = 8 . An increase in angle of attack to
0 = 13 resulted in a decrease in t?2 with increasing M .
It is important to note the magnitude of the non-dimensional period of vibration compared to the cross-correlation results. The four tunnel speeds investigated
had moderate cross-correlation between the two modes at |t? | > 500. In contrast,
the longest characteristic period for the bending mode was an order of magnitude
smaller. The implication is that fluctuations in the vibration amplitude may have

124

been influencing the system long after they occurred.

4.4 Conclusions
The response of the blade over a large range of Mach numbers and at two angle
of attacks was investigated. The response was due primarily to vibration of the
blade at the first two modes. An increase in the angle of attack from 0 = 8 to
0 = 13 resulted in significant changes in the blade response over 0.5 M < 1.
Specifically, excitation of the third and fourth modes were observed at 0 = 13
but were not at 0 = 8 . Additionally, the variation in RMS modal amplitude for
the bending and torsion modes with Mach number was also sensitive to 0 .
The response of the blade in first (bending) mode at 0 = 8 was characterized
by the gradual increase in vibration amplitude with increasing Mach number up to
M = 0.90. A similar increase in the RMS modal velocity with Mach number was
also observed for the second (torsion) mode up to M = 0.60. A local maxima in
the RMS modal velocity was observed at M = 0.75 for first torsion. In addition,
a local maxima for first bending was observed at M = 0.95. An increase in angle
of attack resulted in two local maxima for the torsion mode at M = 0.65 and
M = 0.80. Additionally, the response in bending at 0 = 13 revealed a gradual
increase in RMS modal velocity without a similar local maxima at M = 0.95. All
the local maxima for the bending and torsion modal velocities occurred during
conditions of transonic flow over the blade.
The modal vibration velocity revealed that the response of the blade was highly
unsteady. The envelope of the modal velocity fluctuated, and these fluctuations
were influenced by Mach number and angle of attack. The local maxima observed
for the bending mode at 0 = 8 was associated with a long duration burst-type

125

vibration pattern, which could be described as a long period of high amplitude


vibration interspersed among short periods of low amplitude vibration.
The local maxima of the torsion vibration at 0 = 13 at M = 0.65 was
associated with a burst-type vibration pattern. The second local maxima at M =
0.81 could be characterized by a short duration pulse-type vibration pattern, which
can be described as a short duration period of high vibration amplitude that
occurred at regular intervals.
The auto- and cross correlation coefficients of the fluctuations in modal velocity
envelope between the two modes revealed that the fluctuations were correlated over
long time scales. Many of the flow speeds that were investigated had a pattern of
local maxima in the autocorrelation which occurred at somewhat regular intervals
and over long time scales. The peak correlation coefficients associated with these
maxima were between 0.2 and 0.5.
The characteristic de-correlation time scales, defined as the first zero crossing,
were sensitive to both 0 and M . The time scale for bending increased with
increasing Mach number at 0 = 8 , but decreased with increasing Mach number
at 0 = 13 . Additionally, the de-correlation time scale for the torsion mode was
nearly independent of the Mach number at 0 = 8 , but decreased with increasing
Mach number at 0 = 13 .
Finally, the envelope fluctuations between both modes were correlated. The
non-dimensional time lag at which these fluctuations were correlated was at least
an order of magnitude larger than the period of vibration for either mode. This
implied that there was an effect that influenced blade vibration which acted over
very long time scales.

126

CHAPTER 5
IMPULSE RESPONSE OF AN ISOLATED COMPRESSOR BLADE IN
COMPRESSIBLE FLOW

5.1 Introduction
The objective of this chapter was to explore the response of the cantilevered
blade to a mechanical impulse at different flow conditions. The response of the
blade to a mechanical impulse can depend on the structural properties of the blade,
the fixture on which it is mounted, and the unsteady aerodynamic forces that occur
as a result of blade vibration. In typical axial turbomachinery the combined effect
of all these interactions can determine the vibration characteristics of the blades.
The blades response to the flow was a result of the aerodynamic forces acting
on the blade. These forces can be separated based on their phase relationship with
the blade. Components of the aerodynamic forcing which were in-phase with the
blade vibration will be referred to as aerodynamic excitation, whereas out-of-phase
components are referred to as aerodynamic damping. Aerodynamic damping can
be an important factor in both the natural response and the transient response of
turbomachine blades. The objective of this set of experiments was to investigate
the effective damping of the cantilevered blade

127

5.2 Theory
5.2.1 Model of aerodynamic damping
The aerodynamic damping was defined as the component of aerodynamic forcing that was out-of-phase with the blade vibration displacement. The equations
governing the response of the blade for the bending and torsion modes were presented in chapter 4. They are repeated here for clarity


dZ1
dZ12
2
+ 2(2f 1 )1
+ (2f 1 ) Z1 = L1
M1
dt2
dt
 2

dZ2
dZ2
2
I2
+ 2(2f 2 )2
+ (2f 2 ) Z2 = T2 .
dt2
dt


(5.1)
(5.2)

Note that the subscripts refer to the eigen-modes of the blade. The factor
and

dZ2
dt

dZ1
dt

can be factored out of L1 and T2 respectively. Separating the real and

imaginary components of lift and moment and rearranging terms yields




dZ12
1
dZ1
2
M1
+ 2(2f 1 ) 1 = [L1 ]
+ (2f 1 ) Z1 = < [L1 ] (5.3)
dt2
M1 4f 1 dt



 2
dZ2
1
dZ2
2
I2
+ 2(2f 2 ) 2 = [T2 ]
+ (2f 2 ) Z2 = < [T2 ] , (5.4)
dt2
I2 4f 2 dt


where the aerodynamic contribution to damping can be expressed as

=[L1 ]
,
M1 4f 1
=[T2 ]

.
I2 4f 2

A

1

(5.5)

A
2

(5.6)

128

5.2.2 Estimation of the damping ratio from measurements


The response of the blade to a mechanical impulse can depend on many factors, such as the damping due to rubbing of the mechanical interfaces (Coulomb
damping) which attach the blade to the test section and the unsteady aerodynamic forces associated with the natural response. The combined effect of all
such interactions defines the response of the blade, which, ideally would result in
vibrations which are rapidly attenuated. The response of the blade was characterized by an effective damping ratio, , which for this work was defined assuming
that each blade mode responds to the mechanical impulse as an under-damped
single degree of freedom system

Zp (t) = A0 ejf p t(1j)

(5.7)

where A0 is the initial amplitude of vibration. This assumption implied that the
modal forcing due to the natural unsteady aerodynamics were small compared
to the forces generated as a result of the mechanical impulse. Such a situation
may occur immediately following a large mechanical impulse. The estimates of
the effective damping ratio use the first seven cycles of vibration for the first mode
to estimate the aerodynamic damping. The duration used was selected based on
observations of the blade vibration after the impulse had been applied.
The effective damping ratio was estimated using the approach described in
chapter 2, section 2.2.3. It will be repeated here for convenience. Given the
time-resolved modal velocity of the blade, Zp (t), Matlabs Hilbert function was
applied to make the signal analytic. This operation is denoted by the operator H[].
Assuming a form for Zp (t) described by equation 5.7, the magnitude of H[Zp (t)]

129

is given by

|H[Zp (t)]| = A0 ef p p t .

(5.8)

The phase of H[Zp (t)] can be expressed as

tan

= [H[Zp ]]
< [H[Zp ]]


= 2f p t.

(5.9)

Equations 5.8 and 5.9 were used to estimate the effective modal damping, p .
It was further assumed that the damping ratio and frequency of vibration were
approximately constant over the duration of the impulse response. This assumption is valid for simple vibration energy dissipation mechanisms such as viscous
damping and the aerodynamic damping described by linear theory. Note that this
assumption may not be valid if non-linear sources of vibration dissipation are significant, such as Coulomb or slip damping (cf. [32]) or non-linear fluid-structure
interactions that result in hysteretic damping (such as dynamic stall, cf. [55] and
[12]).
The results presented in the following sections characterize the blade motion
in terms of modal velocity, p , rather than modal displacement, Zp . Note that
for an impulse response, the modal velocity () differs from the position (Z) by
a factor of 2f p (j p ). It was discovered that p << 1 for both modes and
all flow conditions investigated. As a result, the difference between the modal
velocity and displacement is the amplitude and phase. Recall that the damping is
depended only on the normalized decay rate, and so the error associated with using
modal velocity in lieu of displacement can be considered negligible. In general,
this relation is not valid for complex modal amplitude variations or large damping
ratios.
130

5.3 Results
5.3.1 Variation in effective damping ratio for the first two modes
Figure 5.1 shows the effective modal damping ratio of the blade at 0 = 8 as
a function of Mach number. The figure presents the damping ratio from single
impulse events (in dots) and the Mach-averaged damping ratio as a solid curve.
Despite the high repeatability of the impulse response in quiescent air, the data
show significant scatter at non-zero fluid velocities. The scatter is very large for
the bending vibration at transonic Mach numbers (recall Mcrit. 0.62). The
modal damping ratio for bending increased with increasing Mach number up to
M = 0.65. A moderate decrease in damping to slightly below the value measured
in quiescent air was observed between 0.65 M 0.80. A sharp increase was
then observed between 0.80 M 0.85, followed by a large decrease. The Machaveraged damping was nearly zero at M = 0.95. A slight increase was observed
between 0.95 M 1.00. The ensemble variation also grew with increasing Mach
number up to M = 0.90. A reduction was observed in the ensemble variation
between 0.90 M < 1.
The damping ratio of the torsion mode had less variation with Mach number
than the bending mode. The scatter in the individual damping ratio estimates
was nearly uniform over nearly the entire Mach number range and was larger than
the scatter observed for quiescent air. Additionally, the damping ratio was very
nearly zero over the entire range of Mach numbers investigated.
The effective damping ratio under quiescent flow conditions (M = 0) was
significant with 1 0.0128 and 2 0.0078. An investigation of the change
in damping ratio as a function of atmospheric pressure was performed using a
similar blade. The objective of this experiment was to estimate the contribution

131

of the quiescent air to the damping of the blade. A description of this experiment
and the results can be found in Appendix A. It was found that the quiescent
air contributed very little to the aerodynamic damping. Thus, it may be likely
that most of the damping observed under quiescent conditions for these tests were
associated with the fixture holding the blade rather than due to the presence of
the fluid.

(a)

(b)

0.1

0.02

0.015

0.01

0.05

0.005

0.005

0
0.01

0.015

0.05
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.02
0

Mach number

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Mach number

Figure 5.1. Effective damping ratio vs Mach number for 0 = 8 . (a)


Mode 1; (b) Mode 2. calculated from individual response,
Average at fixed Mach number.

The effective damping ratio of the blade at 0 = 13 is shown in figure 5.2. The
following observations may be made. The damping of the bending mode increased
slightly with increasing Mach number at a rate similar to the 0 = 8 results for
M < 0.80. A slight decrease in damping was observed between 0.80 M 0.95.
The damping for the torsion mode showed little variation over a large range of

132

Mach numbers. Similar to the 0 = 8 results, the torsion modal damping ratio
was also very nearly zero. Note however that an increase in damping and a
reduction in scatter was observed between 0.50 M 0.60.

(a)

(b)

0.1

0.02

0.015

0.01

0.05

0.005

0.005

0
0.01

0.015

0.05
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.02
0

Mach number

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Mach number

Figure 5.2. Effective damping ratio vs Mach number for 0 = 13 . (a)


Mode 1; (b) Mode 2. calculated from individual response,
Average at fixed Mach number.

A comparison of the ensemble-averaged change in damping ratio for the first


two modes at both angle of attacks as well as the theoretical predictions based
on thin airfoil theory are shown in figure 5.3. Note that the change in damping is
shown. This was estimated by subtraction of the damping observed for the modes
in quiescent air. Second order effects such as angle of attack were neglected in the
theoretical model.
The variation in damping of the bending mode agreed reasonably well with
the linear model for subsonic flow. The damping ratio for the bending mode at

133

(b)
0.05

0.05

0.04

0.04

0.03

0.03

0.02

0.02

(a)

0.01

0.01

0.01

0.01

0.02

0.02

0.03
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.03
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Mach number

Mach number

Figure 5.3. Ensemble-averaged change in damping ratio vs Mach


number for both angles of attack. (a) Mode 1; (b) mode 2; 0 = 8 ;
0 = 13 ; aerodynamic damping from unsteady thin airfoil theory.

0 = 8 and 13 had nearly identical variation with Mach number below M < 0.55.
The measured damping ratios at high Mach number deviated from the theoretical
prediction. The theoretical model was derived from a linearized version of the
equations of motion, and so it was expected that it would be inaccurate at high
Mach number.
The variation in damping with Mach number for the torsion mode is shown
in figure 5.3 (b). The theoretical and experimental aerodynamic damping ratios
were in reasonable agreement and very nearly zero for low Mach numbers.The
theoretical damping ratio diverged from the measurements beyond M > 0.70.

5.3.2 Time-resolved blade response


The time-resolved blade response due to a mechanical impulse can also yield
important insights into the damping. Figure 5.4 presents the time-resolved modal
vibration velocity for the first two modes at M = 0.65. Also shown is the ap-

134

proximate exponential decay envelope calculated from the instantaneous damping


ratio. Recall that the damping was estimated from the first 7 cycles of vibration
after the impulse was applied.
The response at 0 = 8 and M = 0.65 (figure 5.4 (a)) revealed that the bending mode had high damping but the torsion mode showed very little damping. The
bending response could be approximated by an exponentially decaying sinusoid
quite well. A deviation in the periodicity of the response can be observed around
t = 34 ms. Specifically, an elongation of the vibration period can be observed
between 34 t 36 ms. The initial torsion response could be described using a
sinusoid with nearly constant amplitude. Note that at t = 35 ms, the vibration
amplitude was abruptly attenuated. Beyond t = 35 ms, modest growth in the
torsion amplitude was observed coupled with a low magnitude oscillation.
The response at 0 = 13 was noticeably different from the response at 0 = 8 .
The bending modal damping ratio was much lower when the angle of attack was
increased. The decay rate was approximately exponential. The torsion modal
damping ratio was higher than the damping observed at 0 = 8 . In addition, the
envelope of vibration showed more deviations from an exponential decay. Beyond
t = 27 ms, the envelope fluctuation was nearly periodic.
An increase in Mach number to M = 0.75 resulted in a substantial change in
the blade response. Recall that this Mach number corresponded to a local minima
in the bending modal damping at 0 = 8 . It can be seen that the bending modal
damping at 0 = 8 (figure 5.5 (a)) was smaller than was observed at M = 0.65.
The amplitude could be approximated by an exponential decay only for the first
5 cycles (15 t 30 ms). The vibration was observed to grow beyond t = 30 ms.
The torsion response over the same period showed moderate growth over the same

135

(b)

0.6

0.6

0.4

0.4

0.2

0.2

1 [m/s]

1 [m/s]

(a)

0
0.2
0.4

0.4

10

15

20

25

30

35

40

45

50

0.5

0.5

3 [m/s]

3 [m/s]

0
0.2

0
0.5

10

15

20

10

15

20

25

30

35

40

45

50

25

30

35

40

45

50

0
0.5

10

15

20

25

30

35

40

45

50

Time [ms]

Time [ms]

Figure 5.4. Time series of blade response to mechanical impulse at


M = 0.65. (a) 0 = 8 ; (b) 0 = 13 .

time period, implying negative aerodynamic damping. The general trend could be
approximately described by an exponential growth, but instantaneous vibration
amplitude was also characterized by a quasi-periodic oscillation. Finally, a large
spike was observed at t = 18 ms, immediately followed by a sharp reduction in
vibration amplitude.
The effect of increasing blade angle of attack at M = 0.75 resulted in noticeable
changes in the response. The difference between the two cases, shown in figure
5.5 (a) and 5.5 (b) can be described as follows. The bending modal damping at
0 = 13 was larger than observed at 0 = 8 . The bending response had low
magnitude deviations from an exponential decay. The torsion modal damping was
positive, and the response deviated from an exponential decay. These deviations
were quasi-periodic.
The blade impulse response at M = 0.81 is shown in figure 5.6. Recall that
at this Mach number a local minima was observed for the torsion modal damping
at 0 = 13 and for the bending modal damping at 0 = 8 . The 0 = 8

136

(b)

0.6

0.6

0.4

0.4

0.2

0.2

1 [m/s]

1 [m/s]

(a)

0
0.2
0.4

0.4

10

15

20

25

30

35

40

45

50

0.5

0.5

3 [m/s]

3 [m/s]

0
0.2

0
0.5

10

15

20

10

15

20

25

30

35

40

45

50

25

30

35

40

45

50

0
0.5

10

15

20

25

30

35

40

45

50

Time [ms]

Time [ms]

Figure 5.5. Time series of blade response to mechanical impulse at


M = 0.75. (a) 0 = 8 ; (b) 0 = 13 .

case (figure 5.6) could be described as follows. The vibration amplitude remained
nearly constant for a long period after the mechanical impulse was applied. The
resultant total damping was approximately zero, implying that the aerodynamic
damping was negative. Note that at t = 47 ms, an abrupt decrease in vibration
amplitude was observed. The torsion amplitude over the same time period had
an exponential decay for the first 15 ms after the mechanical impulse. A sharp
decrease in vibration amplitude was observed between 30 t 36 ms, followed by
a period of rapid growth. This period of growth transitioned into another period
of decay.
The response at M = 0.81 and 0 = 13 (figure 5.6) also had some noteworthy
features. The bending vibration amplitude had an initial exponential decay with
a correspondingly low damping ratio. This damping ratio was lower than the
damping observed at M = 0.75. A sharp decrease in vibration amplitude was
observed at t = 36 ms, followed by a period of nearly constant amplitude vibration.
The torsion response could be characterized by an exponentially growing vibration.

137

Note that large deviations from this growth rate were observed.

(b)

0.6

0.6

0.4

0.4

0.2

0.2

1 [m/s]

1 [m/s]

(a)

0
0.2
0.4

0.4

10

15

20

25

30

35

40

45

50

0.5

0.5

3 [m/s]

3 [m/s]

0
0.2

0
0.5

10

15

20

10

15

20

25

30

35

40

45

50

25

30

35

40

45

50

0
0.5

10

15

20

25

30

35

40

45

50

Time [ms]

Time [ms]

Figure 5.6. Time series of blade response to mechanical impulse at


M = 0.81. (a) 0 = 8 ; (b) 0 = 13 .

The blade response at M = 0.95 is shown in figure 5.7. Recall that at this
Mach number, a local minima in bending modal damping was observed at 0 = 8 .
This is illustrated in figure 5.7 (a), which revealed that the bending vibration
was nearly constant over most of the observed response. A period of decaying
vibration amplitude was observed beyond t = 38 ms. The torsion response could
be characterized by a period of rapid growth followed by high amplitude vibrations.
The instantaneous amplitude fluctuated, but the mean value was nearly constant
between 25 t 45 ms.
The response at high angle of attack and M = 0.95 is shown in figure 5.7
(b). The bending vibration amplitude had an approximately exponential decay,

138

though moderate deviations from this curve were observed. The decay rate was
much larger than was observed at 0 = 8 and M = 0.95 as well as at 0 = 13
and M = 0.81. The torsion vibration amplitude had little decay over the same
time period. Note that the amplitude had moderate fluctuations about the nearly
constant mean value.

(b)

0.6

0.6

0.4

0.4

0.2

0.2

1 [m/s]

1 [m/s]

(a)

0
0.2
0.4

0.4

10

15

20

25

30

35

40

45

50

0.5

0.5

3 [m/s]

3 [m/s]

0
0.2

0
0.5

10

15

20

10

15

20

25

30

35

40

45

50

25

30

35

40

45

50

0
0.5

10

15

20

25

30

35

40

45

50

Time [ms]

Time [ms]

Figure 5.7. Time series of blade response to mechanical impulse at


M = 0.95. (a) 0 = 8 ; (b) 0 = 13 .

5.4 Conclusions
The response of the blade to a mechanical impulse was investigated as a function of Mach number and blade angle of attack. It was found that the effective
damping, as inferred from the blade vibration decay rate, varied significantly with
tunnel speed. The bending modal damping ratio had a similar variation with

139

Mach number for the two angles of attack investigated between 0 M < 0.6.
This variation agreed reasonably well with a linear model for the aerodynamic
damping. The model diverged from the measurements beyond M 0.60. This
suggests that most of the variation in the response of the blade to a mechanical
impulse for M 0.6 could be attributed to changes in the aerodynamic damping
as described by linearized unsteady thin airfoil theory.
The ensemble-averaged effective damping ratio for first bending showed a gradual increase with Mach number up to M = 0.65 for both angle of attacks investigated. Beyond this Mach number, the damping ratio at 0 = 8 had a substantial
decrease followed by a recovery at M = 0.90. At M >= 0.90, the damping ratio decreased again resulting in a total blade damping ratio of nearly zero. This
decrease coincided with a local maxima for the bending modes natural response.
This behavior was not observed at 0 = 13 . Instead, a moderate decrease was
observed between 0.65 M 1.0.
The ensemble-averaged effective damping for the torsion mode showed less
variation than the bending mode. At 0 = 8 , the aerodynamic damping decreased
between 0 M 0.4 such that the total modal damping was very nearly zero.
The damping ratio then remained very nearly constant over 0.4 M 1.0. In
contrast, the damping at 0 = 13 for this mode remained very nearly constant
at zero over 0 < M < 1.
The time-resolved blade response to a mechanical impulse revealed that the
response at 0 = 8 for bending could be described well by an exponential decay.
This implied that the viscous damping force model for the aerodynamic damping
(cf. chapter 4) was a fair approximation). In contrast, the bending response at
0 = 13 deviated from the exponential decay.

140

The time-resolved blade response for the torsion mode differed substantially
from the bending response. Specifically, the majority of the responses observed
had an exponential decay (or growth) like behavior only for the first few cycles
after the mechanical impulse was applied to the blade. This variation was observed
at both angle of attacks and over a wide range of Mach numbers. This suggested
that a linear model for the modal damping was not appropriate for this mode.

141

CHAPTER 6
DEVELOPMENT AND UNCERTAINTY ANALYSIS OF BLADE IMAGE
VELOCIMETRY

6.1 Introduction
This chapter discusses the theory and bench-top validation of a novel noncontact blade vibration measurement technique. This technique was developed
as an alternative to the two conventional vibration measurement techniques commonly employed in turbomachine aeromechanical validation : strain gages and
Blade Tip Timing.
A description of the general implementation of combined BIV/PIV will be
presented in the following section along with a demonstration which shows the
aeroelastic response of a flat plate in incompressible flow. This will be followed by
a detailed analysis of the accuracy of the PIV algorithms for measuring the structural velocity. The calculations used for the determination of the modal amplitude
as well as the associated uncertainty analysis will then be provided. This will be
followed by a discussion of two sets of experiments used to both demonstrate and
validate the BIV measurement technique. The first experiment was a benchtop
validation of the measurement technique against a commercially available Laser
Doppler Vibrometer using a harmonically forced cantilevered beam. The second
set of experiments applied the BIV measurement technique to a high-speed axial

142

compressor rotor. A second validation was performed at a low speed operating


point where blade vibrations were negligible. Then, BIV was used to estimate
blade vibrations at high shaft speed. At 13,531 RPM and at high mass flow, a
low amplitude blade vibration in first flexure was measured. The compressor was
also operated in deep stall at the same shaft speed. Excitation of the first three
blade vibration modes were detected.

6.2 PIV/BIV implementation


6.2.1 Aeroelastic demonstration of simultaneous PIV/BIV
A demonstration of the BIV technique with simultaneous PIV measurements
was performed. This experiment was conducted to demonstrate a unique feature
of the BIV measurement technique : the ability to acquire simultaneous fluid
and structure velocity measurements. The demonstration used a standard single
camera PIV system [54] imaging the tip velocity and flow field of a cantilevered
flat plate undergoing flutter. A schematic of the setup is shown in Figure 2.7. The
model shown is a cantilevered beam which nominally represents a generic axial
compressor blade geometry. Large amplitude torsional vibration was observed at
U = 12 m/s and 0 = 20 . A description of the setup can be found in chapter 2.
Figure 6.1 shows a sample image acquired by the PIV camera as well as the
computed velocity vectors. The x0 -z 0 coordinates are oriented such that the freestream fluid velocity is in the x0 direction. The vectors representing the structural
velocity (colored red) are scaled by a factor of 50 with respect to the fluid velocity
vectors. All images were processed using the approach described in section 2.3.2.
The vector resolution was 1.5 vectors/mm. Note that in figure 6.1(b) every 4th
fluid velocity vector and every 3rd structure velocity vector are displayed for clarity.

143

(a)
40
35
30
25

z 0 [mm]

20
15
10
5
0
5
10
15

20

10

10
20
x0 [mm]

30

40

50

Vorticity [rad/s103 ]
2.5

(b)
40
35

30

1.5

25

z0 [mm]

20

0.5

15
0
10
0.5

1.5

10
15

20

10

10
20
x0 [mm]

30

40

50

2.5

Figure 6.1. (a) Image of seeded fluid and structure for combined
BIV/PIV measurements. (b) Computed velocity and vorticity field.
fluid velocity; structure velocity 50; contours indicate fluid vorticity.

144

The BIV/PIV image pairs were acquired at 1 Hz. The low sampling rate with
respect to the frequency of the structural vibration (53 Hz) implied that each
velocity field obtained by combined BIV/PIV was an independent realization.
However, dynamic information can be obtained by phase averaging the velocity
measurements with respect to the blade motion. The phase-state of the vibration
was determined by measuring the angle of the blade with respect to the approach
flow as well as the angular velocity of the blade. The angle of the blade tip was
determined from a single blade image. The angular velocity was derived from
the velocity measurements. A phase-space type diagram of 300 realizations of
the structural motion is shown in figure 6.2. The resulting limit cycle shown is
approximately elliptical. The solid ellipse shown in the figure represents a phase
averaged representation of the quasi-periodic blade motion. The major and minor
axes represent the amplitude of the phase-averaged angular displacement and
velocity, respectively. The ratio of the major axis to the minor axis of this ellipse
yields an estimate for the frequency of vibration, and was found to be 48 Hz.
The phase-averaged velocity and vorticity fields at two points in the response
are shown in figure 6.3. An isocontour representing velocity magnitude of 11 m/s
is shown in order to illustrate the spatial extent of the wake. In figure 6.3(a),
the plate is shown pitching upward and the general structure of the flow field is
similar to the instantaneous measurement presented in figure 6.1(b). Note that
the velocity field shown is a planar section of a highly three dimensional flow at
the tip of the plate. The velocity shows a separation point at 30% chord (as
measured from the leading edge). The isocontour shows that the separation region
grows in the chord-normal direction as the flow progresses downstream. Aft of
50% chord, the isocontour becomes nearly parallel to the free-stream velocity.

145

3
/10

[radians/s]

0.05

(a)
0

(b)

0.05
0.05

0.05

[radians]

Figure 6.2. Phase space behavior of the flat plate subject to uniform
flow. Points marked represent phase-state locations used in figure 6.3.

There are significant differences in the flow field as the plate pitched downward
(figure 6.3(b)). The separation point moved forward to approximately 15 percent
of the chord, and two regions within the wake were observed. Specifically, the
isocontour is nearly parallel to the blade over the mid section of the chord, and
nearly parallel to the free stream direction near the trailing edge. Analysis of the
full cycle of the phase-averaged response revealed an interesting dynamic between
the plate and the fluid. As the plate pitches upward (figure 6.3(a)), the separation
point retreats toward the trailing edge and the chord-normal extent of the wake
grows. A new separation region then forms upstream of the pre-existing wake
region as the plate begins to pitch downward. The new separation region grows
until the lowest angle of attack is reached. A video illustrating the phase-averaged
velocity field over the entire limit cycle is available from the authors1 .
Figures 6.1 - 6.3 demonstrate that significant insight into fluid-structure inter1

http://www.archive.org/details/BIV_PIV_Phase_average_Mikrut_Morris_09

146

Vorticity [rad/s103 ]
2.5

(a)
40
35

30

1.5

25

z0 [mm]

20

0.5

15
0
10
0.5

1.5

10
15

20

10

10
20
x0 [mm]

30

40

50

Vorticity [rad/s103 ]
2.5

(b)
40
35

30

1.5

25

20
z0 [mm]

2.5

0.5

15
0
10
0.5

1.5

10
15

20

10

10
20
x0 [mm]

30

40

50

2.5

Figure 6.3. Phase averaged flow around a blade undergoing aeroelastic


oscillations. See figure 6.2 for the corresponding locations in phase
space. fluid velocity; color p
contours represent vorticity; structure
velocity 50; isocontour of Ux20 + Uz20 = 11 m/s; (a) 0.0145 rad,
10.5 rad/s; (b) 0.0144 rad, 10.5 rad/s.

147

actions can be obtained using the BIV/PIV technique in aeroelastic flows. The
following section will focus on the accuracy of the structural velocity measurements using the PIV hardware. This will be followed by a description of the
mathematics and uncertainty related to using the BIV measurements to estimate
the discrete modal amplitudes.

6.2.2 Uncertainty in structural velocity


The accuracy of the structural velocity measurements was evaluated experimentally using harmonic forcing of the structure. A blade-like geometry was
utilized in which the natural frequency of the first torsional mode was 782 Hz.
Details of the geometry and mode shapes are provided in Section 2.3.2. A Laser
Doppler Vibrometer (LDV) was used as a reference sensor that was assumed to
provide the true velocity at one point on the blade. A schematic of the experimental setup is shown in figure 6.4.

Figure 6.4. Schematic of the BIV calibration experiment.

148

Velocity measurements obtained from BIV were spatially averaged in the vicinity of the LDV measurement point. The LDV measurements were time averaged
over the 200s required to obtain two BIV images. The optical arrangement and
vector processing were similar to that described in the previous section.
The velocity measured by the BIV system is shown as a function of the LDV
measured velocity in Figure 6.5(a). The corresponding difference between the two
measured velocities are shown in Figure 6.5(b) in pixel equivalent displacement.
The LDV velocity was converted to pixel equivalent displacement by applying
the PIV camera calibration to the LDV data. The peak locking effect[65], a common error in PIV measurements, is evident in the data. The standard deviation
of the peak-locking effect is approximately 0.045 pixels. This corresponds to an
accuracy of 1.2% of the full-scale measurement, or 9.1 mm/s over a range of 757
mm/s. These results will be used in section 6.5 for estimating the uncertainty of
the blade modal amplitude.

6.3 Determination of vibration modal amplitude


This section describes the mathematical operations necessary to obtain both
instantaneous and mean-squared amplitudes of discrete vibration modes. Consider
the cantilevered blade shown in figure 2.7. The velocity of the blade can be
composed of a time-averaged component (due to rotation if the blade is part of
a turbomachine rotor) and a fluctuating component due to blade vibration. For
simplicity of presentation, the fluctuating velocity of the blade, v , is assumed to be
scalar valued representing the chord-normal component of the structural velocity.
This is not a limitation of the technique since the image correlations provide two
components of tip velocity. The velocity is given at discrete spatial locations

149

(a)

(b)
0.5

0.2

0.4

0.15

VLDV VPIV [pixels]

0.3

VPIV [m/s]

0.2
0.1
0
0.1
0.2

0.1
0.05
0
0.05
0.1

0.3
0.15

0.4
0.5
0.5

0.2
2

0.5

VLDV [m/s]

XLDV [pixels]

Figure 6.5. Accuracy of the correlation algorithms in measuring


structural velocity: (a) BIV measured velocity as a function of LDV
measured velocity. The solid line indicates a slope of 1 with zero offset;
(b) Difference in measured velocity as a function of LDV measured
velocity. Results are presented as pixel equivalent velocity.

(xi , yj ) and time (tn ), [vv]i,j = v(xi , yj ). Here i = 1, j = 1 represents the leading
edge of the blade tip. The trailing edge of the blade is represented by i = M ,
where M is determined by the number of velocity vectors along the chord that are
provided by the PIV processing. The notation [ ]i,j denotes the element of the
argument corresponding to the ith row and j th column. The number of subscripts
denotes the dimension of the argument; in this case a matrix with 2 dimensions.
The velocities can be represented by a summation of eigenmodes as

[vv]i,j =

P
X

[]i,j,p []p .

(6.1)

p=1

where is a time-varying vector of modal amplitudes with dimensions [P 1],


where P is the number of modes used to represent the structural vibration. The
150

eigenmode shapes of the structure are given by the three dimensional matrix .
Because the BIV technique provides velocity measurements at the tip of the structure, it will be useful to provide a more specific notation to represent these values.
Specifically, the measurements along the chord of the blade tip are represented as

[V]i v(xi , yj=1 ).

(6.2)

Similarly a matrix is defined to represent the mode shapes at the tip as

[]i,p []i,j=1,p .

(6.3)

such that is a two dimensional [M P ] matrix with rows corresponding to the ith
measurement point along the chord-wise direction and the columns corresponding
to the pth vibration mode evaluated at yj=1 . Unlike the full eigenmode shapes
given by , the columns of the matrix are not generally orthogonal.
The relationship between the observed velocities and the modal amplitudes is
now given by

V = ,

(6.4)

This relation can be inverted to obtain the modal amplitudes directly from the
measured velocities:

= [T ]
where the superscript

T V

(6.5)

denotes the matrix transpose. The only constraint is

that the columns must be linearly independent ( must be full column rank).
Equation 6.5 can be used with equation 6.1 to estimate the full blade velocity v .
151

The mean squared amplitude of the various modes is most often of interest, and
can be evaluated directly from an ensemble of velocity measurements (V) through
equation 6.5. However, it will be useful to consider the mean squared amplitudes
of the modes based on the statistics of the measured velocities in order to simplify
the uncertainty analysis to follow. The ensemble of (tn ) can be represented as
the matrix



(t1 ) (t2 ) (t3 ) (tN )

(6.6)

where the discrete time tn is shown explicitly for clarity. Note that is a [P N ]
matrix, where N is the number of samples. The vector of mean-squared amplitudes
of the P modes is estimated by

1 T


(6.7)

where the operator D () selects the diagonal elements of the argument. Note that
2 is a vector of length P .
The measured velocities can be used to define the ensemble of observations as


=


V(t1 ) V(t2 ) V(t3 ) V(tN )

= ,

(6.8)

The two-point correlation matrix C is defined as

1
1
T = T T .
N
N

152

(6.9)

The vector of mean-squared amplitudes 2 can be found by taking the left


pseudoinverse of , L [T ]1 T and the right pseudoinverse of T , R
1

[T ] :
2 = D (LCR)

(6.10)

Equation 6.10 relates the mean-squared amplitudes of the structural modes given
in 2 to the two-point correlation matrix of the observed velocities.
The results obtained from application of equation 6.10 may be validated
by Proper Orthogonal Decomposition (POD) of the observed tip velocities, .
POD performs an eigen-decomposition of the two point correlation matrix of an
ensemble of velocity observations. Feeny and Kappagantu [23] demonstrated that
for structural vibration, the eigen-modes obtained from POD approximate the
mode-shapes of a given structure. Therefore, if the tip projection of the active
modes of vibration are nearly orthogonal, then the eigen-modes obtained from
POD should agree with the active tip-mode shapes, . Conversely, if the observed
tip velocity is due primarily to measurement noise, then the POD eigen-modes
will differ significantly from .
Note that the POD eigen-modes are an orthogonal least-squares approximation to the tip projection of the active blade modes, which may be non-orthogonal.
The difference in the mode shapes can result modal-amplitude estimates which
differ significantly between BIV and POD analyses. The amplitude estimates obtained from BIV will reflect the true modal amplitude because BIV uses the true
tip projection, rather than an orthogonal approximation.

153

6.4 Uncertainty analysis


There are two types of errors which can arise in the estimation of 2 . The
first is noise corruption of V. The second type of error is associated with the
discretization of the velocity by measuring displacement over a small time interval. This particular error is present because of the method by which PIV (and
as a consequence, BIV) estimates velocity. These two errors will be discussed
separately in the following subsections.

6.4.1 Noise corruption of the correlation matrix


Assume that V is corrupted by additive noise. The observation matrix is
modified as

= + T ,

(6.11)

where T is the ensemble of additive noise. The correlation matrix becomes

C=

1
( T T + T T T + + T ) .
N

Applying the pseudoinverse operators L and R to the modified correlation matrix


C yields

D (LCR) = 2 +

1
D (L T T + R + L T R) .
N

154

(6.12)

The left hand side of equation 6.12 represents the measured quantity. The first
term on the right hand side of equation 6.12 contains the true values of the modal
amplitudes. The remainder of the terms are errors due to the additive noise. The
second and third terms on the right hand side of equation 6.12 are noise-signal
cross correlations, while the last is the autocorrelation of the noise. The signalnoise correlation terms can be finite in magnitude for a given experiment, but
have an expected value of zero so long as the noise is not related to the mode
shapes. However, the error due to noise autocorrelation is positive definite which
leads to an expected bias. This error will be defined as

[A ]p

i 

1 h
E [D (L T R)]p / 2 p ,
N

(6.13)

where E[] is the expected value of the argument. The error is can be simplified to

h 
i


[A ]p = D [T ]1
2 / 2 p ,

(6.14)

where the variance of the noise is given by


 
1 T

.
E D
N

(6.15)

Equation 6.12 demonstrates that a non-zero positive bias error is introduced


for finite noise. However, the magnitude of additive noise is generally not known
a priori. A method for estimating the magnitude of noise corruption can be
obtained from the statistics of the blade tip velocity measurements. Consider the

155

scalar quantity , defined as

N
M
1
M
X
1 X 1
1 X

N n=1 M 1 i=1
M i=1

M
1 X T
T

M i=1

!
,

(6.16)

which is the ensemble-averaged chordwise variance of the measured tip velocity.


This quantity depends on the statistics of the true blade tip velocity and the
additive noise. If the blade is not vibrating, will be equal to the variance of
additive noise. Alternatively, if there is blade vibration and negligible noise, will
be determined by the mode shapes and their respective amplitudes. Substituting
equation 6.11 and performing the summations, this can be simplified as
P
X
 2  h 2i
=
p + 2

(6.17)

p=1

where

 2
p

M
1
M
X
1
1 X
[]i,p
[]i,p
M 1 i=1
M i=1

M
1 X T
T
[ ]p,i
[ ]p,i
M i=1

!
(6.18)

is the variance of the pth mode along the chord. Note that 2 was assumed
constant for all modes such that is a scalar quantity. Equations 6.12 and 6.17
represent a system of P + 1 equations with P + 1 unknowns which is solvable for
the mean-squared modal amplitudes 2 and the noise 2 .
6.4.2 Bias error due to velocity discretization
The bias error associated with the discretization of the velocity when using
BIV will now be considered. The BIV measurement technique computes velocity
156

based on displacement measurements with a specified time interval. This can


be considered as a discrete approximation to the velocity field using a central
difference approach. Note that this discretization does not effect the mode shapes
. Therefore, according to equation 6.4, the bias error which manifests in the
velocity estimation is proportional to . Let the true value of at time tn be of
the form

[]p

q




2 [ 2 ]p cos 2 [f ]p tn ,

(6.19)

where f is a vector of frequencies associated with the mean-squared amplitudes


2 . Let the time between an image pair be t. The approximation of by BIV
is


h i


q
sin [f ]p t
2


2 [ ]p cos 2 [f ]p tn .
[f ]p t
p

(6.20)

The normalized error in the estimate of will be defined as


h i
[]p
[B ]p =

[]p

(6.21)

Using equations 6.19 and 6.20, this can be simplified as

[B ]p = 1

sin( [f ]p t)
.
[f ]p t

(6.22)

Thus, this error results in an underestimation of the amplitude of vibration. This


error can be corrected for if the frequencies of oscillation f and the time between
images t are known.

157

6.5 Experimental validation of uncertainty


The BIV measurement technique and the derived equations for the modal
uncertainty were evaluated using a harmonically excited cantilevered beam. A
description of the experimental setup can be found in section 2.3.2.
The blade tip motion was recorded with one thousand image pairs (N = 1000).
The mean squared amplitudes of vibration were computed using equation 6.10.
A single point LDV was positioned near the blade tip and used as the reference
transducer.
The beam was harmonically forced only at first torsion at two different amplitudes to demonstrate the interaction of the two types of error. The t was
varied among different tests to vary the influence of random and bias errors on
the measured amplitude of vibration. Figure 6.6 presents both normalized errors
as well as the experimentally measured error as a function of (t)f . The solid
line in the figure represents the summation of the noise correlation error (A ) and
the velocity discretization error (B ). The data show substantial agreement with
the derived uncertainty values over the range of (t)f tested. As expected, the
lower values of (t)f show a positive bias in the BIV measured modal amplitudes
compared to the true value observed from the LDV. This was a direct result of
the noise correlation bias due to pixel locking in the digital image correlations.
At higher (t)f , the BIV measured amplitudes were lower than the true values
indicating that the velocity discretization error is dominant. For the specimen
tested there was an optimal range of (t)f 0.1 where both types of error were
found to be small.
The technique was also tested with two modes active. The t was 150 s. The
first torsion ([f ]1 782 Hz) and second bending ([f ]2 959Hz) modes were ex-

158

0.1

0.05

0.05

0.1

0.15

0.2

0.25

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

(t)f

Figure 6.6. Error in amplitude


estimate

for a single mode as a function


of (t)f . B ; 2A ; B + 2A ; E from BIV with peak
amplitude of 2.3 m/s; 4 (E ) from BIV with peak amplitude of 4.2
m/s.

159

TABLE 6.1
COMPARISON OF MODAL AMPLITUDE ESTIMATES FOR A
BEAM UNDER TWO COMPONENT HARMONIC FORCING.

LDV [m/s]

BIV [m/s]

Mode 1

0.0015

N/A

Mode 2

1.4159

1.4097

Mode 3

0.7468

0.7338

cited simultaneously with a corresponding (t)f of 0.117 and 0.144, respectively.


The amplitude estimates for first torsion and second bending were expected to
be biased low by 2.25% and 3.37% low respectively. Table 6.1 shows the amplitudes of the modes obtained from the LDV measured autospectra compared to
those obtained by BIV when one corrects for the velocity discretization error. The
amplitude estimate obtained from the corrected BIV technique differ from those
found by the LDV by 0.7% for first torsion and 1.75% for second bending. This
agreement demonstrates the ability of the BIV measurements to obtain accurate
estimates of the modal amplitudes when multiple modes are active.

6.6 Conclusions
BIV has been developed for measuring structural vibration using a commercial PIV system. The method is non-intrusive and can measure structural velocity
with high accuracy. Furthermore, PIV measurements can also be obtained simultaneously using this technique.

160

There are a number of potential advantages compared to traditional structural measurements. First, the instantaneous amplitude of the blade mode shapes
can be obtained because a spatial distribution of structural velocity is measured.
When coupled with PIV measurements of the fluid velocity, the causal relationships between the fluid and structural motion can be studied. An example was
provided using a thin flat plat in a low-speed wind tunnel. A limit-cycle behavior was found in which the first torsional vibration mode was closely linked with
a moving separation point. A second advantage of BIV is the ability to obtain
aeroelastic measurements on rotating structures. This will allow measurements of
both forced excitation and flutter in turbomachinery blades.
The theory relating the fluctuating velocity at the free-end of a blade (the
tip) to the modal amplitudes of the blade was presented. The measurements can
provide independent realizations of the modal amplitude as long as the observable
mode shapes are linearly independent. The equations were developed in order to
describe the variance of the modal amplitudes in terms of the two-point correlation
matrix of the measured velocity. This allowed for an analysis of the error due to
noise corruption in the velocity measurements.
Two sources of error in the estimation of modal amplitudes were identified.
The first is associated with the noise corruption of the velocity estimates and is
due to the peak-locking effect, an error common to PIV. This results in an overestimation of the modal amplitudes. The second error is due to the central difference
approximation of the velocity. Specifically, when the time between successive images is large the velocities are underestimated which leads to a systematic error
in the modal amplitudes.
A specimen was created to represent a compressor blade geometry. The blade

161

vibration modes were excited using harmonic acoustic forcing and compared with
laser vibrometry measurements that were acquired simultaneously with the BIV.
The results confirmed that the noise in the BIV measurements were dominated by
peak-locking error. This led to an over estimation of the modal amplitudes when
the displacement of the blade was small ((t)f < 0.1). The magnitude of the
error agreed well with the predicted values based on the estimates of peak-locking
noise. The error due to the central difference approximation of the velocity was
found to dominate at large blade displacements ((t)f > 0.1). Good agreement
was found between the experimentally measured magnitude and the theoretical
prediction.

162

CHAPTER 7
APPLICATION OF BIV TO AN AXIAL COMPRESSOR

7.1 Introduction
This chapter discusses the application of the BIV measurement technique to
a high speed axial compressor rotor. A description of the axial compressor and
data analysis algorithms used in this investigation can be found in chapter 2. The
theory used in determining the modal vibration velocity was described in chapter
6.
Two rotor shaft speeds were investigated using the BIV measurement technique. The first was at low shaft speed (2,886 RPM), and was used to validate
the measurement technique. The second was at high shaft speed (13,531 RPM).
The first set of experiments were conducted at low shaft speed and were performed to validate the accuracy of the measurement system against a known rotor
tip velocity. The rotor shaft speed was held constant at 2,886 RPM, corresponding
to a blade tip velocity of 65.5 m/s. The magnetic bearings were used to oscillate
the rotor axially. The axial oscillation was synchronous to the shaft rotation with
a frequency of 96.2 Hz and an amplitude of 49 mm/s. The frequency of the axial
oscillation was well below the natural frequencies of the blades and so the motion
of the tip due to blade vibration was assumed to be negligible. Three blades were
analyzed, corresponding to different points within a cycle of oscillation. Blade

163

number 5 was used as the reference case for the a-posteriori calibration. Fivehundred realizations of the tip velocity were acquired for each blade.
The second set of experiments were conducted at a shaft speed of 13,531 RPM,
corresponding to a blade tip velocity of 322 m/s. The magnetic bearings maintained the axial position of the shaft to 30m, and controlled the whirl orbits to
within 6m. BIV data were acquired at two operating points. The first was at the
aerodynamic design point of the compressor. The second was a low air mass flow
condition where the compressor was operating in deep stall. Large unsteady aerodynamic loads were expected at this flow condition and the BIV technique was
used to determine if these unsteady loads would induce a detectable level of blade
vibration. A single blade was investigated and an a-posteriori calibration based on
the ensemble-averaged tip velocity was applied at each operating condition. One
thousand realizations of the tip velocity were acquired for each operating point.

7.2 Results
This section presents the results of the application of BIV to the rotor of the
axial compressor. The results of the low shaft speed test will be discussed first,
followed by the analysis of the high-speed unsteady blade tip velocity.

7.2.1 Results, 2,886 RPM shaft speed with magnetic bearing excitation
A snapshot of the blade tip for the low shaft speed test is shown in figure
7.1. Also shown is the estimated tip velocity. Note that the leading and trailing
edges of the adjacent blades were removed for simplicity of presentation. It can be
seen that the major component of tip velocity is in the circumferential direction,
due to shaft rotation.

164

(b)
0

10

10

20

20

30

30
RT [mm]

RT [mm]

(a)

40

40

50

50

60

60

70

70
80

80
0

10

20

30

X [mm]

10
20
X [mm]

30

Figure 7.1. (a) image of a blade tip used for BIV measurements. The
shaft speed was 2,881 RPM. (b) Tip velocity estimated from the BIV
technique.

x 10

4.5

(V RT )/RT

4
3.5
3
2.5
2
1.5
1
0.5
0
0

20

40

60

80

100

% Chord

Figure 7.2. Chordwise distribution of error in ensemble-averaged


circumferential tip velocity for a shaft speed of 2,886 RPM.  blade 1;
blade 4;  blade 5; blade 5, single-frame double-exposure.

165

The RMS error in the measured circumferential velocity as a function of


chord is shown in figure 7.2. The error for all three blades are shown along with
the error when the PIV system was operating in single-frame double-exposure
mode. The RMS error was defined as


1/2
(V RT ) < (V RT )2 >
,

(7.1)

where <> denotes the ensemble average of the argument, RT is the radius of the
blade tip and is the shaft speed as estimated from the OPR sensor. Note that
good agreement was obtained for all four cases analyzed. The typical RMS error in
the BIV measured velocity for a single blade was 2.1 104 RT , corresponding
to a velocity of 14 mm/s. This was about 0.29 pixels displacement in the cameras
field of view. The uncertainty in the shaft speed and tip radius resulted in an
uncertainty of 4 mm/s in the true circumferential velocity, and so it can be
seen that the RMS error for the BIV measurement was near the practical limit
for digital image correlation velocity estimates (cf. Westerweel [65]).
The RMS error in the axial velocity estimate as a function of chord is shown
in figure 7.3. The error was defined as


1/2
(V x Vx ) < (V x Vx )2 >
,

(7.2)

where Vx is the axial velocity estimated from the shaft axial position sensor installed in the magnetic bearings. Note that good agreement was obtained with a
typical RMS error of 1.6 104 RT , corresponding to a velocity of 11 mm/s,
or about 0.22 pixel. This was similar in magnitude to the circumferential velocity
error. The similar magnitude between the circumferential and axial velocity errors
suggest a similar source: inaccuracies in the sub-pixel velocity estimate associated
166

x 10

4.5
4
(Vx Vx )/RT

3.5
3
2.5
2
1.5
1
0.5
0
0

20

40

60

80

100

% Chord

Figure 7.3. Chordwise distribution of RMS error in axial tip velocity for
a shaft speed of 2,886 RPM.  blade 1; blade 4;  blade 5; blade 5,
single-frame double-exposure.

with BIV.
The axial velocity induced by the magnetic bearings was calculated from
the phase-averaged velocity of several blades. Figure 7.4 shows the reconstructed
axial velocity using both phase-averaged BIV and the axial position sensor. It
can be seen that the axial velocity obtained from BIV agreed with the true axial
velocity of the hub. This also validates the assumption that tip velocity due to
blade vibration was negligible at this operating point.
In summary, the BIV measurement technique is capable of accurately resolving two components of tip velocity of a rotating blade with high accuracy.
Furthermore, the accuracy associated with single-frame, double-exposed images
was similar to double-frame, double-exposed images.

167

60

Axial Velocity [mm/s]

40

20

20

40

60
0

1/2

1
Phase [radians]

3/2

Figure 7.4. Phase-averaged axial velocity vs. reference time. shaft speed
was 2,886 RPM and axial velocity was induced by magnetic bearings.
BIV measurement; axial velocity from axial position sensor.

7.2.2 Results, 13,531 RPM


Figure 7.5 shows a comparison of the chord-normal RMS velocity at the tip
for the three blade FEM modes and the first three POD eigen-modes when the
compressor was operated near the design mass-flow rate at high shaft speed. Also
shown is the RMS velocity due to noise. The chord-normal velocity was calculated
by multiplying the RMS modal amplitude by the chord-normal projection of the
2-component mode shapes. The RMS velocity due to noise was the RMS noise
amplitude obtained using the approach described in section 6.4. It can be observed
that the magnitude of the fluctuating tip velocity was approximately 0.04% of the
average tip speed, suggesting that any blade vibration at this operating point was
very small. Note also that the tip velocity distribution for all three FEM blade
vibration modes are nearly equal to or less than the RMS noise level.
The velocity distribution for the first FEM mode show good agreement with
the first POD eigen-mode. The comparison between the RMS velocity distribution

168

0.5
0.4
0.3

VN [m/s]

0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
0

20

40

60

80

100

% Chord

Figure 7.5. A comparison of RMS chord-normal fluctuating tip velocity


as determined from POD and BIV analyses. The compressor was
operating at 13,551 RPM and at design air mass flow. FEM mode 1;
FEM mode 2; FEM mode 3; first POD eigen-mode; second
POD eigen-mode; third POD eigen-mode. The gray box denotes the
RMS noise level.

for the second POD eigen-mode and the second FEM mode are less favorable, but
qualitatively similar. Finally, it can be seen that the velocity distribution associated with the third POD eigen-mode does not resemble the velocity distribution
associated with the third FEM mode. The good agreement between the chordnormal RMS velocity distributions associated with the first POD mode and the
first FEM mode implies that a significant portion of the observed fluctuating tip
velocity was due to blade vibration in the first mode (cf. figure 2.13). Additionally, the poor agreement between the third POD eigen-mode and the third FEM
mode implies that the measurement technique could not distinguish between any
tip velocity due to excitation of this mode and the random noise associated with
errors in the tip velocity measurement. Finally, it can be concluded that the
second mode of vibration was active. The deviation of the POD chord-normal

169

TABLE 7.1
COMPARISON OF RMS AMPLITUDES OF THE POD
EIGEN-MODES AND THE FEM EIGEN-MODES USING THE BIV
TECHNIQUE FOR A SHAFT SPEED OF 13,531 RPM NEAR
DESIGN CONDITIONS.

Mode Number

[m/s]

Max. Tip Deflection [m]

0.54 0.13

15 7

0.31 0.13

87

velocity distribution from that observed using the FEM tip mode-shape implies
that the vibration amplitude was very close to lower limits of the measurement
technique.
Table 7.1 summarizes the RMS modal amplitudes and the maximum tip
deflection for each mode. The uncertainty in the modal amplitude was assumed
to be equal to the RMS noise estimated from the approach discussed in section 6.4.
The uncertainty in the displacement was estimated by multiplying the RMS noise
by the t used to measure the tip velocity. It can be seen from the maximum tip
deflections that the blade vibrations at this compressor operating point were very
small, and very near the limits of measurement accuracy for the BIV technique.
Figure 7.6 shows a comparison of the chord-normal RMS velocity at the tip
for the three blade FEM modes and the first three POD eigen-modes when the
compressor was operated in stalled conditions at high shaft speed. Also shown is
the RMS noise estimate. The chord-normal velocity attributed to blade vibration

170

0.5
0.4
0.3

VN [m/s]

0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
0

20

40

60

80

100

% Chord

Figure 7.6. A comparison of RMS chord-normal fluctuating tip velocity


as determined from POD and BIV analyses. The compressor was
operating at 13,551 RPM and in deep stall. FEM mode 1; FEM
mode 2; FEM mode 3; first POD eigen-mode; second POD
eigen-mode; third POD eigen-mode. The gray box denotes the RMS
noise level.

was much higher when the compressor was operated under stalled conditions than
at design mass-flow. The RMS velocity for the first two FEM modes increased by
a factor of three each. Furthermore, excitation of the third FEM was detected. A
comparison between the velocity distribution using the FEM tip modes and the
POD eigen-modes shows good agreement between both analyses, confirming that
a large portion of the observed fluctuating tip velocity was due to blade vibration.
Table 7.2summarizes the RMS modal amplitudes and the maximum tip deflection for each mode. It can be seen that all three vibration modes were much
larger than the estimated noise. Additionally, a comparison with table 7.1 reveals
that the amplitude of the first two modes of vibration increased by 330% and 345%
respectively. Finally, it should be mentioned that in spite of this dramatic increase
in vibration amplitude, the maximum tip deflection due to blade vibration was

171

TABLE 7.2
COMPARISON OF RMS AMPLITUDES OF THE POD
EIGEN-MODES AND THE FEM EIGEN-MODES USING THE BIV
TECHNIQUE FOR A SHAFT SPEED OF 13,531 RPM OPERATING
AT STALLED CONDITIONS.

Mode Number

[m/s]

Max. Tip Deflection [m]

1.70 0.12

47 6

1.09 0.12

27 6

0.70 0.12

15 6

very small.

7.3 Conclusions
A novel method for the estimation of blade vibration amplitudes on rotating
turbomachinery blades has been presented. The method, termed Blade Image Velocimetry, utilized equipment common to a standard PIV system. The theory and
uncertainty analysis of the measurement technique were presented. A method for
the estimation of the noise corruption of the measured tip velocities was presented.
Additionally, it was shown how a POD analysis could independently validate the
results obtained from BIV.
The measurement technique was applied to a high-speed, low pressure axial
compressor. The accuracy of the velocity measurements was evaluated at low
shaft speed using a known shaft rotation rate and axial position. The max-

172

imum error in the two-component tip velocity measurement was found to be


2.1104 RT , or about 14 mm/s. It was also found that images can be acquired
using a single-frame, double-exposure approach instead of a double-frame, double exposure recording approach. This allows for the use of conventional digital
cameras, which are less expensive and higher in resolution.
The BIV technique was used to investigate the unsteady tip velocity of the
rotor operating at high speed and at two operating points. The first corresponded
to a high air mass flow through the machine, and a low level excitation of the first
blade eigen-mode was found. The RMS amplitude of this excitation was found to
be 0.53 m/s. The second operating point was at low air mass flow, corresponding
to compressor stall. Excitation of the first three blade eigen-modes was found.
The RMS amplitudes of the eigen-modes were 1.70, 1.09 and 0.70 m/s, whereas
the RMS of the measurement noise was found to be 0.14 m/s.

173

CHAPTER 8
CONCLUSIONS AND FUTURE WORK

8.1 Objectives of dissertation


The objective of this dissertation was to address two specific issues related
to axial turbomachinery durability. The first issue was the fundamental fluidstructure interactions between a vibrating blade and a high speed flow. The second
issue was the measurement of blade vibration on axial turbomachine rotors. A
discussion of the major results will be presented for both topics. Areas of future
research will also be suggested.

8.2 Investigation of fluid-structure interactions between an isolated compressor


blade and a high-speed flow
The first area that this dissertation addressed was the unsteady fluid-structure
interactions of a isolated compressor blade in a high speed flow. The motivation for
this work was to obtain a better understanding of the fluid-structure interactions
for a flexible blade in a high speed flow. The current theoretical models describing
these aeroelastic interactions are typically limited to subsonic flows because of the
significant non-linearity associated with shock waves on the blade surface. CFD
calculations, such as those described by Farhat et al. [22] and Vahdati et al. [61],
can provide good insight into transonic aeroelastic interactions. However, there
174

are also significant limitations to these studies which have precluded widespread
routine use on problems of practical interest.

8.2.1 Discussion of results


The flow about the blade at mid-span was measured using PIV. Five conditions
were investigated : incompressible, subsonic and transonic flow at 0 = 8 and
incompressible and transonic flow at 0 = 13 . It was found that at subsonic flow
conditions for 0 = 8 the steady flow field could be modeled with reasonable
accuracy using a linear theory. The flow at 0 = 13 and M = 0.16 had less
agreement with the linear theory. Additionally, the flow field far from the blade
surface was qualitatively similar to the observations at 0 = 8 and M = 0.16.
The unsteady flow due to the vibrating blade manifested in the wake region
and near the suction side of the blade. This suggests that the suction side aerodynamics may play an important role in the generation of unsteady forces. This
intuitively makes sense at high 0 ; boundary layer separation, which can dramatically reduce the lift produced by the blade, is more likely to occur on the suction
side as a result of large adverse pressure gradients. A variation in spatial extent
or fluid momentum deficit of the region of separated flow could result in large
variations in aerodynamic loads, resulting in blade vibrations.
The flow at transonic conditions was very unsteady at both angles of attack
investigated. Most of the velocity fluctuations were concentrated near the wake
and suction side of the blade. The wake structure was also different from the
observations at slower tunnel speeds. Specifically, a large increase in the chordnormal thickness was observed on the upper side of the wake (associated with the
suction side of the blade). Pockets of high and low momentum flow were observed

175

aft of the blade trailing edge. Note that anomalous, shock-free deceleration was
observed at these conditions, and so the PIV results should be interpreted as a
qualitative description of the flow.
The PIV measurements indicated that the flow field at transonic conditions
was significantly different than at subsonic conditions. This change most likely
influenced the mechanisms by which the blade exchanged energy with the flow.
The literature review presented in chapter 1 showed that shock-boundary layer
interaction may be a key mechanism that can result in large blade vibrations. According to Lee [44], shock oscillation typically occurs when the local Mach number
upstream of the shock was between 1.1 and 1.5. A steady panel method analysis
of the airfoil transformation predicts a peak Mach number of approximately 1.4
at 30% chord on the suction side of the airfoil at a free-stream Mach number of
M = 0.75, which implies that an oscillating shock could have been present [44].
The response of the blade was characterized into two types. The first, termed
the natural response, described the quasi-steady vibration of the blade as a result
of steady flow. The second response described the transient response of the blade
as a result of the steady flow and a mechanical impulse, and was quantified using
the modal damping ratio.
The natural response was dominated by the first two eigen-modes of the blade.
These modes, termed bending (f 1 = 349 Hz) and torsion (f 2 = 1540 Hz), were
sensitive to both Mach number and angle of attack. The RMS modal velocity for
the bending mode tended to increase with increasing Mach number and a local
maxima was observed at 0 = 8 and M 0.95. This local maxima occurred over
a narrow range of Mach numbers. The RMS modal velocity for the torsion mode
also increased with increasing Mach number and a local maxima was observed at

176

0 = 8 and M = 0.75. The natural response of the bending mode at 0 = 13


had a gradual increase in RMS modal velocity wit free-stream Mach number. The
local maxima at 0 = 8 and M = 0.95 was not observed at 0 = 13 . The torsion
RMS modal velocity had two distinct local maxima at M = 0.65 and M = 0.81.
These maxima also occurred over a narrow band of Mach numbers. All local
maxima in RMS modal velocity occurred above the critical Mach number of the
blade, which indicated that they were associated with transonic flow.
All local maxima in the RMS modal velocity were observed at transonic flow
conditions. The maxima for bending occurred when the free-stream velocity was
nearly sonic. This implied that a strong shock may have been present. Note that
an increase in 0 from 8 to 13 removed this maxima. This provides a unique
method for attenuating bending mode vibrations at transonic conditions for this
system. If possible, an increase in 0 could reduce bending mode vibrations. Note
that for axial compressors, the risk for surge due to an increase in incidence angle
must be weighed against the risk of bending vibrations.
The effect of increasing 0 from 8 to 13 resulted in increased torsion mode
vibration levels over a wide range of Mach numbers. Additionally, two large
local maxima were observed when 0 was increased. Recall that a single, low
amplitude maxima was observed for the torsion mode at 0 = 8 . All observed
torsion local maxima occurred at Mach numbers lower than the bending maxima
(0.65 M 0.81). Thus, the approach for mitigating high bending vibration
may result in high torsion vibrations.
A comparison between the measured response and predictions based on unsteady compressible thin airfoil theory revealed that the bending response was
under-predicted at low flow speeds and was over-predicted at high speeds for

177

both angles of attack investigated. In contrast, the torsion response was underpredicted for all conditions investigated.
The amplitude of the modal velocity was highly unsteady between 0.65
M 1 for both angle of attacks investigated. The vibration for both modes could
be described as a series of bursts and pulses of vibration. The bursts were
periods of high amplitude vibration, whereas the pulses were shorter in duration.
The pulses and bursts of vibration occurred at regular intervals over a large range
of transonic conditions for both angles of attack investigated. The duration of the
bursts and pulses were sensitive to both Mach number and angle of attack. Furthermore, it was discovered that these vibration patterns were correlated between
the first two blade modes. The orthogonal mode shapes and large difference in
natural frequency for the the modes indicated that the correlation may have been
a result of modal coupling through the fluid. The characteristic time scales over
which both modes were correlated were much longer than the vibration period for
either mode. This indicated a coupling between the modes through the unsteady
fluid dynamics.
The correlation between the bending and torsion modes may provide an advantage for flutter control strategies. The cross-correlation between the torsion
and bending modes indicates that a control strategy that targets one mode may
also result in control of the second mode for this system. Additionally, the relatively long time scales observed for the blade vibration indicate that an effective
control system may only need to operate intermittently to promote conditions
which attenuate blade vibrations.
The variation in damping was also investigated for the first two modes. The
bending modal damping ratio varied significantly between single impulses when

178

the free-stream tunnel speed was M > 0.60. The scatter in the damping ratio
is important for the aeroelastic design of turbomachinery blades because it was
so large. A large variation in aerodynamic damping can be incorporated into
aeromechanical design by providing sufficient design margin for modal damping.
The Mach-averaged aerodynamic contribution to damping for the bending
mode increased with increasing tunnel speed for subsonic flow. This variation
was in fair agreement with the model predictions using unsteady thin airfoil theory. A decrease in this damping was observed for transonic flow conditions, with
negative aerodynamic damping being observed at M > 0.95 when 0 = 4 . In
contrast, the Mach-averaged aerodynamic damping for the torsion mode was insensitive to tunnel speed at M > 0.4. This behavior deviated from the model
predictions.
The time-resolved vibration decay due to a mechanical impulse was also observed. It was found that the bending vibration could be described well by a
sinusoid with exponentially decaying (or growing) amplitude. This observation
was valid for both angle of attacks and all Mach numbers investigated. The torsion modal velocity amplitude after the mechanical impulse deviated from the
characteristic exponential decay.

8.2.2 Future directions


The system investigated in this series of experiments provided useful insight
into the fluid-structure interactions of an isolated compressor blade. The results
presented here suggest several areas of future research. Two areas will be reviewed. They are the complete fluid velocity mapping of the test section and the
investigation of cascade effects.

179

The PIV measurements indicated that a substantial change to the flow field
occurred for transonic conditions. Specifically, the presence of pockets of high and
low momentum flow were observed aft of the blade and did not appear to be correlated with the rest of the flow field. Obtaining insight into the physics associated
with these observations may yield additional insight into the blades response at
transonic flow conditions. The complete description of this flow field at a fixed
tunnel speed and blade angle of attack through experimental measurements shall
be referred to as fluid velocity mapping. Such a mapping could be accomplished
by a variety of measurement techniques. However, non-intrusive techniques (such
as Laser Doppler Velocimetry or PIV) may be suitable for this study because of
their minimal influence on the flow.
A second area for future work is the investigation of cascade effects. The experimental setup specifically targeted and isolated compressor blade to minimize
experiment complexity. This allowed for a straightforward investigation of the
fluid-structure interactions. A natural extension of this work would be to repeat
the experiment using a cascade. A flexible test blade would be installed in a
cascade of stiff blades. Ideally, only the flexible test blade would be allowed to
vibrate, thus providing insight into how the steady aerodynamics of a cascade influences the unsteady fluid-structure interactions. The use of stiff cascade blades
surrounding the flexible test blade is recommended to remove adjacent blade vibration velocities as a potentially uncontrolled variable. The results from this set
of experiments would help guide future studies of more complex systems, such as
fully flexible cascades and turbomachine rotor/stator stages. Furthermore, the
experimental data provided by these simple studies could provide well-defined
benchmark problems for theoretical and computational model development.

180

8.3 Blade vibration measurement technology


The second area that this dissertation addressed was the measurement of blade
vibration on rotors. Blade vibration measurements are critical to aeromechanical
design validation. The long service life of the engine and high vibration frequency
can result in high cycle fatigue, and so accurate measurements are necessary to
validate engine durability. These measurements can be difficult to obtain using
traditional approaches such as strain gages or Blade Tip Timing (BTT). A novel
measurement technique was developed that can provide a viable alternative to
either strain gages or BTT.

8.3.1 Discussion of results


A novel measurement technique was developed that used an off the shelf PIV
system. The PIV system obtained estimates of the velocity of the tip of a cantilevered blade, such as a typical axial turbomachine rotor blade. The theory by
which the tip motion may be related back to the blade modal amplitudes was
presented along with an uncertainty analysis.
This technique was distinct from Blade Tip Timing in that velocity estimates
were obtained over a large spatial area, allowing for discrimination of modes based
on their shapes at the blade tip. Blade tip timing is usually configured to measure
only one location along the blade chord, and must therefore rely upon the aliased
frequencies to determine modal amplitudes. This approach can yield ambiguous
results because the rate at which measurements are made is typically much lower
than the blade natural frequencies.
The BIV technique also differed from strain gage measurements in that it
indirectly inferred blade vibration from observations of the tip. Strain gages,

181

in contrast, can yield accurate, time-resolved estimates of blade vibration at a


single location along the blade. BIV, in contrast, provides a set of statistically
independent observations of the tip velocity, and so only a statistical description
of the blade vibration is currently available using BIV.
The BIV technique was validated using a series of bench-top experiments. The
camera and laser system were validated using a Laser Doppler Vibrometer, and
the accuracy of the system was found to agree well with the uncertainty analysis.
The theory, uncertainty analysis and bench-top validation were published in the
journal of Measurement Science and Technology [49].
The technique was also applied to a single stage, high speed axial compressor.
A second round of validation was performed at low speed using the magnetic
bearing system to oscillate the rotor axially. The BIV technique was able to track
the axial motion of the rotor well. The technique was then used to investigate
blade vibration at high shaft speed. Two operating conditions were considered.
The first was at design air mass flow and had resulted in low level response for
the first two modes. The vibration amplitudes measured were 0.53 0.13 m/s for
mode 1 and 0.31 0.13 m/s for mode 2. This corresponded to a blade vibration
whose maximum tip deflections were 15 7m and 8 7m respectively. The
second operating point was at low air mass flow, corresponding to compressor stall.
Excitation of the first three blade eigen-modes was found. The RMS amplitudes
were 1.70 0.14 m/s for mode 1, 1.09 0.14 m/s for mode 2 and 0.70 0.14
m/s for mode 3. The corresponding tip deflections were 47 6m, 27 6m and
15 6m respectively. Note that the accuracy of the technique was comparable
with conventional tip timing approaches.

182

8.3.2 Future directions


The measurement technique at its current state of development is sufficiently
mature to be used for blade vibration measurements on axial turbomachinery in
laboratory settings. However, there are still areas which could be targeted for
future research. The two major areas for future development are measurement of
degenerate modes, and the simultaneous measurement of fluid and structural
velocity.
Degenerate modes are classified in this context as two or more blade modes
which have the same tip projection. The bending modes of a flat plate are examples of degenerate modes. The first two bending modes of a flat plate can
be distinguished by observation of the location of the nodal lines of the mode.
However, the mode shape at the tip are identical between the two, and so it is
impossible to differentiate the two modes from tip motion alone. A solution to
this problem is an additional measurement of the span-wise distribution of blade
velocity. Measurements of both span-wise and chord-wise blade velocity distributions provide more information and reduce the probability of encountering a
degenerate mode significantly. The practical implementation of the secondary velocity measurement is less straightforward. A second optical system, similar in
configuration as conventional BIV but targeting either the blade leading or trailing edge could be used simultaneously with the current BIV setup. Alternatively,
a tomographic imaging approach (whereby a 3D volume is imaged using multiple
cameras) may yield an acceptable solution to this problem. Here, a tomographic
imaging approach could provide blade velocity measurements over the entire blade
surface which could eliminate the possibility of a degenerate mode.
A final area of research is the simultaneous measurement of fluid and structure

183

velocity. The benefits of such a measurement system were presented in chapter 2,


where the flutter of a thin flat plate in incompressible flow was linked to an unsteady flow separation over the suction side of the plate. These observations were
a direct result of phase-averaged BIV-PIV measurements using a single camera
and laser and offered a large amount of insight for a relatively simple measurement
system. These types of measurements can allow for the identification of critical
phenomena which can drive unsteady fluid-structure interactions.
There are some technical challenges which must be addressed before such measurements can be applied to high speed turbomachinery. One challenge is the
difference in t required for PIV compared with BIV measurements. High speed
flows require a t much shorter than those used for BIV. The mismatch is due
to the small vibration velocities encountered on the blade and the high fluid velocities. A short t is necessary for PIV to minimize loss of correlation errors
[54], due to particles moving out of the light sheet or field of view. A long t
is necessary for BIV to allow the imaging system to accurately measure blade
tip deflection. In addition, the optics of the illumination system have different
requirements between both BIV and PIV measurements. Nearly all the BIV measurements presented were illuminated with a cone of laser light projected along
the span of the blade. PIV measurements typically use a sheet of light to illuminate a single plane of particles. Note that the PIV measurement plane is typically
orthogonal to the blade span. Obtaining both a light sheet for fluid measurements
and a cone for blade tip measurements can be difficult to achieve using a single
illumination source.
The differences in illumination requirements suggests that two separate systems be employed: one optimized for the fluid velocity and a second one optimized

184

for the structural velocity measurements. Both systems would be synchronized to


some common trigger such that the fluid velocity and blade tip velocity could be
referenced to some common timing event. Such a measurement system is feasible
for axial turbomachinery using off-the-shelf hardware.

185

APPENDIX A
VARIATION IN BLADE DAMPING RATIO IN QUIESCENT AIR AS A
FUNCTION OF STATIC PRESSURE

A.1 Introduction
This appendix describes the investigation of the effect of static pressure on the
measured damping ratio of a cantilevered blade in quiescent air. This test was
performed to provide perspective on the relative contribution of both the quiescent
air and the fixture holding the blade to the observed damping ratio.

A.2 Experimental setup


The experiments were performed using a four-bar mechanism similar to the
unit used in the experiments described in chapter 5. The device is shown schematically in figure A.1. The general principle of operation can be described as follows. The impactor initially deflects the blade by a pre-determined amount. The
change in load when the impactor slips off the tip of the blade approximates a step
function. At moderate impactor speeds this loading approximates a mechanical
impulse.
All data presented here was acquired using a single point Laser Doppler
Vibrometer (LDV). Reflective tape was affixed to the surface of the blade to
improve measurement accuracy. The area of the tape was much smaller than the
186

Figure A.1. Schematic of impulsive loading mechanism and general


orientation with respect to the test blade.

planform area of the blade (the area ratio is 1 : 22 tape:blade). Consequently,


the effect of the reflective tape can be considered negligible. The accuracy of the
LDV is a function of the frequency of oscillation of the surface. The minimum
accuracy for all measurements presented here is 170m/s.
The geometry of the blade tested is shown in figure A.2. Note that the crosssection of the blade was thicker and the chord was shorter than the blade used
in chapters 3-5. As such, a direct correlation between the results presented here
and the observations of the damping ratio described in chapter 5 is not possible.
Recall that the objective of this appendix was to provide perspective on the relative
contribution of the quiescent air to the observed damping ratio.
Figure A.3 shows the shape of the first two eigenmodes of vibration for this
geometry. The cross-section of the blade is constant over the outer 75% of the span.
The blade was constructed using 6061-T6 aluminum. The blade was attached to
a steel block whose mass was O(102 ) greater than the blade itself. The steel block

187

(b)
10
8
6

z [mm]

4
2
0
2
4
6
8
10

10

x [mm]

Figure A.2. Geomtery of airfoil used in exploratory measurements. All


dimensions in mm. (a) Isometric view of blade geometry. (b) Blade cross
section at 0 = 0o . indicates camber line, is the approximate
location of the elastic axis for the cross-section.

188

Figure A.3. First two mode shapes of the blade. (a) Mode 1, f 1 = 1150
Hz. (b) Mode 2, f 2 = 3820 Hz.

189

and impulsive loading mechanism were placed within a vacuum chamber. The
structural response was measured using the single point LDV. The deflection of
the blade by the impactor was held constant throughout the test.

A.3 Results
Figure A.4 displays the damping ratio for the first two modes as a function of
static pressure. The effect of increasing static pressure resulted in a minor increase
in the damping ratio. The increase was slightly larger in mode 1 than in mode 2.
The damping ratio for both modes is extremely small. This result is consistent
with the results of Kielb and Abhari [39].
A cubic polynomial was fit to the data to determine the damping of the structure in a complete vacuum. The change in damping between vacuum and atmospheric pressure can provide an estimate of the aerodynamic damping due to
quiescent air. Table A.1 displays the measured damping ratios at atmospheric
pressure and the estimated damping ratio under vacuum conditions. It can be
seen that most of the damping was due to the fixture rather than the quiescent
air.

A.4 Conclusions
The damping ratio of a cantilevered blade was investigated as a function of
static pressure under quiescent conditions. It was found that the modal damping
ratio for the first two modes decreased as static pressure was decreased. The
damping ratio at vacuum conditions was estimated by polynomial extraction.
It was found that the change in damping ratio between vacuum conditions and

190

Figure A.4. Equivalent linear damping ratio of the aluminum blade as a


function of static pressure.  - Mode 1, - Mode 2.

TABLE A.1
MODAL DAMPING RATIO OF THE BLADE AT ATMOSPHERIC
PRESSURE AND VACUUM

Atmospheric Pressure

Vacuum

Mode 1

3.2 104

2.7 104

Mode 2

0.7 104

0.4 104

191

atmospheric pressure was O(105 ) for the first two modes.

192

APPENDIX B
FIRST 10 EIGEN-MODES OF THE ISOLATED COMPRESSOR BLADE

The first 10 eigen-modes of the blade used in these experiments are shown
in figures B.1 - B.10. The frequencies, which are included in the captions, are
repeated in table B.1 for quick reference to those observed in the autospectral
density plots.

193

||

0.3

Z [mm]

10

0.25

0.2

10
0.15

50
40
30

0.1

20
10
20

0
10

Y [mm]

0
20

20

X [mm]

0.05

Figure B.1. First eigen-mode of the blade. F0 = 349 Hz

Figure B.2. Second eigen-mode of the blade. F0 = 1, 540 Hz

194

Figure B.3. Third eigen-mode of the blade. F0 = 1, 767 Hz

Figure B.4. Fourth eigen-mode of the blade. F0 = 3, 804 Hz

195

Figure B.5. Fifth eigen-mode of the blade. F0 = 3, 985 Hz

Figure B.6. Sixth eigen-mode of the blade. F0 = 4, 990 Hz

196

Figure B.7. Seventh eigen-mode of the blade. F0 = 6, 976 Hz

Figure B.8. Eight eigen-mode of the blade. F0 = 8, 570 Hz

197

Figure B.9. Ninth eigen-mode of the blade. F0 = 8, 649 Hz

Figure B.10. Tenth eigen-mode of the blade. F0 = 10, 084 Hz

198

TABLE B.1
UNDAMPED NATURAL FREQUENCIES OF THE FIRST TEN
EIGEN-MODES OF THE BLADE AS COMPUTED BY FEA.

Mode #

Frequency [Hz]

349

1,540

1,767

3,804

3,985

4,990

6,976

8,570

8,649

10

10,084

199

APPENDIX C
INTEGRAL TIME SCALE

C.1 Introduction
The results presented in chapter 4 indicate that the characteristics of the response of the blade for the first two modes varied with free-stream Mach number.
The auto-correlation coefficients of the fluctuations in the modal velocity envelope also varied with Mach number. A way to characterize the changes in the
correlation coefficients is by observation of the integral time scale.

C.2 Approach
The integral time scale was defined as
Z

t?C

rxx ( )d

(C.1)

t?C

where rxx is the autocorrelation coefficient of the fluctuations in the modal velocity
envelope and t?C is the non-dimensional time where rxx (t? ) = 0.5. Recall that t?
represents a characteristic through-flow time, t? tU /c. This time-scale can
indicate changes in the non-dimensional time over which the envelope fluctuations
first become de-correlated. These in turn can yield insight into the time-scales
associated with the fluid-structure interactions exciting the blade.

200

(b)

300

300

250

250

Integral time scale

Integral time scale

(a)

200

150

100

50

0
0

200

150

100

50

0.2

0.4

0.6

0.8

0
0

Mach Number

0.2

0.4

0.6

0.8

Mach Number

Figure C.1. Integral time-scale of the first two modes at 0 = 8 . Time


scale calculated over a 0.6 second window; ensemble-averaged
time-scale. (a) mode 1; (b) mode 2.

C.3 Results
Figure C.1 shows the integral time scale of the first two modes of vibration as
a function of Mach number at 0 = 8 . The integral time scale for both modes
were much higher than the characteristic through-flow time, indicating that the
vibration for each mode was correlated over a long time-scale. The time-scale for
both modes increased with increasing Mach number. Additionally, the time-scale
over a 0.6 second window varied significantly. This scatter indicated that the
envelope fluctuations were not statistically converged over the 0.6 second window.
This was consistent with the observations of the RMS modal velocity presented in
chapter 4. This variation was especially large for the bending mode near M = 1.
The nearly linear growth in the ensemble-averaged time-scale with Mach number for both modes at M < 0.60 indicated that the characteristic decorrelation
time for the envelope fluctuations was proportional to Mach number. This may

201

(b)

300

300

250

250

Integral time scale

Integral time scale

(a)

200

150

100

50

0
0

200

150

100

50

0.2

0.4

0.6

0.8

0
0

Mach Number

0.2

0.4

0.6

0.8

Mach Number

Figure C.2. Integral time-scale of the first two modes at 0 = 13 .


Time scale calculated using a 0.6 second window; ensemble-averaged
time-scale. (a) mode 1; (b) mode 2.

indicate that the duration of the characteristic pulses and beats observed in
the time-series data varied proportionally with Mach number. Note that a local
maxima in the integral time scale for the first mode was observed at M = 0.95.
These flow conditions corresponded to a local maxima in the first modes RMS
velocity; which was characterized by long periods of high vibration.
Figure C.2 shows the integral time scale of the first two modes of vibration as
a function of Mach number at 0 = 13 . The following differences between these
results and the 0 = 8 results may be observed. First, for the bending mode the
individual time scales at M > 0.90 had less variation at 0 = 13 than at 0 = 8 .
The ensemble-averaged time-scale at these conditions for the bending mode was
also lower at 0 = 13 than at 0 = 8 . The variation at M < 0.9 was higher
for 0 = 13 . Finally, it should be noted that the variation in the time-scale for
the second mode at 0 = 13 had a local maxima near M = 0.67 followed by a
local minima at M = 0.75. This variation was observed for both the individual

202

(b)

250

250

200

200

Integral time scale

Integral time scale

(a)

150
100
50
0
0

150
100
50

0.2

0.4

0.6

0.8

0
0

0.2

Mach Number

0.4

0.6

0.8

Mach Number

Figure C.3. Ensemble-averaged integral time-scale of the first two modes


as a function of Mach number. (a) mode 1; (b) mode 2. 0 = 8 ;
0 = 13 . Error bars denote ensemble RMS.

estimates and the ensemble averaged data, and was not observed at 0 = 8 . It
is interesting to note that the time-series of modal velocity for the second mode
at M = 0.65 were characterized by long periods of high vibration, whereas the
the modal velocity at M = 0.81 was characterized by many short periods of high
vibration.
Figure C.3 provides a graphical comparison of the ensemble-averaged integral
time scale for both modes at both angles of attack investigated as a function
of Mach number. The following may be observed. First, the time-scale for the
bending mode at 0 = 13 was equal to or higher than 0 = 8 when M < 0.90.
It may be concluded that an increase in angle of attack resulted in an increase
in the characteristic time scale of the fluid-structure interactions exciting the first
mode. The variation with Mach number for the first mode was similar, but not
identical between both angles of attack investigated.
The variation in integral time-scale for the second mode with Mach number
and angle of attack is shown in figure C.3 (b). In general, the ensemble-averaged
203

time-scales for the second mode were shorter than those observed for the first
mode. With the exception of 0.65 M 0.80, the integral time scales had
similar magnitudes between both angles of attack investigated. An increase in the
ensemble-RMS may be observed for 0 = 13 relative to 0 = 8 .

C.4 Conclusions
The fluctuations in the modal velocity envelope were investigated by considering the characteristic integral time scale of the vibration. This time scale was
defined using the autocorrelation coefficient of the fluctuations in the modal velocity envelope. This time scale can provide additional insight into the characteristics
of the blade vibration.
The integral time scale calculated from 0.6 second segments of vibration data
were not statistically converged. The ensemble-averaged integral time scales increased with increasing Mach number for the first two modes at 0 = 8 . A local
maxima was observed for the first mode at M = 0.95 and 0 = 8 . This implied
that the duration of the characteristic pulses and bursts were highest at these
conditions. This observation was consistent with the time-series data presented
in chapter 4. A local maxima and minima of the integral time scale was observed
for the second mode at 0 = 13 and at M = 0.65 and M = 0.75 respectively.
This was consistent with the time-series data, which indicated a vibration pattern
at M = 0.65 characterized by long periods of high vibration interspersed among
short periods of low vibration. Additionally, the time-series at M = 0.81 indicated
that the vibration could be described as a series of high amplitude, short duration
periods of vibration.

204

BIBLIOGRAPHY

1. A. Abdel-Rahim, F. Sisto, and S. Thangam. Computational study of stall


flutter in linear cascades. J. of Turbomachinery, 115:157-166, 1993.
2. T. Akai and H. Atassi. Aerodynamic and aeroelastic characteristics of oscillating loaded casscades at low Mach number, Part 2: Stability and flutter
boundaries. J. Engineering for Power, 102:352-356, 1980.
3. B. Al-Bedoor. Blade vibration measurement in turbomachinery: current status. Shock and vibration digest, 34(6):455-461, 2002.
4. Y. Alayli, S. Topcu, D. Wang, R. Dib, and L. Chassagne. Applications of a
high accuracy optical fiber displacement sensor to vibrometry and profilometry. Sensors and Actuators A, 116:85-90, 2004.
5. H. Atassi. Unsteady aerodynamics of vortical flows : early and recent developments. In K.-Y. Fung, editor, Symposium on aerodynamics and aeroacoustics,
pages 121-172. World Scientific, 1993.
6. H. Atassi and T. Akai. Aerodynamic and aeroelastic characteristics of oscillating loaded cascades at low Mach number, Part 1: Pressure distribution,
forces and moments. J. Engineering for Power, 102:344-351, 1980.
7. A. Balakrishnan. Unsteady aerodynamics - subsonic compressible inviscid
case. Contractor Report 206583, NASA, 1990.
8. J. Bell and S. Rothberg. Laser vibrometers and contacting transducers, target
rotation and six degree-of-freedom vibration : what do we really measure? J.
Sound and Vibration, 237(2):245-261, 2000.
9. J. Belz and H. Hennings. Unsteady Aerodynamics, Aeroacoustics and Aeroelasticity of Turbomachines, chapter Experimental flutter investigations of an
annular compressor cascade: influence of reduced frequency on stability, pages
77-91. Springer, 2006.
10. J. Bendat and A. Piersol. Random Data. Wiley Interscience, 2000.

205

11. R. Bisplinghoff, H. Ashley, and R. Halfman. Aeroelasticity. Dover, 1996.


12. P. Bowles. Wind tunnel experiments on the effect of compressibility on the
attributes of dynamic stall. Dissertation, The University of Notre Dame, 2012.
13. K. Breuer, J. Bird, G. Han, and J. Westin. Infrared Diagnostics for the measurement of fluid and solid motion in micromachined devices. In 2001 ASME
Mechanical Engineering Congress and Exposition, New York, NY, November,
2001. ASME.
14. D. Buffum, V. Capece, A. King, and Y. El-Aini. Oscillating cascade aerodynamics at large mean incidence. Technical Memo 107247, NASA, 1996.
15. A. Burner, T. Liu, and R. DeLoach. Uncertainty of videogrammetric techniques used for aerodynamic testing. In 22nd Aerodynamic Measurement Technology and Ground Testing, St. Louis, MO, June 2002. AIAA. Paper No.
2002-2794.
16. J. Cameron. Stall inception in a high-speed axial compressor. Dissertation,
The University of Notre Dame, 2007.
17. I. Carrington, J. Wright, J. Cooper, and G. Dimitriadis. A comparison of blade
tip timing data analysis methods. Proc. Instn. Mech. Engrs., 125G:301-312,
2001.
18. T. Chen, P. Vasanthakumar, and L. He. Analysis of unsteady blade row
interaction using nonlinear harmonic approach. J. Propulsion and Power,
17(3):651-658, 2002.
19. F. Claveau, P. Fortier, S. Lord, and D. Gingras. Mechanical vibration analysis
using an optical sensor. In Electrical and Computer Engineering, pages 876879. IEEE, 1996.
20. R. Dib, Y. Alayli, and P. Wagstaff. A broadband amplitude-modulated fibre
optic vibrometer with nanometric accuracy. Measurement, 35:211-219, 2004.
21. E. Dowell, editor. A modern course in aeroelasticity. Kluwer, 4th edition,
2005.
22. C. Farhat, M. Lesoinne, and P. LeTallec. Load and motion transfer algorithms
for fluid/structure interaction problems with non-matching discrete interfaces:
momentum and energy conservation, optimal discretization and application
to aeroelasticity. Computer methods in applied mechanics and engineering,
157:95-114, 1998.

206

23. B. Feeny and R. Kappagantu. On the physical interpretation of Proper Orthogonal Modes in vibrations. J. of Sound and Vibration, 211(4):607-616,
1998.
24. P. Ferrand, J. Boudet, and J. Caro. Analysis of URANS and LES capabilities
to predict vortex shedding for rods and turbines. In K. Hall, R. Kielb, and J.
Thomas, editors, Unsteady Aerodynamics, Aeroacoustics and Aeroelasticity of
Turbomachines, pages 381-393. Springer, 2006.
25. J. Gallego-Garrido, G. Dimitriadis, and J. Wright. A class of methods for
the analysis of blade tip timing data from bladed assemblies undergoing simultaneous resonances - part 1: theoretical development. Int. J. of Rotating
Machinery, 2007, 2007. Article ID 27247.
26. J. Gill, V. Capece, and R. Fost. Experimental methods applied in a study of
stall flutter in an axial flow fan. Shock and vibration,11:597-613, 2004.
27. M. Goldstein, W. Braun, and J. Adamczyk. Unsteady flow in a supersonic
cascade with strong in-passage shocks. J. Fluid Mech., 83:569-604, 1977.
28. J. Gomes and H. Lienhart. Fluid-Structure Interaction, volume 53, chapter
Experimental study on a fluid-structure interaction reference test case, pages
356-370. Springer, 2006.
29. S. Gorrell and W. Copenhaver. DPIV measurements of the flow field between a
transonic rotor and an upstream stator. In K. Hall, R. Kielb and J. Thomas,
editors, Unsteady Aerodynamics, Aeroacoustics and Aeroelasticity of Turbomachines, pages 505-519. Springer, 2006.
30. B. Gr
uber and V. Carstens. The impact of viscous effects on the aerodynamic
damping of vibrating transonic compressor blades - a numerical study. J. of
Turbomachinery, 123:409-417, 2001.
31. K. Hall, J. Thomas, and W. Clark. Computation of unsteady nonlinear flows
in cascades using a harmonic balance technique. AIAA, 40(5):879-886, 2002.
32. C. Harris and A. Piersol. Harris Shock and Vibration Handbook. McGraw-Hill,
5th edition, 2002.
33. S. Heath, and M. Imregun. A survey of blade tip-timing measurement techniques for turbomachinery vibration. J. of Gas Turbines and Power, 120:784791, 1998.
34. S. Heath. A new technique for identifying synchronous resonances using tiptiming. J. Gas Turbines and Power, 122:219-225, 2000.

207

35. F. Hild and S. Roux. Digital image correlation: from displacement measurement to identification of elastic properties - a review. Strain, 42:69-80, 2006.
36. D. Hoying, C. Tan, H. Vo, and E. Greitzer. Role of blade passage flow
structures in axial compressor rotating stall inception. J. of Turbomachinery,
121:735-742, 1999.
37. M. Jakobsen, T. Dewhirst, and C. Greated. Particle image velocimetry for
predictions ofacceleration fields and force within fluid flows. Meas. Sci. Tech.,
8(12):1502-1516, 1997.
38. D. Japikse and N. Baines. Introduction to Turbomachinery. Concepts ETI,
1994.
39. J. Kielb and R. Abhari. Experimental study of aerodynamic and structural
damping in a full scale rotating turbine. J. Gas Turbines and Power, 125:102,
2003.
40. J. Kielb, R. Abhari, and M. Dunn. Experimental and numerical study of
forced response in a full-scale rotating turbine. In ASME Turbo Expo, New
Orleans, LA, June 2001. ASME. 2001-GT-0263.
41. M. Kobayashi. Annular cascade study of low back-pressure supersonic fan
blade flutter. J. of Turbomachinery, 112:768-777, 1990.
42. B. Lakshminarayana. Fluid dynamics and heat transfer of turbomachinery,
Wiley, 1996.
43. C. Lawson and P. Ivey. Turbomachinery Blade Vibration Amplitude Measurement Through Tip Timing with Capacitance Tip Clearance Probes. Sensors
and Actuators, 118:14, 2005.
44. B. Lee. Self-sustained shock oscillations on airfoils at transonic speeds.
Progress in Aerospace Sciences, 37:147-196, 2001.
45. J. Lin and K. Illif. Aerodynamic lift and moment calculations using a closedform solution of the Possio equation. Technical Memo 209019, NASA, 2000.
46. T. Liu, L. Cattafesta, and R. Radeztsky. Photogrammetry Applied to WindTunnel Testing. AIAA, 38(6), 2000.
47. X. Liu and J. Katz. Instantaneous pressure and material acceleration measurements using a four-exposure PIV system. Experiments in Fluids, 41(2):227240, 2006.
48. J. Marshall and M. Imregun. A review of aeroelasticity methods with emphasis
on turbomachinery applications. J. of Fluids and Structures, 10:237-267, 1996.
208

49. P. Mikrut, M. Bennington, S. Morris, and J. Cameron. Blade image velocimetry: development and uncertainty analysis. Meas. Sci. Tech., 21, 2010.
50. T. Nicholas. Critical issues in high cycle fatigue. Int. J. of Fatigue, 21:S221,
1999.
51. R. Pappa, J. Black, J. Blandino, T. Jones, P. Danehy, and A. Dorrington. Dotprojection photogrammetry and videogrammetry of Gossamer space structures. J. Spacecraft and Rockets, 40(6):858-867, 2003.
52. M. Platzer and F. Carta, editors. AGARD manual on Aeroelasticity in
axial-flow turbomachines. Volume 1: unsteady turbomachinery aerodynamics.
NATO, 1987. AGARDograph No. 298.
53. M. Platzer and F. Carta, editors. AGARD manual on Aeroelasticity in axialflow turbomachines. Volume 2: structural dynamics and aeroelasticity. NATO,
1987. AGARDograph No. 298.
54. M. Raffel, C. Willert, S. Wereley, and J. Kompenhans. Particle image velocimetry : a practical guide. Springer, 2nd edition, 2007.
55. D. Rival and C. Tropea. Characteristics of pitching and plunging airfoils under
dynamic-stall conditions. J. of Aircraft, 47(1), 2010.
56. A. Sanders, K. Hassan, and D. Rabe. Experimental and numerical study of
stall flutter in a transonic low-aspect ratio fan blisk. J. of Turbomachinery,
126:166-173, 2004.
57. C. Sieverding, T. Arts, R. Denos, and J.-F. Brouckaert. Measurement techniques for unsteady flows in turbomachines. Experiments in fluids, 28:285-321,
2000.
58. A. Srinivasan. Flutter and resonant vibration characteristics of engine blades.
J. of Engineering for Gas Turbines and Power, 119:742-775, 1997.
59. H. Thermann and R. Niehuis. Unsteady Navier-Stokes simulation of a transonic flutter cascade near-stall conditions applying algebraic transition models. J. of Turbomachinery, 128:474-483, 2006.
60. C. Tropea, A. Yarin, and J. Foss, editors. Springer handbook of experimental
fluid mechanics. Springer, 2007.
61. M. Vahdati, A. Syama, J. Marshall, and M. Imregun. Mechanisms and prediction methods for fan blade stall flutter. J. of Propulsion and Power,
17(5):1100-1108, 2001.

209

62. J. Verdon and J. Caspar. A linearized unsteady aerodynamic analysis for


transonic cascades. J. Fluid Mech., 149:403-429, 1984.
63. S. Walker. Experiments characterizing nonlinear shear layer dynamics in a
supersonic rectangular jet undergoing screech. Dissertation, The University of
Notre Dame, 1997.
64. M. Wernet. PIV for turbomachinery applications. Technical Memo 107525,
NASA, 1997.
65. J. Westerweel. Theoretical analysis of the measurement precision in particle
image velocimetry. Exp. in Fluids, 29(7):3, 2000.
66. M. Zellinski and G. Ziller. Noncontact vibration measurements on compressor
rotor blades. Meas. Sci. Tech., 6:847-856, 2000.

This document was prepared & typeset with pdfLATEX, and formatted with
nddiss2 classfile (v3.0[2005/07/27]) provided by Sameer Vijay.

210

Anda mungkin juga menyukai