Copyright

by
Daiju Uehara
2019
 The Dissertation Committee for Daiju Uehara
certifies that this is the approved version of the following dissertation:

Estimation of Helicopter Rotor Loads from Blade

Structural Response

Committee:

Jayant Sirohi, Supervisor

David B. Goldstein

Krishnaswa Ravi-Chandar

Jeffrey K. Bennighof

Mahendra J. Bhagwat
Estimation of Helicopter Rotor Loads from Blade
Structural Response

by

Daiju Uehara

DISSERTATION
Presented to the Faculty of the Graduate School of
The University of Texas at Austin
in Partial Fulfillment
of the Requirements
for the Degree of

DOCTOR OF PHILOSOPHY

THE UNIVERSITY OF TEXAS AT AUSTIN

December 2019
Dedicated to my family, for their endless love, support, and encouragement.
 Acknowledgments

First of all, I would like to express my sincere gratitude to my academic

advisor and mentor in life, Prof. Jayant Sirohi. On the first day I came to
Austin, I was a young Japanese kid who had no experience of living outside
Japan, studying rotorcraft, and even speaking English. Even so, from the first
day to the end of my PhD program, Prof. Sirohi has been patient, and always
listened to me. If he had not been my advisor, I would have gone back to
Japan by the end of the first year without earning any degree. He shared
a variety of his personal stories and experiences from the old-days memories
when he was a young PhD student just like myself. These stories taught me
that even a great professor and teacher, Prof. Sirohi, had gone through tough
challenges similar to those that I have faced during the graduate school. His
advice, as an academic advisor and as a person who had overcome the similar
challenges and achieved success, helped me in all the time.

In addition to my advisor, I would like to acknowledge the rest of

the dissertation committee: Prof. David Goldstein, Prof. K. Ravi-Chandar,
Prof. Jeffrey Bennighof, and Dr. Mahendra Bhagwat, for their invaluable
suggestions and comments to my thesis and during the proposal and defense.
Their broad and critical comments to my research opened up new horizons
that I would have never come up with by myself.

v
 My sincere thanks also goes to Dr. Christopher Cameron and Dr.
Anand Carpatne, who were my senior PhD students in the research group.
Chris and Anand provided significant support not only for academic, research-
related topics, but also for my private life. I do not know how many times
I asked questions to them, how many times I had drink with them, and how
many times we laughed together. I am so lucky that I had Chris and Anand
as my seniors.

I would like to thank my research collaborators: Dr. Roland Feil, Dr.

Juergen Rauleder, and Dr. Larry Matthies. Roland and Juergen provided de-
tailed, thorough understanding of rotorcraft dynamics from the perspective of
numerical modeling, and improved the quality of our collaborative conference
proceedings and peer-reviewed journal publications. Larry gave me an intern-
ship opportunity at NASA JPL in 2018 summer, and provided insights to the
collaborative work, mission modeling of small VTOL aircraft for exploration
of Titan. Thanks to these collaborators, I was able to write and publish 5
research papers in total, based on these collaborations.

I thank my fellow lab colleagues: Young-Joon, Jason, Sergio, Palash,

Marc, Krystal, Patrick, Chloe, and Matt, for the countless discussions and
meetings, for the informal conversations, and for all the fun we have experi-
enced together in the last five and half years. Besides the colleagues in the
research group, I would also like to appreciate all of my friends who I met at
UT Austin. I want to keep in touch with all of them for the rest of my life.

Lastly, of course, I would like to thank my family. My parents and

vi
sisters have always supported me with love. At every turning point of my life,
including when I decided to leave my home town Okinawa and go to Tokyo
after high school, and when I decided to leave my home country Japan and go
to the United States after undergraduate, my parents just encouraged me to
do so. They did not show me how they felt in their heart at all, just gave the
push I needed. I just want to say, thank you.

vii
 Estimation of Helicopter Rotor Loads from Blade
Structural Response

Publication No.

Daiju Uehara, Ph.D.

The University of Texas at Austin, 2019

Supervisor: Jayant Sirohi

Measuring the load distribution along a helicopter rotor blade has been
one of the most challenging tasks in experimental aeromechanics. Conven-
tional loads measurements with on-blade instrumentation, such as pressure
transducers for airloads and strain gages for structural loads, require the ex-
perimentalist to overcome a large number of technical barriers; for example,
sensor integration to the rotor blade structure, sensor failure due to strong
centrifugal forces, and influence of sensor installation on rotor blade dynam-
ics. The goal of this dissertation is to develop a new, combined experimental
and theoretical methodology to estimate helicopter rotor loads without using
these conventional on-blade sensors.

The rotor loads estimation methodology begins with the measurement

of blade structural deformation measurements using non-contact, optical, time-
resolved Digital Image Correlation (DIC). The time-resolved DIC technique

viii
successfully showed its capability of measuring the three-dimensional deforma-
tion time history of a rotating blade for both a small- and a large-scale rotor
in hover. The modal properties (natural frequencies, mode shapes, damping
ratios, and modal coordinates) of the blade in the rotating-frame were then ex-
tracted from the deformation time history using Natural Excitation Technique
- Eigensystem Realization Algorithm (NExT-ERA) and Complexity Pursuit
(CP), which are operational modal analysis (OMA) algorithms. The first three
modes were identified by the OMA algorithms and well correlated with a nu-
merical model. Rotor loads were then finally estimated based on the measured
deformations and blade modal characteristics.

Having validated the present approach incrementally with measure-

ments performed on rotors at different scales, configurations, and operating
conditions, the current study estimated the spanwise lift distribution and in-
tegrated thrust at the hub for a 2 m-diameter, two-bladed, isolated single rotor
in hover. Due to a lack of participating modes (only the first and second flap
modes), the estimated sectional lift distribution did not capture the lift loss
typically observed at regions of the blade tip and induced by trailing tip vor-
tices. Nevertheless, the mean value of the estimated thrust at the rotor hub
was within 5% of the measured value for all the operating conditions.

ix
 Table of Contents

Acknowledgments v

Abstract viii

List of Tables xiii

List of Figures xiv

Chapter 1. Introduction 1
1.1 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Optical deformation measurement . . . . . . . . . . . . 6
1.2.1.1 Grid method . . . . . . . . . . . . . . . . . . . 6
1.2.1.2 Point Tracking . . . . . . . . . . . . . . . . . . 7
1.2.1.3 Digital Image Correlation . . . . . . . . . . . . 9
1.2.2 Operational Modal Analysis . . . . . . . . . . . . . . . . 11
1.2.3 Rotor loads estimation from structural response . . . . . 15
1.3 Present approach . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4 Contribution of the Current Research . . . . . . . . . . . . . . 19
1.5 Organization of the Dissertation . . . . . . . . . . . . . . . . . 20

Chapter 2. Theoretical Development 22

2.1 Digital Image Correlation . . . . . . . . . . . . . . . . . . . . . 22
2.1.1 Blade coordinate system . . . . . . . . . . . . . . . . . . 23
2.1.2 2D DIC . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.1.3 3D DIC . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2 Operational Modal Analysis . . . . . . . . . . . . . . . . . . . 27
2.2.1 NExT-ERA . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.2 Complexity Pursuit . . . . . . . . . . . . . . . . . . . . 34

x
 2.3 Rotor loads estimation . . . . . . . . . . . . . . . . . . . . . . 38
2.3.1 Inertial loads . . . . . . . . . . . . . . . . . . . . . . . . 39
2.3.2 Aerodynamic loads . . . . . . . . . . . . . . . . . . . . . 42
2.3.3 Numerical experiment . . . . . . . . . . . . . . . . . . . 45

Chapter 3. Experimental Setup and Procedure 51

3.1 Hover Test Stand . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.1.1 Extremely Flexible Rotor (small-scale test bench) . . . . 52
3.1.1.1 Rotor configuration . . . . . . . . . . . . . . . . 53
3.1.1.2 Data acquisition . . . . . . . . . . . . . . . . . 55
3.1.2 Single-Bladed or Two-Bladed, Isolated Single or CCR Ro-
tor (large-scale test bench) . . . . . . . . . . . . . . . . 56
3.1.2.1 Drive system . . . . . . . . . . . . . . . . . . . 56
3.1.2.2 Rotor configuration . . . . . . . . . . . . . . . . 58
3.1.2.3 Rotor blade . . . . . . . . . . . . . . . . . . . . 59
3.1.2.4 Rotor hub . . . . . . . . . . . . . . . . . . . . . 61
3.1.2.5 Blade root pitch angle sensor . . . . . . . . . . 63
3.1.2.6 Control system . . . . . . . . . . . . . . . . . . 66
3.1.2.7 Fixed-frame instrumentation . . . . . . . . . . . 72
3.1.2.8 Data acquisition . . . . . . . . . . . . . . . . . 74
3.2 DIC Deformation Measurement . . . . . . . . . . . . . . . . . 74
3.2.1 Small-scale rotor . . . . . . . . . . . . . . . . . . . . . . 74
3.2.2 Large-scale rotor . . . . . . . . . . . . . . . . . . . . . . 81
3.2.3 DIC post-processing . . . . . . . . . . . . . . . . . . . . 85
3.3 Test Envelope and Measurement Procedure . . . . . . . . . . . 87
3.3.1 Test campaign 1 . . . . . . . . . . . . . . . . . . . . . . 88
3.3.2 Test campaign 2 . . . . . . . . . . . . . . . . . . . . . . 89
3.3.3 Test campaign 3 . . . . . . . . . . . . . . . . . . . . . . 91

xi
Chapter 4. Results and Discussion 97
4.1 Rotor blade deformation measurement . . . . . . . . . . . . . . 97
4.1.1 Small-scale rotor . . . . . . . . . . . . . . . . . . . . . . 99
4.1.2 Large-scale rotor . . . . . . . . . . . . . . . . . . . . . . 104
4.1.2.1 Single-bladed CCR rotor (Test campaign 2) . . 105
4.1.2.2 Two-bladed single rotor (Test campaign 3) . . . 111
4.2 Rotor Blade modal characteristics identification . . . . . . . . 121
4.2.1 Small-scale rotor blade characteristics . . . . . . . . . . 122
4.2.1.1 Modal frequency . . . . . . . . . . . . . . . . . 123
4.2.1.2 Mode shapes . . . . . . . . . . . . . . . . . . . 127
4.2.1.3 Modal damping . . . . . . . . . . . . . . . . . . 136
4.2.2 Large-scale rotor blade characteristics . . . . . . . . . . 138
4.2.2.1 Random excitation (Test campaign 2) . . . . . 139
4.2.2.2 Step function excitation test (Test campaign 3) 142
4.3 Rotor loads measurement and estimation . . . . . . . . . . . . 149
4.3.1 CCR rotor loads measurement . . . . . . . . . . . . . . 150
4.3.2 Two-bladed single rotor loads measurement and estimation154
4.3.2.1 Test condition 1 . . . . . . . . . . . . . . . . . . 154
4.3.2.2 Test condition 2 . . . . . . . . . . . . . . . . . . 163
4.3.2.3 Test condition 3 . . . . . . . . . . . . . . . . . . 172

Chapter 5. Summary, Conclusions, and Future Work 183

5.1 Summary and Conclusions . . . . . . . . . . . . . . . . . . . . 183
5.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188

Bibliography 193

Vita 206

xii
 List of Tables

3.1 Rotor system parameters . . . . . . . . . . . . . . . . . . . . . 54

3.2 Summary of rotor parameters for the large-scale hover test stand 59
3.3 Summary of the curve-fit coefficients to the AMR sensor output
signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.4 Specifications of the lead-screw servo actuator . . . . . . . . . 69
3.5 Summary of three test campaigns . . . . . . . . . . . . . . . . 88
3.6 Summary of the test matrix for Test campaign 1 . . . . . . . . 89
3.7 Summary of operating conditions for Test campaign 2 . . . . . 91
3.8 Summary of operating conditions for Test campaign 3 . . . . . 92
3.9 Summary of the design parameters for digital low-pass filters . 94

4.1 Recap: Summary of the test matrix for Test campaign 1 . . . 99

4.2 Summary of the MAC values calculated from the measured and
predicted mode shapes, obtained at θ0 = 30◦ . . . . . . . . . . 135
4.3 Summary of modal damping ratios for different rotational speeds
at 0◦ root pitch angle . . . . . . . . . . . . . . . . . . . . . . . 137
4.4 Summary of the rotating natural frequencies and MAC values
calculated from the measured and predicted mode shapes. . . 142
4.5 Parameters used for the spectrogram analysis . . . . . . . . . 147
4.6 Summary of the experimentally and numerically obtained modal
frequencies and MAC values calculated from the measured and
predicted mode shapes for Test campaign 2 and 3 . . . . . . . 149
4.7 Recap: Summary of operating conditions for Test campaign 3 154
4.8 Summary of cutoff frequencies for low-pass filtering process for
Test campaign 3 . . . . . . . . . . . . . . . . . . . . . . . . . . 174

xiii
 List of Figures

1.1 Comparison between images taken during static display and

flight operation of the CH-53 Sea Stallion [1, 2] . . . . . . . . 2
1.2 An example of complex on-blade strain gage instrumentation . 3

2.1 Blade coordinate system . . . . . . . . . . . . . . . . . . . . . 23

2.2 A schematic of subset search process . . . . . . . . . . . . . . 25
2.3 A schematic of the basic 3D DIC principle . . . . . . . . . . . 27
2.4 A calculation scheme of the 3D displacement vector . . . . . . 28
2.5 Flow chart of the OMA algorithms . . . . . . . . . . . . . . . 38
2.6 Diagram of rotor blade forces . . . . . . . . . . . . . . . . . . 39
2.7 Unprocessed and regularized flapwise blade deformation . . . . 42
2.8 Measured flapwise blade deformation and 3rd-order polynomial
curve fit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.9 A flowchart of the whole approach proposed in this dissertation 45
2.10 Comparison between the original and estimated lift distribution
over the finite element 1D rotating beam . . . . . . . . . . . . 47
2.11 Integrated force comparison between the original and recon-
structed loads . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.12 Comparison between the original and estimated loads as a func-
tion of the number of participating modes . . . . . . . . . . . 50

3.1 Side view of the rotor test article . . . . . . . . . . . . . . . . 53

3.2 The hub and the flexible blade with the fluorescent paint and
random speckle pattern . . . . . . . . . . . . . . . . . . . . . . 55
3.3 The large-scale rotor test stand . . . . . . . . . . . . . . . . . 57
3.4 Side view of the transmission system . . . . . . . . . . . . . . 58
3.5 An exploded view of the rotor blade composite laminate . . . 60
3.6 Primary components of the rotor hub . . . . . . . . . . . . . . 62
3.7 A cross-section view of the blade grip assembly . . . . . . . . 63
3.8 Dynamic response of upper and lower rotor load cells . . . . . 64

xiv
3.9 A schematic of the AMR sensor installation on the blade grip
assembly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.10 The AMR sensor characteristics . . . . . . . . . . . . . . . . . 67
3.11 A drawing of the hydraulic servo actuator assembly . . . . . . 70
3.12 Rotor pitch angle control system assembly with the hydraulic
servo actuators . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.13 A diagram of pitch angle control system for the hydraulic actu-
ation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.14 The setup for small-scale rotor blade deformation measurement. 75
3.15 A schematic of the small-scale DIC measurement setup . . . . 76
3.16 Unprocessed images of the rotor strobed at 16 azimuthal loca-
tions, showing complete rotor disk at 1200 RPM. . . . . . . . 79
3.17 Camera calibration target . . . . . . . . . . . . . . . . . . . . 80
3.18 Vector map processed by DIC for the 30◦ root pitch rotor, spun
at 1500RPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.19 A schematic of camera arrangement for the large-scale DIC test 83
3.20 Stereoscopic camera system for the single-bladed CCR rotor . 84
3.21 Rotor blade painted with orange fluorescent color and black dots 84
3.22 A schematic of the DIC measurement setup . . . . . . . . . . 85
3.23 Unprocessed images taken simultaneously by the high-speed
cameras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.24 Camera calibration setup . . . . . . . . . . . . . . . . . . . . . 86
3.25 A schematic of stereoscopic camera arrangement used during
Test campaign 3 . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.26 Pitch angle variations and the initial digital trigger timing ob-
tained from the three independent measurements . . . . . . . 95

4.1 A diagram of the organization of this chapter . . . . . . . . . 98

4.2 Recap: Side view of the rotor test article . . . . . . . . . . . . 99
4.3 Flap bending deformation over the rotor disk at θ0 = 30◦ . . . 100
4.4 Static blade deformations of each degree of freedom, along the
blade quarter-chord axis at 30◦ root pitch and different rota-
tional speeds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.5 Time history of the pitch angle displacement at the blade tip
for different root pitch angles . . . . . . . . . . . . . . . . . . 104
4.6 Recap: Single-bladed, CCR rotor configuration . . . . . . . . . 106

xv
4.7 Flap bending deformation of the CCR lower rotor blade at CT /σ
= 0.09 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.8 Lower rotor blade tip displacement comparison between mea-
surement and prediction . . . . . . . . . . . . . . . . . . . . . 109
4.9 Maximum and minimum flap bending deformation of the lower
rotor blade as a function of blade radial station at CT /σ = 0.09 110
4.10 Interaction of blade bound circulation in the CCR rotor system 111
4.11 Recap: Two-bladed, isolated single rotor configuration . . . . 112
4.12 Out-of-plane blade deformation over the entire rotor disk . . . 113
4.13 Blade tip displacement as a function of azimuthal angle . . . . 114
4.14 3D static deformation distributed over the entire blade span . 116
4.15 Static blade deformations of each degree of freedom, along the
blade quarter-chord axis . . . . . . . . . . . . . . . . . . . . . 117
4.16 Out-of-plane blade deformation over the entire rotor disk for
the lateral pitch angle input . . . . . . . . . . . . . . . . . . . 118
4.17 Blade tip displacement as a function of azimuthal angle for the
lateral pitch angle input . . . . . . . . . . . . . . . . . . . . . 119
4.18 Blade tip deformations of each degree of freedom as a function
of time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.19 Rotor blades with fluorescent paint and random speckle pattern 122
4.20 Pitch angle displacement at the tip for θ0 = 30◦ , Ω = 1200 RPM125
4.21 Log-scale frequency spectrum of the out-of-plane displacement
measured at the blade tip for θ0 = 30◦ , Ω = 1200 RPM . . . . 126
4.22 Fan plot for θ0 = 0◦ , comparing natural frequencies extracted
with the CP and NExT-ERA algorithms . . . . . . . . . . . . 128
4.23 Fan plot for θ0 = 30◦ , comparing natural frequencies extracted
with the CP and NExT-ERA algorithms, along with numerical
predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.24 The first mode shape of the rotor blade for θ0 = 30◦ at 1200 RPM130
4.25 The second mode shape of the rotor blade for θ0 = 30◦ at 1200
RPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.26 The first and second mode shapes along the quarter-chord axis,
for θ0 = 0◦ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.27 The first and second mode shapes along the quarter-chord axis,
for θ0 = 30◦ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

xvi
4.28 Comparison of the first mode shapes estimated by the CP and
NExT-ERA as well as numerical prediction for θ0 = 30◦ at 1200
RPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.29 Comparison of the second mode shapes estimated by the CP
and NExT-ERA as well as numerical prediction for θ0 = 30◦ at
1200 RPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.30 Thrust measured on the single-bladed single rotor at zero mean
thrust . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.31 Measured variation of tip displacement on the single-bladed sin-
gle rotor at zero mean thrust . . . . . . . . . . . . . . . . . . . 141
4.32 Fan plot with measured natural frequencies (100% = 900 RPM),
compared with numerical predictions . . . . . . . . . . . . . . 142
4.33 Measured rotating mode shapes of the first flap mode at nominal
rotational speed (900 RPM) compared with numerical predictions143
4.34 Measured rotating mode shapes of the first lag mode at nominal
rotational speed (900 RPM) compared with numerical predictions143
4.35 Measured rotating mode shapes of the second flap mode at nom-
inal rotational speed (900 RPM) compared with numerical pre-
dictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.36 Time history of the out-of-plane deformation measured at the
trailing-edge point of the rotor blade tip . . . . . . . . . . . . 145
4.37 Frequency spectra of the three different time periods . . . . . 146
4.38 Spectrogram of the flap response measured at the trailing edge
point of the blade tip . . . . . . . . . . . . . . . . . . . . . . . 148
4.39 Reconstructed (filtered) thrust of the CCR lower rotor, com-
pared to predictions at CT /σ = 0.09 . . . . . . . . . . . . . . 151
4.40 Reconstructed (filtered) pitch link load of the CCR lower rotor,
compared to predictions at CT /σ = 0.09 . . . . . . . . . . . . 153
4.41 Phase-averaged thrust variation as a function of blade azimuthal
location for the steady collective pitch input . . . . . . . . . . 156
4.42 Phase-averaged torque variation as a function of blade azimuthal
location for the steady collective pitch input . . . . . . . . . . 157
4.43 Phase-averaged pitching moment variation as a function of blade
azimuthal location for the steady collective pitch input . . . . 158
4.44 Phase-averaged pitch link force variation as a function of blade
azimuthal location for the steady collective pitch input . . . . 160
4.45 Phase-averaged pitch angle variation as a function of blade az-
imuthal location for the steady collective pitch input . . . . . 162

xvii
4.46 Estimated rotor loads for the steady collective pitch input . . 164
4.47 Phase-averaged thrust variation as a function of blade azimuthal
location for the lateral cyclic pitch input . . . . . . . . . . . . 166
4.48 Phase-averaged torque variation as a function of blade azimuthal
location for the lateral cyclic pitch input . . . . . . . . . . . . 167
4.49 Phase-averaged pitching moment variation as a function of blade
azimuthal location for the lateral cyclic pitch input . . . . . . 168
4.50 Phase-averaged pitch link force variation as a function of blade
azimuthal location for the lateral cyclic pitch input . . . . . . 170
4.51 Phase-averaged pitch angle variation as a function of blade az-
imuthal location for the lateral cyclic pitch input . . . . . . . 171
4.52 Estimated rotor loads for the lateral cyclic pitch input (Test
condition 2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
4.53 Blade root pitch angle as a function of rotor revolutions during
the step pitch input operation . . . . . . . . . . . . . . . . . . 175
4.54 Rotational speed of the rotor as a function of rotor revolutions 176
4.55 Rotor thrust response to the step pitch change as a function of
rotor revolutions . . . . . . . . . . . . . . . . . . . . . . . . . 177
4.56 Rotor torque response to the step pitch change as a function of
rotor revolutions . . . . . . . . . . . . . . . . . . . . . . . . . 178
4.57 Pitch link force response to the step pitch change as a function
of rotor revolutions . . . . . . . . . . . . . . . . . . . . . . . . 179
4.58 The estimated inertial load integrated at the hub as a function
of the number of revolutions for the collective step pitch change
(Test condition 3) . . . . . . . . . . . . . . . . . . . . . . . . . 180
4.59 Comparison between the estimated and measured rotor hub
loads for the collective step pitch change (Test condition 3) . . 182

xviii
 Chapter 1

Introduction

1.1 Problem Statement

Understanding rotor vibratory loads remains an essential task for he-

licopter design. Helicopters typically encounter significant vibration in tran-
sition or high-speed flight regimes, due to aerodynamic phenomena such as
blade-vortex interaction (BVI) and negative loading on the advancing side.
Those loadings induce significant elastic deformations of the rotor blade struc-
ture during helicopter operations as shown in Fig. 1.1. To understand the
nature of rotor loads due to the structural and aerodynamic loadings, a num-
ber of flight and wind tunnel tests have been performed in the past [3, 4, 5];
however, measuring rotor loads is still one of the most challenging assignments
of experimental research in the rotorcraft field. Direct surface pressure mea-
surements using on-blade pressure transducers is a typical methodology for
measuring aerodynamic forces on a rotor blade, while structural loads (or in-
ertial loads) are conventionally measured with on-blade strain gages as shown
in Fig. 1.2. In order to perform these rotor loads measurements, experimen-
talists need to overcome several technical barriers, such as the installation of
sensors and electrical wiring on a rotor blade, usage of expensive slip rings, and
sensor failure due to strong centrifugal forces. Aerodynamics and structural

1
dynamics of the rotor blade can also be significantly altered by installation
and modifications to accommodate the pressure transducers, static pressure
taps, and strain gages.

Figure 1.1: Comparison between images taken during static display and flight
operation of the CH-53 Sea Stallion [1, 2]

These challenges have motived an alternate approach to investigate he-

licopter rotor loads; that is, airloads estimation strategies based on the struc-
tural response of a rotor blade. The capability of estimating rotor loads using
only measured structural response would not only reduce cost and time to
complete measurements, but also help one to understand the complex rela-

2
tionship between the structural and aerodynamic loads on a helicopter rotor.
Bousman [6] provided a good summary of previous airload estimation method-
ologies, as well as his method that estimates blade aerodynamic forces based
on measured flap bending moments and modal parameters of a rotor blade.
This work showed good agreement between the estimated and directly mea-
sured spanwise aerodynamic loadings, except for regions near the blade tip.
Wang et al. [7] discussed a combined analytical-experimental approach to pre-
dict the vibratory loads in an articulated rotor. In this study, the estimated
airloads were obtained based on strain measurements as well as the lifting-line
aerodynamic and finite element blade structural models, with a satisfactory
agreement to the corresponding flight test data. These previous studies have
shown that aerodynamic estimations appeared to be theoretically robust and
feasible for prediction of rotorcraft dynamic behaviors; however, the method-

Figure 1.2: An example of complex on-blade strain gage instrumentation

3
ologies used in these studies rely on precise identification of modal parameters
of a rotating rotor blade (i.e., mode shapes, modal amplitudes, and natural fre-
quencies), as well as data measured by pointwise on-blade sensors (i.e., strain
gages). The former would necessitate additional experiments (modal tests) or
to develop an accurate finite element model of the blade structure, while the
latter would require extensive preparation for measurements with challenges
inherent to operating in a high centrifugal force environment.

In other words, the helicopter rotor loads estimation can be achieved

if the two major problems mentioned above, one relevant to modal parameter
identification and the other relevant to on-blade sensor failure, are addressed.
As a possible approach to modal characteristics extraction, this dissertation in-
troduces a system identification algorithm, called Operational Modal Analysis
(OMA). The OMA, also called output-only modal analysis, is widely recog-
nized as an attractive technique to extract modal parameters of a structure in
its operating condition. The significance of the OMA is that it only requires
measurement of the resultant responses of a structure under ambient or oper-
ational excitation, in contrast to classical modal analysis that requires mea-
surements of both the input and output. This feature makes OMA well-suited
for investigating the dynamic behavior of a rotating rotor blade, especially
considering the challenge of measuring aerodynamic excitation forces.

The other concern for a successful rotor load estimation, on-blade sen-
sors failure, can be resolved by introducing non-contact optical measurement
techniques that requires no troublesome preparation and installation of sen-

4
sors on a target specimen. To date, the optical methodologies have been ap-
plied to a variety of different structural engineering fields, including helicopter
rotor blade deformation and strain measurements in the rotating frame; for
example, photogrammetry, holographic interferometry, and projection moiré
interferometry (PMI). Among quite a few of these methodologies, this disser-
tation selects the Digital Image Correlation (DIC) deformation measurement
technique, since the DIC technique has shown its robustness in measuring
rotor blade deformations at different rotor configurations and scales, as dis-
cussed in [8, 9, 10, 11]. The DIC deformation measurement technique is able to
measure three-dimensional displacements over the entire blade span at a high
spatial resolution with minimal preparation of the test article (i.e., painting
the rotor blade surface with stochastic speckle patterns).

This dissertation presents a new experimental and theoretical method-

ology for helicopter rotor loads estimation based on blade structural response.
Theoretical developments of the three key components of the methodology
(digital image correlation, operational modal analysis, and rotor loads identi-
fication) are first presented. Three different model-scale rotor hover test stands
used for verification of the proposed experimental and theoretical framework
are then described. The results of hover tests on the three different rotor
systems are presented with a focus on the rotor blade deformations, modal
parameter identifications, and rotor loads reconstructions.

5
1.2 State of the Art

This section discusses the literature related to the three core elements of
the present study: the optical deformation measurement techniques, the oper-
ational modal analysis (OMA) algorithms, and rotor loads estimation method-
ologies.

1.2.1 Optical deformation measurement

Digital Image Correlation (DIC) is categorized as a photogrammetric

measurement technique. Photogrammetry is an image-based optical measure-
ment technique that has been commonly applied to structural deformation
monitoring due to its robustness and ability to capture full-field deformation
of the target structure, with the help of digital cameras and powerful com-
puters [12]. In the applications to rotating structures, optical measurement
techniques can be divided into three major categories:

• Grid method

• Point Tracking

• Digital Image Correlation

1.2.1.1 Grid method

The grid method is one of the earliest optical methods for measurement
of structural deformations [13]. The basic procedure of the grid method is as
follows; (1) regular marking patterns are applied to the target surface of a

6
test specimen, (2) diffusive light reflection from those patterns are recorded,
(3) the differences between two grid images are used to calculate the defor-
mations of the structure. The projected grid method was used for the Higher
Harmonic Control Aeroacoustic Rotor Tests (HART) program, performed by
Kube et al. [14]. The method provided flapwise and torsional deflections along
the blade span at selected azimuthal angles, and its results were compared
to signals measured by on-blade strain gages. Electronic Speckle Pattern In-
terferometry (ESPI) and Projection Moiré Interferometry (PMI) are other
non-contact optical approaches. The former electronically observes formation
of fringes and records images of the reference (undeformed) and deformed ob-
jects [15, 16], while the latter evaluates fringe interference patterns between
the deformed projected grid and a computationally phase-shifted grid to calcu-
late out-of-plane displacements [17, 18]. Recently, Sekula [19] employed PMI
to measure blade deformations of a model-scale rotor in hover in the NASA
Langley subsonic wind tunnel and hover test facility, and successfully acquired
the out-of-plane hub height, tip height, and the flap angles of the 66.5 in (1.69
m)-radius target blade. These results showed that measuring blade deflection
using digital images is feasible.

1.2.1.2 Point Tracking

The key idea of the point-tracking optical deformation measurement is

to place reflective markers onto the surface of a target structure and track the
motion of the markers by digital cameras. A few examples of this methodology

7
are shown as follows:

Three-Dimensional Point Tracking (3DPT) technique is an optical tech-

nique that identifies and tracks the coordinates of discrete and reflective points
mounted on the target structure. Ozbek et al. [20, 21, 22] utilized the 3DPT
technique to measure the dynamic response of a 80 m diameter rotor on a 2.5
MW wind turbine using four CCD cameras, and achieved an average accuracy
of 25 mm over a measurement window of 220 m. Schneider [23] introduced the
Stereo Pattern Recognition (SPR) technique for rotor blade position and de-
flection measurements on a 40% Mach-scaled model helicopter rotor in a wind
tunnel and successfully measured the blade displacements using four digital
cameras located underneath the rotor. Abrego et al. [24] performed blade dis-
placement measurements using multi-camera stereo photogrammetry during
the full-scale wind tunnel testing of the UH-60A rotor and compared the ex-
perimental results with CFD/CSD predictions. Baqersad et al. [12] provided a
good summary of photogrammetry and optical methods in the context of struc-
tural dynamics. Lundstrom et al. [25], combined 3DPT and modal parameter
identification to monitor the dynamics of a 10.1-m diameter helicopter main
rotor; they were able to extract the rotating natural frequencies and operating
deflection shapes (ODS). Javh et al. [26] also performed a hybrid measure-
ment, combining photogrammetry and several modal parameter identification
methods and successfully identified the eigenvalues and mode shapes from a
set of very noisy digital images, though this measurement was not performed
on rotary machines.

8
1.2.1.3 Digital Image Correlation

The basic, two-dimensional digital image correlation (2D DIC) tech-

nique was first developed in 1980s following the wide usage of the projected
grid method, in order to measure in-plane deformation and strain field. Since
then, the same technique has been called with different names in the litera-
ture; for example, texture correlation, electronic speckle photogrammetry, and
digital speckle correlation. Peters and Ranson [27] developed the theoretical
basis of the digital correlation of speckle images with a simple measurement
of rigid body translational motion. Theory of DIC was then validated in the
literature through measurements of deflections of a cantilever beam [28, 29].
Chu et al. [30] also corroborated these findings through the measurement of
rigid-body rotation, constant-angular velocity, and uniform-finite strain. Im-
age correlation approach with subdivided small interrogation windows, which
is commonly implemented in modern DIC algorithms, was introduced in late
1990s by McNeill et al. [31] for accurate and rapid camera calibrations and
profiling of target structure images.

In the DIC technique, a random speckle pattern is painted on the sur-

face of target structure undergoing deformation and digital images of the sur-
face are captured before and after deformation. Cross-correlation of these
images in conjunction with calibration of a stereo pair of cameras yields the
three-dimensional deformation of the surface over the whole field of the image.
DIC has the capability of not only capturing a large field of vision, but also
measuring extremely small deformation of a structure on the order of a few

9
microns [32]. Although ESPI also has the capability of measuring small struc-
tural motion on the same order of sensitivity, it is not well-suited for measuring
large deflections of a structure unlike DIC approach. Another advantage of
DIC is its ready applicability to dynamic deformation measurements. These
instantaneous measurements can be analyzed by a dynamic sequence of digital
images taken by high-speed image recording equipment. Schmidt et al. [33]
used a stereoscopic set of high-speed cameras with a pulsed YAG laser as a
light source to measure rapid deformations of a road wheel tire, achieving a
short exposure time of 7 nanoseconds.

Over the past few years, special attention has been paid to the DIC
technique not only in the field of experimental structural and solid mechanics,
but in a large number of different research and engineering fields, such as tensile
testing of a knee tendon, deformation of a frog heart, tissue studies, and testing
of biomimetic materials in microscopic biomechanics [34]. The fundamental 2D
DIC framework itself has been extensively investigated and improved towards
reducing computational cost, widening application range, and achieving high
accuracy and precision. One critical expansion of the application area was to
be capable of measuring out-of-plane, three-dimensional deformation over the
curved surface of a structure.

The three-dimensional DIC (3D DIC) is the technique used in the

present study. The 3D DIC has been applied to different types of rotor blade
deformation measurements. For example, Sirohi and Lawson [8, 9] used the 3D
DIC method to measure the deformation of a model-scale rotor blade at sev-

10
eral azimuthal locations using a phase-averaging technique, that is, the images
are captured at a specific azimuthal location and averaged over multiple rotor
revolutions. The process is repeated at a different azimuthal location, yielding
an average deformation at each azimuthal location [10, 11]. However, due to
the averaging process, these deformations are not correlated in time. Sicard
and Sirohi [9] utilized the 3D DIC technique to measure the deformation of an
extremely flexible rotor blade at a specific azimuthal location with validation
performed using laser displacement sensor measurements and an aeroelastic
numerical simulation tool. The DIC measurement technique was also used
by Cameron et al. [35] to measure the rotating modal properties of a flexible
blade in a reduced-scale coaxial helicopter rotor over a field of view limited to
one quarter of the rotor disk.

Although both the 2D and 3D DIC method have proven to be useful

tools for full-field rotor blade deformation measurement in rotating frame,
none of the past studies have measured the time history of deformation of a
rotating helicopter rotor blade. The present study extends the range of 3D
DIC measurement over a much larger field of view with the spatial resolution
on the order of 0.1% of a rotor radius, in order to obtain the continuous time
history and transient deformations of a model-scale rotor blade in hover.

1.2.2 Operational Modal Analysis

Operational Modal Analysis (OMA) only requires measurement of the

response of a structure under ambient or operational excitation, in contrast to

11
classical experimental modal analysis (EMA) that requires measurements of
both the input and output information. Since the beginning of its development
in civil engineering, OMA methodologies have been proposed in a wide variety
of engineering applications, particularly in situations where measuring input
forcing or exciting the structure is impractical, such as in large-scale bridges,
buildings, aircraft, and wind turbines.

There are a number of different OMA techniques formulated in the

past. Of them, this dissertation focuses on four algorithms that have been
applied to a number of different engineering and research fields including rotary
structures, such as wind turbines and helicopters. A list of the algorithms is
as follows:

• Ibrahim Time Domain (ITD) method

• Eigensystem Realization Algorithm (ERA)

• Natural Excitation Technique (NExT)

• Complexity Pursuit (CP)

The Ibrahim Time Domain (ITD) is categorized as a time-domain

modal identification algorithm and utilizes free-response time histories mea-
sured at several measurement points on a target structure. The theoretical
basis was proposed and developed by Ibrahim in 1977 [36], and has been used
to extract the modal characteristics of several types of structures such as can-
tilever beams or spacecraft payloads [37]. Recently, Rizo-Patron and Sirohi [11]

12
used the ITD method to extract the modal properties of a cantilever beam and
a model-scale, conventional rotor blade under rotation. Cameron et al. [35]
identified the rotating natural frequencies and mode shapes of a rotor blade
on a reduced-scale coaxial, counter-rotating rotor system using the modified
version of ITD algorithm. Two common features in these two previous studies
are: (i) the excitation to the spinning structure was a periodic gust from a
compressed air source, and (ii) the deformation measurements were not con-
tinuous in time. Nevertheless, the ITD algorithm was able to identify the first
three flap-bending natural frequencies and mode shapes.

The Eigensystem Realization Algorithm (ERA) is essentially a system

realization methodology that realizes a linear model for a dynamic system, and
the realized system can be transformed into modal space for modal parameter
identifications. Juang and Pappa [38] first developed the basis of ERA and
applied it to the modal testing on the Galileo spacecraft. Since then this
algorithm has been used for various problems; for example, in fluid mechanics
for flow control [39] and model reduction, and in civil engineering for modal
testing on large-scale structures [40].

The Natural Excitation Technique (NExT) is another methodology of

modal analysis that allows derivation of pseudo-impulse response functions
based only on output information. The NExT is in general employed together
with a time domain modal identification scheme, such as the ERA or the
Poly-reference Time Domain algorithm. Regarding the applications to rotat-
ing machinery, James et al. [41] performed modal testing with the NExT on

13
vertical and horizontal axis wind turbines, and there have been a number of
follow-up studies to this pioneering work in wind turbine research and devel-
opment [42, 43, 44].

The Complexity Pursuit (CP) algorithm belongs to a broad family of

Blind Source Separation (BSS) algorithms. In general, signals measured in any
physical system are a mixture of statistically independent source signals [45].
Blind Source Separation is a methodology that allows identification and sepa-
ration of the sources without any input information. The usefulness of the BSS
techniques has been demonstrated for a diverse variety of data analyses, such
as financial time series of the Japanese stock market [46] and human brain
imaging dynamics [47]. The application to structural analysis was introduced
by Roan et al. [48] for damage detection and condition health monitoring of a
structure, and Kerschen et al. [49] described structural modal analysis using a
BSS technique called Independent Component Analysis, along with numerical
validations. The general concept of the CP algorithm was first formulated by
Stone [50] in the field of neural computation and applied to structural modal
parameter extraction by Yang and Nagarajaiah [51].

As described above, these OMA algorithms have proved their worth

as tools for modal identification of different types of structures, including a
helicopter or wind turbine rotor blade. However, most of the past studies ap-
plied the OMA algorithms to the deformation that was obtained by traditional,
conventional instrumentation, such as strain gages at a limited number of mea-
surement points. Only few studies in the literature have combined OMA and

14
optical deformation measurement techniques, although OMA in conjunction
with the non-contact optical measurement methods seems to have great po-
tential for reducing traditional difficulties of identifying rotor blade structural
dynamic characteristics. Thus, the current study in this dissertation applies
all the four methodologies to rotor blade structural responses measured using
the DIC technique, and evaluates the compatibility and applicability of each
OMA algorithm to the optically-obtained structural deformation.

1.2.3 Rotor loads estimation from structural response

Rotor loads estimation or rotor loads identification is a theoretical

framework for the determination of helicopter rotor loads based on measured
structural response. The identification includes the aerodynamic and struc-
tural loads on the rotor. This concept has been discussed in the rotorcraft
community because the determination of blade structural response is much
more tractable than that of blade airloads. Direct surface pressure measure-
ment with on-blade transducers is still a challenging task for experimentalists
due to extensive preparation for measurements with a high likelihood of sensor
failure caused by high vibration and centrifugal forces.

Although there exist various approaches for rotor load identification,

which are well summarized in Ref. [52], a number of previous studies can
be classified as normal modes or modal superposition methodologies. Somer-
son [53] presented a methodology of estimating rotor inflow and flapwise air-
loads distribution based on the inertial force distribution associated with nor-

15
mal modes of the rotor blade, with measured flap bending strain and blade
root motion. Bousman [6] also proposed the rotor flapwise loads estimation
technique based on a modal expansion of the flapping beam equation of mo-
tion with the aid of flap bending moment measurement. These previous studies
gave the key suggestion that in applications of their methodologies, at least
the first three modes in addition to the rigid body mode should be employed
for accurate estimation of spanwise airloads distribution.

Recent work has involved the growth of accurate computational mod-

eling of a helicopter rotor for estimating rotor loads. Öry and Lindert [54]
introduced a methodology of reconstructing applied forces to rotor blades by
inversely solving the equations of motion in modal space, based on flight test
data on Kamov-26 and Hughes 500E. Wang et al. [7] used a lifting-line aerody-
namic model to derive spanwise aerodynamic force distribution on a rotor blade
based on strain data measured at a few points. In these studies, the required
modal parameters were obtained from the finite element modeling of the rotor
blade. Apart from the modal expansion approaches, ShouShen et al. [55] de-
veloped a force analysis methodology, called Inverse Transfer Matrix Method,
to deal with the inherent issues with modal analysis (i.e., ill-posed problems
must be solved to obtain generalized modal coordinates). The methodology
was verified using the response data obtained during the NASA/Army UH-
60A Airloads Program. Schrage et al. [56] also applied the Inverse Transfer
Matrix Method to the blade out-of-plane and in-plane moments measured by a
large number of on-blade sensors, in order to determine the rotor loads on the

16
YAMAHA R-50/RMAX UAV testbeds. Although the proposed methods and
processes in these studies were verified and have been shown to be useful in
practice, accurate computational models of the rotor, such as a finite element
model for the structure, aerodynamic model, and inflow distribution model,
were required to obtain the necessary information for loads estimation.

From these studies, it appears that the framework of estimating rotor

loads has been well defined, however, the previous studies have some drawbacks
associated with their rotor loads identification approaches, such as: an exces-
sive amount of measured quantities (bending moment, displacement, slope at
blade root, and strain of the structure), or precise computational modeling of
the rotor. Thus, the approach presented in this dissertation makes use of the
basic theoretical framework of the modal expansion techniques as presented in
the literature, and integrates the load estimation methodology with the DIC
deformation measurement technique and the OMA algorithms.

1.3 Present approach

The primary goal of this study is to develop a theoretical and experi-

mental methodology for estimating helicopter rotor loads based on measure-
ments of blade structural response. The present approach consists of three
steps: (i) The continuous deformation time history of a reduced-scale rotor
blade is measured using the DIC technique. (ii) The measured blade deforma-
tion is processed with several OMA algorithms, i.e., Natural Excitation Tech-
nique (NExT), Eigensystem Realization Algorithm (ERA), and Complexity

17
Pursuit (CP), in order to identify modal frequencies and mode shapes in the
rotating frame. (iii) The rotor loads including the structural and aerodynamic
forces are identified based on the measured blade deformations and the modal
parameters obtained in the previous two steps. This combined DIC-OMA ap-
proach would yield a simple and inexpensive strategy to estimate rotor loads
without using on-blade sensors and detailed computational models of the ro-
tor. It is especially valuable because the challenging and costly problem of
performing accurate surface pressure measurements could be avoided.

First, the combined DIC-OMA approach is applied to a 0.46 m-diameter,

two-bladed, extremely flexible rotor for verification of the proposed method-
ology. The time history of the extremely flexible rotor blade deformation in
hover is measured using the DIC technique, and the measured responses are
processed with the NExT, ERA, and CP algorithms. The experimental results
are compared and correlated to an aeroelastic numerical prediction, developed
by Sicard [9].

The approach is then extended to measurements on a larger-scale (2

m-diameter), single-bladed, coaxial counter-rotating (CCR) rotor system. The
measured blade deformation employing the DIC and the identified modal pa-
rameters are compared to numerical analysis, provided from Ref. [57].

Finally, the rotor loads estimation methodology is applied to the blade

response of a 2 m-diameter, two-bladed, isolated single rotor in hover. Three
different rotor operating conditions are tested: (1) steady-state, constant load-
ing at certain rotor thrust level (collective pitch input), (2) periodic loading at

18
once-per-revolution frequency (cosine cyclic pitch input), and (3) step change
in rotor blade pitch angle (step function collective pitch input). Hub load (ro-
tor thrust) is then reconstructed from numerical integration of the estimated
spanwise lift distribution along the blade quarter-chord axis, and is compared
to the time history of rotor thrust directly measured by a load cell installed
on the rotor hub.

1.4 Contribution of the Current Research

There are three major contributions of the present work:

1. Development of an experimental methodology to measure the continuous

time history of helicopter rotor blade deformation over the entire blade
span including flap, lead-lag, and torsional bending degrees of freedom,
with minimum preparation of the test specimen (i.e., painting the surface
of the structure).

2. Identification of rotating frame natural frequencies and mode shapes of

a rotor blade in its operating condition without using conventional on-
blade sensors.

3. Estimation of helicopter rotor loads using only measured structural re-

sponse, resulting in significant reduction of time and cost for measure-
ment.

19
1.5 Organization of the Dissertation

Chapter 1 describes the importance of understanding the nature of he-

licopter rotor loads, and the limitation of current experimental and theoretical
tools for rotor loads estimation. The new approach in the present study to
overcome these limitations is proposed and the key components of the ap-
proach are described. The literature review provides the state-of-the-art in
each key component of the present approach. Additionally, experimental se-
tups and procedures for verification of the proposed approach are introduced
and briefly explained.

Chapter 2 presents the theoretical development and formulation of the

current approach for estimating rotor loads based on blade structural response.
This chapter covers the fundamentals of the non-contact optical blade defor-
mation measurement technique (digital image correlation), operational modal
analysis (NExT, ERA, and CP), and rotor loads estimation methodology used
in the dissertation.

Chapter 3 documents the details of three experimental setups and pro-

cedures used in the present study. The setups include: (i) a 0.46 m-diameter,
two-bladed, extremely flexible single rotor hover test stand, (ii) a 2 m-diameter,
single-bladed, coaxial counter-rotating (CCR) rotor hover test stand, and (iii)
a 2 m-diameter, two-bladed, isolated single rotor hover test stand. This chap-
ter also explains the arrangement of the digital image correlation measurement
for each rotor test bed.

20
 Chapter 4 discusses the results of the DIC blade deformation measure-
ment, modal parameter identification, and rotor loads estimation of each ex-
perimental system. Note that the setups (i) and (ii) are used for verifying the
applicability and robustness of the combined DIC-OMA approach, while the
setup (iii) is used for verifying the entire process of the current methodology,
including the DIC, the OMA, and rotor loads estimation. These individual
measurements on the three different rotor configurations and operating con-
ditions examine whether the methodology proposed in this dissertation is a
general tool for helicopter rotor loads estimation.

Chapter 5 summarizes the key conclusions of the present research, and

provides several suggestions for future directions of study.

21
 Chapter 2

Theoretical Development

This chapter is divided into three sections, each corresponding to the

theoretical development of three key components (DIC, OMA, and rotor loads
estimation methodology) of the present study. The basic concept of the DIC
is introduced and explained in § 2.1. The mathematical frameworks of the
three OMA algorithms (NExT, ERA, and CP) are presented in § 2.2. The
rotor loads estimation methodology based on the modal expansion technique
is described in § 2.3.

2.1 Digital Image Correlation

Digital Image Correlation (DIC) begins with comparing digital images

of a structure taken at different stages of deformation. By tracking subsets
of pixels, the DIC processing algorithm can compute surface displacement
and reconstruct full field 2D or 3D deformation vector fields and strain maps.
This section first defines the blade coordinate system used in the study, then
explains the basic principles of 2D and 3D DIC technique.

22
2.1.1 Blade coordinate system

The deformed position of the rotor blade is described using the blade-
fixed coordinate system, shown in Fig. 2.1. The origin of the coordinate system
is fixed at the center of rotation (center of the rotor hub), and the coordinate
system rotates about the positive z-axis at a constant rotational speed of Ω.
The x-axis is along the quarter-chord axis of the undeformed rotor blade.
Additionally, three degrees of freedom are used in the study: out-of-plane
(flapwise) motion, in-plane (lead-lag) motion, and torsional (feathering) mo-
tion with respect to the hub plane under the action of changing aerodynamic
lift and drag forces. Blade deformation measured using the DIC technique
shown in chapter 4 follows this definition of the blade-fixed coordinate system.

𝑧
𝑦
Ω
𝑥 Quarter-chord axis Torsion

Flap

Rotor blade
Lead-lag

Figure 2.1: Blade coordinate system

23
2.1.2 2D DIC

The basic procedure of the 2D DIC deformation measurement is as

follows:

1. Random speckle patterns are applied onto a surface of the target struc-
ture.

2. Camera calibration is performed to compute a mapping function between

global coordinates and camera image plane coordinates.

3. A sequence of digital images of the surface of the target structure is

obtained by digital cameras, such as CMOS or CCD cameras, before,
during, and after deformation.

4. The full image is divided into a set of small interrogation windows (also
called subsets) with an area of multiple pixels such as 16×16 pixels.

5. Light intensity distribution inside each interrogation window is com-

puted.

6. The algorithm searches for an area of same light intensity distribution

in the image before and after deformation using 2D image correlation.

7. Once the location of the target interrogation window is found in the de-
formed image, a 2D displacement vector corresponding to the motion of
the window is computed, and the search-and-compute process is repeated
for all the interrogation windows on each image.

24
 To search for the same subset match in the reference and deformed im-
age (Step. 6 and 7 in the procedure), the normalized cross-correlation function
C(x0 , y0 , x1 , y1 ) is computed as follows:

P
F (x0 , y0 )G(x1 , y1 )
C(x0 , y0 , x1 , y1 ) = pP (2.1)
F (x0 , y0 )2 G(x1 , y1 )2
where F (x0 , y0 ) and G(x1 , y1 ) represent the light intensity (values of gray levels
of the pixels) inside the area of the subset from the undeformed and deformed
images, respectively. (x0 , y0 ) and (x1 , y1 ) are the coordinates of a central point
on the target subset before and after deformation, respectively. Figure 2.2
shows a schematic of the subset search process.

Reference image Deformed image

𝑥% , 𝑦%

𝑥# , 𝑦#

Displacement
vector
𝑦

Figure 2.2: A schematic of subset search process

25
 Essentially, displacement computations are treated as an optimization
problem, that is, the displacement vector is calculated by searching a set of
the coordinates (x1 , y1 ) after deformation, that maximizes the cross-correlation
function C(x0 , y0 , x1 , y1 ). That is the reason why this technique is called digital
image correlation. Details of the mathematical development of 2D DIC are
available in Ref. [58].

2.1.3 3D DIC

3D DIC generally uses multiple digital cameras and arranges their rel-
ative positions in a stereoscopic manner, taking images of a target structure
from different perspectives. A commercially available software LaVision DaVis
8.4.0 [59] is used for camera control and calibration, image acquisition, and
DIC processing.

First, a physical point in global coordinates (laboratory coordinates,

(x, y, z)) is mapped onto the image plane of cameras 1 and 2, as shown in
Fig. 2.3. The mapping functions M1 (x, y, z) and M2 (x, y, z) are defined during
the camera calibration process as is the case of 2D DIC. Then the correspond-
ing point in image coordinates of camera 1 before and after deformation are
searched by solving the optimization problem with the cross-correlation func-
tion using Eq. 2.1, and a 2D displacement vector 1 is computed. The same
procedure is repeated for the corresponding image from camera 2, and a 2D
displacement vector 2 is computed. With the help of the surface height dis-
tribution H(x, y) and the two mapping functions M1 (x, y, z) and M2 (x, y, z),

26
 Image plane 1 Image plane 2

Camera 1 Camera 2

Deformed Mapping function 2

surface Subset 𝑀& 𝑥, 𝑦, 𝑧
Mapping function 1
𝑀( 𝑥, 𝑦, 𝑧

Surface height 𝑧
𝑦
𝐻 𝑥, 𝑦
Reference 𝑥
surface

Figure 2.3: A schematic of the basic 3D DIC principle

the two 2D displacement vectors are combined to a 3D displacement vector,

as shown in Fig. 2.4. Since there are a large number of experiment-specific
parameters for the DIC deformation measurement, details of its setting, such
as lens magnification, aperture, light source, size of subset, camera resolution,
and camera mount arrangement, will be described in chapter 3.

2.2 Operational Modal Analysis

This section describes the theoretical framework of the two Operational

Modal Analysis (OMA) algorithms used in the dissertation: Natural Excita-
tion Technique & Eigensystem Realization Algorithm (NExT-ERA) and Com-
plexity Pursuit (CP).

27
 Image plane 1 at 𝑡 Image plane 1 at 𝑡 + ∆𝑡

2D vector 1
𝑀% 𝑥, 𝑦, 𝑧

3D vector
Image plane 2 at 𝑡 Image plane 2 at 𝑡 + ∆𝑡

𝑀* 𝑥, 𝑦, 𝑧
2D vector 2

Figure 2.4: A calculation scheme of the 3D displacement vector

2.2.1 NExT-ERA

Since NExT is in general employed together with a time domain modal

identification scheme such as ERA, an overview of the technique is presented
here as the NExT-ERA combined approach.

The first step of NExT-ERA analysis is to compute cross-correlation

functions based on measured structural response. The correlation functions
can be expressed as summation of exponentially decaying sinusoidal signals and
each signal consists of a damped natural frequency and damping ratio, that is
equivalent to that of an associated structural mode. These cross-correlation
functions are thus used as pseudo-impulse response functions and processed
by ERA for extraction of modal properties.

Consider the differential equation of motion for a multi-degree-of-freedom

28
structure
Mẍ(t) + Cẋ(t) + Kx(t) = f (t) (2.2)

where M, C, and K are the mass, damping, and stiffness matrices respectively,
f (t) is the force vector, and ẍ(t), ẋ(t), and x(t) are the acceleration, velocity,
and displacement vectors at time t. Assuming that the system is excited by a
white noise, Eq. 2.2 can be written as:

MẌ(t) + CẊ(t) + KX(t) = F(t) (2.3)

where X(t) is the random response vector and F(t) is the random excitation
vector. Then multiplying Eq. 2.3 by a reference scalar random response vector
Xi (s) at a location i and time s on the system yields

MẌ(t)Xi (s) + CẊ(t)Xi (s) + KX(t)Xi (s) = F(t)Xi (s) (2.4)

Taking the expected value of Eq. 2.4 results in

ME[Ẍ(t)Xi (s)] + CE[Ẋ(t)Xi (s)] + KE[X(t)Xi (s)] = E[F(t)Xi (s)] (2.5)

where E is the expectation operator. Using R as the cross-correlation function

operator and τ = t − s as the time separation between t and s, Eq. 2.5 can be
expressed as:

MRẌXi (τ ) + CRẊXi (τ ) + KRXXi (τ ) = RFXi (τ ) (2.6)

The following relationship holds according to Caicedo et al. [60]:

Rx˙i xj (τ ) = Ṙxi xj (τ ) (2.7)

29
 Rx(m) xj (τ ) = Rx(m)
i xj
(τ ) (2.8)
i

where xi (t) and xj (t) are random response vectors at two different locations
(m)
i and j, and xi is the mth derivative of xi (t) with respect to time, and
(m)
Rxi xj (τ ) denotes the mth derivative of the cross-correlation function Rxi xj (τ )
with respect to τ .

With the assumption that the responses of the system are uncorrelated
to the random excitation F(t), Eq. 2.6 can now be expressed as

MR̈XXi (τ ) + CṘXXi (τ ) + KRXXi (τ ) = 0 (2.9)

In Eq. 2.9, it is obvious that the cross-correlation function RXXi (τ ) satisfies the
homogenous differential equation of motion. As described by James et al. [41],
the cross-correlation function is equivalent to a sum of decaying sinusoids of
the same form as the impulse response of the original structure in time domain.
This cross-correlation function obtained with the NExT analysis is used as an
input to ERA process.

Computing cross-correlation functions is the key step of the NExT al-

gorithm. Based on the fact that the cross-spectral density (frequency domain)
is equivalent to the cross-correlation function (time domain), Rxi xj (τ ) can be
calculated using the inverse Fourier transform of the cross-spectral density of
xi (t) and xj (t) as described in Ref. [61]. First, the responses xi (t) and xj (t) are
divided into data blocks, possibly with overlapping. The cross-spectral den-
sity function is then computed with ensemble averaging and Hanning window

30
of the Fourier Transform of each data block in order to improve the accu-
racy of the computation. Then the cross-correlation function between the two
channels i and j is obtained using
N −1  
1 X 2πkn
Rxi xj (n) = Sx x (k)exp i , n = 0, . . . , N − 1 (2.10)
N k=0 i j N

where k and n represent indices in frequency and time domain, respectively,

and Sxi xj (k) is the discrete cross-spectral density function between the two
channels i and j.

It should be noted that the assumption made for the theoretical devel-
opment, i.e., the white noise excitation is uncorrelated, is not necessarily met
for modal analysis on a real structure. Nevertheless, the operational modal
analysis with NExT has been performed successfully on real structures in past
studies [41, 62, 63] without satisfying the assumption, i.e. the excitation to
the structure is not a white noise; it is typically bandwidth limited and does
not have equal power at all frequencies. For appropriate use of this method,
it is essential for an analyst to determine the number of samples used for the
algorithm, the number of data blocks, the number of overlapping data points,
the length of the data blocks, shape of window, and reference channels for
calculating cross-correlation functions. These are all typically determined by
trial-and-error.

For the formulation of ERA, consider the state-space equations in con-

tinuous time
ẋ(t) = Ac x(t) + Bu(t)
(2.11)
y(t) = Cx(t)

31
where x(t) is the state vector, u(t) is the excitation vector, y(t) is the response
vector to the excitation, and Ac , B, and C are state-space matrices in con-
tinuous time. Assuming that an impulse excitation is applied to the dynamic
system, a solution to these equations has the form

Y(t) = CeAc t B (2.12)

where Y(t) consists of the free responses of the system to an impulse excitation.
In discrete time, the solution is represented as

Y(n) = CAn−1 B (2.13)

where Y(n) is the response at the nth time step, A = eAc ∆t is the discrete-
time state transition matrix and ∆t is the time step of system discretization.
The goal of ERA is to estimate the constant matrices A, B, and C from
the impulse response data Y(n) acquired from measurement. The system
modal parameters are then computed from the eigenvalues and eigenvectors
of the constructed state matrices. Note that the y(t) vector is the output of
the NExT algorithm that yields pseudo-impulse responses from the measured
structural deformation. It is important to mention that the matrices B can-
not be estimated for the present modal identification since the NExT is an
output-only modal analysis technique and the real excitation to the structure
is unknown. However, the modal parameters are obtained from the matrices
A and C.

32
 System realization starts from building the generalized Hankel matrix:
 
Y(n) Y(n+1) ... Y(n+p)
.. ..
 Y(n+1)
H(n − 1) =  . . 
 (2.14)
 .. 
.
Y(n+q) ... Y(n+p+q)

where Y(n) is in this case the N × 1 vector which consists of the pseudo-
impulse responses computed by NExT at the nth time step, and N is the
number of channels corresponding to the number of interrogation windows
during DIC processing. The parameters p and q correspond to the size of the
Hankel matrix. As a rule of thumb for selecting p and q, the analyst should
consider ten times the number of expected modes for the number of columns
p, and twice to three times the number of columns p for the number of rows q
in the Hankel matrix. Normally, the Hankel matrix is formed with block rows
and block columns data shifted in time by one sample from the previous block
row and column.

The Hankel matrix for n = 1 can be decomposed by using the singular

value decomposition as
 
CB CAB ... CAp−1 B
... ..
H(0) = 
 CAB .  = R ST .
 P
(2.15)
..
.
q−1
CA B ... CAp−1+q−1 B

Then the nth order system realization is calculated as

A = n1/2 Rn H(1) Sn n−1/2

P 1/2 P
(2.16)
C = ET Rn n1/2
P

where H(1) is the Hankel matrix for n = 2 and E = [I O] with an identity

matrix I and a null matrix O of appropriate order. These matrices are then

33
transformed to modal coordinates by multiplying with the eigenvector matrix
of A
Λ = Ψ−1 AΨ
(2.17)
Cm = CΨ
where Λ is the diagonal matrix of eigenvalues of A and Ψ is the matrix of
eigenvectors of A. After transforming back to the continuous time domain,
one can obtain the modal damping ratios and damped natural frequencies of
the system from the real and imaginary parts of the eigenvalues, respectively.
The column vectors of Cm correspond to the mode shapes.

However, there are a number of parameters that an analyst needs to

determine for identifying true physical modal parameters for the use of NExT-
ERA. This drawback provides motivation for using another OMA algorithm,
Complexity Pursuit, which will be described in the following section. Further
details and theoretical background of NExT and ERA can be found in Refs. [38,
41].

2.2.2 Complexity Pursuit

The Complexity Pursuit (CP) algorithm is classified under the family

of Blind Source Separation (BSS) algorithms. Consider the differential equa-
tion of motion for a multi-degree-of-freedom structure with mass matrix M,
damping matrix C, and stiffness matrix K

Mẍ(t) + Cẋ(t) + Kx(t) = F(t) (2.18)

34
where F(t) is the force vector and ẍ(t), ẋ(t), and x(t) are the acceleration,
velocity, and displacement vectors at time t. The modal transformation can
be expressed as
x(t) = Φη(t) (2.19)

where Φ and η are the modal matrix and modal coordinate vector of the
system, respectively. Note that the modes of a system form, by definition,
a linearly independent set of basis vectors [64]. Let us introduce the general
concept of BSS in mathematical form as

y(t) = x(t) + n(t) = A s(t) + n(t) (2.20)

where y(t) is the measured displacement vector at time t, n(t) is a noise vector,
s(t) is the source vector, and A, often called “mixing” matrix, is a transfer
matrix between sensors and sources. As shown in Eq. 2.20, the usage of BSS
within the context of operational modal analysis is to estimate the mixing
matrix A and the source vector s from the measured output responses y(t),
and interpret them as the modal matrix Φ and modal coordinates η of the
structural system.

This methodology can significantly simplify the multi-degree-of-freedom

modal parameter identification due to the fact that each separated source
ideally represents the equivalent single-degree-of-freedom response, whose fre-
quency and damping ratio are associated with each structural mode. Mode
shapes are represented by columns of the mixing matrix A. It should be noted
that the dimension of the estimated modal matrix is N × N , where N is the

35
number of measurement points, meaning that BSS separates the measured
data into N different source signals. A number of techniques falling under the
broad spectrum of BSS can be found in the literature. These techniques share
the fact that a priori information about the sources or the input forcing is not
required.

The CP algorithm separates source signals by using a measure of sig-

nal complexity called temporal predictability. The temporal predictability of
a signal describes how well one can predict the value of a signal at the next
time step, knowing only the values of the signal at the previous time steps.
The algorithm is based on the theorem that the temporal predictability of a
mixture of source signals is lower than that of each contained source signal,
i.e., each source signal, or mode, is less complex than the measured signal,
or superposition of modes. Therefore, the algorithm tries to find the mixing
matrix A of the system, so that the temporal predictability of each resulting
source is as high as possible, which basically reduces to an optimization prob-
lem. Details on this method are well summarized in the literature [50]; a brief
overview is given here.

The temporal predictability G(yi ) of a signal is given by

N
Σ [yi (tj ) − y i (tj )]2
j=1
G(yi ) = log N
(2.21)
Σ [yi (tj ) − ŷi (tj )]2
j=1

with yi being a signal at i th measurement location with N time entries, and

ȳi and ŷi being the long-term and short-term predictors respectively; these are

36
expressed as weighted sums of signal values measured up to time tj−1 given by

y i (tj ) = λL y i (tj−1 ) + (1 − λL )yi (tj−1 ), 0 < λL < 1 (2.22)

ŷi (tj ) = λS ŷi (tj−1 ) + (1 − λS )yi (tj−1 ), 0 < λS < 1 (2.23)

Values of λL = 0.99 and λS = 0.5 are used in this study. Minimization of the
function G(yi ) leads to an eigenvalue problem that yields the columns Ai of
the mixing matrix and corresponding eigenvalues Λi as follows:

Ĉ−1 C̄Ai = Λi Ai (2.24)

where Ĉ and C̄ are the short-term and long-term covariance matrices respec-
tively, calculated by:

N
X
Ĉ = (y(tj ) − ŷ(tj ))(y(tj ) − ŷ(tj ))T
j=1
N
(2.25)
X
C̄ = (y(tj ) − ȳ(tj ))(y(tj ) − ȳ(tj ))T
j=1

The biggest advantage of the CP algorithm is ease of implementation;

only two parameters λL and λS need to be chosen and the algorithm is robust
to the choice of these parameters, in contrast to the NExT-ERA approach.
The primary differences are summarized in the flow chart shown in Fig. 2.5.

37
Thus, this present study selected the CP algorithm as a primary scheme to
extract rotating-frame modal properties of a rotor blade.

NExT-ERA Complexity Pursuit (CP)

User inputs Raw data Raw data User inputs

• Resampling
• Windowing
• Averaging NExT CP
• Ref. channel
selection

• Size of Hankel
matrix Spectrum
• Peak-picking
• # of poles ERA analysis
• # of reference
channels

• CMI
• Smoothing User User
mode shapes evaluation evaluation

Modal Modal
parameters parameters

Figure 2.5: Flow chart of the OMA algorithms

2.3 Rotor loads estimation

Rotor loads estimation mainly consists of two components: one is the

estimation of distributed forces along the rotor blade span and the other is
the estimation of integrated loads acting at the rotor hub. Each component
has inertial and aerodynamic terms. This study focuses on the estimation of

38
out-of-plane rotor loads.

The vertical hub reaction force, which is essentially the rotor thrust, is
the net shear at the root of the rotating blade and is theoretically obtained by
integrating the sectional inertial and aerodynamic forces on the blade. Con-
sidering a rigid blade with no hinge offset as shown in Fig. 2.6, the vertical
shear Sz at the blade root is expressed as
Z R Z R
Sz = Fz dr − mz̈ dr (2.26)
0 0

where R is the blade radius, Fz is the sectional aerodynamic force, m is the

mass per unit length, and z̈ is the transverse (out-of-plane) acceleration. Equa-
tion 2.26 is applicable to either articulated or hingeless rotors, since the z axis
is always aligned to the rotational axis of the rotor.

Aerodynamic force
$%
Ω

Centrifugal
force
Hub reaction
force &% Inertial force
Blade #
!#̈
(

Figure 2.6: Diagram of rotor blade forces

2.3.1 Inertial loads

Let us first consider the estimation of the inertial loads distributed

along the rotor blade. The inertial force is expressed in the second term

39
of Eq. 2.26. The transverse acceleration z̈ can be obtained by numerically
differentiating measured out-of-plane displacements z in space, since the DIC
technique measures three-dimensional, full-field, time-resolved deformation of
the blade structure. With the mass properties of rotor blades calculated from
a CAD model, one can compute the distributed inertial forces on the blade as
a function of time (azimuthal angle) and space (radial location).

However, it is known that numerical differentiation of measured data

severely amplifies measurement noise. Thus, it is quite important to apply
smoothing methods to measured data in advance of numerical differentiation,
in order to accurately compute inertial loads. There are a number of method-
ologies available in the literature for measurement data smoothing, each with
advantages and disadvantages. The first methodology tested in the present
study is the method of data smoothing and numerical differentiation by a reg-
ularization approach. In particular, the problem statement in this study falls
into the class of Tikhonov regularization problems [65]. The general concept
of regularization is as follows:

Provided a set of data y(xi ), where i = 1, 2, ...N , the regularization

method defines an objective function Q;
2
xN xN
d2 ŷ
Z Z
2
Q(ŷ) = ŷ(x) − y(x) dx + λ dx (2.27)
x1 x1 dx2

where y(x) is a continuous function that exactly describes the given data
trend. The first term of the equation above is to quantify the goodness of fit
of a ”smooth” function ŷ(x) to the trend y(x), whereas the second term of

40
the equation is to determine the ”roughness” of ŷ(x). The second derivative
of ŷ(x) corresponds to the curvature of the function. Since large curvature
of a function means that the function changes rapidly, constraining the term
to be small makes the desired curve smooth. The goal of the regularization
is to find a function ŷ(x) such that the objective function Q(ŷ) is minimized
with a weighting factor λ. Further details of the mathematical development
as well as several methods for selecting an appropriate value of λ can be found
in Ref. [66].

To evaluate the applicability to rotor blade deformation measured us-

ing DIC, the regularization algorithm was applied to the out-of-plane bending
deformation of a 2 m-diameter rotor blade spun at 900 RPM. Figure 2.7 com-
pares the unprocessed and regularized flapwise bending deformation extracted
from the quarter-chord axis of the rotor blade at a certain time instant. The
unprocessed deformation showed an unusual dip at regions of the blade tip,
possibly due to the fact that the significant blade deformation at the tip intro-
duced out-of-focus blur and poor DIC processing quality in raw digital images.
This non-physical trend must be smoothed out before numerical differentiation
is performed. However, the regularization process with two different weight-
ing factors λ could not either remove the dip in Fig. 2.7a, or preserve the tip
deflection in Fig. 2.7b.

Alternately, polynomial curve fitting was applied to the same set of

data used for the regularization processing. As shown in Fig. 2.8, the large
dip was eliminated and the blade tip deflection was well preserved. Based

41
 140 140
Original Original
120 Regularized, =5 120 Regularized, = 50
Z displacement, [mm]

Z displacement, [mm]
100 100
80 80
60 60
40 40
20 20
0 0
-20 -20
0 10 20 30 40 50 60 70 80 90 100 110 0 10 20 30 40 50 60 70 80 90 100 110
Radial location, [%R] Radial location, [%R]
(a) λ = 5 (b) λ = 50

Figure 2.7: Unprocessed and regularized flapwise blade deformation

on this result, a 3rd-order polynomial curve fit was selected to smoothen the
deformation measured by DIC in this dissertation. Note that the inertial
properties of a rotor blade used in this study are known from an accurate
CAD model, and its details are described in chapter 3.

2.3.2 Aerodynamic loads

The spanwise distribution of airloads along the rotor blade is estimated

using a formulation in modal space [67]. The present approach makes use of
modal parameters identified using the CP algorithm. The partial differential
equation of motion for the out-of-plane bending of a rotating rotor blade is
represented as
Z R
∂2 ∂ 2z
  
∂ 2 ∂z
EI 2 − mΩ ρdρ + mz̈ = Fz (2.28)
∂r2 ∂r ∂r r ∂r

42
 140
Original
120 3rd order polynomial

Z displacement, [mm]
100
80
60
40
20
0
-20
0 10 20 30 40 50 60 70 80 90 100 110
Radial location, [%R]

Figure 2.8: Measured flapwise blade deformation and 3rd-order polynomial

curve fit

where E is the modulus of elasticity of the blade section, I is the area moment
about the chordwise principal axis, and Ω is the rotational speed. Now consider
the free vibration of the rotating rotor blade at frequency ν. The solution for
the homogeneous partial differential equation can be written as

z = η(r)eiνt (2.29)

Substituting Eq. 2.29 to 2.28 and making the right hand side of Eq. 2.28 zero,
the result is
Z R
d2 d2 η
  
d 2 dη
EI 2 − mΩ ρdρ − ν 2 mη = 0 (2.30)
dr2 dr dr r dr

which can be viewed as the equation for vibration in a vacuum. This modal
equation yields an eigenvalue problem for the natural frequency ν and mode
shape η(r) and there exists a series of eigensolutions ηk (r) and corresponding

43
eigenvalues νk of the k-th mode. The out-of-plane deflection z(r, t) is thus
expanded as a series of mode shapes describing the spanwise deformation as
∞
X
z(r, t) = ηk (r)qk (t) (2.31)
k=1

where ηk (r) and qk (t) correspond to the k th mode shape and modal coordinate
of the rotating rotor blade, respectively. Substituting Eq. 2.31 into 2.28 yields
X  d2  d2 ηk  Z R 
d 2 dηk X
EI − mΩ ρdρ q k + mηk q̈k = Fz (2.32)
k
dr2 dr2 dr r dr k

The terms in the first summation in Eq. 2.32 can be replaced by Eq. 2.30,
yielding
X
(νk2 mηk qk + mηk q̈k ) = Fz . (2.33)
k

Thus, if the modal parameters (i.e., natural frequencies νk , mode shapes

ηk , and modal coordinates qk , q¨k ) are known, one can compute the distributed
aerodynamic force Fz along the rotor blade from Eq. 2.33. This modal ap-
proach discussed above is practical if one is considering the fact that a small
number of modes are typically needed to represent the flapwise dynamics of
the rotating rotor blade [67]. With the distributed loads identified, rotor hub
loads can be obtained by integrating the radial distribution of both structural
and aerodynamic forces along the blade span, based on Eq. 2.26. For ver-
ification of the present methodology, the estimated hub loads are compared
to the results of rotor loads measurement, which is simultaneously performed
with the DIC deformation measurement. The outline of the whole procedure
is described in the flowchart shown in Fig. 2.9.

44
 DIC OMA Load estimation Verification
Differentiation in time
Noise-reduced Distributed
3rd-order polynomial displacement inertial forces
curve fitting

Blade deformation Integration Estimated Rotor

measured by DIC in space loads at hub

CP Natural frequencies
Distributed Comparison
Mode shapes
aerodynamic forces
Modal amplitudes Modal expansion
approach Measured rotor thrust

Figure 2.9: A flowchart of the whole approach proposed in this dissertation

2.3.3 Numerical experiment

Before the load estimation based on the measured data was examined,
a numerical experiment was carried out using a finite element model of a one
dimensional (1D) rotating cantilever beam with an arbitrary external forcing.
The numerically-derived blade elastic response and modal parameters were
used as an input to the present framework of rotor loads estimation approach,
and the known external force was compared to that estimated by the present
approach. The purposes of the numerical experiment were; (i) to examine if
the approach proposed in this dissertation was feasible to estimate rotor loads
distribution based on identified modal parameters and blade deformation and
(ii) to evaluate the influence of the participating number of modes on accuracy
of rotor loads estimation.

Figure 2.12 shows the original and estimated force distribution along
the blade span, obtained from the numerical experiment on the 40-element
(80 degrees of freedom) 1D rotating cantilever beam spun at 900 RPM. The

45
shape of the original excitation to the system followed a typical spanwise lift
distribution along a helicopter blade, that is, the lift force increases as the
local flow velocity increases towards the blade tip and the sharp drop occurs
at proximity of the blade tip due to the strong downwash from the trailed
blade tip vortices [68].

As can be seen in Fig. 2.10a, the first three modes were not enough
to reconstruct the unique shape of the typical airload over the blade span,
especially at regions near the blade tip. As the number of participating modes
increased, the accuracy of the load estimation was improved and the over-
all shape was satisfactorily captured with 10 modes, as shown in Fig. 2.10c.
From Fig. 2.10d to 2.10f, there was little improvement after the number of
participating modes went beyond 15 modes.

From this numerical experiment, it was discovered that the proposed

approach would require at least 10 modes to estimate a typical shape of aero-
dynamic loading over a rotating helicopter blade. Considering the general
difficulty of experimentally identifying higher-order structural modes, which
oscillate at high frequencies with low amplitudes, accurate estimation of span-
wise lift distribution over the rotor blade with the identified modes might be
impractical. However, hub loads (thrust), obtained by numerical integration
of the lift distribution along the blade span, could be close to a value directly
measured by a load cell installed in the rotating frame, according to Fig. 2.11,
which compares the area under the distributed lift curves along the blade radial
location for both the original and estimated load with the first three modes.

46
 350 350
Original Original
Force per unit length, [N/m]

Force per unit length, [N/m]

300 3 modes 300 5 modes

250 250

200 200

150 150

100 100

50 50

0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Radial location, [%R] Radial location, [%R]

(a) 3 modes (b) 5 modes

350 350
Original Original
Force per unit length, [N/m]

Force per unit length, [N/m]

300 10 modes 300 15 modes

250 250

200 200

150 150

100 100

50 50

0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Radial location, [%R] Radial location, [%R]

(c) 10 modes (d) 15 modes

350 350
Original Original
Force per unit length, [N/m]

Force per unit length, [N/m]

300 20 modes 300 40 modes

250 250

200 200

150 150

100 100

50 50

0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Radial location, [%R] Radial location, [%R]

(e) 20 modes (f) 40 modes

Figure 2.10: Comparison between the original and estimated lift distribution
over the finite element 1D rotating beam

47
 350
Original force
Force per unit length, [N/m] 300 Reconstructed force with 3 modes

250

200

150

100

50

0
0 10 20 30 40 50 60 70 80 90 100
Radial location, [%R]
Figure 2.11: Integrated force comparison between the original and recon-
structed loads

To quantitatively assess the influence of the number of participating

modes on estimation accuracy, two metrics were used: One was the area dif-
ference between the original and estimated hub loads defined as Eq. 2.34, and
the other was the root-mean-square (RMS) error of the spanwise lift distribu-
tion expressed as Eq. 2.35. Fˆz and Fz correspond to the estimated and original
loads, respectively.
Z R Z R
Area difference = dFˆz − dFz (2.34)
0 0
s
Z R
1
RMS = [Fˆz (x) − Fz (x)]2 dx (2.35)
R 0

48
The area difference normalized by the mean original hub load and the RMS
error are plotted in percentage as a function of the number of modes partic-
ipating in the estimation process, as shown in Fig. 2.12a and 2.12b. Both
plots show the common trend that the difference dramatically decreases from
the first three to ten modes participation, and gradually converges towards
zero. The conclusion from the qualitative analysis, that is, hub loads might be
accurate even with a small number of modes participation, is now reinforced
by the fact that the hub load estimated from the first three modes falls within
approximately 2% deviation from the original value.

49
 2.5

2
Area difference [%]

1.5

0.5

0
3 5 10 15 20 40
Number of modes
(a) Integrated loads difference

30

25

20
RMS

15

10

0
3 5 10 15 20 40
Number of modes
(b) RMS error of the spanwise lift distribution

Figure 2.12: Comparison between the original and estimated loads as a func-
tion of the number of participating modes

50
 Chapter 3

Experimental Setup and Procedure

This chapter describes the experimental setups, test envelopes, and

procedures conducted to validate the rotor loads estimation approach pro-
posed in the dissertation. To verify the applicability of the present theoretical
framework to various rotor configurations, blade deformation and rotor loads
measurements were performed on two different scale rotor hover test stands:

• Small-scale test bench: A 0.46 m-diameter, two-bladed, extremely flexi-

ble rotor

• Large-scale test bench: A 2 m-diameter, single-bladed or two-bladed,

single or coaxial counter-rotating (CCR) rotor

The rotor loads estimation starts with measuring the deformation time history
of the rotor blade using the time-resolved DIC, which requires one to take a
sequence of digital images over an entire rotor disk. The diameter of the rotor
is an essential measurement parameter that determines the whole setup of
the DIC measurement, including camera positioning and optical equipment
arrangement. To show the scalability and flexibility of the methodology, the

51
current study performed the time-resolved DIC measurements on the small-
and large-scale rotor blade.

Using these hover test rigs, three test campaigns were conducted. Test
campaign 1 was performed on the small-scale, extremely flexible rotor hover
test stand, and Test campaigns 2 and 3 were performed on the large-scale,
2 m-diameter rotor test stand at different rotor configurations and operating
conditions. This chapter is divided into three sections. In § 3.1, the design,
specification, and instrumentation of each rotor test stand are documented.
In § 3.2, the camera arrangement, optical settings, and other details of DIC
rotor blade deformation measurement on each rotor test stand are presented.
In § 3.3, the measurement envelope and procedure of the three test campaigns
are summarized.

3.1 Hover Test Stand

3.1.1 Extremely Flexible Rotor (small-scale test bench)

This rotor blade is extremely flexible; it is constructed out of a thin

carbon fiber ribbon whose structural stiffness is negligible. The extremely
flexible rotor blade concept has been proposed to increase the weight efficiency
and survivability of helicopter rotors. In this concept, centrifugal force on
the blades during rotation provides stiffness as well as stability. Because the
elimination of a structural stiffness requirement results in reduced weight, such
rotor blades have been proposed for heavy-lift helicopters. The low stiffness of
the rotor blade precludes conventional methods of deformation measurement;

52
and makes it an ideal testbed for the DIC measurement. This subsection
describes the key components of this extremely flexible rotor hover test stand.

228 mm
0.45
9” m
Flexible blade

Fixed-pitch hub
Flexible blade Fixed-pitch hub
DCmotor
DC motor

Shaft
Shaft encoder
encoder

ATI load cell

Figure 3.1: Side view of the rotor test article

3.1.1.1 Rotor configuration

The two-bladed rotor test article, shown in Fig. 3.1, consists of two
flexible blades and a fixed-pitch rigid hub. The blade structure comprises two
plies of ±45◦ AS4/3501-6 prepreg and a tungsten rod (25.4 mm long, and 2.4
mm diameter) held at the blade tip in a chordwise direction (see Fig. 3.2) to
provide centrifugal stiffening and passive stabilization. A rectangular brass
plate (51 µm thick, 7.6 mm length in chordwise, and 15.2 mm length in span-

53
wise direction) is inserted between the two plies of the composite laminate
at regions of the leading edge, and a brass cylinder is directly soldered onto
the plate. The brass structure with a small amount of epoxy is used to hold
the tungsten mass at the blade tip. The blade is nominally untwisted with a
constant thin circular arc airfoil profile; the thin, open section profile results
in low bending stiffness and negligible torsional stiffness, both dominated by
centrifugal force.

The fixed-pitch hub was designed in CAD and fabricated using a rapid-
prototyping plastic 3D printer. Four different root pitch angles were selected
for hover testing; 0◦ , 10◦ , 20◦ , and 30◦ . The rotor parameters are summarized
in Table 3.1; note the low thickness to chord ratio of the rotor blades. The
whole rotor assembly is placed onto a six-component strain gage load cell (ATI
Mini-40) to measure hub loads. The rotor is driven by a brushless DC motor
(Hacker A50-16S) and the rotational speed is measured by a 1024/rev optical
incremental encoder (US Digital E5), as shown in Fig. 3.1.

Table 3.1: Rotor system parameters

Rotor diameter 456 mm

Root cutout 11 %R
Blade airfoil Thin circular arc
Camber 9%
Chord 26.2 mm
Thickness to chord ratio 1.5%
Number of blades 2

54
 Tip mass
Blade grip Leading edge

Trailing edge

202 mm

Figure 3.2: The hub and the flexible blade with the fluorescent paint and
random speckle pattern

3.1.1.2 Data acquisition

Analog/digital signal data acquisition (DAQ) was obtained by a Na-

tional Instruments (NI) multi-function data acquisition card (NI-PCI-6052E)
with a 68-pin DAQ breakout connector (NI-SCB-68). Both analog signals from
the ATI load cell and digital signals from the US DIGITAL optical encoder
were acquired by the card. A high-speed counter on the DAQ card was used
for measuring the azimuthal locations of the rotor with the incremental A, B,
and one-per-revolution Z index channel from the encoder. Another high-speed
counter on the DAQ card was used to downsample the original incremental
signal from the channel A, and the downsampled digital pulse train was used
for triggering shutter timing of high-speed cameras and laser strobing (for DIC

55
measurement) at integer multiples of the rotational frequency. The one-per-
revolution pulse from the Z-index was used to synchronously start analog data
logging and image acquisition.

3.1.2 Single-Bladed or Two-Bladed, Isolated Single or CCR Rotor

(large-scale test bench)

A 2 m-diameter, rigid rotor hover test stand was designed, fabricated,

and assembled as a large-scale test bench to measure the transient loads and
blade deformations in hover. As shown in Fig. 3.3a and 3.3b, the hover stand
can be configured as a single-bladed or two-bladed, isolated single or coaxial
counter-rotating (CCR) rotor. Note that the rotor stand, shown in Fig. 3.3,
is oriented horizontally to facilitate imaging of the entire rotor disk, needed to
perform the DIC blade deformation measurement.

3.1.2.1 Drive system

A hydraulic power unit drives the rotors through a belt-driven trans-

mission system with synchronous 64-toothed belts and pulleys. As shown in
Fig. 3.4, there are two belt-pulley arrangements in the transmission, each cor-
responding to the upper rotor or lower rotor in CCR rotor configuration. Row
1 in Fig. 3.4 is connected to the inner main shaft, which rotates the lower rotor
in a clockwise direction, whereas row 2 is connected to the outer main shaft,
which rotates the upper rotor in a counter-clockwise direction. In the CCR
rotor configuration, a serpentine belt with an idler causes the upper and the
lower rotors to spin in the opposite directions at the same rotational speed.

56
 Transmission Counter weight

Hydraulic motor

1.016 m

Fluorescent-painted rotor blade

(a) Single-bladed, CCR rotor configuration

Two-bladed
single rotor

Swashplate assembly

A-shaped frame

Transmission

(b) Two-bladed, isolated single rotor configuration

Figure 3.3: The large-scale rotor test stand

57
A hydraulic motor (Rexroth AA2FM-63) attached to the drive shaft spins the
upper and lower pulleys together, and the maximum power provided by the
hydraulic unit is 105 kW at 4000 RPM.

Row 1 Row 2 Inner shaft rotor

Hydraulic motor
Outer shaft rotor

Figure 3.4: Side view of the transmission system

3.1.2.2 Rotor configuration

The CCR rotor system is designed to replicate the X2TD rotor sys-
tem [69], featuring a rigid hub and a closely-spaced CCR rotor. The specifi-
cations of the rotor system are summarized in Table 3.2. For the CCR rotor
system, the blade passages occur at the top and bottom of the rotor disk, i.e.,

58
at azimuthal angles of 0◦ and 180◦ . In single-bladed rotor configurations, one
rotor blade and a counter-weight are attached to each rotor, whereas in two-
bladed rotor configurations two identical blades are attached to each rotor. A
counter-weight is attached to the hingeless hub to balance the location of the
center of mass (CM) of each rotor. The counter-weight assembly consists of a
cylindrical-shaped steel solid piece, a threaded rod, and a rod fixture on the
hub. To adjust the location of CM of the entire rotor assembly, the location
of the steel piece can be readily changed by rotating the threaded rod.

Table 3.2: Summary of rotor parameters for the large-scale hover test stand

Parameter Single Coaxial

Number of blades Nb 1 or 2 1 or 2 (each)
Radius R, m 1.016
Root cutout, m 0.122
Rotor spacing, m - 0.140
Airfoil section VR-12 with 5% trailing edge tab
Precone, deg. 3
Chord c, m 0.080
Solidity σ 0.025 or 0.05 0.05 or 0.10
Vtip , m/s 95.8
Rotational speed Ω, RPM 900

3.1.2.3 Rotor blade

Rotor blades used on the large-scale test stand have the following fea-
tures: no taper, constant chord, uniform VR-12 airfoil section including a
5% trailing edge tab. The blade structure is made of a foam core and for-
ward D spar, wrapped with carbon-epoxy composite materials (a plain-weave

59
AS4/3501-6 prepreg and an IM7/3501-6 uni-directional carbon fiber tape).
The foam core in the D-spar is machined to place tungsten weights along the
leading edge, for adjusting the location of the blade center of gravity. The
±45◦ orientation plain-weave fabric provides increased torsional stiffness. Ro-
hacell IG31-F closed cell foam is used for the core and FM-300K film adhesive
is used to bond the carbon fiber prepreg to the form core. At the blade root, a
rectangular aluminum insert is placed onto the form core to provide compres-
sive strength where the rotor blade is clamped in the blade grip of the hub.
Figure 3.5 shows a breakdown of the blade laminate lay-up.

Film adhesive Fore foam core

Carbon-epoxy Aft foam core

composite

Carbon-epoxy
composite

Tungsten leading
edge mass

Aluminum blade
root insert

Figure 3.5: An exploded view of the rotor blade composite laminate

60
3.1.2.4 Rotor hub

The rotor hingeless hub consists of three key components; single-piece

central hub structure, blade grip and bearing carrier, and six-component load
cell as shown in Fig. 3.6. The single-piece central hub structure is made of
aluminum and designed to be directly mounted onto the load cell. There
are four ports on the side of the central structure, each corresponding to a
rotor blade attachment, i.e., the hub can be configured as a four-bladed rotor
system. At each port, a modular blade grip and bearing assembly is attached
and used to clamp the rotor blade and allow pitch feathering motion. A cross-
section view of the modular blade grip and bearing carrier assembly is shown
in Fig. 3.7. The bearing carrier packages three types of bearings; a roller thrust
bearing to support high axial (centrifugal) load and two needle roller bearings
to support thrust and flap-bending moments.

Below the central hub structure, a custom-modified, six-component

load cell (ATI Omega-160) is installed and measures hub loads in the rotating
frame for each rotor. Each load cell was modified to include on-board signal
conditioning to minimize slip-ring noise when transmitted from the rotating
frame to the fixed frame. High sensitivity silicon-based strain gauges allow for
stiff construction, resulting in large overload capacity and favorable dynamic
response. The resolution of force and moment measurements is 0.75 N and
0.025 Nm, respectively, based on the manufacturer specificaation. In-plane
forces are measured with bias accuracies of ±4 N, while thrust is measured
with a bias accuracy of ±8 N. All moment measurements had bias accura-

61
 Blade grip assembly

Load cell

Hub central structure

Figure 3.6: Primary components of the rotor hub

cies of ±0.5 Nm. While the isolated load cell has a natural frequency above
1000 Hz, the addition of the hub, blade, and counter-weight significantly mod-
ifies the frequency response. An in-situ dynamic calibration was performed
on the vertical (FZ ) component of each load cell using a calibrated impact
hammer. The resulting transfer functions for the upper and lower rotors are
shown in Fig. 3.8. The load cell response remains flat until about 140 Hz, or
7-per-revolution at a rotor speed of 1200 RPM. These transfer functions were

62
 Needle bearing
Thrust bearing

Blade grip

Bearing carrier

Figure 3.7: A cross-section view of the blade grip assembly

used to correct the measured FZ hub loads presented throughout the study.

3.1.2.5 Blade root pitch angle sensor

Two different blade root pitch angle sensors were tested: A linear
Hall effect sensor (Honeywell SS495A1) and an Anisotropic Magneto-Resistive
(AMR) position sensor (KMZ60).

The Hall effect sensor is mounted onto a stationary fixture inside the
central hub structure, and two Neodymium magnets are bonded to the bottom
surface of the blade grip. This root pitch measurement system operates based
on the Hall effect: the sensor measures the variation of magnetic field as the
two Neodymium magnets of opposite polarity rotate with the pitching motion
of the rotor blade grip. Details of the instrumentation and specification of the
Hall effect pitch angle sensor are available in Ref. [70]. The Hall effect sensor

63
 Figure 3.8: Dynamic response of upper and lower rotor load cells

was used for initial tests, however, it was found that the sensor often needed
re-calibration.

Fig. 3.9 shows a schematic of the AMR sensor installation on the blade
grip assembly. The AMR sensor generally consists of a thin film of ferromag-
netic metals such as Ni and Fe. The resistance of the thin film varies according
to the strength of the applied magnetic field along a specific direction. The sen-
sor uses the variation of resistance, which results in voltage change. The AMR
sensor is held on a stationary plate and a Neodymium magnet is bonded to the
face of the blade grip, as shown in Fig. 3.9. The stationary plate is mounted
on to the blade grip carrier in a way such that the feathering axis of the blade

64
grip is aligned to the coordinate system of the AMR sensor. Then by placing
the magnet at the center of the blade grip face, the AMR sensor can detect
the variation of the magnetic field associated with the feathering motion of
the rotor blade.

Magnet Blade grip

Section A
Sensor holder

Isometric view
𝜃

Section A
Section B
AMR sensor

Section B

Rotor hub
𝜃

AMR sensor

Sensor holder

Figure 3.9: A schematic of the AMR sensor installation on the blade grip
assembly

The AMR sensor provides ratiometric sine and cosine analog output
signals and its calibration is performed using an absolute optical inclinometer
(US A2T series). Figure 3.10a shows the variation of the two output voltage
signals from the AMR sensor as a function of rotational angle measured by
the inclinometer. A rotating magnetic field delivers the two sinusoidal output
signals with the double frequency of the mechanical angle between the sensor
and magnetic field direction. The angle can be calculated using the following

65
equation:
arctan (VSIN /VCOS )
α= (3.1)
2
where α is the angle between the sensor and magnetic field direction, VSIN
and VCOS are sine and cosine analog output signals, respectively. A sinusoidal
function f (x) = A sin (Bx + C) + D curve-fitted to the sensor output over the
angle range of interest is shown in Fig. 3.10b and the curve-fitting coefficients
with 95% confidence are summarized in Table 3.3. The uncertainties based
on the standard deviation calculated from three individual repeated measure-
ments are shown as the highlighted area in Fig. 3.10c. The measurement
uncertainty was computed to be ± 19 mV and corresponds to approximately
0.38 deg. Note that the current study used only the sine signal due to the
number of measurement channels available in the data acquisition system.
Table 3.3: Summary of the curve-fit coefficients to the AMR sensor output
signal

f (x) = A sin (Bx + C) + D

A 1.456 ± 0.003
B 1.951 ± 0.005
C 0.119 ± 0.002
D 2.506 ± 0.003

3.1.2.6 Control system

The pitch angle control system consists of two independent swashplates

actuated by three custom servo actuators used to control collective and cyclic
pitch for the upper and lower rotor independently in the CCR rotor configu-
ration. Each swashplate is connected to a pitch horn on the blade grip by a

66
 5 5
Output 1
Output 2 Output signal
4 4 Curve fit
Output amplitude [V]

Output amplitude [V]

3 3

2 2
Curve fit model
f(x) = A sin(Bx+C)+D
1 1

0 0
-180 -120 -60 0 60 120 180 -40 -20 0 20 40 60
Angle [deg.] Angle [deg.]
(a) Sine and cosine output signals from (b) Curve-fit to the voltage variation
the AMR sensor from the AMR sensor
3

2.75
Output Voltage, [V]

2.5

2.25

1.75

1.5
-10 -5 0 5 10 15
Angle [deg.]
(c) Output voltage signals from the AMR
sensor with measurement uncertainty

Figure 3.10: The AMR sensor characteristics

67
pitch linkage. Due to the unique load path and load cell configuration of the
rotor assembly, pitch link loads must be measured and subtracted from load
cell measurements for recovering the true rotor hub loads. As such, tension-
compression load cells with a full-scale range of ±250 N and an accuracy of
±0.5 N are installed in each pitch link to measure the variation of pitch link
forces as a function of blade azimuthal angle and appropriately correct the
hub load cell measurements.

There are two different custom-designed servo actuators installed on

the rotor stand:

• A lead-screw actuator

• A hydraulic actuator

The first design, a lead screw servo actuator, uses a lead-screw rod, driven by
a brushed DC motor through a gear reduction system. The main design target
of this servo actuator was to maintain or change the blade pitch angle under
severe rotor load conditions with high servo resolution and low compliance.
Table 3.4 summarizes the specification of the lead-screw servo actuator, and
further details can be found in Ref. [70].

The second servo design, a hydraulic actuator, was designed to achieve

dynamic rotor pitch variation. As shown in Fig. 3.11, the hydraulic actuator
assembly consists of three components: a main actuator structure, a hydraulic
servo valve, and a piston position sensor. The main actuator structure (a

68
 Table 3.4: Specifications of the lead-screw servo actuator

Parameter Lead-screw servo

Resolution [µm] 16
Repeatability [µm] 25
Maximum pitch angle error [deg] 0.05
Maximum pitch angle velocity [deg/s] 7.7
Force at maximum power [N] 1750

piston and a cylinder body) was designed in-house and fabricated by an exter-
nal manufacturer (SMC Corporation of America). The hydraulic servo valve
(KNR Systems) has a capability of operating 5.5 L/min at pressure of 70 bar at
a bandwidth above 60 Hz. The piston position feedback sensor (SICK MPS-T
position sensor) is mounted on a T-slot outside the cylinder body and con-
tinuously detects the variation of the magnetic field generated from a magnet
installed on the piston. The whole rotor control assembly with the hydraulic
actuation system is shown in Fig. 3.12

A block diagram of the pitch angle control system for the hydraulic ac-
tuation is shown in Fig. 3.13. In the pitch control signal line, input commands
(collective and cyclic pitch angle position) are first provided to a central con-
trol panel. These PWM signals are then sent to a signal distribution panel,
which takes 12 V power from a power supply, and the power and signals are
distributed to three micro-controllers (Pololu jrk 21v3). Each micro-controller
takes a feedback signal from the position sensor and drives the hydraulic servo
valve of each actuator for closed-loop control of the piston position.

The hydraulic unit is controlled by a manufacturer-provided (KNR Sys-

69
 M6 rod end

Piston

Hydraulic servo valve

Cylinder

Piston position sensor

Return

Pressure
M6 rod end

Figure 3.11: A drawing of the hydraulic servo actuator assembly

tems) control software. First, users set a pressure level and hydraulic flow rate
on the software and these parameters are sent to a mobile Hydraulic Power
Unit (mHPU) that supplies the hydraulic flow at the specified pressure and
rate. The fluid flow is distributed by a hydraulic manifold to the three indi-
vidual actuator assemblies. This hydraulic actuation system allows dynamic
variation of rotor blade pitch angle, as compared to the lead-screw servo ac-
tuation system.

70
 Swashplate assembly

Hydraulic line
Pushrod

Rotor hub

Pitch horn

Swashplate

Blade grip Hydraulic actuator

Figure 3.12: Rotor pitch angle control system assembly with the hydraulic
servo actuators

71
 Hydraulic line Pitch angle Pressure and flow rate
Signal line User input User input
Power line
User control panel Pump control software
Power supply

Power supply Hydraulic Pump

Signal distribution panel Pressure distribution manifold

Microcontroller Microcontroller Microcontroller

Feedback Feedback Feedback

Servo valve Position sensor Servo valve Position sensor Servo valve Position sensor
Actuator 1 Actuator 2 Actuator 3

Figure 3.13: A diagram of pitch angle control system for the hydraulic actua-
tion

3.1.2.7 Fixed-frame instrumentation

Fixed-frame instrumentation is designed to monitor several key param-

eters of the hover test stand for safety, including vibration level of the test
stand and bearing temperatures. Also, an optical incremental encoder used
for measuring the rotational speed of the rotor is categorized as fixed-frame
instrumentation, because it is mounted on the drive shaft and the transmission
frame in a stationary manner.

A two-axis MEMS accelerometer (ADXL-330) is mounted onto the

72
transmission for monitoring the fixed-frame vibration. Data from the ac-
celerometer is also used to analyze the dynamic imbalance of the rotor system
before testing, and to monitor dynamic unsteady loads observed during ac-
tual hover testing. Four precision integrated-circuit temperature sensors with
an output voltage linearly-proportional to the Centigrade temperature (Texas
Instruments LM35) are installed onto four bearing housings (two for the main
shaft, and two for the drive shaft) to monitor bearing health.

A 4096 count/rev, hollow-bore, optical incremental encoder (US Dig-

ital HB6M) is mounted to the drive shaft. This encoder provides the one-
per-revolution Z-index signal, in addition to the A and B timing signals. The
primary use of the encoder is; blade azimuth angle and rotational speed mea-
surement, timing reference for synchronous averaging of multiple revolutions
during data post-processing, and triggering the image acquisition equipment
(high-speed cameras and a laser strobe) for the DIC measurements.

To transfer electrical signals from the instrumentation installed in the

rotating frame to the fixed frame DAQ system, two slip rings, each correspond-
ing to the upper and lower rotor instrumentation, are used. A 24-channel self-
contained slip ring, manufactured by Fabricast is selected for the upper rotor,
whereas a 26-channel separate rotor and brush block assembly is used for the
lower rotor.

73
3.1.2.8 Data acquisition

Signal data acquisition (DAQ) from all the instrumentation described

above is conducted by a National Instruments (NI) PXI system. NI PXI sys-
tem generally consists of a PXI chassis and instrument modules; the chassis
supplies power, cooling, and a communication bus for the modules, while in-
strument modules acquire and generate analog/digital signals. The instrument
modules can also be used to trigger and synchronize measurement signals. In
the current test setup, the chassis houses three multi-function DAQ cards, one
PXI-6225 (maximum sample rate = 250 kS/s) and two PXIe-6358 (maximum
sample rate = 1.25 MS/s). The PXI-6225 card was mainly used for the data
acquisition from the fixed-frame instrumentation (temperature sensors and ac-
celerometers), whereas the PXI-6358 cards were used for the instrumentation
in the rotating frame, such as the load cells and root pitch angle sensors.

3.2 DIC Deformation Measurement

3.2.1 Small-scale rotor

DIC blade deformation measurement on the small scale rotor hover test
stand (see § 3.1.1) was performed in the test chamber shown in Fig. 3.14 and its
schematic is shown in Fig. 3.15. To acquire the time history of the rotor blade
deformation, a sequence of digital images over the whole rotor disk must be
taken. This was accomplished by mounting high-speed digital cameras on lin-
ear camera rails above the rotor plane with their axes oriented approximately
45◦ with respect to the rotor disk, as shown in Fig. 3.15. The stereoscopic ar-

74
rangement of two cameras (Phantom Miro M310 with Nikon NIKKOR f/1.8D
35 mm lens) enables measurements of three-dimensional displacement fields
on the entire flexible blade. Knowing the positions of two cameras relative to
each other and the magnifications of lenses, the stereoscopic DIC algorithm
can calculate the absolute three-dimensional coordinates of any point on the
blade surface.

Figure 3.14: The setup for small-scale rotor blade deformation measurement.

75
 Mirror
Camera 2

Camera 1
Diffuser

Beam Rotor
Motor
Laser
Mirror

Optical bench Test chamber

Figure 3.15: A schematic of the small-scale DIC measurement setup

A key step in implementation for a successful DIC measurement is

the preparation of the test article, which in this case is the flexible rotor
blade. A stochastic, high contrast intensity distribution that deforms with
the surface of the structure allows the algorithm to detect the speckle pattern
in images, resulting in good cross-correlation between reference and deformed
images. One way to obtain high contrast digital images is to enlarge the
aperture opening of the cameras to allow more light to reach the CCD image
sensors while illuminating the target surface by a bright light source. The
resultant images would have high light intensity difference between the surface

76
of the structure and the environmental background; however, this approach
may cause out-of-focus images due to shallow depth of focus inherent to a large
aperture setting. Increasing exposure time while maintaining a small aperture
value is an alternate approach, however, increased exposure time would cause
more blurry images. Thus, the present study makes novel use of two techniques
for DIC measurement: fluorescence and pulsed laser illumination.

Fluorescence is a chemical process that emits light at a certain wave-

length, triggered by molecular absorption of photons or other electromagnetic
radiation. Since the emitted light has a specific wavelength, an appropriate
selection of camera lens filters would allow only the light emitted by the fluo-
rescent paint to pass through and reach the camera image sensors. Therefore,
if the surface of the target structure is painted with fluorescent paint and a ran-
dom speckle pattern is painted on the surface with some non-fluorescent color,
one can readily obtain high contrast images of the applied speckle pattern;
the background area unnecessary for DIC processing can be automatically re-
moved by masking algorithms. Moreover, if a pulsed laser is used as a light
source, a very short duration exposure can be achieved to eliminate blur in
the images. This visualization technique is called Laser-Induced Fluorescence
(LIF) or Laser-Stimulated Fluorescence (LSF); it has been used recently on a
variety of materials and structures, such as gossamer, transparent, reflective
aluminum-coated, and polymer membranes [71, 72, 73, 74].

As can be seen in Fig. 3.2, orange fluorescent paint was applied on

the blade surface and a random pattern of black dots was painted on the

77
surface. A Photonics dual-head Nd:YLF DM-30 pulsed laser strobe with a
wavelength of 527 nm and a pulse energy of 30 mJ/pulse routed through
an engineered diffuser was used to illuminate the entire rotor disk at any
desired instant of time (as shown in Fig. 3.15). The short duration pulsed
laser illumination enabled a high f-number (f/8.0D) and a short exposure time
(10µs) for all the images captured for DIC. The optical incremental shaft
encoder on the rotor shaft was used to generate a 16-per-revolution pulse
train, which triggered laser strobing and image acquisition at 16 evenly-spaced
azimuthal locations. Since the maximum rotational speed was set at 25 Hz,
the maximum sampling rate was 400 Hz. At each test condition, images were
taken over 400 rotor revolutions, yielding 6400 images per camera over 16
seconds. Note that reference (undeformed blade) images for DIC calculations
were captured at the same azimuthal locations by rotating the rotor shaft
slowly by hand, i.e. at a negligibly small rotational speed, thereby eliminating
any aerodynamic forcing. Figure 3.16 shows the unprocessed images of the
rotor at 1200 RPM acquired by the high-speed camera at 16-per-revolution,
overlaid on each other to illustrate the complete rotor disk; a typical DIC
interrogation window is also shown for reference.

After the images were captured by a stereo pair of high-speed cameras,

the DIC software (LaVision DaVis 8 Strain Master 3D) was used to calculate
the deformation of the blade. The complete image (of size 1280 x 800 pixels)
was divided into a number of interrogation windows and one displacement vec-
tor was calculated for each interrogation window. The size of the window was

78
 Interrogation
window

Figure 3.16: Unprocessed images of the rotor strobed at 16 azimuthal loca-

tions, showing complete rotor disk at 1200 RPM.

selected as 19 x 19 pixels and the window shift was set to be 5 pixels for map-
ping the interrogation windows to the whole image, resulting in 54% overlap.
The camera scaling factor was approximately 2.24 pixel/mm determined by
the geometry of the test setup as well as the camera lens characteristics. De-
formations at a total of 988 locations (13 chordwise and 76 spanwise locations)
were calculated at each azimuth.

A critical step for obtaining 3D displacements from recorded images is

the calibration of the high-speed camera system prior to the measurements.
Figure 3.17 shows the calibration process for the cameras, using a standard
calibration plate placed at the rotor plane. Information of the size and location

79
of the calibration plate with respect to the rotor plane enables calculation of
a mapping function between the global coordinates and camera image plane
coordinates. With the help of mapping functions calculated during the camera
calibration process, the two two-dimensional vectors from each camera are
combined to yield a three-dimensional vector. The accuracy of the vectors
depends on the size of the interrogation window and is estimated to be 0.01%R
(0.023 mm), while the spatial resolution of the displacement field is 0.2%R
(0.45 mm). A sample processed result for the 30◦ root pitch rotor, rotating at
1500 RPM, is shown in figure 3.18.

Calibration target plate

Figure 3.17: Camera calibration target

80
 10

Blade root 9

Out-of-plane displacement, [mm]

8

2
Blade tip
1

Figure 3.18: Vector map processed by DIC for the 30◦ root pitch rotor, spun
at 1500RPM

3.2.2 Large-scale rotor

In addition to the small-scale rotor measurements, the time-resolved

DIC technique with LIF strobing was applied to the large-scale rotor test
stand (see § 3.1.2). The basic concept of large-scale DIC is the same as that
of small-scale DIC described in § 3.2.1, that is, a stereoscopic camera system
captures the whole area of the rotor disk so that the deformation time history
of the blade structure can be obtained. However, due to the scale-up of the
rotor diameter (from 0.46 m to 2 m), several modifications were required for a
successful large-scale DIC blade deformation measurement. One of these mod-
ifications was the horizontal orientation of the rotational axis that facilitates

81
imaging of the whole rotor disk as shown in Fig. 3.3. Other modifications are
discussed in this subsection.

Figure 3.19 and 3.20 show one of the stereoscopic camera arrangements
used for the large-scale DIC measurement. In the setup, three high-speed
cameras were used to image the entire rotor disk. The first and second cam-
eras (Phantom Miro M310 with Nikon NIKKOR f/1.8D 35 mm lenses) were
mounted so that each camera captured the entire rotor disk, and the third
camera was mounted on a tripod located between the first and the second
camera. At some azimuthal locations, images taken by the left and right cam-
eras (Camera No. 1 and Camera No. 3) were distorted due to large parallax.
These images were corrected using images taken by the middle camera (Cam-
era No. 2).

As was the case for the small-scale DIC measurement, orange fluores-
cent paint with stochastic speckle patterns was applied to the blade surface as
shown in Fig. 3.21, and the LIF strobing was achieved by a digital pulse train
generated from the optical encoder installed onto the drive shaft (see § 3.1.2).
A schematic of the whole DIC measurement setup is shown in Fig. 3.22. Cam-
era exposure time was maintained at 10 µs. The short exposure time with
laser strobing enables a high contrast (almost pure black and white), blur-free
image acquisition. Figure 3.23 shows unprocessed images taken by the three
cameras that were triggered at a certain azimuthal angle.

High-speed camera calibration was performed using a calibration plate

placed at the rotor plane in order to obtain a mapping function between global

82
Figure 3.19: A schematic of camera arrangement for the large-scale DIC test

and camera image plane coordinate system. To cover the larger field of interest
than that for the small-scale DIC, a large calibration plate was used, as shown
in Fig. 3.24. The large calibration plate was moved around to cover the whole
area of the rotor disk. The displacement accuracy for the large-scale rotor
blade was estimated to be 0.01%R (0.1 mm), and the spatial resolution of the
displacement field was estimated to be 0.2%R (2 mm).

83
 Rotor Camera 3

Camera 1

Camera 2

Figure 3.20: Stereoscopic camera system for the single-bladed CCR rotor

1.016 m

0.122 m

𝒚
𝒙

Figure 3.21: Rotor blade painted with orange fluorescent color and black dots

84
 Rotor Wake direction
Camera 1, 3

Diffuser
Camera 2

Laser Mirror
Beam

Support column Optical bench

Figure 3.22: A schematic of the DIC measurement setup

3.2.3 DIC post-processing

It should be highlighted here that output data obtained from the com-
mercial DIC software must be post-processed by the analyst to recover the
continuous time history of the rotor blade deformation. There are two fac-
tors that must be addressed: First, a whole area of each digital image is put
into a Cartesian grid and is divided into a number of interrogation windows.
Thus, the windows placed over the rotor blade structure do not correspond
to the blade-fixed coordinate system, that is, the measurement points are not

85
 Left image Right image

Center image

Figure 3.23: Unprocessed images taken simultaneously by the high-speed cam-

eras

Calibration target

Rotor blade

Figure 3.24: Camera calibration setup

86
uniformly distributed in either chordwise or spanwise direction over the blade
structure. Second, output data structure from the software is not organized
in time order because the DIC processing in the software is not performed
between images taken at different azimuthal locations but performed between
images taken at a certain azimuthal location at different time steps. This is
because of the fact that in-plane displacements between different azimuthal
locations are too large to compute with the DIC algorithm.

To deal with these concerns, output data are first interpolated to a

uniform grid over the rectangular blade span with cubic spline function, and
the interpolated data structure are reorganized in chronological order. This
uniform grid in the blade-fixed coordinate system is used for all the azimuthal
locations and time steps. This post-processing enables deformation data anal-
ysis from a variety of different perspectives; for example, one can track a single
point at the blade tip and plot its displacements as a function of time to ex-
amine the temporal trend of the particular point at the tip. Another example
could be that the analyst can average the 3D displacements measured at a
particular point on the blade over the entire time history to investigate the
mean deformation at the point with its standard deviation.

3.3 Test Envelope and Measurement Procedure

With the experimental setups described in § 3.1 and § 3.2, three indi-
vidual test campaigns were performed in the present study. Test campaign 1
was performed on the small-scale rotor hover test stand, whereas Test cam-

87
paigns 2 and 3 were performed on the large-scale hover test stand at different
rotor configurations and operating conditions. Table 3.5 summarizes the mea-
surement setups of the three test campaigns, and details of each campaign are
described in the following subsections.

Table 3.5: Summary of three test campaigns

Campaign Test scale Rotor configuration

1 Small Two-bladed, extremely flexible rotor
2 Large Single-bladed, single or coaxial rotor
3 Large Two-bladed, single rotor

3.3.1 Test campaign 1

The main purpose of this campaign was to verify if the methodology

of the time-resolved DIC with LIF strobing could be used to measure the
small-scale deformation of the extremely flexible rotor blades. The two-bladed
flexible rotor was tested at four different blade root pitch angles: 0◦ , 10◦ , 20◦ ,
and 30◦ , using the fixed-pitch hubs shown in Fig. 3.1. Additionally, the rotor
was spun at four different rotational speeds in each test in order to understand
the dynamic behavior of the blade with respect to the rotational speed. For
the 0◦ collective case, the rotor was spun at 600, 900, 1000, and 1500 RPM,
whereas for the rest of the test cases the rotor was spun at 600, 900, 1200, and
1500 RPM. The complete test matrix is shown in Table 3.6.

The collective pitch angle was first set by selecting the fixed-pitch hub,
and the rotor with the extremely flexible blades was spun at a target rotational
speed. Having synchronized the timing of two high-speed cameras and laser

88
 Table 3.6: Summary of the test matrix for Test campaign 1

Condition Root pitch angle Ω [RPM]

1 θ0 = 0◦ 600, 900, 1000, 1200
2 θ0 = 10◦ 600, 900, 1200, 1500
3 θ0 = 20◦ 600, 900, 1200, 1500
4 θ0 = 30◦ 600, 900, 1200, 1500

strobe, 6400 images were continuously taken at 16 azimuthal locations at each

rotational speed. This procedure was repeated for different collective pitch an-
gles. After the collective pitch and rotational speed sweep, undeformed images
were taken by rotating the motor by hand slowly (approximately 60 RPM) and
used as reference images for the DIC deformation calculation process.

3.3.2 Test campaign 2

During Test campaign 2, the time-resolved DIC blade deformation and

rotor loads measurements were performed on the large-scale rotor hover test
stand (see § 3.1.2) at several operating conditions. The primary objectives
of Test Campaign 2 were: (1) to expand the scale of the time-resolved DIC
technique with laser strobing, (2) to simultaneously measure the blade defor-
mation and rotor loads, and (3) to correlate these experimental results with
a numerical model. First, the single-bladed, isolated single rotor balanced
by a counter-weight was tested at a slightly negative collective pitch angle to
produce almost zero mean thrust. In this condition, the wake was in close
proximity to the rotor blades so that the force on the rotor blade was domi-
nated by unsteady random excitation, which is one of the conditions needed for

89
operational modal parameter identification. Then, the single-bladed isolated
rotor was tested at a moderate thrust, where the rotor produced 70 N corre-
sponding to a blade loading coefficient of CT /σ = 0.07. For all test conditions,
the rotational speed Ω was set at 900 RPM.

Additionally, the single-bladed CCR rotor system, shown in Fig. 3.3a,

was tested at a mean thrust of the lower rotor of 90 N (CT /σ = 0.09). This
rotor configuration was also spun at 900 RPM and the coaxial system was
trimmed to maintain torque balance. Note that a mean torque of both rotors
was set to be approximately 9.0 N-m and a mean thrust of the upper rotor
was 124 N at this operating condition. For all of the test cases, both the
blade deformation and rotor loads were simultaneously measured in the rotat-
ing frame. Three-dimensional deformation of the rotor blade was calculated at
a total of 2020 (20 chordwise and 101 spanwise) locations with the size of the
interrogation window being 21 pixels × 21 pixels. Since the digital pulse train
for triggering camera shutter was downsampled to 32/rev and the nominal
rotational speed was 900 RPM (15 Hz), the sampling rate of the time-resolved
DIC measurement was 480 Hz. At each test condition, images were continu-
ously recorded over 300 rotor revolutions yielding 9600 images per camera over
20 seconds. Due to the geometrical constraints of the experimental setup, the
blade deformation measurement was performed on the lower rotor only. All
the operating conditions tested during the Test campaign 2 are summarized
in Table 3.7.

90
 Table 3.7: Summary of operating conditions for Test campaign 2

Condition Rotor CT /σ Ω [RPM]

1 Isolated single 0 900
2 Isolated single 0.07 900
3 CCR 0.09 (lower rotor) 900

3.3.3 Test campaign 3

Test campaign 3 was an extension of the second test campaign. The

main purpose of this test campaign was to apply a variety of rotor pitch angle
control inputs to the 2 m-diameter, large-scale rotor in hover and evaluate if
the rotor loads estimation methodology developed in the dissertation could
be used in different operating (loading) conditions. To achieve the goal, sev-
eral improvements were made on the rotor configuration, pitch angle control
system, and DIC measurement setup.

While the single-bladed, single or CCR rotor with the lead-screw pitch
actuation system was used during the second test campaign, the two-bladed,
isolated single rotor with the hydraulic pitch actuation system was used dur-
ing Test campaign 3. The two-bladed rotor stand and its hydraulic actuation
system are shown in Fig. 3.3b and 3.12, respectively. The hydraulic servo actu-
ators enabled three different rotor pitch control inputs: (1) a steady collective
pitch input at a blade loading (CT /σ) of 0.125, (2) a lateral cyclic pitch input
(θ1C = 2◦ ), and (3) a collective step pitch input (2◦ step function). For all
of these conditions, the rotor was spun at 900 RPM. As was the case of Test
campaign 2, the blade deformation time history and rotor loads were simul-

91
taneously measured at each operating condition. These three test conditions
are summarized in Table. 3.8.

Table 3.8: Summary of operating conditions for Test campaign 3

Condition CT /σ Trim θ0 θ1c

◦
1 0.125 Constant pitch 9 -
◦
2 0.125 1/rev pitch 9 2◦
◦ ◦
3 - Step pitch 9 + 2 step -

Regarding the DIC measurement setup, a new stereoscopic camera sys-

tem was used for Test campaign 3 as shown in Fig. 3.25. The major difference
from the second test campaign was the orientation of the laser beam with
respect to the rotational axis. As compared to the schematic of the DIC mea-
surement during Test campaign 2 (see Fig. 3.22), the center axis of the diffused
laser beam was aligned to the rotational axis of the rotor stand, resulting in
a uniform distribution of the illumination intensity over the whole rotor disk
which improved the quality of images and DIC post-processing.

Multiple independent experiments were performed to assess repeata-

bility for each test condition. Data obtained during the tests 1 and 2 can be
phase-averaged over revolutions because the rotor was controlled by a steady
collective pitch input or a lateral 1/rev periodic pitch input. This phase-
averaging process is based on the assumption that the mean of each revolution
is treated as a statistically independent measurement sample. The standard
deviation of the sample is used to compute precision uncertainty of measured
quantities.

92
 Camera 1
Top view

Diffuser

45°

Rotor Mirror

A-shaped frame
Camera 2

Side view
Camera 1, 2

Wake direction

Beam

Mirror Laser

Optical bench

Figure 3.25: A schematic of stereoscopic camera arrangement used during Test

campaign 3

On the other hand, three independent data sets were obtained for the
test condition 3, each recorded with the digital pulse triggering the step pitch

93
actuation. The initial intention was to use the trigger as a reference point for
phase-averaging data sets taken from multiple individual experiments; how-
ever, the pitch angle responses were not consistent over the three experiments.
Figure 3.26 shows the pitch angle variations, measured by the AMR root pitch
angle sensors, for three independent measurements. The rise time of the col-
lective increase and the time lag between the trigger and blade pitch increase
were calculated to be approximately 200 and 75 milliseconds for all the three
different experiments, respectively. However, the magnitudes of the pitch an-
gle response were not consistently maintained; for example, the blade pitch
was changed from 9.5◦ to 11.5◦ during the first experiment, whereas the pitch
◦
angle was changed from 9.5 to 11.0◦ during the second experiment as shown
in Fig. 3.26. Hence, phase-averaging was not performed on the measured quan-
tities for the test condition 3, and results from the first experiment at which
the blade pitch successfully responded to the 2◦ collective step pitch input will
be primarily discussed in this dissertation.

Table 3.9: Summary of the design parameters for digital low-pass filters

Parameter Low-pass filter

Type IIR
Domain Time
Passband ripple 0.5 dB
Design method Chebyshev type I

Instead of phase-averaging, data measured during the step pitch input

operation were processed by digital low-pass filters. The design parameters
for the low-pass filter are summarized in Table 3.9. Different cut-off frequen-

94
 12 12 12
Blade 1
Pitch angle, [deg] Blade 2
11 11 11

10 10 10

9 9 9

8 8 8
27 28 29 30 31 32 35 36 37 38 39 40 24 25 26 27 28 29

1 1 1
Digital pulse

Digital pulse

Digital pulse
0 0 0
27 28 29 30 31 32 35 36 37 38 39 40 24 25 26 27 28 29
N [revs] N [revs] N [revs]

Figure 3.26: Pitch angle variations and the initial digital trigger timing ob-
tained from the three independent measurements

cies were selected for each measured quantity, thus, the actual frequencies
used for this filtering process will be described along with experimental results
in the next chapter. Since an IIR low-pass filter introduced a frequency-
dependent delay and phase shift between input and output signals, a specific
post-processing algorithm was applied to the output signal in time domain to
compensate for the delays. The algorithm essentially applies the digital low-
pass filter to the input data in both the forward and backward directions in
time to remove the delay.

In summary, this chapter covered the details of the hover test stands,

95
the blade deformation measurements using the time-resolved DIC, the rotor
loads measurements, and the test envelopes and procedures for the small- and
large-scale rotor configurations.

96
 Chapter 4

Results and Discussion

This chapter discusses the results of the rotor blade deformation mea-
surements, the operational modal identification, and the rotor loads estimation
for several different rotor systems and operating conditions. First, the results
of the time-resolved DIC deformation measurements performed on the small-
and large-scale rotor test rigs (see chapter 3) are discussed in § 4.1. Second, the
rotor blade modal characteristics identification with several OMA algorithms
based on the blade deformation time history are described in § 4.2. Finally,
the results of the rotor loads measurements and estimation are summarized
in § 4.3. A block diagram shows the organization of this chapter in Fig. 4.1.

4.1 Rotor blade deformation measurement

Results of blade deformation measurements are presented in the fol-

lowing order: (1) the small-scale DIC measurement on the 0.45 m-diameter,

Portions of this chapter were previously published as “Blade Passage Loads and De-
formation of a Coaxial Rotor System in Hover” [57] in the AIAA Journal of Aircraft and
“Full-field optical deformation measurement and operational modal analysis of a flexible
rotor blade” [75] in the Journal of Mechanical Systems and Signal Processing. All writ-
ing and figures included in this chapter are the original work of the author, incorporating
computational data from Dr. Roland Feil and Dr. Juergen Rauleder, with editing by Dr.
Jayant Sirohi.

97
 Section 4.1 Subsection

• Two-bladed, extremely flexible rotor (small-scale test rig)

Rotor blade deformation measurement • Single-bladed, single or CCR rotor (large-scale test rig)
• Two-bladed, single rotor (large-scale test rig)

Section 4.2 Subsection

• Two-bladed, extremely flexible rotor (small-scale test rig)

Rotor blade modal characteristics identification • Single-bladed, single or CCR rotor (large-scale test rig)
• Two-bladed, single rotor (large-scale test rig)

Section 4.3 Subsection

Rotor loads estimation and measurement • Single-bladed, single or CCR rotor (large-scale test rig)
• Two-bladed, single rotor (large-scale test rig)

Figure 4.1: A diagram of the organization of this chapter

two-bladed, extremely flexible rotor in § 4.1.1 and (2) the large-scale DIC mea-
surement on the 2.0 m-diameter, single-bladed or two-bladed, single or coaxial
rotor in § 4.1.2. The goals of these measurements were to verify and vali-
date if the time-resolved DIC technique with laser strobing and fluorescence
paint, developed in the dissertation, could be applied to different measurement
scales, rotor configurations, and operating conditions. Note that the correct-
ness of the 3D DIC deformation in static, non-rotating frame has been proved
by comparing results with those obtained by a laser displacement sensor, as
thoroughly described in [70].

98
4.1.1 Small-scale rotor

Figure 4.2 and Table 4.1 shows a recap of the rotor setup and test
matrix used for this small-scale DIC test, respectively. The blade deformation
measurements were performed at four different blade root pitch angles and
rotational speeds during Test campaign 1 (see in § 3.3).

228 mm
0.45
9” m
Flexible blade

Fixed-pitch hub
Flexible blade Fixed-pitch hub
DCmotor
DC motor

Shaft
Shaft encoder
encoder

ATI load cell

Figure 4.2: Recap: Side view of the rotor test article

Table 4.1: Recap: Summary of the test matrix for Test campaign 1

Condition Root pitch angle Ω [RPM]

1 θ0 = 0◦ 600, 900, 1000, 1200
2 θ0 = 10◦ 600, 900, 1200, 1500
3 θ0 = 20◦ 600, 900, 1200, 1500
4 θ0 = 30◦ 600, 900, 1200, 1500

Figure 4.3 shows the flap bending deformation of the rotor blade at 30◦
root pitch angle spun at 1200 RPM over one revolution. Note that images were
continuously taken over 400 revolutions at 16 evenly-spaced azimuthal loca-
tions, yielding 6400 images per camera at each test condition. The variation
of the blade flap bending over one revolution shown in Fig. 4.3 was obtained

99
by three steps: (i) The time history of blade deformation over 400 revolutions
was phase-averaged to yield the mean blade bending deformation as a function
of radial station at each of the 16 azimuthal locations. (ii) Displacement vec-
tors along the quarter-chord axis were extracted from the phase-averaged data
set. (iii) Data was interpolated between adjacent azimuthal angles, yielding
the out-of-plane bending deformation over the entire rotor disk; note that the
phase-averaging process eliminates most of the stochastic noise in the mea-
surement.

Figure 4.3: Flap bending deformation over the rotor disk at θ0 = 30◦

The rotor blade at a root pitch angle of 30◦ experienced a positive flap
bending deformation, increasing along the radial direction, and reached ap-
proximately 7 mm at the blade tip. The contour plot shown in Fig. 4.3 clearly

100
presents the azimuthal variation of the blade flapping motion, regardless of
the fact that blade pitch angle was fixed at the root. Over the range between
Ψ = 0◦ and 180◦ , the blade continuously flapped down from 7 mm to 4.5 mm
measured at the tip and started flapping up during the other half of the rev-
olution. This 1/rev motion was possibly caused by the combination of three
major factors: (i) the extreme torsional flexibility of the rotor blade structure,
(ii) no active stabilization mechanism, and (iii) some mean asymmetric air flow
in the test chamber due to flow recirculation. Although there was a passive
stabilization control system (i.e., a blade tip mass inducing propeller moment
as shown in Fig. 3.2), the rotor blade experienced significant elastic twist due
to the low blade torsional rigidity, resulting in azimuthal variation of blade
sectional angle of attack.

From the phase-averaged deformation over one revolution, the mean

flap bending, lead-lag bending, and torsional deformation at the blade quarter-
chord were calculated; these correspond to the static equilibrium deformation
of the rotor blade. While the flap bending and lead-lag bending were measured
from the (non-rotating) reference state of the blades, the torsional deforma-
tion was obtained by calculating slopes of the blade chord between deformed
and undeformed images at each spanwise measurement channel (76 spanwise
locations). Figure 4.4a to 4.4c shows the static deformation of each degree of
freedom of the rotor blade at 30◦ root pitch angle, spinning at four rotational
speeds. As the rotational speed increased, the magnitude of both flap and
lead-lag bending increased, which was due to the increase in lift and drag, re-

101
spectively. The negative twist along the rotor blade increased with rotor speed
due to the centrifugal moments on the tip mass; at 600 RPM, the blade had
a spanwise nose-down twist of about 1◦ , whereas at 1500 RPM it increased to
a nose-down twist of nearly 10◦ . The increase in pitch angle at the blade tip
is due to the presence of the tip mass. The shaded area in Fig. 4.4a to 4.4c
represents the estimated DIC measurement uncertainty based on the standard
deviation and the size of interrogation window as specified in the software
documentation.

During the hover testing, instabilities were observed at some operating

conditions. Figure 4.5 shows the pitch angle displacement at the tip of the
blade as a function of time, measured at different operating conditions. At 600
RPM, the amplitude of pitch angle displacements at θ0 = 0◦ and 30◦ are much
greater than at the other two root pitch angles, indicating the flutter-type
instability that occurred at these operating conditions. On the other hand,
as shown in Fig. 4.5b, the amplitudes of the tip pitch angle variation at 1500
RPM for all three cases are small, indicating stable regimes.

As described in chapter 3, the current study upgraded the conventional

DIC technique for rotating rotor blade deformation measurement by combin-
ing the following features: expansion of camera field of images to the entire
rotor disk, Laser-Induced Fluorescence (LIF) technique to obtain high con-
trast images, and camera shutter and laser strobing synchronization using a
rotary encoder. The results shown in this subsection prove that the new, time-
resolved DIC technique developed in this dissertation can be used to measure

102
 8 1
600 RPM
600 RPM
7 900 RPM
900 RPM
0 1200 RPM
1200 RPM 1500 RPM
6 1500 RPM
Deflection [mm]

Deflection [mm]
-1
5

4 -2

3
-3
2
-4
1

0 -5
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
Radial location [mm] Radial location [mm]

(a) Flap bending deformation (b) In-plane deformation

32

30
Blade pitch angle [deg]

28

26

24 600 RPM
900 RPM
1200 RPM
22 1500 RPM

20
0 20 40 60 80 100 120 140 160 180 200
Radial location [mm]

(c) Sectional blade pitch angle

Figure 4.4: Static blade deformations of each degree of freedom, along the
blade quarter-chord axis at 30◦ root pitch and different rotational speeds

deformation time history of the small-scale (0.46 m-diameter) rotor blade at

different operating conditions on the order of a few millimeters.

103
 0
= 0°
30 30 = 10°
= 10° 0
0 = 20°
25 = 20° 25 0
0 = 30°
20 0
= 30° 20 0

15 15
10 10
[deg]

[deg]
5 5
0 0
-5 -5
-10 -10
-15 -15
-20 -20
-25 -25
-30 -30
0 0.1 0.2 0.3 0.4 0.5 0 0.05 0.1 0.15 0.2
time [s] time [s]
(a) Rotor spun at 600 RPM (b) Rotor spun at 1500 RPM

Figure 4.5: Time history of the pitch angle displacement at the blade tip for
different root pitch angles

4.1.2 Large-scale rotor

Now, to expand the applicability of the time-resolved DIC technique,

several measurements were performed on the large-scale rotor hover test rig
at different rotor configurations and operating conditions. The large-scale (2
m-diameter) test rig can be configured to the single-bladed or two-bladed,
isolated single or CCR rotor system as shown in Fig. 3.3.

First, the results of the blade deformation measurements performed on

the single-bladed CCR rotor during Test campaign 2 are presented along with
numerical predictions provided from research collaborators at the Technical
University of Munich; these predictions are described in detail in Ref. [57]. The
primary goal of the testing was to verify if the time-resolved DIC technique
developed in this dissertation can be applied to the large-scale rotor blade

104
deformation measurements, by comparing the experimental results with the
numerical predictions. Note that the blade structural properties (the bending
and torsional stiffness) used in the numerical modeling were obtained from
multiple DIC deformation measurements in the non-rotating frame, resulting
from static loadings in out-of-plane, in-plane, and torsional degrees of freedom.
Further details of the structural characteristics identification can be found in
Ref. [70].

Next, results from additional measurements performed on the two-

bladed isolated single rotor during Test campaign 3 are discussed. The purpose
of the additional measurements was to apply a variety of pitch angle control
inputs to the large-scale rotor and demonstrate if the DIC technique is able
to capture the dynamic resultant motion of the rotating blade responding to
different rotor loading conditions. Measurement procedures and test envelopes
on the large-scale test rig are summarized in § 3.3.

4.1.2.1 Single-bladed CCR rotor (Test campaign 2)

The single-bladed CCR rotor system, shown in Fig. 4.6 as a recap, was
tested at a mean thrust of the lower rotor of 90 N (CT /σ = 0.09). The DIC
deformation measurements were carried out only on the lower rotor of CCR
system (the orange-painted blade shown in Fig. 4.6), due to the geometrical
limitation of experimental setup. This rotor was spun at 900 RPM and the
coaxial system was trimmed to achieve torque balance, that is, the total yaw
moment was trimmed to be zero.

105
 Transmission Counter weight

Hydraulic motor

1.016 m

Fluorescent-painted rotor blade

Figure 4.6: Recap: Single-bladed, CCR rotor configuration

The measured and predicted out-of-plane lower rotor blade deforma-

tion over the entire rotor disk at a blade loading of 0.09 (CT /σ = 0.09) are
shown in Fig. 4.7a and 4.7b, respectively. Figure 4.7a were obtained by the
same processing method as the contour plot shown in Fig. 4.3 in the small-
scale DIC measurement. Overall, there was excellent agreement between the
measured and predicted flap bending (out-of-plane) deformation. The blade
tip deflection reached 75 mm, which is approximately 10 times larger than the
tip deflection that the small-scale, 0.46 m-diameter rotor experienced. Con-
sidering the fact that the radius of the large-scale rotor is 1.016 m, which is
also 4.45 times larger than that of the small-scale rotor, the time-resolved DIC
successfully demonstrates its capability of measuring out-of-plane deformation
time history of a rotating blade at different measurement scales.

106
 (a) Measured CCR lower rotor blade flap bending de-
formation
180°
150° 210° 80

70
120° 240° 60

Deflections, [mm]
50
0.2 270° 40
90° 0.6 0.4

30

60° 300° 20

10
30° 330° 0
= 0°

(b) Predicted CCR lower rotor blade flap bending de-

formation

Figure 4.7: Flap bending deformation of the CCR lower rotor blade at CT /σ
= 0.09

107
 Figure 4.8 shows the measured and predicted CCR lower rotor blade
tip displacement extracted at the blade quarter chord location as a function
of azimuthal angle at CT /σ = 0.09. The trend revealed the significant 2/rev
characteristics, that are associated with the blade crossings located at ψ =
0◦ and 180◦ . This 2/rev characteristic reflects the lower rotor blade response
to the transient loads generated by the upper–lower rotor blade crossing. For
this investigated thrust level (i.e., a blade loading coefficient of CT /σ = 0.09),
vibrations around the mean deflection of 69.6 mm occurred with a maximum
deflection of 74.6 mm and a minimum deflection of 66.4 mm. The maximum
peaks of blade tip deflection were seen at approximately ψ = 30◦ and 210◦ ,
whereas the minimum peaks were observed at approximately ψ = 150◦ and
290◦ . The 2/rev half-peak-to-peak magnitudes of 4.1 mm corresponded to
5.9% of the mean deflection. Predicted results correlated satisfactorily with
these measurements in both magnitude and phase. The numerical predictions
showed that the upper rotor blade had a greater mean deformation, which
is consistent with its greater thrust share compared to the lower rotor in a
torque-balanced trim condition.

Figure 4.9 shows the measured and predicted blade deformation of the
lower rotor in the coaxial system along the blade quarter-chord axis as a func-
tion of radial station. The two lines correspond to the azimuthal locations of
maximum and minimum tip deflection that were previously determined from
Fig. 4.8. There was excellent agreement between the measured and predicted
flap bending (out-of-plane) deflection, with the largest discrepancy limited to

108
 90 Coaxial lower measurement
Coaxial lower prediction
Coaxial upper prediction
85

80

75

70

65

60
0 60 120 180 240 300 360

Figure 4.8: Lower rotor blade tip displacement comparison between measure-
ment and prediction

less than 3% (2.5mm) at the approximately 55% of the radial station on the
minimum deflection curve. Those errors were also more apparent over the
inboard portion due to smaller magnitude deflections.

The phase lag between the maximum and minimum deflection and the
blade crossings (i.e., ψ = 0◦ and 180◦ ), shown in Fig. 4.8, can be explained by
the influence of rotor blade bound circulation. As the upper and lower rotor
blades approach each other, the bound vortex on one blade induces interac-
tional upwash on the other blade, resulting in an increase in effective angle
of attack of the other blade, as shown in Fig. 4.10. The increase in angle of
attack causes a 2/rev flap bending deflection of the rotor blade. Aerodynamic

109
 80
Measurement
Prediction
60
max.

40 min.

20

0
0 0.2 0.4 0.6 0.8 1

Figure 4.9: Maximum and minimum flap bending deformation of the lower
rotor blade as a function of blade radial station at CT /σ = 0.09

angles of attack increase up to the azimuthal angle of blade crossing, reach

a peak just before crossing, and eventually drop to a minimum after blade
passage due to the change in sign of interactional induced velocity (upwash to
downwash). These blade deformation measurements indicate that the tran-
sient vibration of the coaxial rotor system is due to the interaction mechanism
of blade bound circulation.

Over all, the large-scale time-resolved DIC measurements were success-

fully carried out on the single-bladed coaxial lower rotor with good agreement
between the experimental and numerical results. Additionally, the measure-
ment data revealed the significant 2/rev, transient response of the rotor blade

110
 Before crossing After crossing

Γ"##$% Γ"##$%
Upwash Downwash
Upper

Γ&'($% Γ&'($%
Free stream
velocity

Lower Resultant velocity

Airflow induced by
upper rotor
Airflow induced by
lower rotor

Figure 4.10: Interaction of blade bound circulation in the CCR rotor system

due to the blade crossings in the CCR rotor system, which could only be
identified by analyzing the continuous time history of the blade deformation.

4.1.2.2 Two-bladed single rotor (Test campaign 3)

In addition to the blade deformation measurements on the single-bladed

coaxial rotor system as described above, the time-resolved DIC technique was
used to measure the blade responses to a variety of rotor loading conditions
during Test campaign 3. The large-scale rotor was reconfigured to the two-
bladed, isolated single rotor with the hydraulic actuation system as shown in
Fig. 4.11 as a recap from chapter 3. The hydraulic servo actuators enabled
three different rotor pitch control inputs: (1) a steady collective pitch input at
a blade loading (CT /σ) of 0.125, (2) a lateral cyclic pitch input (θ1C = 2◦ ), and
(3) a collective step pitch input (2◦ step function). For all of these conditions,
the rotor was spun at 900 RPM. Details of the hydraulic, high-speed rotor
pitch actuation system are available in § 3.1.2.

111
 Two-bladed
single rotor

Swashplate assembly

A-shaped frame

Transmission

Figure 4.11: Recap: Two-bladed, isolated single rotor configuration

The first operating condition tested on the two-bladed single rotor was
a steady collective pitch input (θ0 = 9◦ ) at a blade loading CT /σ of 0.125.
The deformation measurement was carried out on one of the two rotor blades
during this test campaign. Note that the deformation data shown in the
current section were computed based on the averaged reference image taken
at the blade root pitch angle of 0◦ , spun at the same rotational speed of 900
RPM.

First, the out-of-plane blade deformation over the entire rotor disk is
shown in Fig. 4.12. The flap bending deflection reached nearly 90 mm at the
blade tip with little azimuthal variation. The axisymmetric trend represents

112
the fact that the constant collective pitch angle of 9◦ was properly applied
to the rotor and steady-state loading over the rotor disk was achieved. The
phase-averaged blade tip displacement over one revolution was then extracted
from the contour plot and is shown in Fig. 4.13, representing the data both
from the 32 measurement points and from the spline interpolation. A slight
fluctuation over the entire range was clearly observed in Fig. 4.13, especially
over the range from 60◦ to 180◦ where a large rise-and-fall between 85 mm and
95 mm occurred. One possible cause of this dip is that one of three hydraulic
servo actuators might have had non-ideal free-play in the linkage path, such
as a rod end, resulting in the ± 5 mm fluctuation that periodically occurred
over the range.

Figure 4.12: Out-of-plane blade deformation over the entire rotor disk

113
 110
32 data points
105 Interpolation

100
Tip path, [mm]

95
90
85
80
75
70
0 60 120 180 240 300 360
Azimuth, [deg]
Figure 4.13: Blade tip displacement as a function of azimuthal angle

Recall that one of the advantages of the DIC technique is that it can
measure the three-dimensional (3D) full-field deformation on the surface of
a target structure. Figure 4.14 shows the 3D static deformation distributed
over the blade span. This 3D deformation with high spatial resolution cannot
be obtained if conventional on-blade sensors, that can only perform pointwise
measurements, are used. From the 3D plot, deformations of each degree of free-
dom (flap, lead-lag, and torsion) along the quarter-chord axis were extracted
and plotted in Fig. 4.15.

The spanwise variations of the blade flap and lag deformation shown in
Fig. 4.15a and 4.15b represent the out-of-plane and in-plane static deflection at

114
the given rotor pitch angle and rotational speed, respectively. One can clearly
see a flap-up and lagging motion corresponding to increase in lift and drag
forces as the blade angle of attack increases. While the flapwise and chordwise
deformation along the blade quarter-chord axis was directly interpolated from
the 3D distributed plot, the sectional blade pitch angle, shown in Fig. 4.15c,
was obtained based on the 1st-order polynomial fitting to the 3D surface height
distribution at each spanwise station. The relationship between the sectional
pitch angle and the slope of the fitting is as follows:

wi (yi ) = P1 yi + P0
(4.1)
θi = arctan P1
where wi , yi , and θi are out-of-plane displacement, chordwise location, and
local torsional deformation at a spanwise station i, respectively. P1 and P0
represent the slope and y-intercept of the 1st-order polynomial fit.

As can be seen in Fig. 4.15c, the elastic twist of the rotor blade was
small, considering the fact that the blade root pitch was trimmed to the
mean collective angle θ0 = 9◦ measured in real time using the AMR sensor
(see § 3.1.2). There was a dip observed in regions of the blade tip, however,
its magnitude was within a fraction of a degree and on the same order of
measurement uncertainty. Note that highlighted areas in these plots represent
uncertainties of each measured quantity.

For the second operating condition employed during Test campaign 3,

the rotor was trimmed at a steady collective pitch + a lateral cyclic pitch.

115
 Figure 4.14: 3D static deformation distributed over the entire blade span

The collective pitch was adjusted so that the mean blade loading of this test
condition remained the same as for the first operating condition (θ0 = 9◦ ).
The lateral cyclic pitch θ1C was set to be ±2◦ , by introducing a swashplate in-
clination with the hydraulic-driven actuators. Rotor track was achieved using
the real-time hub loads measurement, specifically hub pitching and rotating
moment in the rotating frame.

The contour plot in Fig. 4.16 shows the phase-averaged flap bending
deformation over one revolution. Due to the lateral pitch input and the gy-
roscopic phase shift, the whole rotor disk tilted in the lateral direction. This
behavior was more pronounced in the blade tip displacement, as shown in
Fig. 4.17. The sinusoidal motion over one revolution as well as the phase shift

116
 110 0
100
-2.5

Leag-lag deformation, [mm]

90
-5
Flap bending, [mm]

80
70 -7.5
60
-10
50
40 -12.5
30 -15
20
-17.5
10
0 -20
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Radial location, [%R] Radial location, [%R]
(a) Out-of-plane deformation (b) In-plane deformation
10.5
Torsional deformation, [deg.]

10

9.5

8.5

8
20 30 40 50 60 70 80 90 100
Radial location, [%R]
(c) Torsional deformation

Figure 4.15: Static blade deformations of each degree of freedom, along the
blade quarter-chord axis

can be clearly seen in the tip displacement.

In Fig. 4.17, the maximum flap bending deflection occurred at the az-
imuthal angle of 250◦ , whereas the minimum occurred at 60◦ . The tip deflec-
tion difference between the maximum and minimum reached approximately 75
mm. In other words, it was demonstrated that this large sinusoidal oscillatory

117
Figure 4.16: Out-of-plane blade deformation over the entire rotor disk for the
lateral pitch angle input

motion of the 2 m-diameter rotor at 1/rev frequency (15 Hz) was successfully
captured by the DIC technique.

The last operating condition tested during Test campaign 3 was a col-
lective step change input. The rotor pitch angle was first set to be at a con-
stant collective angle θ0 = 9◦ , corresponding to the blade loading (CT /σ) of
0.125. Once the rotor reached a steady-state condition, the collective pitch was
rapidly increased by approximately 2◦ to examine the rotor transient responses
to the dynamic pitch control from an aeroelastic perspective. This type of step
(practically ramp) pitch input excites the structure such that the responses at
different modal frequencies could be studied, and may help to extract higher

118
 140
32 data points
Tip displacement, [mm] Interpolation
120

100

80

60

40
0 60 120 180 240 300 360
Azimuth, [deg]
Figure 4.17: Blade tip displacement as a function of azimuthal angle for the
lateral pitch angle input

modes with the operational modal analysis. As for the other test conditions,
the time history of the rotor blade deformation during the collective step pitch
input was measured using the time-resolved DIC technique.

Figure 4.18 represents the rotor blade deformation of each degree of

freedom (flapwise, chordwise, and torsional deformation) measured at the tip
as a function of the number of revolution. To clearly observe the effect of the
step input on the blade dynamics, a low-pass filter was applied to the measured
blade deformation with a 3dB cutoff frequency of 13 Hz. Note that the cutoff
frequency was manually selected by trial-and-error process. These responses

119
were initiated during the 28th revolution and the rise time of the blade flap,
lag, and torsion was computed to be about 280 milliseconds (4.2 revolutions)
from Fig 4.18a to 4.18c.

Due to the step change, the resultant flapwise deflection at the blade
tip increased by 20 mm and reached nearly 115 mm in Fig. 4.18a. The step
response in chordwise degree of freedom due to the pitch angle change was also
observed, and the change was approximately 10 mm according to Fig. 4.18b.
Additionally, the local pitch angle at the blade tip was computed using Eq. 4.1
and is shown in Fig. 4.18c as a function of the number of revolution. First, the
local twist at the blade tip was settled at approximately 9◦ , then increased to
11.5◦ at maximum immediately after the step change, and stabilized at nearly
11◦ .

In summary, the time-resolved DIC technique developed in this dis-

sertation successfully measured the blade deformations in the following rotor
configurations and operating conditions: (i) the small-scale, extremely flexible
rotor at four different root pitch angles and rotational speeds, (ii) the large-
scale, single-bladed, coaxial lower rotor at a blade loading of 0.09 with good
agreement to numerical simulations, (iii) the large-scale, two-bladed, isolated
single rotor with a constant collective pitch input, a 1/rev periodic pitch input,
and a collective step pitch input.

120
 130 -10
Out-of-plane deformation, [mm]

In-plane deformation, [mm]

120 -15

110 -20

100 -25

90 -30

80 -35
20 25 30 35 40 45 50 55 60 65 70 20 25 30 35 40 45 50 55 60 65 70
N [rev] N [rev]
(a) Out-of-plane deformation (b) In-plane deformation
13
Sectional pitch angle, [deg.]

12

11

10

8
20 25 30 35 40 45 50 55 60 65 70
N [revs]
(c) Torsional deformation

Figure 4.18: Blade tip deformations of each degree of freedom as a function of

time

4.2 Rotor Blade modal characteristics identification

Based on the rotor blade deformations measured by the DIC technique

as described in the previous section, modal properties of the small- and large-
scale rotor blade in the rotating frame were identified using several OMA
algorithms. Figure 4.19a and 4.19b show the fluorescence-painted small- and

121
large-scale rotor blade, respectively. Results of the modal parameters iden-
tification on the 0.46 m-diameter rotor blade are first presented in § 4.2.1,
and results of the OMA analysis on the large-scale rotor blade are discussed
in § 4.2.2.

Tip mass
Blade grip Leading edge

Trailing edge

202 mm

(a) Recap: Small-scale rotor blade

1.016 m

0.122 m

𝒚
𝒙

(b) Recap: Large-scale rotor blade

Figure 4.19: Rotor blades with fluorescent paint and random speckle pattern

4.2.1 Small-scale rotor blade characteristics

Modal identification deals with three parameter: natural frequencies,

mode shapes, and damping ratios. The OMA algorithms used for the small-

122
scale rotor blade characteristics were the combined NExT-ERA approach and
CP algorithm. Each parameter of the small-scale rotor blade at different op-
erating conditions extracted by OMA processing will be separately addressed
in this section.

Additionally, an aeroelastic analysis of the flexible rotor blade was per-

formed using a finite element based numerical model; details of this analysis
can be found in Ref. [76]. This computational prediction tool numerically
solves non-linear aeroelastic equations of motion including coupled axial elon-
gation, lead-lag (in the rotor plane) bending, flap (normal to the rotor plane)
bending and torsional degree of freedom. The aerodynamic part of the analy-
sis includes unsteady terms based on two-dimensional strip theory. The modal
frequencies and mode shapes identified by OMA are compared to predictions
from this numerical model.

The primary objectives of this process were: (1) to examine how many
modes can be identified using the OMA algorithms based on deformation data
measured by the time-resolved DIC technique, (2) to compare the results iden-
tified by the OMA algorithms, with numerical simulations, and (3) to evaluate
the NExT-ERA and CP algorithm regarding accuracy, usability, and robust-
ness of each identification process.

4.2.1.1 Modal frequency

First, identification of the rotating frame natural frequencies is dis-

cussed. Before proceeding to the operational modal analysis, the time history

123
and frequency spectrum of the local pitch angle displacement measured at
the blade tip for θ0 = 30◦ case, rotating at 1200 RPM, were first plotted in
Fig. 4.20a and 4.20b. The time history of the pitch angle variation at the tip
shows periodic motion, corresponding to the oscillation at 1/rev frequency.
It can be seen that in this operating condition the spectrum peaks are clear
enough to identify which peaks are associated with operating frequencies or
modal frequencies. The first peak right next to the 1/rev and second peak
between 2/rev and 3/rev harmonics correspond to the first and second modal
frequencies as shown in Figure 4.20b. Even if the excitation to the system is
random, there are always 1/rev harmonics on any rotating system.

Figure 4.21 shows the frequency spectrum of the blade tip out-of-
plane displacement in frequency domain (non-dimensionalized by the rotor
frequency), in log scale. This is basically another way of looking at the data
taken at 30◦ root pitch angle spun at 1200 RPM. It appears that the tip dis-
placement variation was a combination of random and periodic motion as also
observed in both Fig. 4.20b and 4.21. There were obvious peaks observed in
the frequency spectrum, corresponding to the harmonics of rotational speed
(i.e., 1200 RPM) as well as the rotating natural frequencies of the rotor blade.
Note that the presence of harmonic components might be an issue if the eigen-
frequencies of the test specimen are quite close (on the order of 0.1 Hz) to the
harmonic excitation, which is not the case in this study. If the frequency of
input harmonic excitation is close enough to a natural frequency of the struc-
ture to cause failure of the OMA-based modal identification, one would need

124
 10

Pitch angle at blade tip [deg]

7.5
5
2.5
0
-2.5
-5
-7.5
-10
0 0.1 0.2 0.3 0.4 0.5
time [s]
(a) Time domain

0 10 20 30 40 50 60 70 80 90 100 110 120

(b) Frequency domain

Figure 4.20: Pitch angle displacement at the tip for θ0 = 30◦ , Ω = 1200 RPM

125
special treatments to account for the harmonic components; see Mohanty and
Rixen [77].

101
Tip displacement [mm]

10-1

10-3

10-5
0 1 2 3 4 5 6 7 8
N/rev
Figure 4.21: Log-scale frequency spectrum of the out-of-plane displacement
measured at the blade tip for θ0 = 30◦ , Ω = 1200 RPM

Figure 4.22 and 4.23 show the modal frequencies identified by the OMA
algorithms as a function of rotor speed; these plots are known as fan plots.
In addition, the fan plot for θ0 = 30◦ shows the comparison between these
experimental results and predictions from the aeroelastic numerical analysis.
Note that these fan plots are presented based on measured data in air, i.e.,
aerodynamic damping is included. Predictions for θ0 = 0◦ were not available
because the numerical model could not converge for this root pitch angle. One

126
possible explanation for the divergence is instability of rotor blade motion with
zero-root pitch angle; during measurements, unstable behaviors of the rotor
blade were observed for all the rotational speeds. However, instability analysis
is beyond the scope of the current study.

There was good agreement between the CP and NExT-ERA up to the

second mode for θ0 = 0◦ and θ0 = 30◦ over a range of different rotational
speeds; the extracted modal frequencies for θ0 = 30◦ also agreed well with
numerical predictions. These results indicate that this measurement method-
ology, that is, the DIC deformation measurement in conjunction with OMA
algorithm, is quite a useful tool for identifying the modal frequencies of a
rotating rotor blade without any input excitation.

4.2.1.2 Mode shapes

Mode shapes of the entire rotor blade, extracted by the CP algorithm,

at θ0 = 30◦ spinning at 1200 RPM are plotted in Fig. 4.24 and 4.25. Figure 4.24
corresponds to the first mode and Fig. 4.25 shows the second mode, respec-
tively. The first mode shape was mainly dominated by the flap bending while
the second mode was influenced by the torsional deformation. This bending
coupling is because the sectional center of gravity and the elastic center are
not coincident over the entire span of the rotor blade.

Figure 4.26 and 4.27 shows the components of each degree of freedom
along the blade quarter-chord line for the first and second mode shapes, for
θ0 = 0◦ and θ0 = 30◦ root pitch angles at 1200 RPM, respectively. The

127
 100
4/rev
1st mode CP
90 2nd mode CP
1st mode NExT/ERA
80
2nd mode NExT/ERA
3/rev
70
Frequency [Hz]

60
50
2/rev
40
30
1/rev
20
10
0
0 300 600 900 1200 1500
[RPM]

Figure 4.22: Fan plot for θ0 = 0◦ , comparing natural frequencies extracted

with the CP and NExT-ERA algorithms

amplitudes of the mode shapes are normalized by the maximum value among
the flap, lead-lag, and torsional degrees of freedom, to identify the contribution
of each degree of freedom to each mode.

It can be seen that at θ0 = 0◦ , the lead-lag bending degree of freedom

contributed to the first mode shape much less than the flap and torsion degrees
of freedom. The contribution of lead-lag bending for the second mode at this
same root pitch angle case was also small compared to the other two degrees
of freedom, which suggests that the first and second modes of the rotor blade

128
 100
1st mode CP 4/rev
90 2nd mode CP
1st mode NExT/ERA
80 2nd mode NExT/ERA
70 1st mode prediction 3/rev
2nd mode prediction
Frequency [Hz]

60
50
2/rev
40
30
1/rev
20
10
0
0 300 600 900 1200 1500
[RPM]

Figure 4.23: Fan plot for θ0 = 30◦ , comparing natural frequencies extracted
with the CP and NExT-ERA algorithms, along with numerical predictions

mainly consisted of coupled flap and torsional deformations. Now comparing

the amplitude of the flap mode to that of the torsional mode, the largest
contribution for the first mode was the flap bending while that for the second
mode was the torsional deformation. The trend of the mode shapes for θ0 = 0◦
and θ0 = 30◦ was almost the same, except for the amplitude of tip deflection
of the second flap mode.

The mode shapes extracted using the two different OMA algorithms
are compared to the numerically-predicted results for θ0 = 30◦ , operated at

129
Figure 4.24: The first mode shape of the rotor blade for θ0 = 30◦ at 1200 RPM

1200RPM. Figures 4.28 and 4.29 show the first and second mode shapes of all
three degrees of freedom, identified by CP, NExT-ERA, and the computational
prediction tool, respectively. For the first mode shown in Fig. 4.28, the lead-
lag and torsional mode shapes identified by the CP algorithm appear to be
in excellent agreement with those predicted by the numerical model; however,
there is a minor discrepancy between the flap mode shape obtained by CP
and the computational tool. On the other hand, the mode shapes obtained
by the NExT-ERA algorithm are quite different from those predicted by the
numerical model for all the degrees of freedom. For the second mode shown in

130
Figure 4.25: The second mode shape of the rotor blade for θ0 = 30◦ at 1200
RPM

Fig. 4.29, the CP-based mode shapes of all three degrees of freedom agree quite
well with the numerical results, as was observed for the first mode comparison.
However, the mode shapes obtained by the NExT-ERA process are not well
correlated with the numerical prediction, except for the lead-lag degree of
freedom.

To quantify the differences between the shapes identified by the OMA

algorithms and the numerical model, the Modal Assurance Criterion (MAC)
was used in this study; details can be found in Ref. [78]. The MAC is one of

131
 1 1st mode
2nd mode

Flap
0

-1
0 0.25 0.5 0.75 1
1
Lead-lag

-1
0 0.25 0.5 0.75 1
1
Torsion

-1
0 0.25 0.5 0.75 1
Radial location

Figure 4.26: The first and second mode shapes along the quarter-chord axis,
for θ0 = 0◦

the most common tools for the quantitative comparison of two modal vectors
and is defined in terms of the normalized scalar product of the two vectors
{ϕ1 } and {ϕ2 } as:

|{ϕ1 }T {ϕ2 }|2

M AC = . (4.2)
({ϕ1 }T {ϕ1 })({ϕ2 }T {ϕ2 })

The MAC is calculated for two types of comparisons: (1) between the
NExT-ERA and the numerical model, and (2) between the CP and the numer-
ical model. Table 4.2 summarizes the MAC values calculated from the mode

132
 1
1st mode

Flap
2nd mode
0

-1
0 0.25 0.5 0.75 1
1
Lead-lag

-1
0 0.25 0.5 0.75 1
1
Torsion

-1
0 0.25 0.5 0.75 1
Radial location

Figure 4.27: The first and second mode shapes along the quarter-chord axis,
for θ0 = 30◦

shapes identified by the OMA algorithms and the mode shapes predicted by
the numerical model. These mode shapes were obtained from the operating
condition where the blade root pitch angle was 30◦ . Based on the MAC val-
ues, the mode shapes extracted by the CP algorithm correlated quite well
with the predicted mode shapes (MAC values larger than 0.9 indicate con-
sistent correspondence between two modal vectors). However, the low MAC
values between mode shapes extracted by NExT-ERA and predicted mode
shapes indicate poor resemblance of these two mode shapes, except for the

133
 CP
1 NExT/ERA
Prediction

Flap
0

-1
0 0.25 0.5 0.75 1
1
Lead-lag

-1
0 0.25 0.5 0.75 1
1
Torsion

-1
0 0.25 0.5 0.75 1
Radial location

Figure 4.28: Comparison of the first mode shapes estimated by the CP and
NExT-ERA as well as numerical prediction for θ0 = 30◦ at 1200 RPM

1st and 2nd lag modes. Note that this study uses Partial Modal Assurance
Criterion (PMAC) where only a selected degree-of-freedom from a complete
modal vector is used to compute the consistency of identified modes. In this
way, linearity of a mode shape of each degree-of-freedom can be evaluated.

The discrepancies between the mode shapes identified by NExT-ERA

and the numerical model are mainly associated with the difference in the num-
ber of inflection points. For example, for the first torsional mode shape, there
are two inflection points on the NExT-ERA result, whereas there is no inflec-

134
Table 4.2: Summary of the MAC values calculated from the measured and
predicted mode shapes, obtained at θ0 = 30◦

1st mode 2nd mode

Flap Lag Torsion Flap Lag Torsion
CP 0.874 0.855 0.991 0.932 0.942 0.956
NExT-ERA 0.117 0.835 0.226 0.041 0.586 0.022

tion point on the CP and numerically-predicted results. One possible reason

of the poor mode shape estimation by NExT-ERA is that the fictitious and
real mode obtained by NExT-ERA analysis were so close to each other in fre-

1
Flap

-1
0 0.25 0.5 0.75 1
1
Lead-lag

-1
0 0.25 0.5 0.75 1
1
Torsion

-1
0 0.25 0.5 0.75 1
Radial location

Figure 4.29: Comparison of the second mode shapes estimated by the CP and
NExT-ERA as well as numerical prediction for θ0 = 30◦ at 1200 RPM

135
quency domain that the real mode could not be properly extracted. Another
possible reason is that the mode shape extraction by means of NExT-ERA
requires a number of parameters and post-processing steps that must be cho-
sen by an analyst, such as applying polynomial fits or removing outliers; these
post-processing manipulations could result in poor mode shape identification.
A block diagram in Fig. 2.5 shown in § 2.2 summarizes the advantages and
disadvantages of these two OMA algorithms. In contrast, the mode shapes
identified by CP algorithm without any post-processing showed good correla-
tions with the numerical prediction results.

4.2.1.3 Modal damping

Modal damping has been reported to be the most difficult value among
the modal parameters to be estimated with OMA approaches [79, 80]. The
present study calculated the damping ratios using a curve fit based on a single-
degree-of-freedom impulse response to the auto-correlation function of the sep-
arated source signal. Table 4.3 summarizes the estimated modal damping of
the rotor blade at a root pitch angle of θ0 = 0◦ for different rotational speeds.
For all the cases, the values of modal damping were too small if considering
the influence of aerodynamic damping on the rotor blade motion. Table 4.3
also shows the estimated damping values by NExT-ERA [81]. All the damping
values from the NExT-ERA analysis were also less than 3% as is the case of
the CP analysis.

One of the typical sources of poor damping extraction is the presence

136
of harmonic excitations on a target structure, as discussed by Mohanty and
Rixen [77]. If the harmonic frequency is very close to the modal natural
frequency, the OMA algorithm may select this harmonic component of the
structural response as one of the eigenvalues. In the current study, however,
each mode was selected after excluding the sources with frequencies that were
integer multiples of once-per-revolution frequency, as well as the corresponding
mode shape. Thus, the identified mode should not be associated with the
harmonic excitations, and the cause for the low modal damping is still unclear.
These results imply that the modal damping identification process must be
further refined to improve accuracy.

Table 4.3: Summary of modal damping ratios for different rotational speeds
at 0◦ root pitch angle

Ω [RPM] 600 900 1000 1200

Modal damping, CP [%] 2.88 0.37 0.11 0.16
Modal damping, NExT-ERA [%] 1.01 0.83 0.75 0.54

In summary, there are three key conclusions from this modal identifi-
cation processing based on the small-scale rotor blade deformation measure-
ment: (1) The first two modes were identified including the natural frequen-
cies, mode shapes, and modal damping ratios. (2) Modal parameters identified
by CP algorithm showed good agreement with the results provided from the
numerical model. (3) Identification process with CP algorithm was more ef-
ficient and user-friendly, as compared to that with the combined NExT-ERA
algorithm. Only the first two modes were identified; this could be because

137
higher-frequency deformations tend to be small and might approach the noise
floor of the DIC technique itself. Increasing the signal-to-noise ratio of the
DIC measurements is essential to enable identification of higher modes. The
conclusions (2) and (3) are somewhat related, that is, the complicated manual
steps with NExT-ERA algorithm resulted in poor quality of modal parameter
identification.

4.2.2 Large-scale rotor blade characteristics

Modal parameter identification based on the small-scale rotor blade re-

sponses measured by the DIC technique was completed in the previous section.
Now, it must be verified if this combined DIC-OMA approach is applicable
to the large-scale rotor blade characteristics identification before the ultimate
goal of this current study is pursued, i.e., rotor loads estimation based on blade
deformation and modal properties. Considering the lessons learned from the
small-scale rotor blade characteristics identification, CP algorithm was used
to identify the modal parameters for the large-scale rotor blade. As was the
case of rotor blade deformation discussed in § 4.1.2, experimental results are
compared to numerical predictions provided from Ref. [57].

To extract the modal parameters, two different types of excitation to

the rotor blade were tested: (i) random excitation at zero thrust and (ii)
collective step input excitation. The former was applied to the single-bladed,
single rotor during Test campaign 2, whereas the latter was applied to the
two-bladed, single rotor during Test campaign 3. This section describes and

138
compares the results of each test case with the numerical model.

4.2.2.1 Random excitation (Test campaign 2)

Random vibration tests were performed on the single-bladed, isolated

single rotor rotating at 900 RPM. Measured random responses were processed
with CP algorithm to extract modal parameters of the rotor blade. The ran-
dom forcing was obtained by setting a slightly negative rotor collective pitch to
achieve almost zero mean thrust with random fluctuations. This would meet
the assumption for the OMA algorithm, that is, the deformation input to the
algorithm should be the response to random excitation.

Before discussing the OMA results, frequency content of the measured

rotor thrust and blade tip displacement are examined. Figures 4.30a and 4.30b
show the rotor thrust variation in time and frequency domain, respectively. It
appeared that the thrust variation was random in the time domain. However,
there were obvious peaks observed in the frequency spectrum. These peaks cor-
responded to the harmonics of N -per-revolution frequencies, which is inherent
to any rotating system and is exacerbated by the presence of a counter-weight
(see Fig. 4.6), i.e., dissimilar rotor blades. Importantly, the magnitude of the
thrust increased with frequency and was fairly constant at high frequencies,
which lends confidence to the assumption of random excitation.

The tip displacement in time and frequency domain, shown in Figs. 4.31a
and 4.31b, also exhibited a trend similar to that for the rotor thrust response.
The harmonic components of the rotating frequency were the significant con-

139
 25 1
10
20
0
15 10

10
10-1

Thrust [N]
Thrust [N]

5
0 10-2
-5
-10 10-3

-15
10-4
-20
-25
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Time [s] N/rev

(a) Time domain (b) Frequency domain

Figure 4.30: Thrust measured on the single-bladed single rotor at zero mean
thrust

tributors to the response, as observed in the results of the hover testing on

the extremely flexible rotor system; (see § 4.2.1). However, since the modal
frequencies were not close to the harmonic excitations (on the order of 0.1 Hz),
no special treatment was required.

Now, the CP algorithm was used to extract the rotating natural fre-
quencies and mode shapes of the rotor blade. The fan plot in Fig. 4.32 shows
the natural frequencies identified by the CP algorithm (at the nominal rota-
tional frequency of 900 RPM) along with numerical predictions. Note that
the natural frequencies at 300 RPM, 600 RPM, and 1200 RPM were obtained
in previous experiments by Cameron et al. [35], who used a different modal
identification algorithm on the same experimental setup. It can be seen that
the extracted natural frequencies agree well with the numerical predictions as
well as with previous measurements over different rotational frequencies.

140
 6 2
10
5
4 101
Tip displacement [mm]

Tip displacement [mm]

3
100
2
1 10-1
0
-1 10-2
-2
-3
-3 10
-4 -4
10
-5
-6
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Time [s] N/rev

(a) Time domain (b) Frequency domain

Figure 4.31: Measured variation of tip displacement on the single-bladed single

rotor at zero mean thrust

Figure 4.33 to 4.35 shows the first flap, the first lag, and the second
flap rotating mode shapes compared to predicted results, respectively. There
was good agreement between the numerically predicted characteristics and the
results obtained by the CP algorithm for all three mode shapes. The MAC
was also used for the results of the large-scale rotor blade modal parameter
identification, to quantify the differences between the measured and predicted
mode shapes; see Eq. 4.2. Table 4.4 summarizes the MAC values of the mode
shapes as well as the rotating frame natural frequencies for both identified and
predicted parameters. This excellent agreement between the CP-derived and
numerically-obtained modal parameters indicates that the DIC deformation
measurement in conjunction with the CP algorithm can also be used to identify
the rotating frame natural frequencies and modes shapes of the large-scale
rotor blade.

141
 200
10/rev 9/rev
Measurement
180 Cameron et al. [35] 8/rev
Prediction
160 3rd flap
7/rev
140
6/rev
Frequency, Hz

120
5/rev
100
4/rev
80
2nd flap 3/rev
60
1st lag
40
2/rev

20 1/rev
1st flap
0
0 25 50 75 100 125 150
Rotational Speed, % nominal

Figure 4.32: Fan plot with measured natural frequencies (100% = 900 RPM),
compared with numerical predictions

Table 4.4: Summary of the rotating natural frequencies and MAC values cal-
culated from the measured and predicted mode shapes.

1st Flap 1st Lag 2nd Flap

Freq., Hz MAC Freq., Hz MAC Freq., Hz MAC
Measurement 17.9 39.0 66.0
0.999 0.998 0.992
Prediction 19.6 39.8 66.2

4.2.2.2 Step function excitation test (Test campaign 3)

Although the combined DIC-OMA approach was able to extract modal

parameters from the random response obtained during the Test campaign 2
as described above, the identified modes were limited to only the first flap,

142
 1
Prediction
0.8
Modal displacement
Measurements
0.6

0.4

0.2

-0.2
0 0.2 0.4 0.6 0.8 1
Radial station r/R

Figure 4.33: Measured rotating mode shapes of the first flap mode at nominal
rotational speed (900 RPM) compared with numerical predictions

0.8
Modal displacement

0.6

0.4

0.2

-0.2
0 0.2 0.4 0.6 0.8 1
Radial station r/R

Figure 4.34: Measured rotating mode shapes of the first lag mode at nominal
rotational speed (900 RPM) compared with numerical predictions

the first lag, and the second flap bending mode. This was the motivation of
performing modal parameter identification based on another data set, which
was the blade response to the step function excitation.

143
 1

Modal displacement 0.5

-0.5

-1
0 0.2 0.4 0.6 0.8 1
Radial station r/R

Figure 4.35: Measured rotating mode shapes of the second flap mode at nom-
inal rotational speed (900 RPM) compared with numerical predictions

First, frequency spectrum analysis was conducted to examine the fre-

quency components of the blade structural responses to the 2◦ collective step
pitch input. As a set of data samples for this analysis, the time history of
the out-of-plane deformation measured at the trailing-edge point of the rotor
blade tip was extracted and is shown in Fig. 4.36. The data set was then
divided into three periods: (i) before the step pitch input was applied, (ii)
during the transient response, and (iii) after the response settled down to the
steady-state condition.

Figure 4.37 shows the frequency spectrum of the three different time
periods up to 120 Hz (8/rev). The overall trend commonly observed in the
spectra was the dominant periodic peaks corresponding to the integer multi-
ples of rotational frequency. There were some smaller peaks between those
multiples of rotational frequency, however, it was challenging to distinguish

144
 130

Out-of-plane deformation, [mm] ( ii ) ( iii )

120

110

(i)
100

90

80
20 25 30 35 40 45 50 55 60 65 70
N [rev]
Figure 4.36: Time history of the out-of-plane deformation measured at the
trailing-edge point of the rotor blade tip

which peaks were associated with fictitious noise or the actual rotating frame
natural frequencies. Since the time period (iii) had the largest number of sam-
ples, the frequency spectrum (iii) in Fig. 4.37 has a higher resolution in the
frequency domain than in the spectra in cases (i) and (ii).

Noise level of each time period seems to be different in Fig. 4.37, how-
ever, that is due to the number of sample sizes used for the Fourier Transform.
Since the time period (iii) had a largest amount of samples, the frequency
spectrum (iii) appears to contain more noise than the other two sets of data.

To address this challenge, another analysis approach in frequency do-

145
 102
( i ) Before step pitch input
0
10

10-2

10-4
0 10 20 30 40 50 60 70 80 90 100 110 120

102
z-displacement [mm]

( ii ) During the transient response

100

10-2

-4
10
0 10 20 30 40 50 60 70 80 90 100 110 120

102
( iii ) After the response settled down
to the steady-state condition
100

10-2

-4
10
0 10 20 30 40 50 60 70 80 90 100 110 120
Frequency [Hz]

Figure 4.37: Frequency spectra of the three different time periods

main, called spectrogram, was used and applied to the same data set. A
spectrogram is defined as a visual representation of frequency spectrum of a
time-dependent signal and can estimate the time-localized frequency contents

146
using the short-time Fourier transform. The signal is typically divided into
several segments for windowing, so that one can see how frequency compo-
nents in the signal change in time. Figure 4.38 shows the spectrogram of
the same signal analyzed in Fig. 4.37. The parameters used for the spec-
trum are summarized in Table 4.5. Note that the sampling rate of the DIC
measurement was 15 Hz (rotational speed) × 32 azimuthal resolution = 480
Hz. At regions where the step change occurred (around 2.5 s) in the spectro-
gram, wide-bandwidth vibration was propagated in the power contour and it
immediately died out after the step change. This frequency content can be
interpreted as a proof of the successful excitation to the blade structure. On
the other hand, the periodic harmonic components attributed to the integer
multiples of the rotational frequency (15 Hz) remained constant over the entire
measurement period.

Table 4.5: Parameters used for the spectrogram analysis

sample size 8000

sample rate 480 Hz
length of each segment 500
segment overlap 490
windowing function Hamming
data points for FFT 500

With these spectrum analyses completed, the complexity pursuit al-

gorithm was applied to the blade response. Table 4.6 summarizes the modal
parameters identified based on the responses to two different excitations, along
with the numerical model. The OMA algorithm only extracted the first two

147
 120
20
105

Power/frequency (dB/Hz)
90 0
Frequency [Hz]

75
-20
60
45 -40

30
-60
15
0 -80
0.5 1.5 2.5 3.5 4.5 5.5 6.5 7.5
Time [s]
Figure 4.38: Spectrogram of the flap response measured at the trailing edge
point of the blade tip

flap modes, which was less than the number of modes identified during the
Test campaign 2. Additionally, the MAC for the second flap mode identified
during Test campaign 3 was low, indicating that the second mode shape was
not well correlated with either the numerical prediction or the mode shape ex-
tracted during Test campaign 2. The original intention of this test condition
was to excite the rotor blade structure by the rapid collective pitch change
(step function as an excitation source), so that the resultant response might
contain a large number of structural modes at higher amplitudes as compared
to the response to the random, turbulence-driven, ambient aerodynamic load-

148
ing achieved during Test campaign 2.
Table 4.6: Summary of the experimentally and numerically obtained modal
frequencies and MAC values calculated from the measured and predicted mode
shapes for Test campaign 2 and 3

1st Flap 1st Lag 2nd Flap

Freq., Hz MAC Freq., Hz MAC Freq., Hz MAC
Test campaign 2 17.9 0.999 39.0 0.998 66.0 0.992
Test campaign 3 17.8 0.992 - - 66.3 0.641
Prediction 19.6 N/A 39.8 N/A 66.2 N/A

There could be several reasons why the OMA processing of the data
measured in this test condition failed to extract the higher structural modes:
First, the sampling rate was not large enough to capture the step-induced os-
cillation whose decay rates were so large as shown in the spectrogram (see in
Fig. 4.38). In other words, the decay of the transient response to the collective
step change was to rapid to capture with the current sampling rate (480 Hz).
Second, the magnitude of the step change (corresponding to 2◦ increase) was
not strong enough to excite the higher modes of the blade structure during
rotation. These results of the modal parameter identification must be investi-
gated further in future work.

4.3 Rotor loads measurement and estimation

Results of rotor loads measurements and estimation are presented in

this section. The current study focused on the large-scale rotor loads measure-
ment and estimation based on the blade deformations and modal parameters
discussed in § 4.1 and § 4.2. This section is divided into two subsections:

149
In § 4.3.1, measured thrust and pitch link force of the CCR lower rotor are
compared with the numerical model provided from Ref. [57], and measured
rotor loads are compared with results of loads estimation process for the three
different operating conditions (a constant collective input, a periodic 1/rev,
and a collective step input) in § 4.3.2.

4.3.1 CCR rotor loads measurement

The blade deformation and rotor loads were simultaneously measured

during the Test campaign 2 on the single-bladed CCR rotor system at a high
blade loading coefficient that was still below stall limits (CT /σ = 0.09). The
frequency spectrum of the vibratory thrust on the lower rotor of the CCR
rotor system is shown in Fig. 4.39a up to 9/rev. It can be seen that the 1/rev
component of ±20 N was the largest contributor to the coaxial lower rotor
thrust. Although 2/rev harmonics are the dominant characteristic of vibratory
loads associated with blade passage as shown in Fig. 4.10, a contribution from
odd harmonics can also be seen.

Figure 4.39b shows the measured lower rotor thrust as a function of

rotor azimuthal angle compared to numerical predictions. The shaded area
corresponds to measurement uncertainty. Overall, it can be seen that the
thrust increased as the upper and lower rotor blades approached each other,
and it decreased after the blade passage. This can be explained by the mecha-
nism of bound circulation interaction that is shown schematically in Fig. 4.10.
That is, upwash induced by the upper rotor blade on the lower rotor blade

150
 25
Measurement
CAMRAD II
20

Thrust [N] 15

10

0
1 2 3 4 5 6 7 8 9
n/rev
(a) Frequency domain

140
Experiment
Prediction

120
Thrust [N]

100

80

60
0 90 180 270 360
[deg]
(b) Azimuthal domain

Figure 4.39: Reconstructed (filtered) thrust of the CCR lower rotor, compared
to predictions at CT /σ = 0.09

increased the effective angle of attack of the lower rotor blade, resulting in the
increase in rotor thrust, and vice versa. While the magnitude of the predic-

151
tions correlated well with the measurements, there was some discrepancy in
the phase. For the first blade passage, peaks occurred at about ψ = 15◦ and
ψ = 30◦ for prediction and measurement, respectively, whereas for the second
blade passage, peaks occurred at about ψ = 195◦ and ψ = 210◦ for prediction
and measurement, respectively. Recall that the blades crossed each other at
ψ = 0◦ and ψ = 180◦ .

Figure 4.40a shows the amplitudes of each harmonic component of mea-

sured pitch link loads, revealing a large 1/rev component similar to the thrust
variation. Figure 4.40b compares the measured and predicted pitch link loads
as a function of blade azimuthal angle. Good correlation of magnitude between
measurement and prediction was observed, whereas a phase shift of ψ = 60◦
between the measured and predicted trends was also observed.

Overall, the vibratory thrust was approximately 10% of the mean thrust,
which is consistent with previous studies by Cameron et al. [35, 82] where the
vibratory thrust was 11% of the mean thrust in hover. More significantly, the
vibratory pitch link load was found to be nearly 30 % of the mean pitch link
load, which can have a profound impact on the design of the rotor control
system. Thus, for the large-scale rotor hover test rig, it is concluded that the
three major components of measured quantities for the rotor loads estimation
methodology proposed in the dissertation, i.e., blade deformation, blade modal
properties, and rotor loads, were captured well by the numerical prediction.

152
 20
Measurement
Prediction

Pitch link loads [N]

15

10

1 2 3 4 5 6 7 8 9
n/rev
(a) Frequency domain

10
Experiment
Prediction
0
Pitch link load [N]

-10

-20

-30

-40

-50
0 90 180 270 360
[deg]
(b) Time domain

Figure 4.40: Reconstructed (filtered) pitch link load of the CCR lower rotor,
compared to predictions at CT /σ = 0.09

153
4.3.2 Two-bladed single rotor loads measurement and estimation

With the success of validation process between the experimental results

and numerical simulations during Test campaign 2, the final step of the current
study is to estimate rotor loads based on the blade deformation measured
by the DIC technique and the blade modal parameters identified by OMA
processing. Rotor loads include the spanwise lift distribution along the blade
and the integrated hub loads. To verify if the proposed approach for rotor loads
estimation holds in a variety of loading scenarios, three different operating
conditions were tested during Test campaign 3 and summarized in Table 4.7.
Note that among the three modes identified through the OMA processing,
only the first and second flap modes were used to reconstruct the rotor loads
because the lag modes does not contribute to the out-of-plane force.

Table 4.7: Recap: Summary of operating conditions for Test campaign 3

Condition CT /σ Trim θ0 θ1c

◦
1 0.125 Constant pitch 9 -
2 0.125 1/rev pitch 9◦ 2◦
3 - Step pitch 9◦ + 2◦ step -

4.3.2.1 Test condition 1

The first operating condition tested on the two-bladed single rotor was
a constant collective pitch input at a blade loading CT /σ of 0.125. Measured
quantities include: hub loads, pitch link loads, and blade root pitch angles.

The rotor thrust variation, measured by the six-component load cell

154
and phase-averaged over 250 revolutions, is shown in time (azimuthal) and
frequency (non-dimensionalized by the rotational frequency, 15 Hz) domain
in Fig. 4.41a and 4.41b, respectively. Note that all the loads variations in
this section are measured in the rotating frame. The frequency spectrum
clearly represents the general relationship between the mean and vibratory
components of hub loads, that is, vibratory loads of a single rotor in hover
are much smaller than the steady mean component of rotor loads. In this test
case, there was little vibration in hub loads (less than approximately 1% of
the mean thrust), since a constant collective pitch input was applied to the
rotor.

A similar trend can be seen in the rotor torque; the variation in time
and frequency domain are shown in Fig. 4.42a and 4.42b, respectively, with
an oscillatory behavior at the 3/rev harmonic visible in both domains. The
magnitude ratio of the 3/rev component to the mean torque is approximately
8.5%, which is relatively large compared to a typical vibration level in hover.
The 3/rev variation can also be seen in more profound manner in the rotating
frame rolling moment, as shown in Fig. 4.43. One possible reason for this large
3/rev vibration is the proximity of the 1st lag mode frequency to the 3/rev
blade crossing (see the fan plot shown in Fig. 4.32).

Pitch link loads were measured using the axial tension-compression

load cell for both blades. Pitch link force measurements enable investigation
of rotor blade pitch moment dynamics, as well as determination of maximum
allowable blade loading for safe operation during hover testing. Figure 4.44

155
 260
250
240

Thrust, [N] 230

220
210
200
190
180
170
0 60 120 180 240 300 360
Azimuth, [deg]
(a) Time domain

240
210
180
Thrust [N]

150
120
90
60
30
0
Mean 1 2 3 4 5 6 7 8 9
N/rev
(b) Frequency domain

Figure 4.41: Phase-averaged thrust variation as a function of blade azimuthal

location for the steady collective pitch input

156
 25
Torque
22.5

Torque, [Nm] 20

17.5

15

12.5

10
0 60 120 180 240 300 360
Azimuth, [deg]
(a) Time domain

20
17.5
15
Torque [Nm]

12.5
10
7.5
5
2.5
0
Mean 1 2 3 4 5 6 7 8 9
N/rev
(b) Frequency domain

Figure 4.42: Phase-averaged torque variation as a function of blade azimuthal

location for the steady collective pitch input

157
 7
6 My

5
4
My, [Nm] 3
2
1
0
-1
-2
-3
-4
0 60 120 180 240 300 360
Azimuth, [deg]
(a) Time domain

2.5

2
My [Nm]

1.5

0.5

0
Mean 1 2 3 4 5 6 7 8 9
N/rev
(b) Frequency domain

Figure 4.43: Phase-averaged pitching moment variation as a function of blade

azimuthal location for the steady collective pitch input

158
shows the azimuthal variation and its frequency spectrum of pitch link loads
for both blades. In Fig. 4.44a, there was a clear phase difference of nearly
180◦ between the pitch link (blade) 1 and 2, implying that there might exist a
slight inclination of the swashplate that caused the out-of-phase pitch moment
variation for the two blades. Most of the vibratory components were contained
in the first three harmonics for both blades, with a 1/rev component as the
largest contributor to the pitch link force variation as shown in Fig. 4.44b.
A possible source of the slight magnitude difference of the 1/rev harmonic
component was the presence of viscous damping in the bearing mechanism
(grease lubrication) for rotor feathering motion.

Blade root pitch angles for both rotor blades were measured using the
Anisotropic Magneto-Resistive (AMR) sensors installed on the blade grips
(see § 3.1.2 for more details). The real-time measurement of root pitch an-
gles enables trimming the rotor at a target operating condition during hover
testing.

Figure 4.45 shows the phase-averaged root pitch angle of both blades as
a function of azimuthal position over one revolution. The overall trend of the
blade pitch 1 corresponds to the trim target of this test condition 1 (a steady,
mean collective pitch θ0 = 9◦ ). However, there was a large fluctuation in the
blade pitch 2 measurement over the range of 60 to 180 deg, possibly due to
the free play of the corresponding pitch link or needle bearings that hold the
blade grip. It is interesting to note that the mean component of root blade
pitch 1 (about 9.27◦ from Fig. 4.45a) was nearly the same as the torsional

159
 25
Pitch link 1
20 Pitch link 2

Pitch link force, [N]

15
10
5
0
-5
-10
-15
0 60 120 180 240 300 360
Azimuth, [deg]
(a) Time domain

9
Pitch link 1
8 Pitch link 2
7
Pitch link force [N]

6
5
4
3
2
1
0
Mean 1 2 3 4 5 6 7 8 9
N/rev
(b) Frequency domain

Figure 4.44: Phase-averaged pitch link force variation as a function of blade

azimuthal location for the steady collective pitch input

160
displacement over the blade span calculated using the DIC results shown in
Fig. 4.15c. This result implies that the rotor blade used for this hover testing
experienced little elastic twist at this blade loading condition, and the DIC
measurement agrees well with the results measured by the AMR pitch angle
sensor.

The frequency spectrum of the pitch angle variations is shown in Fig 4.45b,
revealing that the observed fluctuation on the pitch angle measurement 2 was
associated with the 1/rev frequency component. The normalized amplitude of
the vibratory component was less than 2% of the mean collective pitch, thus
it is fair to say that the intended rotor trim control was achieved for the test
case A.

Now the results of rotor loads estimation for the test condition 1 are
discussed. Rotor loads include the spanwise lift distribution along the quarter-
chord blade axis, and the hub loads (thrust) obtained by numerical integration
of the lift distribution in spanwise direction. For rotor load estimation, the
first and second flap modes identified from Test campaign 2 (see § 4.2.2) were
used.

Figure 4.46a shows the estimated sectional lift distribution as a function

of blade radial and azimuthal location. This plot was obtained by applying
the rotor load estimation methodology to the phase-averaged blade flap de-
formation over one revolution, and only illustrates the variation of the airload
of one rotor blade, whose deformation was measured using the DIC technique
(Recall that one of two blades on the two-bladed rotor was used as a test spec-

161
 10
Pitch angle 1
Pitch angle 2

9.5
Pitch angle, [deg]
9

8.5

8
0 60 120 180 240 300 360
Azimuth, [deg]
(a) Time domain

11
10 Pitch angle 1
Pitch angle 2
9
Pitch angle [degree]

8
7
6
5
4
3
2
1
0
Mean 1 2 3 4 5 6 7 8 9
N/rev
(b) Frequency domain

Figure 4.45: Phase-averaged pitch angle variation as a function of blade az-

imuthal location for the steady collective pitch input

162
imen for the DIC deformation measurement). As expected from the numerical
experiment, the sectional load monotonically increased up to the blade tip
without capturing the decrease (tip loss) due to the influence of tip vortices
(see Fig. 2.12). The lack of participating modes (only two modes) was the
primary cause for the poor estimation at the blade tip. On the other hand,
the integrated hub load, shown in Fig. 4.46b appeared to be well correlated
to the directly-measured thrust, especially for the steady, mean component of
the thrust variation. Although the vibratory loads seemed to have a slight
discrepancy, the mean thrust difference was observed within 5%.

4.3.2.2 Test condition 2

For the test condition 2, the rotor was trimmed at a constant collective
plus a lateral cyclic pitch angle. The collective pitch θ0 was set to be 9◦ so
that the mean blade loading remained the same as for the test condition 1.
The periodic 1/rev lateral pitch θ1C was set to be ±2◦ . Measured quantities
are the same as the previous test condition, including hub loads, pitch link
loads, and root pitch angle variation.

The phase-averaged thrust variation in the rotating frame as a function

of blade azimuthal location and its frequency spectrum are shown in Fig. 4.47a
and 4.47b, respectively. The 2/rev vibratory load can be clearly seen in az-
imuthal domain as compared to the steady thrust for the test condition 1
(steady collective pitch input), and the relative magnitude of the component
normalized by the mean thrust was calculated to be nearly 3.2%. Figure 4.48a

163
 (a) Estimated lift distribution over one revolution

260
Measured
250 Estimated
240
230
Thrust, [N]

220
210
200
190
180
170
0 60 120 180 240 300 360
Azimuth, [deg]
(b) Comparison between the estimated and measured
rotor hub loads

Figure 4.46: Estimated rotor loads for the steady collective pitch input

164
and 4.48b show the torque variation, revealing the large 3/rev vibratory com-
ponent (relative amplitude of 11%) that was also observed for the test condi-
tion 1. The presence of the 3/rev harmonics is consistent with both the test
condition 1 and 2, indicating that the source of this vibration might depend on
the rotor blade modal parameters, as discussed in the previous test condition.

The primary difference between the test condition 1 and 2 was found
in the rotating frame rolling moment My as shown in Fig. 4.49. Due to the
asymmetric lift distribution over the rotor disk, the rotor experienced the large
variation in the rolling moment over one revolution. The clean sinusoidal trend
at the 1/rev frequency corresponds to the lateral cyclic pitch input (θ1C = 2◦ )
to the rotor control. As can be seen in Fig 4.16, the tilting motion of the
rotor disk was consistently associated with the trend of the rolling moment,
meaning that the tilting direction of the rotor disk appropriately reflected the
sign change of the rolling moment.

The frequency spectrum shown in Fig. 4.49b helped to identify that

the 1/rev vibratory component was the largest contributor to the moment
variation, which was consistent with the trend in azimuthal domain shown in
Fig. 4.49a. Note that comparing the vibratory loads for this condition 2 to
that for the condition 1, the magnitudes of the 3/rev harmonic component
were on the similar order of 2 Nm, implying that this 3/rev vibratory load
consistently observed in most of the loads data was possibly attributed to a
common source in the experimental setup.

The pitch link force variation for both rotor blades is shown in Fig 4.50.

165
 270
260
250
Thrust, [N] 240
230
220
210
200
190
0 60 120 180 240 300 360
Azimuth, [deg]
(a) Time domain

250
225
200
175
Thrust [N]

150
125
100
75
50
25
0
Mean 1 2 3 4 5 6 7 8 9
N/rev
(b) Frequency domain

Figure 4.47: Phase-averaged thrust variation as a function of blade azimuthal

location for the lateral cyclic pitch input

166
 25
Torque
22.5

Torque, [Nm] 20

17.5

15

12.5

10
0 60 120 180 240 300 360
Azimuth, [deg]
(a) Time domain

20
17.5
15
Torque [Nm]

12.5
10
7.5
5
2.5
0
Mean 1 2 3 4 5 6 7 8 9
N/rev
(b) Frequency domain

Figure 4.48: Phase-averaged torque variation as a function of blade azimuthal

location for the lateral cyclic pitch input

167
 30
25 My
20
15
My, [Nm] 10
5
0
-5
-10
-15
-20
-25
-30
0 60 120 180 240 300 360
Azimuth, [deg]
(a) Time domain

25
22.5
20
17.5
My [Nm]

15
12.5
10
7.5
5
2.5
0
Mean 1 2 3 4 5 6 7 8 9
N/rev
(b) Frequency domain

Figure 4.49: Phase-averaged pitching moment variation as a function of blade

azimuthal location for the lateral cyclic pitch input

168
Figure 4.50a presents a significant 1/rev sinusoidal trend for both pitch links
and the clear azimuthal phase offset of approximately 180◦ between the two
rotor blades. The 1/rev vibratory component directly corresponds to the vari-
ation of the blade pitching moment due to the swashplate inclination corre-
sponding to the lateral cyclic pitch. On the other hand, Fig. 4.50b compares
the vibratory components of the two pitch link loads in frequency domain. The
magnitudes of the 1/rev component for the pitch link 1 and 2 were respectively
19.4 N and 15.7 N, which were the dominant component consistently for both
linkages; however, the values were shown to have a discrepancy of approxi-
mately 5 N. The mean, steady component seems to have a difference on the
same order of magnitude as well. The constant offset between two linkages
can also be seen in Fig. 4.44 for the test case condition 1, indicating that
there must be some common source of this difference, regardless of rotor trim
condition, such as free-play in rod ends or grease lubrication in the feathering
bearings.

The phase-averaged blade root pitch angle variations are shown in

Fig. 4.51a and 4.51b in blade azimuthal and frequency domain, respectively.
This demonstrates successful pitch control using the swashplate and servo-
actuators; the collective (mean) pitch θ0 was set to be equal to 9◦ , and the
cyclic pitch θ1C was set to be ±2◦ .

The spanwise distribution and integrated hub load for the test condition
2 were estimated and compared to the measured data. For the test condition
2, the rotor was trimmed at the steady collective pitch θ0 = 9◦ + the lateral

169
 50
Pitch link 1
40 Pitch link 2

Pitch link force, [N]

30
20
10
0
-10
-20
-30
0 60 120 180 240 300 360
Azimuth, [deg]
(a) Time domain

20
Pitch link 1
17.5 Pitch link 2
Pitch link force [N]

15
12.5
10
7.5
5
2.5
0
Mean 1 2 3 4 5 6 7 8 9
N/rev
(b) Frequency domain

Figure 4.50: Phase-averaged pitch link force variation as a function of blade

azimuthal location for the lateral cyclic pitch input

170
 12
Pitch angle 1
Pitch angle 2
11

Pitch angle, [deg]

10

6
0 60 120 180 240 300 360
Azimuth, [deg]
(a) Time domain

11
10 Pitch angle 1
Pitch angle 2
9
Pitch angle [degree]

8
7
6
5
4
3
2
1
0
Mean 1 2 3 4 5 6 7 8 9
N/rev
(b) Frequency domain

Figure 4.51: Phase-averaged pitch angle variation as a function of blade az-

imuthal location for the lateral cyclic pitch input

171
cosine cyclic input θ1C = 2◦ . Figure 4.52a shows the sectional lift distribution
based on the phase-averaged blade flap bending variation over the entire rotor
disk. The contour clearly represents the aerodynamic force variation at once-
per-revolution frequency with the consistent phase, however, the lift tip loss
was not captured for this case, either. Figure 4.52b compares the integrated
hub loads for the lateral cyclic pitch input, revealing the pronounced 2/rev
vibratory content in the estimated thrust variation. The 2/rev component of
the estimated thrust agreed well with that of the measured thrust in terms of
both magnitude and phase. There was a mean thrust discrepancy of nearly
5%, which was also observed in Fig. 4.46 for the test condition 1.

4.3.2.3 Test condition 3

The test condition 3 was a rapid increase of collective pitch angle by 2◦ ,

so that the rotor would experience dynamic, transient responses to the step-
function excitation. It should be noted that most of the measured data shown
in this section were processed with a low-pass filter at a cutoff frequency that
was specifically selected for each measured quantity, in order to clearly observe
trends of the response to the collective step. These cutoff frequencies were
mainly selected by trial-and-error. Table 4.8 summarizes the cutoff frequencies
selected for this low-pass filtering process.

First, the measured blade root pitch angle is plotted as a function of

rotor revolution in Fig. 4.53. Note that the horizontal axis of plots shown in the
following subsections was converted from time to the number of revolutions.

172
 (a) Estimated lift distribution over one revolution

260
Measured
250 Estimated

240
Thrust, [N]

230
220
210
200
190
180
0 60 120 180 240 300 360
Azimuth, [deg]
(b) Comparison between the estimated and measured
rotor hub loads

Figure 4.52: Estimated rotor loads for the lateral cyclic pitch input (Test
condition 2)

173
Table 4.8: Summary of cutoff frequencies for low-pass filtering process for Test
campaign 3

Measured quantity Cutoff frequency [Hz]

Deformations 13
Blade root pitch 13
Rotor rotational speed 42
Thrust 25
Torque 25
Pitch link load 8

To clearly show the pitch variation, the low-pass filter was applied with a
3 dB corner frequency of 13 Hz. In Fig. 4.53, the blade root pitch angle
started increasing from approximately 9.5 degree during the 28th revolution,
and reached the steady state value of about 11.5 degree after completing the
35th revolution.

There was an additional measured quantity specifically available for

this test condition, which is the rotational speed of the rotor resulting from the
rapid increase of rotor torque. Figure 4.54 shows the rotational speed variation
of the rotor shaft as a function of time, low-pass filtered at a cutoff frequency
of 42 Hz. The maximum speed drop was nearly 25 RPM, and the ringing
behavior lasted for approximately 40 revolutions. This low-frequency oscilla-
tion possibly contaminates other measured quantities, that is, data might no
longer represent the rotor physics operated at constant rotational speed of 900
RPM. However, the magnitude of the oscillation might be small (nearly 2.5%)
and it has little influence of the analysis of the current study, because the pri-
mary purpose of this test condition was to excite the modes of the rotor blade

174
 13

12
Pitch angle, [deg]

11

10

Pitch angle 1
9 Pitch angle 2

8
20 25 30 35 40 45 50 55 60 65 70
N [revs]
Figure 4.53: Blade root pitch angle as a function of rotor revolutions during
the step pitch input operation

structure as much as possible by applying the rapid collective pitch change.

Now the rotor thrust and torque responses to the step pitch increase
are examined. The time history of the low-pass filtered thrust and torque
are shown in Fig. 4.55 and 4.56, respectively. The increase in thrust started
during the 28th revolution and maintained the rate of change up until the
30th revolution, then the slope changed to a lower value. This slope transition
was consistent to the variation of other quantities including the pitch angle
variation and blade tip deflection. The rotor mean thrust was first at about
230 N at θ0 = 9◦ , then increased by approximately 40 N due to the 2◦ collective

175
 920

Rotational speed, [RPM]

910

900

890

880

870
20 25 30 35 40 45 50 55 60 65 70
N [revs]
Figure 4.54: Rotational speed of the rotor as a function of rotor revolutions

step input and eventually stabilized at about 265 N.

On the other hand, the trend of the rotor torque was slightly different
from that of the rotor thrust. Figure 4.56 shows the torque variation as a
function of time, low-pass filtered at the 3dB cutoff frequency of 25 Hz. The
rate of change of the rotor torque variation appeared to be monotonic unlike
the trend of the thrust variation, and the torque reached its peak immediately
after passing the 35th revolution and died out at around the 45th revolution.
A possible reason for this trend difference is that there was a close correlation
between the torque and the rotational speed variation and these two quantities
had a pronounced influence on each other. The low-frequency rotor speed

176
 300

280
Thrust, [N]

260

240

220

200
20 25 30 35 40 45 50 55 60 65 70
N [revs]
Figure 4.55: Rotor thrust response to the step pitch change as a function of
rotor revolutions

ringing hence appeared more in the torque response, than in the responses of
other measured quantities.

The pitch link loads variations, low-pass filtered at the 3 dB cutoff

frequency of 8 Hz, for the two blade linkages are shown in Fig. 4.57. Recall
that the negative slope (compression) and positive slope (tension) of the pitch
link load variation correspond to the nose-up and -down motion of the rotor
blade, according to the linkage mechanism (see § 3.1.2). In Fig. 4.57, the pitch
link force first decreased at around the 28th revolution, associated with the
beginning of the swashplate motion pushing the rotor blade (compression) to

177
 30

27.5

25
Torque, [Nm]

22.5

20

17.5

15
20 25 30 35 40 45 50 55 60 65 70
N [revs]
Figure 4.56: Rotor torque response to the step pitch change as a function of
rotor revolutions

increase the blade pitch angle. Then, the pitch link force started increasing
after passing the 30th revolution and overshot at about the 35th revolution,
because of the increase in blade pitch moment in nose-down direction (tension).
Hence, the trends observed in Fig. 4.57 adequately represent the blade pitch
moment (aerodynamic and inertial) attributed to the dynamic collective step
change.

The loads estimation methodology first computed the integrated iner-

tial load at the rotor hub as a function of the number of revolutions, as shown
in Fig. 4.58. The inertial load was obtained by the numerical differentiation

178
 15
Pitch link 1
Pitch link force, [N] 12.5 Pitch link 2

10
7.5
5
2.5
0
-2.5
-5
20 25 30 35 40 45 50 55 60 65 70
N [revs]
Figure 4.57: Pitch link force response to the step pitch change as a function
of rotor revolutions

of the out-of-plane deformation time history in time, multiplied by the mass

property of the blade structure. Note that the time history of the flap bending
deformation was smoothed by 3rd-order polynomial curve fit prior to the nu-
merical differentiation. In Fig. 4.58, there was a small fluctuation before the
step pitch input was applied, and the maximum load of 7.5 N was observed
immediately after the step input applied during the 28th revolution. The os-
cillation induced by the step lasted for approximately 300 milliseconds, which
was equivalent to 4.5 revolutions.

The time history of the integrated hub load for the test condition 3

179
was estimated and compared to the measured thrust variation, as shown in
Fig. 4.59. The resultant increase in rotor thrust due to the collective step
change was found to be consistently 40 N in both the measured and estimated
profile; however, there was a constant magnitude offset of approximately 15
N and a slight time lag of step response between the measured and estimated
thrust. The former concern was consistently observed for all three operating
conditions (the constant collective pitch input, the 1/rev periodic pitch input,
and this collective step pitch input) on the same order of magnitude (10-15
N), which could possibly be explained by the area difference under the original

10
7.5
5
Inertial load [N]

2.5
0
-2.5
-5
-7.5
-10
20 25 30 35 40 45 50 55 60 65 70
N [revs]
Figure 4.58: The estimated inertial load integrated at the hub as a function of
the number of revolutions for the collective step pitch change (Test condition
3)

180
and estimated spanwise lift curve; (see Fig. 2.11). The latter concern could be
because of the nature of this load estimation methodology, i.e., the estimated
loads were purely based on rotor blade structural deformation measured by
the non-contact optical technique (DIC), whereas the hub loads measured by
the load cell contained a large number of experiment-specific factors, such as
linkage mechanism, free-play and lubrication in bearings, non-ideal aerody-
namic interaction with the experimental setup, or the recirculation effect in
the hover test chamber. These factors might cause the constant magnitude off-
set and time lag between the measured and estimated thrust response; further
investigation must be performed as future work.

In addition to the constant shift in mean thrust, the estimated thrust

exhibits a steeper jump than the measured trend. A possible cause for the
trend difference is the fact that the measured quantities in this test case were
contaminated by the variation of rotational speed as mentioned before (see
Fig. 4.54). Although the blade deformation measured during the test was also
affected by the rotor speed change, the influence was less significant on the
blade structural response measured by the non-contact optical DIC technique
than that on the rotor loads measured by the load cell directly mounted on
the rotor stand.

In conclusion, the integrated hub load estimation showed satisfactory

agreement with the measured thrust variation for all the operating (loading)
conditions, although the lack of the higher modes participation caused poor
estimation of the spanwise lift distribution, particularly at regions of the blade

181
tip. Nevertheless, it must be emphasized that this rotor load estimation was
not made by complicated, troublesome on-blade sensors but achieved by pro-
cessing the sequence of digital images taken during the hover tests, and the
only preparation on the blade structure was the orange fluorescent paint with
random speckle patterns using a spray lacquer and a black-ink marker.

300

280

260
Thrust, [N]

240

220
Measured thrust
Estimated thrust
200

180
20 25 30 35 40 45 50 55 60 65 70
N [revs]
Figure 4.59: Comparison between the estimated and measured rotor hub loads
for the collective step pitch change (Test condition 3)

182
 Chapter 5

Summary, Conclusions, and Future Work

5.1 Summary and Conclusions

A helicopter rotor loads estimation methodology based on blade struc-

tural response was formulated and validated. The primary objective of this
study was to develop an experimental and theoretical methodology to obtain
helicopter rotor loads, without using conventional on-blade sensors. The the-
oretical framework of the methodology consists of three key components: (1)
blade deformation time history measurement using a non-contact optical tech-
nique called Digital Image Correlation (DIC), (2) rotor blade modal properties
identification using Operational Modal Analysis (OMA), and (3) rotor loads
estimation methodology based on blade structural deformation and modal pa-
rameters. To verify the applicability and robustness of the proposed method-
ology, a series of measurements were performed in hover on a 0.46 m-diameter,
small-scale rotor hover test rig and a 2 m-diameter, large-scale rotor hover test
rig at several rotor configurations and operating conditions.

The 0.46 m-diameter, small-scale rotor consisted of two extremely flex-

ible blades and a fixed-pitch hub. The three-dimensional (3D) deformation
time history of the small-scale rotor blade was measured by the time-resolved

183
DIC technique at four different root pitch angles (0◦ -30◦ ) and rotational speeds
(600-1500 RPM). The time-resolved DIC was achieved by (i) taking a sequence
of digital images over the whole rotor disk by stereoscopic high-speed cameras,
(ii) illuminating the surface of the rotor blade painted with fluorescent orange
and stochastic speckle pattern by laser light, and (iii) triggering image ac-
quisition and laser strobe timing with a downsampled (16/rev) digital pulse
train generated from an optical encoder. Blade deformation measurements
consisted of the flap, lead-lag, and torsional deformation over 400 revolutions
at approximately 900 measurement channels over the entire rotor span at 16
evenly-spaced azimuthal locations. The measurements revealed that the time-
resolved DIC technique developed in this dissertation can be used to measure
the rotating blade deformation on the order of a few millimeters.

To expand the applicability of the time-resolved DIC, several measure-

ments were performed on the 2 m-diameter, large-scale rotor hover test rig
spun at 900 RPM at different rotor configurations and operating conditions.
The same methodology as the small-scale testing, i.e., the time-resolved DIC
with laser strobing and fluorescence paint, was applied to the large scale rotor
blade deformation measurement. First, the DIC technique was applied to the
single-bladed, coaxial counter-rotating (CCR) rotor blade deformation mea-
surement and these experimental results were compared with a well-validated
numerical model. Overall, there was excellent agreement between the mea-
sured and predicted flap bending deformations. The maximum tip deflection
of the CCR lower rotor blade reached 74.6 mm, which is approximately 10

184
times larger than the tip deflection of the small-scale rotor. The time-resolved
DIC successfully demonstrated its capability of measuring the deformation
time history of a rotating blade at different measurement scales.

Additional blade deformation measurements were performed on the

large-scale, two-bladed, isolated single rotor with the hydraulic actuation sys-
tem. The goal was to verify if the time-resolved DIC can measure rotor blade
responses to a variety of loading conditions, including (i) a steady collective
pitch input at a blade loading of 0.125, (ii) a 1/rev periodic pitch input, and
(iii) a collective step pitch input. For the first test condition, deformations
were measured and examined from various perspectives: out-of-plane blade
deformation over the entire rotor disk, 3D static deformation distributed over
the blade span, static deformations of each degree-of-freedom (flap, lead-lag,
and torsion) along the blade quarter-chord axis, and blade tip displacement as
function of azimuthal angle. For the second test condition, the tip deflection
difference between the maximum and minimum reached approximately 75 mm,
thus it was demonstrated that this large sinusoidal oscillatory motion of the 2
m-diameter rotor at 1/rev frequency (15 Hz) was successfully captured by the
DIC technique. For the third test condition, out-of-plane, in-plane, and local
blade pitch responses to the collective step input were successfully captured.

The measured deformations were input to two OMA algorithms (NExT-

ERA or CP algorithm) to identify the modal frequencies, mode shapes, and
damping ratios in the rotating frame. The modal parameter identification was
first performed on the small-scale rotor blade. The first two modes were identi-

185
fied using these two OMA algorithms. To quantify the differences between the
measured and predicted mode shapes, Modal Assurance Criterion (MAC) was
used as a measure of agreement between two vectors. The modal frequencies
and mode shapes identified by the CP algorithm agreed well with numerical
predictions, however, there was some discrepancy between mode shapes iden-
tified by NExT-ERA and numerical prediction. All the modal damping values
extracted by the NExT-ERA analysis were less than 3% as was the case of the
CP analysis. While the NExT-ERA required a number of steps and choices of
parameters based on experience or trial-and-error, the CP algorithm was quite
simple to implement and did not require any tuning of parameters. Thus, the
CP algorithm was selected as a primary OMA algorithm for modal identifica-
tion of the large-scale rotor blade.

For the large-scale rotor blade modal identification, two different types
of excitation to the rotor blade were tested: (i) random excitation at zero mean
thrust and (ii) collective step input excitation. Overall, the modal extraction
algorithm based on the random response successfully identified the first flap,
first lag, and second flap frequencies and mode shapes. Using the MAC, it
was demonstrated that there was excellent agreement between mode shapes
extracted by the CP algorithm and the numerical model. On the other hand,
the CP analysis extracted only the first two flap modes from the response to
the collective step change. The spectrogram of the same data set showed that
wide-bandwidth vibration was propagated when the step change occurred,
however, this response immediately died out after the step. Hence, the first

186
three modes identified in the test case (i) (random excitation) were used for
rotor loads estimation.

Rotor loads measurement was first performed on the large-scale, single-

bladed CCR lower rotor at a blade loading of 0.09, simultaneously with the
blade deformation measurement. A significant 2/rev component was observed
in the lower rotor thrust and pitch link load with satisfactory agreement of
the magnitude to numerically predicted results. The vibratory thrust was
approximately 10% of the mean thrust, whereas the vibratory pitch link load
was found to be nearly 30% of the mean pitch link force. With these results
from measurements on the large-scale rotor test rig, it is found that the three
essential measured quantities (blade deformation, blade modal properties, and
rotor loads) were well correlated with the numerical prediction.

Finally, rotor loads were estimated based on the measured blade de-
formation and modal parameters identified by the CP algorithm. Rotor loads
estimation includes the spanwise lift distribution along the blade quarter-chord
axis and hub loads (thrust) obtained by numerical integration of the lift distri-
bution in the spanwise direction. Note that the modal parameters for the first
and second flap modes were used for rotor loads estimation because the lag
mode does not contribute to the out-of-plane lift force reconstruction. Data
sets taken at (1) a constant pitch, (2) a 1/rev periodic pitch, and (3) a col-
lective step pitch on the large-scale, two-bladed rotor were used. For all the
test conditions, the estimated sectional lift distribution failed to capture the
typical trend of lift loss in regions of the blade tip induced by the trailed tip

187
vortices, due to the limited number of participating modes for load reconstruc-
tion. However, the integrated thrust agrees well with the directly-measured
value for the three test cases, possibly due to the fact that the areas under
the estimated and original lift curve are close to each other even if only a
few modes are used for the estimation. In conclusion, the rotor loads estima-
tion methodology is a powerful tool to obtain a rough estimate of the rotor
hub loads without using on-blade instrumentation. To accurately estimate the
spanwise lift distribution, a larger number of modes must be included.

5.2 Future work

Overall, the time-resolved DIC technique sufficiently shows its applica-

bility to blade deformation measurement at a variety of different rotor config-
urations and operating conditions in this study. For further validation of the
DIC measurement technique, simultaneous deformation measurements using
different experimental approaches should be performed; for example, the pitch
angle at the blade tip can be relatively easily measured using in-plane pho-
togrammetry which locates high-speed cameras in the rotor plane and captures
images of the rotor blade cross-section at the tip. Comparing results with the
DIC-obtained pitch angle, one can validate the correctness of the DIC blade
deformation measurement to some extent.

Additionally, regarding the DIC deformation measurement, the current

study mainly focused on the blade deformation along the blade quarter-chord
axis or the particular measurement point at the blade tip. The 3D DIC de-

188
formation measurement sometimes requires an analyst to deal with an unnec-
essarily large number of data points in space over the entire blade span. To
effectively utilize the significant amount of measured information, spatial av-
eraging over the neighborhood is suggested; instead of tracking one particular
measurement point that corresponds to a single interrogation window, aver-
aging 3D deformations over the neighborhood points in both spanwise and
chordwise directions might effectively eliminate noise and outliers in raw data.

The DIC-OMA combined approach failed to identify structural modes

higher than the third mode for both the small- and large-scale rotor blade.
Since higher modes oscillate with small amplitudes, deformations might ap-
proach the noise floor of the DIC technique. Increasing the signal-to-noise
ratio for improvement in resolution is essential to enable the identification of
higher modes. One simple suggestion for measuring vibration for the higher-
order modes could be to increase input excitation forcing to the rotor blade
structure so that large elastic deformations can be obtained. Another ap-
proach could be to implement post-processing algorithms, such as the motion
magnification technique, to magnify small deformation so that vibration over
the specified bandwidth can be emphasized and easily picked.

The cause of low values of modal damping extraction is unclear; there is

room for improvement in modal damping identification. However, specifically
for the case of helicopter blade motion, the flap damping term comes from
aerodynamic damping and is very high; for example, a typical articulated
rotor (Lock number = 8) has a 50% flap damping ratio in hover. Flapping

189
motion tends to die out rapidly due to the high damping. Thus, the estimation
of damping remains the most difficult part of modal analysis in helicopter rotor
blade dynamics.

Investigating the reason for the discrepancy between results of NExT-

ERA analysis and numerical prediction remains an important area of future
work, and underlines the sensitivity of this technique to the selection of pro-
cessing parameters and post-processing steps.

The measured deformation and loads of the single-bladed CCR rotor

during the Test campaign 2 contained a large 1/rev component along with
significant odd harmonics, although only even harmonics are expected for a
single-bladed CCR rotor system. The 1/rev and odd harmonics could be
attributed to aerodynamic interference with the rotor stand support structure
as well as the effect of having dissimilar rotor blades, on each single-bladed
rotor with a counter-weight. Additionally, the measured quantities obtained
during the Test campaign 3 (the two-bladed, isolated single rotor) contained
some non-negligible fluctuations. Mean flow recirculation in the closed hover
test chamber might also be one of the sources for these non-ideal fluctuations
observed in data. Another possible source is the presence of viscous damping
in the bearing mechanism and its lubrication for rotor feathering motion, as
well as some free-play in blade pitch linkage mechanism. Future work includes
improvements in the experimental setup to minimize the unnecessary harmonic
contents and fluctuations in the measured blade deformations and rotor loads.

Regarding the rotor loads estimation, the tip lift loss induced by the

190
trailed tip vortices was not quite captured due to the lack of participating
modes in the estimation process. However, the tip loss might be relatively
easily corrected by introducing an additional well-known shape function, such
as the Prandtl tip loss function, to the estimation process, instead of struggling
to identify the higher-order structural modes. The tip loss correction could
not only help one to accurately reconstruct the spanwise lift distribution, but
also address the consistently-observed 5% mean value offset in the estimated
hub loads estimation.

One possible application of the current methodology is to derive and

estimate the blade structural properties based on the estimated modes. This
study used uniform, untwisted, simple blade structure for proof of concept.
However, actual helicopter blades generally have non-uniform cross-section,
complicated composite blade structure whose properties cannot be readily ob-
tained. If the DIC-OMA approach can be scaled up to a full-scale, actual
helicopter blade, its structural properties are obtained by only processing a
sequence of digital images of the blade structure. This process is a typical
usage of the operational modal analysis, that is, the primary goal of modal
parameter identification (or modal testing) is to fine-tune a finite element
model of a structure using experimentally-obtained modal properties. Thus,
the DIC-OMA combined approach has the potential for significantly reducing
the time and cost of modal testing on full-scale helicopter rotor blades.

Improvement of rotor loads estimation will eventually lead to further

validation of a computational model. Accurate prediction of transient heli-

191
copter rotor loads will provide a foundation to develop methods to mitigate
vibration in future helicopter design.

192
 Bibliography

[1] An image of a static display of the CH-53 Sea Stallion. Downloaded from
https://www.military.com/equipment/ch-53d-sea-stallion.

[2] An image of the CH-53 Sea Stallion taken under operation. Downloaded
from https://commons.wikimedia.org/wiki/File:CH-53 Sea Stallion -
Side View (Balikatan 2016).JPG.

[3] J. P. Jr. Rabbott, A. A. Lizak, and V. M. Paglino. A Presentation

of Measured and Calculated Full-scale Rotor Blade Aerodynamic and
Structural Loads. USAAVLABS Technical Report 66-31, 1966.

[4] R. M. Heffernan and M. Gaubert. Structural and Aerodynamic Loads

and Performance Measurements of an SA349/2 Helicopter with an Ad-
vanced Geometry Rotor. NASA Technical Memorandum 88370, 1986.

[5] R. M. Kufeld, D. L. Balough, J. L. Cross, K. F. Studebaker, C. D. Jen-

nison, and W. G. Bousman. Flight Testing of the UH-60A Airloads Air-
craft. Proceedings of the 50th Annual Forum of the American Helicopter
Society, Washington, D. C., May 11-13, 1994.

[6] W. G. Bousman. Estimation of Blade Airloads from Rotor Blade Bending

Moments. NASA Technical Memorandum 100020, 1987.

193
 [7] G. Wang, A. Abhishek, I. Chopra, N. Phan, and D. Liebschutz. Ro-
tor Loads Prediction Using Estimated Modal Participation from Sensors.
39th European Rotorcraft Forum Proceedings, Paris, France, Sept. 9-11,
2010.

[8] J. Sirohi and M. S. Lawson. Measurement of helicopter rotor blade

deformation using digital image correlation. Optical Engineering, 51(4),
pp. 043603, 2012.

[9] J. Sicard and J. Sirohi. Measurement of the deformation of an extremely

flexible rotor blade using digital image correlation. Measurement Science
and Technology, 24(6), pp. 065203, 2013.

[10] G. C. Cameron, D. Uehara, and J. Sirohi. Transient Hub Loads and Blade
Deformation of a Mach-Scale Coaxial Rotor in Hover. 56th AIAA/ASCE/
AHS/ASC Structures, Structural Dynamics, and Materials Conference,
Kissimmee, Florida, Jan. 5-9, 2015.

[11] S. S. Rizo-Patron and J. Sirohi. Operational Modal Analysis of a Heli-

copter Rotor Blade Using Digital Image Correlation. Experimental Me-
chanics, 57(3), pp. 367–375, 2017.

[12] J. Baqersad, P. Poozesh, C. Niezrecki, and P. Avitabile. Photogrammetry

and optical methods in structural dynamics – A review. Mechanical
Systems and Signal Processing, 86(B), pp. 17–34, 2017.

194
[13] J. P. Sevenhuijsen. The photonical, pure grid method. Optics and Lasers
in Engineering, 18(3), pp. 173–194, 1993.

[14] R. Kube, W. R. Splettstoesser, W. Wagner, U. Seelhorst, Yu. Y. H.,

C. Tung, P. Beaumier, J. Prieur, G. Rahier, P. Spiegel, A. Boutier, T. F.
Brooks, C. L. Burley, D. D. Boyd Jr., E. Mercker, and K. Pengel. HHC
aeroacoustic rotor tests in the German-Dutch wind tunnel: Improving
physical understanding and prediction codes. Aerospace Science and
Technology, 2(3), pp. 177–190, 1998.

[15] A. Macovski, S. D. Ramsey, and L. F. Schaefer. Time-lapse interferome-

try and contouring using television systems. Applied Optics, 10(12), pp.
2722–2727, 1971.

[16] O Lekberg. Electronic Speckle Pattern Interferometry. Physics in Tech-

nology, 11(1), pp. 16–22, 1980.

[17] G. A. Fleming and S. A. Gorton. Measurement of Rotorcraft Blade De-

formation Using Projection Moiré Interferometry. Shock and Vibration,
7(3), pp. 149–165, 2000.

[18] G. A. Fleming and S. Gorton. Projection Moiré Interferometry for rotor-

craft applications: deformation measurement of active twist rotor blades.
Proceedings of the 58th Annual Forum of the American Helicopter Society,
Montreal, Quebec, June 11-13, 2002.

195
[19] M. K. Sekula. The Development and Hover Test Application of a Pro-
jection Moiré Interferometry Blade Displacement Measurement System.
Proceedings of the 68th Annual Forum of the American Helicopter Society,
Fort Worth, TX, May 1-3, 2012.

[20] M. Ozbek, D. J. Rixen, O. Erne, and G. Sanow. Feasibility of monitoring

large wind turbines using photogrammetry. Energy, 35(12), pp. 4802–
4811, 2010.

[21] M. Ozbek and D. J. Rixen. Optical Measurements and Operational

Modal Analysis on a Large Wind Turbine: Lessons Learned. 29th IMAC,
a Conference on Structural Dynamics, Jacksonville, Florida, Jan. 31 -
Feb. 3, 2011.

[22] M. Ozbek and D. J. Rixen. Operational modal analysis of a 2.5 MW wind

turbine using optical measurement techniques and strain gauges. Wind
Energy, 16(3), pp. 367–381, 2013.

[23] O. Schneider. Analysis of SPR Measurements from HART II. Aerospace

Science and Technology, 9(5), pp. 409–420, 2005.

[24] A. L. Abrego, L. E. Olson, E. A. Romander, D. A. Barrows, and A. W.

Burner. Blade Displacement Measurement Technique Applied to a Full-
Scale Rotor Test. Proceedings of the 68th Annual Forum of the American
Helicopter Society, Fort Worth, TX, May 1-3 2012.

196
[25] T. Lundstrom, J. Baqersad, and C. Niezrecki. Monitoring the Dynam-
ics of a Helicopter Main Rotor with High-Speed Stereophotogrammetry.
Experimental Techniques, 40(3), pp. 907–919, 2016.

[26] J. Javh, J. Slavic, and M. Boltezar. High frequency modal identifica-

tion on noisy high-speed camera data. Mechanical Systems and Signal
Processing, 98, pp. 344–351, 2018.

[27] W. H. Peters and W. F. Ranson. Digital Imaging Techniques in Ex-

perimental Stress Analysis. Optical Engineering, 21(3), pp. 427–431,
1982.

[28] M. Sutton, W. J. Wolters, W. H. Peters, W. F. Ranson, and S. R. Mc-

Neil. Determination of Displacements Using an Improved Digital Image
Correlation Method. Image and Vision Computing, 1(3), pp. 133–139,
1983.

[29] Z. L. Kahn-Jetter and T. C. Chu. Three-Dimensional Displacement Mea-

surements Using Digital Image Correlation and Photogrammic Analysis.
Experimental Mechanics, 30(1), pp. 10–16, 1990.

[30] T. C. Chu, W. F. Ranson, and M. A. Sutton. Applications of Digital

Image Correlation Techniques to Experimental Mechanics. Experimental
Mechanics, 25(3), pp. 232–244, 1985.

[31] S. R. McNeill, M. A. Sutton, Z. Miao, and J. Ma. Measurement of Surface

Profile Using Digital Image Correlation. Experimental Mechanics, 37(1),

197
 pp. 13–20, 1997.

[32] J. Tyson, T. Schmidt, and K. Galanulis. Advanced Photogrammetry

for Robust Deformation and Strain Measurement. SEM 2002 Annual
Conference Proceedings, Milwaukee, WI, June 10-12, 2002.

[33] T. Schmidt, J. Tyson, and K. Galanulis. Full-Field Dynamic Displace-

ment and Strain Measurement Using Advanced 3-D Image Correlation
Photogrammetry: Part 1. Experimental Techniques, 27(3), pp. 47–50,
2003.

[34] J. Tyson, T. Schmidt, and K. Galanulis. Biomechanics Deformation

and Strain Measurements with 3D Image Correlation Photogrammetry.
Experimental Techniques, 26(5), pp. 39–42, 2002.

[35] C. Cameron, J. Sirohi, R. Feil, and J. Rauleder. Measurement of transient

loads and blade deformation in a coaxial counter-rotating rotor. Proceed-
ings of the 73rd Annual Forum of the American Helicopter Society, Fort
Worth, Texas, May 9-11, 2017.

[36] S. R. Ibrahim and E. C. Milkulcik. The Experimental Determination

of Vibration Test Parameters from Time Responses. The Shock and
Vibration Bulletin, 46(5), pp. 187–196, 1976.

[37] S. R. Ibrahim and R. S. Pappa. Large Modal Survey Testing Using the
Ibrahim Time Domain Identification Technique. Journal of Spacecraft
and Rocket, 19(5), pp. 459–465, 1982.

198
[38] J. N. Juang and R. S. Pappa. An Eigensystem Realization Algorithm
for Modal Parameter Identification and Model Reduction. Journal of
Guidance, Control, and Dynamics, 8(5), pp. 620–627, 1985.

[39] Z. Ma, S. Ahuja, and C. W. Rowley. Reduced-order models for control

of fluids using the Eigensystem Realization Algorithm. Theoretical and
Computational Fluid Dynamics, 25(1-4), pp. 233–247, 2011.

[40] Q. Qin, H. B. Li, L. Z. Qian, and C.-K. Lau. Modal identification of tsing
ma bridge by using improved eigensystem realization algorithm. Journal
of Sound and Vibration, 247(2), pp. 325–341, 2001.

[41] G. H. III James, T. G. Carne, and J. P. Lauffer. The Natural Excita-

tion Technique (NExT) for Modal Parameter Extraction from Operating
Wind Turbines. Technical Report SAND 92-1666, Sandia National Labs,
Albuquerque, NM, 1993.

[42] M. H. Hansen, K. Thomsen, P. Fuglsang, and T. Knudsen. Two methods

for estimating aeroelastic damping of operational wind turbine modes
from experiments. Wind Energy, 9(1-2), pp. 179–191, 2006.

[43] T. G. Carne and G. H. James III. The inception of OMA in the develop-
ment of modal testing technology for wind turbines. Mechanical Systems
and Signal Processing, 24(5), pp. 1213–1226, 2010.

[44] E. D. Lorenzo, G. Petrone, S. Manzato, B. Peeters, W. Desmet, and

F. Marulo. Damage detection in wind turbine blades by using operational

199
 modal analysis. Structural Health Monitoring, 15(3), pp. 289–301, 2016.

[45] P. Comon and C. Jutten. Handbook of Blind Source Separation: Inde-

pendent Component Analysis and Applications. Academic Press, Oxford,
UK, 2010.

[46] A. D. Back and A. S. Weigend. A First Application of Independent

Component Analysis to Extracting Structure from Stock Returns. Inter-
national Journal of Neural Systems, 8, pp. 473–484, 1997.

[47] T.-P. Jung, S. Makeig, M. J. McKeown, A. J. Bell, T.-W. Lee, and T. J.

Sejnowski. Imaging Brain Dynamics Using Independent Component
Analysis. Proceedings of the IEEE, 89(7), pp. 1107–1122, 2001.

[48] M. J. Roan, L. H. Erling, and L. H. Sibul. A New Non-linear, Adaptive,

Blind Source Separation Approach to Gear Tooth Failure Detection and
Analysis. Mechanical Systems and Signal Processing, 16(5), pp. 719–740,
2002.

[49] G. Kerschen, F. Poncelet, and J. C. Golinval. Physical Interpretation of

Independent Component Analysis in Structural Dynamics. Mechanical
Systems and Signal Processing, 21(4), pp. 1561–1575, 2007.

[50] J. Stone. Blind Source Separation Using Temporal Predictability. Neural

Computation, 13, pp. 1559–1574, 2001.

200
[51] Y. Yang and S. Nagarajaiah. Blind Modal Identification of Output-only
Structures in Time-domain Based on Complexity Pursuit. Earthquake
Engineering Structural Dynamics, 42, pp. 1885–1905, 2013.

[52] M. Chierichetti, C. McColl, D. Palmer, M. Ruzzene, and O. Bauchau.

Combined Analytical and Experimental Approached to Rotor Compo-
nents Stress Predictions. The 1st Joint International Conference on
Multibody System Dynamics, Lappeenranta, Finland, May 25-27, 2010.

[53] H. G. Somerson. Approximation of Rotor Inflow and Flapwise Airload

Distributions on Hinged Blades from Measured Strain and Motion Data.
Proceedings of the 13th Annual Forum of the American Helicopter Society,
Washington, D. C., May 8-11, May, 1957.

[54] H. Öry and H. W. Lindert. Reconstruction of Spanwise Air Load Distri-

bution on Rotorblades from Structural Flight Test Data. 18th European
Rotorcraft Forum, Avignon, France, Sept. 15-18, 1992.

[55] L. ShouShen, J. P. Higman, and D. P. Schrage. Coupled Flap-Lag Rotor

Blade Load Identification. Proceedings of the 51st Annual Forum of the
American Helicopter Society, Fort Worth, TX, May 9-11, 1995.

[56] D. P. Schrage, Y. K. Yillikci, S. Liu, J. V. R. Prasad, and S. V. Hanagud.

Instrumentation of the Yamaha R-50/RMAX Helicopter Testbeds for Air-
loads Identification and Follow-On Research. 25th European Rotorcraft
Forum, Rome, Italy, Sept. 14-16, 1999.

201
[57] D. Uehara, J. Sirohi, R. Feil, and J. Rauleder. Blade Passage Loads and
Deformation of a Coaxial Rotor System in Hover. Journal of Aircraft,
2019.

[58] S. Yoneyama and G. Murasawa. Digital Image Correlation. Experimen-

tal Mechanics, 207, 2009.

[59] LaVision 2017 Strain Master 3D, Software Package Version 8.4. LaVi-
sion GmbH, Göttingen, Germany.

[60] J. M. Caicedo, S. J. Dyke, and E. A. Johnson. Natural Excitation Tech-

nique and Eigensystem Realization Algorithm for Phase I of the IASC-
ASCE Benchmark Problem: Simulated Data. Journal of Engineering
Mechanics, 130, pp. 49–60, 2004.

[61] P. D. Welch. The Use of Fast Fourier Transform for the Estimation of
Power Spectra: A Method Based on Time Averaging Over Short, Modified
Periodograms. IEEE Transactions on Audio and Electroacoustics, 15, pp.
70–73, 1967.

[62] C. R. Farrar and G. H. III. James. System Identification from Ambient

Vibration Measurements on a Bridge. Journal of Sound and Vibration,
205(1), pp. 1–18, 1997.

[63] C. R. Farrar and D. A. Jauregui. Comparative Study of Damage Iden-

tification Algorithms Applied to a Bridge II: Numerical Study. Smart
Materials and Structures, 7(5), pp. 721–731, 1998.

202
[64] L. Meirovitch. Fundamentals of Vibrations. McGraw-Hill Book, New
York, NY, 2000.

[65] P. H. C. Eilers. A Perfect Smoother. Analytical Chemistry, 75(14), pp.

3631–3636, 2003.

[66] J. J. Stickel. Data Smoothing and Numerical Differentiation by a Regu-

larization Method. Computers and Chemical Engineering, 34, pp. 467–
475, 2010.

[67] W. Johnson. Rotorcraft Aeromechanics. Princeton University Press,

1980.

[68] J. G. Leishman. Principles of Helicopter Aerodynamics. Cambridge

University Press, Cambridge, UK, 2006.

[69] A. Bagai. Aerodynamic Design of the X2 Technology DemonstratorTM

Main Rotor Blade. Proceedings of the 64th Annual Forum of the Ameri-
can Helicopter Society, Montreal, Quebec, April 29 - May 1, 2008.

[70] C. G. Cameron. Comprehensive Aeromechanical Measurements of a

Model-Scale, Coaxial, Counter-Rotating Rotor System. Doctoral dis-
sertation, 2016.

[71] R. S. Pappa, J. T. Black, J. R. Blandino, T. W. Jones, P. M. Danehy, and

A. A. Dorrington. Dot-Projection Photogrammetry and Videogrammetry
of Gossamer Space Structures. Journal of Spacecraft and Rockets, 40(6),
pp. 858–867, 2003.

203
[72] T. W. Jones, A. A. Dorrington, P. L. Brittman, and P. M. Danehy. Laser
Induced Fluorescence for Photogrammetric Measurement of Transparent
or Reflective Aerospace Structures. 49th Annual International Instru-
mentation Symposium, Orlando, FL, May, 2003.

[73] A. A. Dorrington, T. W. Jones, P. M. Danehy, and R. S. Pappa. Laser-

Induced-Fluorescence Photogrammetry for Dynamic Characterization of
Membrane Structures. AIAA Journal, 42(10), pp. 2124–2129, 2004.

[74] A. A. Dorrington, T. W. Jones, and P. M. Danehy. Photophysics of

Laser Dye-Doped Polymer Membranes for Laser-Induced Fluorescence
Photogrammetry. Applied Optics, 43(36), pp. 6629–6638, 2004.

[75] D. Uehara and J. Sirohi. Full-field optical deformation measurement and

operational modal analysis of a flexible rotor blade. Mechanical Systems
and Signal Processing, 133, pp. 106265, November, 2019.

[76] J. Sicard and J. Sirohi. Modeling of the Large Torsional Deformation

of an Extremely Flexible Rotor in Hover. AIAA Journal, 52(8), pp.
1604–1615, 2014.

[77] P. Mohanty and D. J. Rixen. Operational Modal Analysis in the Presence

of Harmonic Excitation. Journal of Sound and Vibration, 270, pp. 93–
109, 2004.

[78] A. Randall, J. The Modal Assurance Criterion – Twenty Years of Use

and Abuse. Sound and Vibration, 37(8), pp. 14–23, 2003.

204
[79] D. Tcherniak, S. Chauhan, M. Rossetti, I. Font, J. Basurko, and O. Sal-
gado. Output-only Modal Analysis on Operating Wind Turbines: Appli-
cation to Simulated Data. European Wind Energy Conference, Warsaw,
Poland, April 20-23, 2010.

[80] P. T. Brewick and A. W. Smyth. Exploration of the Impacts of Driving

Frequencies on Damping Estimates. Journal of Engineering Mechanics,
141(3), pp. 30–48, 2015.

[81] D. Uehara, C. Cameron, , and J. Sirohi. Deformation Measurement and

Modal Identification of an Extremely Flexible Rotor Blade. 6th Asian-
Australian Rotorcraft Forum/Heli Japan, Kanazawa, Japan, Nov. 7-9,
2017.

[82] C. Cameron, A. Karpatne, and J. Sirohi. Performance of a Mach-Scale

Coaxial Counter-Rotating Rotor in Hover. Journal of Aircraft, 53(3), pp.
746–755, 2016.

205
 Vita

Daiju Uehara was born in Okinawa, Japan. He received the Bachelor of

Eng. in Aerospace Engineering at the University of Tokyo in 2014. Same year,
he joined the UT Aeroelasticity Lab under Prof. Jayant Sirohi with a focus
on an experimental study of a coaxial, counter-rotating rotor helicopter. He is
a two-time winner of the Vertical Flight Foundation (VFF) scholarship from
the American Helicopter Society/Vertical Flight Society, and also a recipient
of the University of Texas Graduate Continuing Fellowship for 2018-2019. His
recent work presented at AHS Forum 74 was recognized as the best paper for
the Dynamics Sessions. He was a student intern at the NASA/Jet Propulsion
Laboratory during summer 2018, and worked on the flight power modeling of
small VTOL aircraft for exploration on Titan. Besides, he loves soccer.

Contact: udaiju9@gmail.com

This dissertation was typeset with LATEX† by the author.

† A
LT EX is a document preparation system developed by Leslie Lamport as a special
version of Donald Knuth’s TEX Program.

206