Vibration Reduction in Rotorcraft Using Active Flow

Control

by

Ryan P. Patterson

A dissertation submitted in partial fulfillment

of the requirements for the degree of
Doctor of Philosophy
(Aerospace Engineering)
in the University of Michigan
2021

Doctoral Committee:

Professor Peretz P. Friedmann, Chair

Professor Carlos E. S. Cesnik
Professor Bogdan I. Epureanu
Professor Krzysztof J. Fidkowski
 Ryan P. Patterson

[email protected]

ORCID iD: 0000-0002-8301-2001

© Ryan P. Patterson 2021

 Acknowledgments

There are several people I must acknowledge, for their instrumental roles in shaping my
path towards this milestone.
The support of my advisor, Prof. Peretz Friedmann, has been a constant driving force
during my time as a graduate student. I am grateful for his commitment, not only to my
success, but to the success of all his students. I would like to recognize the most recent
cohort from Prof. Friedmann’s research group, with whom I had pleasant, insightful, and
often downright silly interactions: Michael Chia, Daning Huang, Elliot Kimmel, Ashwani
Padthe, Abhinav Sharma, and Puneet Singh. My sincerest thanks go to Puneet Singh, for
his insight and friendship, especially during a global pandemic.
I attribute my academic success largely to the caliber and enthusiasm of my teachers.
Professors Carlos Cesnik, Karthik Duraisamy, Chris Fidkowski, Anouck Girard, and Ken
Powell are all outstanding in this regard. I cannot overstate the significance of my interac-
tion with Prof. James Forbes, who supported my undergraduate research, showed genuine
excitement for my work, and pushed me to consider graduate school. I am also grateful for
the support and encouragement from Prof. Todd Griffith, who was my mentor during an
internship at Sandia National Labs.
My Ph.D. research has been continuously funded through the Vertical Lift Research
Center of Excellence (VLRCOE) at Georgia Tech, which Michigan is a partner in. Fund-
ing was provided by the U.S. Government under Agreement No. W911W6-17-2-0002. I
am thankful to have had the opportunity to collaborate with and learn from Yuehan Tan,
Prof. Ari Glezer, and Prof. Marilyn Smith throughout the project.
My friends have been responsible for keeping my head on straight and my spirits high.
Honorable mentions include Kelly Carew, John Jasa, Prince Kuevor, Zachary Lahay, Rehan
Nawaz, Conor VanderHoff, and Chris Wentland. I would also like to acknowledge the
students I became acquainted with through the Graduate Student Advisory Committee.
I admire the work they do to make the Aerospace Department a more welcoming and
inclusive place.
Lastly, I am truly grateful for the love and support of my family. My grandmother, Dot-
tie Patterson, is always eager to hear about how I spend my time in Ann Arbor and to share
her own fond memories of the University of Michigan. Her final semester as an Aerospace

ii
Engineering student coincided with a defining moment in aerospace history: the first suc-
cessful launch of an Earth-orbiting satellite (Sputnik 1, October 1957). During the last few
months of my own studies, we have witnessed the first powered and controlled flight of an
aircraft on another planet (Ingenuity, April 2021). Imagine what can be accomplished in
the next 64 years! I have shared many laughs with my sisters, Karen and Rachel. I cherish
the shared experiences that we derive our convoluted sense of humor from. Finally, I can-
not fully express how grateful I am for my parents, Mike and Diana Patterson. They ignited
my passion for learning at an early age and have supported me through every step I have
taken to get here. I am in awe of the time, energy, money, and numerous other resources
they have sacrificed for me, so that I might find my own happiness and success.

iii
 TABLE OF CONTENTS

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

List of Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii

List of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii

List of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xx

Chapter

1 Introduction, Background, and Objectives . . . . . . . . . . . . . . . . . . . 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Literature Review on Helicopter Vibration Reduction Approaches . . . . 4
1.3 Literature Review on Active Flow Control and Its Applications . . . . . . 7
1.3.1 Overview of Engineering Goals . . . . . . . . . . . . . . . . . . 8
1.3.2 Overview of Actuator Classifications . . . . . . . . . . . . . . . 10
1.3.3 Experimental Studies . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.4 Flight Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.5 Numerical Simulations . . . . . . . . . . . . . . . . . . . . . . . 22
1.4 Objectives of the Dissertation . . . . . . . . . . . . . . . . . . . . . . . . 25
1.5 Key Contributions of the Dissertation . . . . . . . . . . . . . . . . . . . . 26
1.6 Dissertation Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2 CFD Simulations of Fluidic Actuation and Their Experimental Validation . . 28
2.1 Bi-Directional Pulsed Fluidic Actuation . . . . . . . . . . . . . . . . . . 28
2.2 CFD Model of the Flow Control Device . . . . . . . . . . . . . . . . . . 31
2.2.1 Description of the Flow Solver . . . . . . . . . . . . . . . . . . . 31
2.2.2 Geometry and Boundary Conditions . . . . . . . . . . . . . . . . 32
2.2.3 Grids Employed for Modeling the Airfoil/Actuator Combination . 33
2.3 Characterization of Flow Control Performance on a Static Airfoil . . . . . 37
2.3.1 Low Mach Simulations and Comparison with Experiments . . . . 38
2.3.2 Simulations at Higher Mach Numbers . . . . . . . . . . . . . . . 54

iv
 2.4 Flow Control Performance on a Moving Airfoil . . . . . . . . . . . . . . 57
2.4.1 Sinusoidal Pitching Motion . . . . . . . . . . . . . . . . . . . . 57
2.4.2 Rotor Blade Motion in an Unsteady Freestream . . . . . . . . . . 59
3 Reduced Order Modeling of Unsteady Aerodynamic Loads Induced by Flow
Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.1 Training Data Generation . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.1.1 Freestream Conditions . . . . . . . . . . . . . . . . . . . . . . . 63
3.1.2 Unsteady Aerodynamic Training Data . . . . . . . . . . . . . . . 66
3.2 ROM Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.3 Model Scheduling and Interpolation . . . . . . . . . . . . . . . . . . . . 72
3.4 ROM Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4 Aeroelastic Rotor Simulations Using the AVINOR Code . . . . . . . . . . . . 80
4.1 Structural Dynamics Model . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.2 Aerodynamic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.2.1 CFD-Based RFA Model for Attached Flow . . . . . . . . . . . . 81
4.2.2 ONERA Dynamic Stall Model for Separated Flow . . . . . . . . 85
4.2.3 Free-Wake Model for 3D Wake Calculation . . . . . . . . . . . . 87
4.2.4 Reverse Flow Model . . . . . . . . . . . . . . . . . . . . . . . . 91
4.2.5 Sectional Aerodynamic Loads . . . . . . . . . . . . . . . . . . . 92
4.3 Coupled Aeroelastic Response and Rotor Trim Solution . . . . . . . . . . 96
4.4 Previous Validation Studies . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.5 Flow Control ROM Implementation . . . . . . . . . . . . . . . . . . . . 97
4.5.1 Comment on Code Timing . . . . . . . . . . . . . . . . . . . . . 98
5 Control Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.1 The Higher Harmonic Control Algorithm . . . . . . . . . . . . . . . . . 100
5.1.1 Classical HHC . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.1.2 Adaptive HHC . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.2 Actuator Saturation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.3 Implementation for a Microflap . . . . . . . . . . . . . . . . . . . . . . . 105
5.4 Implementation for Discrete AFC Operation . . . . . . . . . . . . . . . . 107
6 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.1 Helicopter Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.2 Baseline Results Without Control . . . . . . . . . . . . . . . . . . . . . . 115
6.3 Open-Loop Control Studies . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.3.1 Vibratory Loads . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.3.2 Angle of Attack . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.3.3 Normal Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.3.4 Rotor Torque . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.3.5 Sensitivity to Actuator Strength . . . . . . . . . . . . . . . . . . 125
6.4 Examination of the Control Sensitivity Matrix for HHC . . . . . . . . . . 127
6.5 Closed-Loop Control of Vibratory Loads . . . . . . . . . . . . . . . . . . 130
6.5.1 Effect of the Actuator Saturation Limit . . . . . . . . . . . . . . 131

v
 6.5.2 Comparison with the Microflap . . . . . . . . . . . . . . . . . . 132
6.5.3 Rotor Performance Penalty . . . . . . . . . . . . . . . . . . . . . 135
6.5.4 Effect of the Rotor Advance Ratio . . . . . . . . . . . . . . . . . 139
7 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
7.1 Conclusions and New Contributions . . . . . . . . . . . . . . . . . . . . 143
7.2 Recommendations for Future Research . . . . . . . . . . . . . . . . . . . 147
Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

vi
 LIST OF FIGURES

FIGURE

1.1 Illustration of the complex aerodynamic environment of a helicopter . . . . . . 2

1.2 Conceptual schematics of various fluidic actuators for active flow control . . . 12
1.3 B-18 airplane with test section for laminar flow control [1] . . . . . . . . . . . 17
1.4 Northrop X-21 active laminar flow control airplane [1] . . . . . . . . . . . . . 17
1.5 Boeing 707 wing modifications for boundary layer control [2] . . . . . . . . . 18
1.6 Helicopter rotor blade boundary layer control research by Cessna Aircraft [3] . 19
1.7 Circulation control flight research for STOL applications [4] . . . . . . . . . . 20
1.8 Bell XV-15 during AFC test flights and the arrangement of actuators [5] . . . . 21

2.1 Trailing edge pulsed fluidic actuators for SS and PS blowing on a tabbed VR-
12 airfoil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2 Array of trailing edge actuators installed on VR-12 airfoil model . . . . . . . . 30
2.3 Simplification of internal actuator geometry for CFD simulations . . . . . . . 31
2.4 VR-12 airfoil and details of simplified trailing edge fluidic actuator geometry . 32
2.5 Structured computational grid for 2D CFD simulations of VR-12 airfoil with
pulsed fluidic actuators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.6 Structured computational grid for 3D CFD simulations of VR-12 airfoil with
pulsed fluidic actuators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.7 Computational grid for 3D-wing geometry with fluidic actuators installed across
0.1𝑐 of the span . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.8 Unsteady aerodynamic response due to PS normal blowing (𝐶 𝜇 = 0.03);
changes in lift, moment, and drag coefficients are relative to jet off condition
at 𝛼 = 0◦ , time normalized as 𝑡¯ = 𝑡/𝑇conv . . . . . . . . . . . . . . . . . . . . . 39
2.9 Initial transient due to step input PS normal blowing (𝐶 𝜇 = 0.03), depicted by
contours of instantaneous normalized spanwise vorticity; time normalized as
𝑡¯ = 𝑡/𝑇conv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.10 Periodic shedding due to steady PS normal blowing (𝐶 𝜇 = 0.03), depicted by
contours of instantaneous normalized spanwise vorticity; time normalized as
𝑡¯ = 𝑡/𝑇conv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.11 Steady state aerodynamic response due to PS and SS normal blowing at vari-
ous jet strengths; changes in lift, moment, and drag coefficients are relative to
jet off condition at 𝛼 = 0◦ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

vii
2.12 RMS magnitude and frequency of lift oscillations due to PS (solid lines) and
SS (dashed lines) normal blowing at various jet strengths; frequency normal-
ized as 𝑓¯ = 𝑓 𝑐/𝑈∞ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.13 Time-averaged flow fields due to steady normal blowing at different jet strengths;
streamlines are plotted over contours of normalized spanwise vorticity . . . . . 45
2.14 Steady state lift and moment coefficients due to normal and tangential blowing
(𝐶 𝜇 = 0.03) applied on the pressure side and suction side . . . . . . . . . . . . 46
2.15 Steady state changes in drag coefficient, relative to jet off condition, due to
normal and tangential blowing (𝐶 𝜇 = 0.03) applied on the pressure side and
suction side . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.16 Unsteady aerodynamic response obtained from 2D and 3D CFD simulations
of PS normal blowing (𝛼 = 4◦ , 𝐶 𝜇 = 0.03); time normalized as 𝑡¯ = 𝑡/𝑇conv . . . 50
2.17 Spanwise distribution of sectional lift coefficient due to PS normal blowing
(𝛼 = 4◦ , 𝐶 𝜇 = 0.03); span normalized as 𝑧¯ = 𝑧/𝑐 . . . . . . . . . . . . . . . . 52
2.18 Time-averaged flow fields obtained from 2D and 3D simulations of steady PS
normal blowing (𝐶 𝜇 = 0.03); streamlines are plotted over contours of normal-
ized spanwise vorticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.19 Time-averaged 3D flow field due to steady PS normal blowing (𝐶 𝜇 = 0.03),
depicted by isosurfaces of Q-criterion colored by normalized spanwise vortic-
ity; freestream flow direction is along the positive 𝑥 axis . . . . . . . . . . . . 54
2.20 Steady state aerodynamic response due to PS and SS normal blowing (𝐶 𝜇 =
0.005) at various freestream Mach numbers . . . . . . . . . . . . . . . . . . . 55
2.21 Unsteady aerodynamic response due to sinusoidal pitching motion with 4/rev
alternating PS/SS normal blowing (𝐶 𝜇 = 0.005); time normalized as 𝑡¯ = 𝑡/𝑇conv 58
2.22 Unsteady aerodynamic response due to coupled pitching/plunging motion in
an unsteady freestream representing helicopter blade motion at 75% span with
4/rev alternating PS/SS normal blowing . . . . . . . . . . . . . . . . . . . . . 61

3.1 AFC-integrated rotor blade schematic for aeroelastic simulations . . . . . . . . 64

3.2 Local Mach number and angle of attack calculated at aerodynamic stations
along the rotor blades, obtained from forward flight simulations in AVINOR
at 𝜇 = 0.2, 0.3 for 𝐶𝑊 = 0.005 . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.3 Expanded training data range (AVINOR data ±2◦ ) filled with 34 evenly-spaced
ROM training points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.4 Time and frequency domain representations of a pseudorandom ternary sig-
nal; time normalized by clock period 𝑇𝑐 and frequency normalized by clock
frequency 𝑓𝑐 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.5 Process model estimation from CFD simulation of pseudorandom ternary sig-
nal, 𝑀 = 0.7, 𝛼 = 0.6◦ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.6 Kriging interpolation of ROM parameters for lift coefficient due to PS blow-
ing, based on Mach number and angle of attack . . . . . . . . . . . . . . . . . 77
3.7 Unsteady aerodynamic response to rotor blade motion with 4/rev alternating
PS/SS normal blowing, predicted using CFD and ROM . . . . . . . . . . . . . 79

4.1 Normal velocity distributions corresponding to generalized airfoil motions . . 82

viii
4.2 Airfoil undergoing pitching and plunging motion . . . . . . . . . . . . . . . . 82
4.3 Vortex-lattice approximation for rotor wake model . . . . . . . . . . . . . . . 89
4.4 Single peak circulation distribution model and the resulting far wake approxi-
mation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.5 CAMRAD/JA dual peak model and the resulting far wake approximation . . . 90
4.6 Position of the reverse flow region in forward flight . . . . . . . . . . . . . . . 92
4.7 Orientation of tangential and perpendicular air velocities and aerodynamic loads 93
4.8 Validation of unsteady aerodynamic loads obtained from AVINOR with HART
experimental data at 87% span [6] . . . . . . . . . . . . . . . . . . . . . . . . 97

5.1 The microflap [7] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

5.2 Schematic of closed-loop ROM simulation for converting a continuous signal
to a discrete-valued signal using delta modulation . . . . . . . . . . . . . . . . 111
5.3 Conversion of a continuous control input signal to a discrete-valued signal
using delta modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.4 Non-dimensional normal force (𝑀 2𝐶𝑛 ) evaluated at 73% blade span due to
continuous and discrete AFC operation; 𝜇 = 0.3, 𝐶𝑇 = 0.005 . . . . . . . . . . 113
5.5 Vibratory hub load reduction due to continuous and discrete AFC operation;
𝜇 = 0.3, 𝐶𝑇 = 0.005 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

6.1 Baseline uncontrolled 4/rev hub loads during trimmed level flight at various
advance ratios; results are non-dimensional . . . . . . . . . . . . . . . . . . . 116
6.2 Trim variable settings for level flight at various advance ratios . . . . . . . . . 117
6.3 Vibratory hub loads and rotor power for uncontrolled case (baseline) and open-
loop AFC (𝐶 𝜇,avg = 0.0013) applied at 2, 3, 4, and 5/rev harmonics with phase
offset swept from 0◦ to 360◦ in 45◦ increments (𝜇 = 0.30) . . . . . . . . . . . 119
6.4 Variation of thrust coefficient in forward flight (𝜇 = 0.30) due to alternating
PS/SS blowing, before retrimming the rotor . . . . . . . . . . . . . . . . . . . 120
6.5 Effect of re-trimming the rotor in forward flight (𝜇 = 0.30) with 4/rev alter-
nating PS/SS blowing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.6 Local angle of attack; oncoming flow direction is left to right (𝜇 = 0.30) and
blades rotate counter-clockwise . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.7 Local normal force; oncoming flow direction is left to right (𝜇 = 0.30) and
blades rotate counter-clockwise . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.8 Local torque; oncoming flow direction is left to right (𝜇 = 0.30) and blades
rotate counter-clockwise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
6.9 Sensitivity of vertical shear force vibrations and rotor power to actuation strength
(4/rev actuation) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
6.10 Sensitivity of 4/rev hub loads to actuation strength (4/rev actuation with 𝜙offset =
180◦ ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.11 Variation of 4/rev hub load sensitivities to AFC actuation applied at 2, 3, 4,
and 5/rev harmonics; results are non-dimensional . . . . . . . . . . . . . . . . 130
6.12 Sensitivity of closed-loop vibratory load reduction to actuator saturation limit
(𝐶 𝜇,avg ) enforced during HHC; variation of actuation power required is also
shown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

ix
6.13 Continuous and discrete AFC input signals computed by HHC with various
saturation limits on the actuation power 𝑃jet . . . . . . . . . . . . . . . . . . . 133
6.14 Comparison of 4/rev vibratory hub load reduction using the microflap and
AFC with a saturation limit of 𝑃jet = 1.2 kW/blade . . . . . . . . . . . . . . . 134
6.15 Non-dimensional normal force evaluated at 73% blade span from closed-loop
simulations using the microflap and AFC with a saturation limit of 𝑃jet = 1.2
kW/blade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.16 Non-dimensional normal force coefficient for uncontrolled case (baseline) and
closed-loop control using the microflap and AFC with a saturation limit of
𝑃jet = 1.2 kW/blade; oncoming flow direction is left to right (𝜇 = 0.30) and
blades rotate counter-clockwise . . . . . . . . . . . . . . . . . . . . . . . . . 136
6.17 Non-dimensional torque coefficient for uncontrolled case (baseline) and closed-
loop control using the microflap and AFC with a saturation limit of 𝑃jet = 1.2
kW/blade; oncoming flow direction is left to right (𝜇 = 0.30) and blades rotate
counter-clockwise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
6.18 Performance penalty associated with vibration reduction using closed-loop
AFC, including contributions from in-plane drag and power required for actu-
ation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.19 Baseline and controlled 4/rev hub loads obtained using closed-loop AFC at
various advance ratios; results are non-dimensional . . . . . . . . . . . . . . . 140
6.20 Continuous and discrete AFC input signals computed by HHC algorithm for
vibration reduction at various advance ratios . . . . . . . . . . . . . . . . . . 141
6.21 Amplitude and phase of 2, 3, 4, and 5/rev harmonics computed by HHC algo-
rithm for closed-loop vibration reduction using AFC at various advance ratios . 142
6.22 Vibration reduction and associated performance penalty using closed-loop AFC
at various advance ratios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

B.1 Transformation from the “0” system to the “1” system . . . . . . . . . . . . . 154
B.2 The transformation from the “2” system to the “4” system . . . . . . . . . . . 155
B.3 The transformation from the “3” system to the “5” system . . . . . . . . . . . 156
C.1 A schematic of the helicopter in descending flight . . . . . . . . . . . . . . . . 177

x
 LIST OF TABLES

TABLE

2.1 Convergence of steady lift, moment, and drag coefficients obtained from 2D
and 3D CFD simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2 Convergence of mean lift, moment, and drag coefficients obtained from 2D
and 3D CFD simulations of PS normal blowing . . . . . . . . . . . . . . . . . 36
2.3 Comparison of aerodynamic loads and shedding frequencies obtained from 2D
and 3D CFD simulations of PS normal blowing (𝛼 = 4◦ , 𝐶 𝜇 = 0.03) . . . . . . 51

6.1 Rotor parameters used in computations . . . . . . . . . . . . . . . . . . . . . 115

A.1 Sum and product tables for ternary finite field . . . . . . . . . . . . . . . . . . 150

xi
 LIST OF APPENDICES

APPENDIX

A Pseudorandom Signal Generation . . . . . . . . . . . . . . . . . . . . . . . . 149

B Structural Dynamic Model and Generalized Loads . . . . . . . . . . . . . . . 152

C Solution Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

xii
 LIST OF ABBREVIATIONS

2D Two-Dimensional
3D Three-Dimensional
AFC Active Flow Control
ATR Active Twist Rotor
AVINOR Active Vibration and Noise Reduction
BVI Blade-Vortex Interaction
CAMRAD/JA Comprehensive Analytical Model of Rotorcraft Aerodynamics and Dy-
namics, Johnson Aeronautics
CCW Counterclockwise
CFD Computational Fluid Dynamics
CRANE Control of Revolutionary Aircraft with Novel Effectors
CW Clockwise
DARPA Defense Advanced Research Projects Agency
DES Detached Eddy Simulations
HHC Higher-Harmonic Control
IBC Individual Blade Control
LES Large Eddy Simulations
MBB Messerschmitt-Bölkow-Blohm
NASA National Aeronautics and Space Administration
OBC On-Blade Control
ONERA Office National d’Etudes et de Recherches Aérospatiales (French na-
tional aerospace research center)
PS Pressure Side
/rev Per Rotor Revolution
RFA Rational Function Approximation
ROM Reduced-Order Model
RSB Retreating Side Blowing
SS Suction Side
STOL Short Take-Off and Landing
URANS Unsteady Reynolds-Averaged Navier-Stokes

xiii
 LIST OF SYMBOLS

𝑎 𝑗 , 𝑎 𝑗0 , 𝑎 𝑗2 coefficients used in the ONERA dynamic stall model

A, B, C coefficient matrices defined for flow control reduced-order model
𝑏 rotor blade semi-chord length
𝑐 rotor blade chord length
𝐶𝑙 , 𝐶 𝑚 , 𝐶 𝑑 sectional lift, moment, and drag coefficients
Δ𝐶𝑙ds , Δ𝐶𝑚ds , Δ𝐶𝑑ds changes in sectional lift, moment, and drag used by the ONERA dy-
namic stall model
Δ𝐶𝑙fc , Δ𝐶𝑚fc , Δ𝐶𝑑fc changes in sectional lift, moment, and drag predicted by flow control
reduced-order model
𝐶𝑑𝑓 helicopter fuselage drag coefficient
𝐶𝑛 local normal force coefficient
𝐶𝑄 local torque coefficient
𝐶𝑇 helicopter thrust coefficient
𝐶𝑊 helicopter weight coefficient
𝐶𝜇 jet momentum coefficient
𝐶 𝜇,avg jet momentum coefficient evaluated at the average Mach number at 75%
blade span
C0 , C1 , D, E, R coefficient matrices in computational-fluid-dynamics-based rational
function approximation model
C𝑂 𝑗 , C 𝐼 𝑗 , C𝑁𝑊𝑖 𝑗 influence coefficients used in the free-wake calculation
𝐷 drag force
𝐷𝑓 parasitic drag of the fuselage
D(𝜓, 𝜙) wake distortion
𝑒 blade root offset from the center of rotation
ê𝑥𝑖 , ê𝑦𝑖 , ê𝑧𝑖 unit vectors associated with the 𝑖th coordinate system
𝐸 𝑗 , 𝐸 𝑗2 coefficients used in the ONERA dynamic stall model
𝐸 𝐼𝜂𝜂 , 𝐸 𝐼 𝜁 𝜁 principal bending stiffnesses of the blade cross-section
𝑓 frequency
𝑓¯ non-dimensional frequency
𝑓𝑐 clock frequency of pseudorandom multi-level signal
𝑓𝑓 equivalent flat plate area of helicopter fuselage
𝑓max maximum frequency of interest for system identification problem
f vector of aerodynamic loads predicted by computational-fluid-
dynamics-based rational function approximation model
f𝑏 ( · ) blade motion equations
f𝑡 ( · ) trim equations

xiv
Δffc vector of aerodynamic loads predicted by flow control reduced-order
model
𝐹𝑥4 , 𝐹𝑦4 , 𝐹𝑧4 4/rev components of rotor hub shear
F matrix of basis functions associated with Kriging regression model
F𝐻 instantaneous hub shear force
F 𝑅𝑘 resultant blade root force for the 𝑘th blade
𝑔 acceleration due to gravity
𝑔𝑆 𝐿 , 𝑔𝑆 𝐹 , 𝑔𝑆𝑇 distributed structural damping factors in lag, flap, and torsion
g𝑎 ( · ) aerodynamic state equations
g𝑎𝑅 ( · ) reduced dorm of the aerodynamic state equations
𝐺𝐽𝑏 torsional stiffness of the blade cross-section
ℎ plunge displacement at 1/4-chord
h generalized motion vector
𝐻 horizontal force
𝐼MB2 , 𝐼MB3 principal mass moments of inertia of the blade cross-section
I identity matrix
𝐽 quadratic cost function for higher-harmonic control algorithm
𝐾 static gain associated with process model
𝐾 𝑁𝑊 number of azimuthal steps for near wake retained
𝐿𝑏 blade length
𝑚¤ jet flow control jet mass flow rate
𝑀 local Mach number
𝑀𝑏 blade mass
𝑀jet flow control jet Mach number
𝑀𝑥4 , 𝑀𝑦4 , 𝑀𝑧4 4/rev components of rotor hub moment
𝑀∞ freestream Mach number
𝑀 𝑝𝑡 rotor pitching moment about the hub center
𝑀 𝑟𝑙 rotor rolling moment about the hub center
𝑀 𝑦𝑤 rotor yawing moment about the hub center
M𝐻 instantaneous hub moment
M 𝑅𝑘 resultant blade root moment for the 𝑘th blade
𝑛𝐿 number of aerodynamic lag terms in the rational function approximation
approach
𝑁𝑏 number of rotor blades
𝑁basis number of basis functions associated with Kriging regression model
𝑁sp number of sample points in training data
𝑁𝑥 number of input variables
𝑝 𝑜,jet flow control jet stagnation pressure
p𝐴 distributed aerodynamic load per unit length of the blade
p𝐷 distributed structural damping load per unit length of the blade
p𝐺𝑏 distributed gravitational load per unit length of the blade
p 𝐼𝑏 distributed inertial load per unit length of the blade
𝑃jet flow control jet power
𝑞 𝑤𝑖 , 𝑞 𝑣𝑖 , 𝑞 𝜙𝑖 coefficients of the flap, lead-lag, and torsional mode shapes

xv
q induced velocity
q𝑏 vector of blade degrees of freedom
q𝑡 vector of trim variables
q𝐴 distributed aerodynamic moment per unit length of the blade
q𝐷 distributed structural damping moment per unit length of the blade
q𝐺𝑏 distributed gravitational moment per unit length of the blade
q 𝐼𝑏 distributed inertial moment per unit length of the blade
𝑄𝑡 tail rotor torque
Q aerodynamic transfer matrix
Q̃ approximation of Q
Qz weighting matrix on the vibratory hub loads
𝑟 radial distance from rotor hub
𝑟 𝑗 , 𝑟 𝑗0 , 𝑟 𝑗2 coefficients used in the ONERA dynamic stall model
𝑅 rotor blade radius
𝑅 𝐿𝑀 reverse flow parameter for lift and moment
𝑅krg spatial correlation function associated with Kriging model
r𝑏 position of the blade when it generates the wake element
rkrg spatial correlation vector used in Kriging
r𝑝 position vector of a point on the deformed blade
r𝑤 position of a wake element
r𝐸 𝐴 position vector for a point on the elastic axis
R weighting matrix on the control input
R0 position vector of deformed elastic axis
Rkrg spatial correlation matrix used in Kriging
R𝑝 position vector of a point on the blade before deformation
R𝑡 vector of trim residuals
𝑅𝑒 𝑐 Reynolds number based on airfoil chord
𝑠 Laplace variable
𝑠¯ non-dimensional Laplace variable
𝑡 dimensional time
𝑡¯ reduced time
𝑡𝑘 time of controller update
𝑇 axial tension
𝑇𝑐 clock period of pseudorandom multi-level signal
𝑇conv convective time scale, 𝑐/𝑈∞
𝑇𝑡 tail rotor thrust
T sensitivity matrix relating control input to the plant output
T̂LS least-squares estimate of the sensitivity matrix T
𝑢, 𝑣, 𝑤 displacements of a point on the blade’s elastic axis due to blade defor-
mation
𝑢 cont , 𝑢 disc continuous and discrete flow control input signals
𝑢 𝑁𝑐 , 𝑢 𝑁 𝑠 𝑁/rev cosine and sine amplitudes of continuous AFC input signal
𝑢 SS , 𝑢 PS suction side and pressure side flow control inputs
u control input vector

xvi
ufc vector of flow control inputs
𝑈 local oncoming flow velocity magnitude
𝑈jet flow control jet velocity magnitude
𝑈𝑃 , 𝑈𝑇 components of local flow velocity normal to the hub plane and in the
hub plane
𝑈∞ freestream velocity magnitude
V𝐴 total air velocity vector
V 𝐴1 airflow velocity due to forward flight, blade rotation, and induced inflow
V 𝐴2 airflow velocity due to blade dynamics
V𝐸 𝐴 velocity of a point on the elastic axis of the blade due to blade dynamics
𝑤 jet chordwise width of flow control jet orifice on airfoil surface
w disturbance to the plant
𝑊 helicopter weight
𝑊0 , 𝑊1 generalized normal velocity distributions on the airfoil
W matrix relating plant response to disturbance
𝑥𝑟𝑒𝑣 reverse flow boundary
x vector of aerodynamic states associated with computational-fluid-
dynamics-based rational function approximation model
xfc vector of flow control reduced-order model states
𝑋𝐴 offset between the aerodynamic center and the elastic axis
𝑋FA , 𝑍FA longitudinal and vertical offsets between rotor hub and helicopter aero-
dynamic center
𝑋FC , 𝑍FC longitudinal and vertical offsets between rotor hub and helicopter center
of gravity
𝑋Ib offset of the blade cross-sectional center of mass from the elastic axis
𝑋IIb offset of the blade cross-sectional center of area from the elastic axis
𝑋𝑡 , 𝑍𝑡 longitudinal and vertical offsets between the tail rotor center and the
rotor hub center
𝑦¯ 0𝑏 , 𝑧¯0𝑏 coordinates of a point on a blade cross-section relative to the elastic
axis, measured in the “5” system
𝑌 lateral force
z plant output vector

Greek Symbols

𝛼 angle of attack
𝛼𝑐𝑟 critical angle of attack of the airfoil used for the separation criterion in
the ONERA dynamic stall model
𝛼𝐷 descent angle
𝛼eff effective angle of attack
𝛼𝑅 rotor shaft angle
𝛽𝑝 blade precone angle
𝜷 vector of regression coefficients associated with Kriging model
𝛾 Lock number
𝛾𝑛 poles for rational function approximation

xvii
𝛾𝑠 strength of shed vorticity
𝛾𝑡 strength of trailed vorticity
Γ(𝑟) circulation distribution
Γ 𝐼 , Γ𝑂 inboard and outboard circulation peaks in dual wake model
Γ𝑙 , Γ 𝑚 , Γ 𝑑 states of the ONERA dynamic stall model
𝛿 microflap deflection
𝛿limit saturation limit on control amplitude
𝛿 𝑁𝑐 , 𝛿 𝑁 𝑠 𝑁/rev cosine and sine amplitudes of the microflap deflection
𝜖 small dimensionless magnitude of the order of the blade slopes
𝜖 dc error of discrete output relative to continuous output
𝜃0 main rotor collective pitch
𝜃 0𝑡 tail rotor collective pitch
𝜃 1𝑐 main rotor lateral cyclic pitch
𝜃 1𝑠 main rotor longitudinal cyclic pitch
𝜃 geom geometric angle of attack
𝜃 tw blade pretwist distribution
𝜽 vector of correlation parameters associated with Kriging model
𝜆𝑡 tail rotor uniform inflow ratio
𝜆𝑥 , 𝜆 𝑦 , 𝜆 𝑧 main rotor inflow ratio components
𝜇 helicopter advance ratio
𝜌 air density
𝜌𝑏 density of blade structure
𝜌𝐼 radial position of interior circulation peak
𝜎 rotor solidity
𝜎𝑡 tail rotor solidity
2
𝜎var variance of Gaussian process
𝜏 time constant associated with process model
𝜏𝑙ds , 𝜏𝑚ds , 𝜏𝑑ds time delays for lift, moment, and drag associated with ONERA dynamic
stall model
𝜙inflow inflow angle
𝜙offset phase offset angle
𝜙𝑅 fuselage roll angle
𝜙torsion elastic torsional deformation of the blade
𝜙𝑤 wake age
Φ nonlinear mapping function
Φ̂ surrogate mapping function
𝜓 azimuth angle
𝜔F , 𝜔L , 𝜔T blade flap, lag, and torsional natural frequencies
𝜔𝛼 pitching frequency of airfoil
Ω rotor angular speed

Superscripts and Subscripts

( ) −1 inverse
( )T transpose

xviii
(¤) derivative with respect to time
(¥) second derivative with respect to time
( ),𝑥 partial derivative with respect to 𝑥
( ),𝑥𝑥 second partial derivative with respect to 𝑥
( ),𝑥𝑥𝑥 third partial derivative with respect to 𝑥
(¯) average over one rotor revolution

xix
 ABSTRACT

This work presents a comprehensive computational analysis examining the potential of

active flow control (AFC) for helicopter rotor vibration reduction in forward flight. The
initial phase of the work focused on characterization of a novel fluidic actuator concept
which employs pulsed air jets on the upper and lower surfaces of the blade near the trailing
edge. Computational fluid dynamics (CFD) simulations were developed to evaluate the
unsteady aerodynamic effects introduced by AFC on an airfoil in a two-dimensional set-
ting. The CFD simulation approach was validated by comparison with recent wind tunnel
experiments performed in a parallel experimental study at Georgia Tech. Subsequently, a
reduced-order model (ROM) was developed for calculating unsteady aerodynamic loads
induced by AFC at reduced computational cost when compared to CFD. The ROM was
constructed based on training data obtained from CFD simulations performed across a
broad range of flow conditions representing the helicopter rotor operating environment.
Validation of the ROM against direct CFD simulations demonstrated the ROM’s ability to
accurately predict unsteady loads in a fraction of the computational run time.
Next, the effect of AFC on rotor blade vibrations was examined by incorporating the
flow control ROM into a comprehensive aeroelastic rotor analysis code. The comprehen-
sive code models the fully-coupled flap-lag-torsional rotor blade dynamics and features
an unsteady aerodynamic model comprised of three elements: (1) a CFD-based rational
function approximation model for calculating sectional blade loads in attached flow, (2) the
ONERA dynamic stall model for calculating sectional blade loads in separated flow, and (3)
a free-wake model for calculating the influence of the three-dimensional wake on the rotor

xx
inflow velocity distribution. Open-loop control simulations employing AFC actuation were
used to gain insight into the types of actuation signals that are most effective for vibration
reduction. Closed-loop control simulations based on the higher-harmonic control (HHC)
algorithm were developed to account for the discrete operating characteristics of the AFC
actuators, since each actuator, in practice, can only be on or off at a fixed jet strength. The
sensitivity of the closed-loop vibration reduction to the actuation power available, enforced
as a saturation limit in the HHC algorithm, was examined.
Results have demonstrated the effectiveness and control authority of the AFC concept
for vibration reduction: up to 83% reduction of the vibratory hub loads is possible. Fur-
thermore, the vibration reduction capability was shown to be comparable to that obtained
using an electromechanically-operated microflap. The effect of AFC on the overall rotor
performance during closed-loop vibration control was also calculated, to illustrate the fun-
damental trade-off that exists in rotorcraft applications. Finally, the closed-loop control ap-
proach was employed at several different rotor advance ratios. The simulations displayed a
consistent level of vibration reduction in a variety of flow conditions, including high-speed
flight where dynamic stall produces high vibratory loads.

xxi
 CHAPTER 1

Introduction, Background, and Objectives

1.1 Introduction

Helicopters are a unique class of aircraft. The ability to hover and perform precise maneuvers
at low speeds enables them to carry out a number of challenging tasks in areas that are
otherwise inaccessible to fixed-wing aircraft and ground-based vehicles. Helicopters are
indispensable for applications such as search-and-rescue, medical transport, firefighting, and
military operations.
To remain aloft, a helicopter rotor must operate in a complex unsteady aerodynamic
environment, depicted in Fig. 1.1. This leads to several undesired effects, particularly when
operating in forward flight. An unavoidable issue is the asymmetry that develops in the
rotor plane due to the combination of blade rotation and forward flight. The rotor blades
encounter faster oncoming air on the advancing side of the rotor and slower oncoming air
on the retreating side. This asymmetry becomes more apparent – and more problematic – as
the forward flight speed of the helicopter increases. Large variations in the Mach number
along the blade span produce several different flow regimes which coexist on the rotor. On
the advancing side, transonic flow is encountered near the blade tips where shock waves
may develop. On the retreating side, reverse flow is encountered near the blade root where
flow passes over the blades from trailing edge to leading edge. Dynamic stall, which is
characterized by unsteady flow separation, may also be encountered on the retreating side of

1
 Figure 1.1: Illustration of the complex aerodynamic environment of a helicopter

the rotor. Moreover, the shed wakes and tip vortices produced by the rotor blades yield a
complex and unsteady wake geometry which influences the rotor inflow and the aerodynamic
loading on the blades.
The unsteady aerodynamic loads, when combined with elastic forces due to blade
flexibility and inertial forces due to rotation, induce a dynamic structural response in the
rotor blades. Furthermore, the individual blade responses are combined at the rotor hub and
are transmitted to the helicopter fuselage as vibrations. Vibrations have an adverse effect
on the fatigue life of the blades, as well as other mechanical and electronic components,
and thus affect maintenance and operating costs. High vibration levels also cause crew
and passenger discomfort. Vibrations are present throughout the entire flight envelope of a
helicopter, but they are particularly troublesome in the presence of blade-vortex interaction
(BVI) encountered during low-speed descending conditions and dynamic stall encountered
during high-speed conditions.
Reduction of vibratory loads is a critical consideration in helicopter design and devel-
opment. The goal of achieving the same “jet-smooth” ride one can expect in a fixed-wing
aircraft has motivated sustained research on passive and active vibration reduction ap-

2
proaches over the last four decades. Active control approaches implemented on the rotor
are the most effective because vibrations are eliminated before they propagate through the
fuselage. Various approaches have been considered for active vibration control, including
higher-harmonic control implemented at the swashplate, pitch-link actuated individual blade
control, on-blade control using actively controlled trailing-edge flaps or microflaps, and the
active twist rotor. However, a helicopter employing any of these approaches has not yet made
it into production. The hesitancy towards active vibration control is driven primarily by cost
of implementation since such a system would introduce additional mechanical complexity,
weight, and power requirements, and manufacturers have been unable to justify the cost.
Among the broader aerospace and wind energy communities, there has been a growing
interest in augmenting, or altogether replacing, mechanical control surfaces with active flow
control technology. Active flow control (AFC) usually involves some form of air blowing or
suction to induce local fluid motion near a surface, so as to modify the aerodynamic loads.
Numerous applications of AFC technology have been considered, such as lift enhancement,
drag reduction, separation control, and circulation control. Furthermore, myriad AFC
devices have been developed throughout the literature and are often tailored to meet a
specific objective. The potential advantage of AFC compared to mechanical actuation is the
elimination of moving parts. Additionally, AFC systems have practically no influence on
the exterior shape of the surfaces they are embedded in.
This dissertation focuses on an entirely new application of AFC: helicopter rotor vi-
bration reduction. The overall goal is to demonstrate active on-blade control with similar
performance as an electromechanical device, like the microflap, but with the added benefit
of low mechanical complexity. In this chapter, two relevant bodies of literature are reviewed.
A review of the available helicopter vibration reduction approaches is presented in Section
1.2. A review of active flow control and its various applications is presented in Section
1.3. Furthermore, the objectives of this dissertation are defined in Section 1.4 and the novel
contributions of this research are summarized in Section 1.5.

3
1.2 Literature Review on Helicopter Vibration Reduction

Approaches

A comprehensive review of helicopter vibration control approaches was presented in Ref. [8],
emphasizing the use of on-blade control effectors to reduce vibrations at the source, i.e.,
the rotor. The rotor hub acts as a filter on the oscillatory loads generated by the rotor.
Consequently, only integer multiples of the one-per-revolution (1/rev) frequency are present.
When the blades are identical, or when 1/rev oscillations have been removed by properly
balancing the rotor, the oscillatory loads occur only at integer multiples of the blade-passage
frequency 𝑁 𝑏 /rev, where 𝑁 𝑏 is the number of blades. Hence, helicopter vibration reduction
approaches primarily focus on reducing 𝑁 𝑏 /rev loads.
Passive approaches represent the most mature and widely implemented form of vibration
reduction. A survey of these approaches was presented in Ref. [9]. Passive devices that
are typically added to the helicopter include absorbers mounted on the rotor hub or blades,
isolators mounted between the rotor and gearbox, and attenuators mounted at various
locations in the fuselage. The primary drawback of these devices is the significant weight
penalty they impose. Structural optimization of the mass and stiffness properties of the
rotor blades represents another passive approach to vibration reduction [10, 11] which offers
significantly lower weight penalty. The main advantage of rotor blade structural optimization
compared to other passive approaches is that vibration levels are reduced at the source before
they are propagated to the airframe. Multidisciplinary optimization approaches such as
Ref. [12] have incorporated full comprehensive aeroelastic rotor analyses into the blade
optimization framework, so as to optimize for minimum vibration in BVI or dynamic stall
conditions.
The remainder of this literature review focuses on active control approaches that reduce
vibration levels by modifying the unsteady airload distribution along the blade span.
The oldest approach for active vibration control is higher harmonic control (HHC).

4
The HHC approach modifies the pitch angle on the blades through the non-rotating lower
component of the conventional swashplate. The shortcoming of this approach is that the
same pitch is imposed on all the blades. The pilot controls the helicopter through the
0/rev collective and 1/rev cyclic pitch inputs. In the HHC approach, higher harmonic pitch
inputs at 𝑁 𝑏 /rev are superimposed on top of the pilot commands in order to reduce 𝑁 𝑏 /rev
vibrations. In the rotating frame, this translates to pitch inputs at the (𝑁 𝑏 − 1)/rev, 𝑁 𝑏 /rev,
and (𝑁 𝑏 + 1)/rev harmonics. A review of the HHC approach was presented in Refs. [13]
and [14]. Furthermore, HHC has been shown to be effective for vibratory load reduction in
computational simulations, wind tunnel tests, and flight tests [8, 15].
Individual blade control (IBC) is an alternative approach to HHC that modifies the
pitch of each blade in the rotating frame. In the IBC approach, each blade is controlled
independently of the other blades by pitch-link actuators located between the swashplate
and the blade. Similar to the HHC approach, the pitch inputs intended for vibration control
are superimposed on top of the pilot’s collective and cyclic pitch commands. This approach
is far less restrictive than the HHC approach, since arbitrary time-dependent control signals
can be applied to each blade. The vibration reduction capability of the IBC approach
has been demonstrated in several wind tunnel investigations [16], computational studies
[17], and flight tests [18]. As noted in Ref. [19], IBC has been considered for several
other applications besides vibration reduction, including gust alleviation, blade lag stability
augmentation, and stall flutter suppression.
A different form of active rotor control considered for vibration reduction is the Active
Twist Rotor (ATR). The NASA/Army/MIT ATR employs piezoelectrically actuated fibers
embedded in the blade that modify the blade twist distribution when a voltage is applied
to the actuators. The vibration reduction potential of the concept was tested in several
wind tunnel experiments for open-loop and closed-loop control in forward flight [20–22].
The main advantages of the ATR are that no moving parts are involved and the blades are
aerodynamically clean. However, the high voltages required for actuation and high blade

5
fabrication costs are major challenges that remain unresolved.
On-blade control (OBC) concepts such as trailing-edge flaps and microflaps have also
been considered for vibration reduction. The advantage of on-blade control, compared to
HHC implemented at the swashplate or pitch-link actuated IBC, is the relatively low power
required for actuation. For OBC, the actuators cover only part of the blade span, usually near
the 75% span location. Partial-span trailing-edge flaps are the most mature type of actuators
considered for on-blade control since these have been studied in aeroelastic simulations
[23–25], wind tunnel tests on Mach-scaled [26, 27] and full-scale [28] rotors, and flight tests
[29]. Microflaps have also been studied extensively in CFD simulations [30] and aeroelastic
simulations [7, 31, 32] for vibration reduction. The small size of a microflap is a potentially
attractive feature for vibration control since this allows for high-bandwidth actuation with
low power requirements.
Comprehensive aeroelastic rotor simulations play an important role in evaluating the
effectiveness of active control devices. A key component of comprehensive analyses is the
calculation of unsteady aerodynamic loads in an accurate, yet computationally inexpensive
way. The computational cost associated with high-fidelity CFD simulations is excessive
when performing control simulations. However, it was shown in Ref. [8] that CFD-based
predictions are necessary for accurate computation of rotor vibrations. Reduced-order
modeling (ROM) approaches that utilize a limited amount of high-fidelity CFD data to
construct an approximate model of the unsteady aerodynamics are essential for obtaining a
solution suitable for active control of vibrations. The AVINOR code [33, 34], developed at
the University of Michigan, employs a rational function approximation (RFA) model, based
on CFD data generated offline, to predict the aerodynamic effects introduced by trailing
edge flaps and microflaps installed on the rotor blades [7, 31]. Other ROM approaches
based on CFD data considered for rotary-wing applications include the proper orthogonal
decomposition [35] and the surrogate-based recurrence framework [36]. Additional CFD-
based ROM approaches suitable for fixed-wing stability and control applications were

6
surveyed in Ref. [37]. Furthermore, ROMs can reduce computer run time by several orders
of magnitude when predicting unsteady aerodynamic loads, compared to full-order CFD
simulations.
For closed-loop control involving on-blade control devices, the control surface deflection
is constrained, due to practical limitations of the driving actuation system. The effect of
actuator saturation on the vibration reduction achieved using trailing edge flaps was first
examined in numerical simulations in Ref. [38]. Actuator saturation was implemented by
an auto-weighting approach, which iteratively adjusted the control weighting matrix in
the HHC algorithm to properly constrain the flap deflection to within ±4◦ . This method
showed excellent vibration reduction capability compared to two other saturation approaches
considered, namely truncation and scaling of the flap deflection. An improved method for
incorporating actuator saturation into the HHC algorithm involving constrained non-linear
optimization was implemented in Ref. [31] for vibration and noise reduction using mi-
croflaps. This approach has several advantages over the auto-weighting approach, including
improved computational efficiency, especially when utilizing more than one control surface
on each blade. A similar optimization-based approach for constraining the deflection of
individually-controlled trailing edge flaps on a model-scale rotor was presented in Ref. [39].

1.3 Literature Review on Active Flow Control and Its Ap-

plications

Flow control is a term broadly used in fluid mechanics to describe a process or feature
that is designed to alter the natural development of a fluid flow. There are several ways
to classify different flow control approaches [40], and the first distinction often made is
whether a system is passive or active. Passive approaches do not require any form of power
input and usually consist of modifications to the shape and/or texture of an aerodynamic
surface. Passive flow control approaches also include any form of excitation that is activated

7
by the flow itself [41]. Active flow control (AFC) approaches, however, interact with the
surrounding flow by using actuators. This implies that an external power source is required
for AFC systems. AFC can be employed in a variety of open- and closed-loop control
schemes, and therefore it has received considerably more attention in the recent literature
than passive flow control. The following literature review focuses primarily on active flow
control and its applications.

1.3.1 Overview of Engineering Goals

The modern era of flow control is often attributed to the pioneering work of Prandtl, who
in 1904 developed boundary layer theory to explain flow separation in viscous fluids [42].
Prandtl’s experiments included the use of active suction to control the boundary layer on
a cylinder and prevent its separation. Viscous flow effects were not well understood prior
to the introduction of boundary layer theory, and thus earlier flow control improvements
were based on empirical observations. An interesting example of passive flow control that
originated in the “empirical era” is the use of surface roughness patterns on golf balls to
increase their range [43]. For aeronautical applications, flow control typically involves
a deliberate modification of the boundary layer to improve aircraft performance. Hence,
modern flow control approaches rely on an understanding of the phenomena governing
wall-bounded flows, which has improved significantly since the time of Prandtl.
Flow control can be used to accomplish a number of different engineering goals. The
most widely studied is separation control – that is, control of boundary layer separation.
Separation occurs when an adverse streamwise pressure gradient is present within the
boundary layer. The pressure gradient decelerates flow near the wall to the point of flow
reversal, which displaces the shear layer away from the wall. It is desirable to avoid
separation, since it can lead to significant loss of lift and increase in form drag. There are
several factors that influence the severity of flow separation on an airfoil, including angle of
attack, Reynolds number, and Mach number. At low angles of attack, flow generally remains

8
fully attached; however, as the angle of attack increases, flow begins to separate near the
trailing edge. At higher angles, the point of separation moves forward towards the leading
edge, until the stall angle is reached and flow becomes fully separated. At low Reynolds
numbers, the boundary layer is less likely to transition from a laminar state to a turbulent
state. Furthermore, a laminar boundary layer is more susceptible to flow separation than a
turbulent boundary layer. At high Mach numbers, especially in the transonic regime, the
presence of shocks on the airfoil can also induce separation. Moreover, separation control is
intended to prevent or delay separation and/or promote reattachment of the boundary layer,
so as to improve lift and reduce form drag.
Laminar flow control is another application of flow control that involves manipulation
of the boundary layer. The aim of laminar flow control is to keep the boundary layer in a
laminar state as long as possible, so as to reduce skin friction drag. Although a laminar
boundary layer is more susceptible to separation, it produces lower shear stress at the wall
than a turbulent boundary layer, leading to reduced skin friction. Reduction of skin friction
drag offers the potential to reduce fuel consumption in fixed-wing transport applications
[1, 44], since flow separation is not typically encountered during cruise conditions.
A different form of flow control, which has been examined primarily for high-lift
applications, is circulation control [45]. Circulation control involves modifying the flow
near the trailing edge of an airfoil to actively control the wake deflection angle, or Kutta
condition. This yields significant control authority over the airfoil circulation, and thus
the lift. The conventional sharp trailing edge is replaced by a rounded trailing edge, and
tangential blowing slots are placed on the upper and/or lower surfaces of the airfoil. The
combination of the tangential jets and the rounded trailing edge is designed to exploit the
Coandă effect, i.e., the tendency of a turbulent jet to remain attached to a surface. A similar
effect can be achieved by perpendicular blowing near the trailing edge. In this situation, the
airfoil trailing edge can be sharp rather than rounded. Circulation control works by blowing
a tangential jet over a rounded trailing edge. However, perpendicular blowing works by

9
entraining (or “pulling”) flow from the opposite side of the airfoil around the trailing edge
to modify the Kutta condition. This aerodynamic effect resembles the effect of a microflap
placed on the airfoil [46].
Several other fluid dynamics problems have been addressed using flow control, however
these will not be discussed in detail. For example, increased turbulent mixing in the
combustion element of a jet engine is beneficial for improving fuel burn efficiency and
reducing emissions [47]. Also, control of turbulent pressure fluctuations has implications
for reducing acoustic noise [48]. Lastly, a subset of flow control actuators, namely fluidic
oscillators, rely on precise control of the internal fluid dynamics within the actuators to
produce a sweeping jet output [49].

1.3.2 Overview of Actuator Classifications

A wide variety of actuators have been developed for AFC applications. Each device is
typically designed with attention to the desired objective, the flow regime in which it
operates, space constraints, and power requirements. A survey of various AFC devices was
presented in Ref. [50], where actuators were divided into three categories based on their
function: fluidic actuators, moving object/surface actuators, and plasma actuators.
Fluidic devices use blowing and/or suction through slots or orifices in a surface to modify
the surrounding flow. This form of actuation can be used for several different purposes.
Fluid blowing through a surface introduces high-momentum flow into the boundary layer.
The high-momentum flow mixes with the slow-moving flow near the surface to effectively
re-energize the boundary layer and delay flow separation. Conversely, suction through
a surface removes slow-moving flow from the boundary layer. This is also effective for
delaying flow separation.
A special class of fluidic actuators are synthetic jets, depicted in Fig. 1.2a, which
alternately perform blowing and suction in an oscillatory manner. Alternating blowing and
suction is achieved by oscillation of a flexible membrane which modifies the volume of

10
a hollow cavity beneath the surface. These devices are also known as zero-net-mass-flux
devices, since the cycle of blowing and suction yields no net mass injection or removal from
the surrounding flow.
Another type of fluidic actuator is the combustion powered actuator, depicted in Fig. 1.2b.
It produces a high-velocity pulsed jet by combustion of a fuel-air mixture within a cubic-
centimeter-scale chamber. This device has superior control authority over synthetic jets,
especially when operating in the compressible flow regime.
Fluidic oscillators are a different type of fluidic actuator which exploit naturally unstable
fluid dynamics within the actuator to produce a sweeping jet output from a steady flow input.
These include the feedback and feedback-free variants, depicted in Figs. 1.2c and 1.2d.
Another class of flow control devices are moving object/surface actuators. These include
vibrating wires, ribbons, and flaps, as well as morphing surfaces. Whereas fluidic actuators
involve blowing or suction to add or remove momentum from the surrounding flow, moving
object/surface actuators work by generating disturbances within the boundary layer through
small-scale structural vibrations. Theses devices are useful for suppressing flow separation,
since they can influence the transition process from laminar to turbulent flow.
A third class of flow control devices are plasma-based actuators. Plasma actuators
introduce momentum into the boundary layer by ionizing air molecules near the surface
to form a thin plasma layer. The ionized molecules within the plasma layer transfer their
momentum to the surrounding flow through collisions with neutral particles, thereby re-
energizing the boundary layer. Plasma actuators are an attractive solution for controlling
flow separation because they have no moving parts and have low weight penalty. However,
plasma actuators lose control authority as the freestream flow speed increases and are
ineffective in the compressible flow regime.

11
 (a) Synthetic jet actuator [50] (b) Combustion powered actuator [51]

(c) Feedback fluidic oscillator [49] (d) Feedback-free fluidic oscillator [49]

Figure 1.2: Conceptual schematics of various fluidic actuators for active flow control

1.3.3 Experimental Studies

In recent years, the principal thrust of AFC research for helicopter applications has been
aimed at controlling the adverse effects of dynamic stall. Dynamic stall is a result of unsteady
flow separation and reattachment on the blades that produce large periodic variations in
angle of attack, resulting in high vibration levels. This phenomenon occurs primarily on
the retreating blade region of the rotor. Reference [52] provides a thorough description
of dynamic stall on helicopter rotor blades. Flow control approaches aimed at controlling
dynamic stall involve manipulating the separated flow that develops as the airfoil exceeds its

12
static stall margin. Wind tunnel experiments performed in Ref. [53] focused on suppressing
the dynamic stall vortex, which forms during the pitch-up portion of the dynamic stall cycle,
by employing leading edge suction. By contrast, other experimental studies have employed
some form of blowing on the upper surface of the airfoil near the leading edge. In Ref. [54],
continuous blowing was shown to have some advantages for suppressing the dynamic
stall vortex compared to employing a leading-edge slat or a drooped leading edge airfoil.
Continuous blowing was also examined in Ref. [55] for alleviating shock-induced dynamic
stall occurring at a freestream Mach number of 0.5. Moreover, continuous blowing was
compared with pulsed blowing actuation in Refs. [56–59]. Pulsed blowing showed similar or
better performance when suppressing dynamic stall compared to continuous blowing at the
same blowing pressure. The additional benefit of pulsed blowing is a significant reduction
in the overall mass flow required. Pulsed blowing was implemented in Refs. [60–62] using
combustion-based actuators. The spark-ignition of a fuel-air mixture within a miniaturized
combustion chamber created the necessary pressure rise to generate a high-momentum jet
with very low overall mass flow rate. Another approach to reducing the mass flow rate
requirements for dynamic stall control was examined in Ref. [63], where an oscillating
valve was implemented to provide continuous blowing over a limited portion of the dynamic
stall cycle. The oscillating valve was synchronized with the pitch oscillation of the airfoil.
It was found that blowing was only required over 11% of the pitch cycle to achieve the
same benefits as continuous blowing over the entire cycle. Furthermore, AFC has also been
examined for dynamic stall control on thick airfoil sections typically used on wind turbine
blades [64, 65].
Two significant wind tunnel tests of flow control on a full-scale helicopter rotor have
been performed [66, 67]. The first was performed in 1960 at the NASA Ames Research
Center [66]. Two different rotors were tested for their ability to delay retreating blade stall
at advance ratios in the range of 0.30 ≤ 𝜇 ≤ 0.46: one employing leading-edge blowing and
the other employing mid-chord blowing. In both cases, compressed air was blown through

13
the rotor drive shaft to a manifold above the rotor hub which connected to the ducted rotor
blades. The blowing slots were located between 60 to 95% blade span. While mid-chord
blowing produced no benefit, leading-edge blowing yielded a 25% increase in the stall
boundary of the rotor. Furthermore, the improvement gained by using continuous blowing
was replicated using cyclic blowing on the retreating side of the rotor only, which required
approximately half the total air flow compared to the continuous case. The second major
wind tunnel test of flow control on a helicopter rotor was performed in 2017 by Sikorsky
[67]. A passive retreating side blowing (RSB) concept intended for dynamic stall alleviation
was tested. The passive RSB concept relied on centrifugal effects to naturally drive flow
from the inboard region of the blade outwards towards the blade tip through an internal
duct inside the blade. It was a passive system because no additional power was supplied to
increase the air flow rate through the blades. Similar to Ref. [66], leading-edge blowing slots
were located approximately 60 to 95% blade span. However, the passive RSB system did
not produce sufficient jet strength to have any effect on dynamic stall or rotor performance.
Reduction of fuselage drag is another application of AFC that has been considered
within the rotorcraft community. Flow separation from the aft body is a major source of
drag for helicopters with large rear loading ramps. Wind tunnel experiments performed
in Refs. [68–70] demonstrated drag reduction on bluff-body models resembling generic
helicopter fuselages by employing fluidic actuation on the bottom of the rear fuselage. Steady
blowing and pulsed blowing actuation were compared in Refs. [68] and [69]. Synthetic jet
actuators were also compared in Ref. [68]. Furthermore, pulsed combustion actuators were
employed in Ref. [70].
In fixed-wing applications, another goal of AFC has been to control flow separation
on high-lift devices such as trailing edge flaps. Flow separation occurs when the flaps are
deflected at high incidence angles (e.g., during approach and landing), resulting in a loss of
lift. Wind tunnel experiments performed in Ref. [71] demonstrated on several airfoil/trailing-
edge-flap configurations that periodic blowing is an effective method for reducing flow

14
separation and enhancing high-lift capability. The blowing slots were typically located near
the leading edge of the flapped portion of the airfoil. Furthermore, a broad survey of periodic
excitation concepts for separation control was presented in Ref. [72]. In Ref. [73], flow
separation on an inclined flat surface, representing a flap, was controlled by periodic acoustic
excitation. A woofer speaker was used to generate periodic flow disturbances inside a cavity
located at the flap leading edge. This produced an alternating cycle of blowing and suction
resembling the operation of a synthetic jet actuator. In Refs. [74, 75], periodic excitation
was applied using sweeping-jet fluidic oscillators, which alternate the blowing direction of a
steady jet in a periodic fashion. The experiments performed in Ref. [74] demonstrated 2D
separation control on high-lift airfoil configurations such as a simple flap and a Fowler flap.
Moreover, the experiments performed in Ref. [75] tested a similar AFC concept in a 3D
setting, where the actuators were installed on the wing of a scaled common research model
resembling a commercial transport aircraft. A closed-loop feedback approach to separation
control on a trailing edge flap was demonstrated in Ref. [76]. Pulsed blowing actuation
from the leading edge of the flap was applied based on hot-film sensor measurements of the
wall shear stress at various locations on the flap surface. Furthermore, the wall shear stress
measurements were used to formulate an indicator for flow separation.
Several studies have examined normal blowing (i.e., perpendicular to the airfoil surface)
near the trailing edge as a method for modifying aerodynamic loads at low and moderate
angles of attack where the baseline flow is fully attached. Experiments performed in
Ref. [77] employed continuous normal blowing on the lower surface of a symmetric airfoil
at the mid-chord location. It was observed that, for a wide range of jet strengths tested, the
change in lift coefficient was proportional to the square root of the jet momentum coefficient:
p
Δ𝐶𝑙 ∝ 𝐶 𝜇 . In Ref. [78], normal blowing was applied on the upper surface of an airfoil
as a means of reducing the lift coefficient for gust alleviation. In Ref. [79], the ability of
lower-surface normal blowing to increase the lift coefficient of an airfoil was compared
with that of a Gurney flap placed at the same location. A Gurney flap is a small flap with

15
a height of typically 1 to 5% chord attached to the pressure surface of the airfoil near the
trailing edge in order to increase the lift. It was shown that at low angles of attack, normal
blowing produced a lower drag penalty compared to the Gurney flap when producing the
same lift increment. A comparison of normal blowing jets and actively deployable Gurney
flaps for controlling aerodynamic loads on wind turbine blades was performed in Ref. [80].
Both devices were tested on the upper and lower surfaces, so as to decrease or increase
the lift coefficient. Furthermore, a survey of active control devices, including microjets
and microtabs, that have been considered for wind turbine applications was presented in
Ref. [81].
There are several other applications where AFC has been used to control aerodynamic
forces. For example, an array of sweeping-jet fluidic oscillators was used to control flow
separation on a full-scale vertical tail model from a Boeing 757 aircraft in Ref. [82]. This
study was a precursor to a series of flight tests on a full-scale demonstrator [83]. The
reduction of flow separation on the vertical tail enhanced the control authority of the rudder
at high sideslip angles. In Ref. [84], wind tunnel experiments were used to characterize the
unsteady lift response and flow re-attachment process of a statically stalled semi-circular
wing when pulsed blowing actuation was applied from the leading edge. Lastly, synthetic
jet actuators were used to control the aerodynamic forces on a generic 3D axisymmetric
bluff body in Ref. [85].

1.3.4 Flight Tests

A thorough history of early active flow control research in the U.S., Germany, and Great
Britain was documented in Ref. [1]. The primary objective of AFC during this period (late
1930s through mid 1960s) was skin friction drag reduction via laminar flow control. The
earliest flight tests of an AFC system were performed in 1941 at the Langley Aeronautical
Laboratory, where a section of a B-18 airplane wing was equipped with suction slots from
20 to 60% chord length, as illustrated in Fig. 1.3. These early flight tests demonstrated an

16
 Figure 1.3: B-18 airplane with test section for laminar flow control [1]

Figure 1.4: Northrop X-21 active laminar flow control airplane [1]

improvement in the extent of laminar flow over the test section, although not to the extent
predicted in earlier wind tunnel experiments. Similar wind tunnel and flight tests employing
laminar flow control on a partial-wing section were performed throughout the 1950s. In
the 1960s, Northrop began developing the X-21 aircraft (Fig. 1.4), an experimental aircraft
which had active suction slots directly integrated into the wings of the airplane. The program
was supported by the U.S. Air Force in an effort to improve high-altitude long-endurance
capabilities. Although the X-21 was capable of achieving laminar flow over 95% of the area
intended for laminarization under ideal conditions, its performance suffered in the presence
of ice crystals in the atmosphere. The program ultimately lost support at the start of the
Vietnam war.

17
 Figure 1.5: Boeing 707 wing modifications for boundary layer control [2]

Active flow control was also examined in the 1960s by Boeing as a means for enhancing
the high-lift capabilities of commercial transport aircraft [2]. A Boeing 707 prototype
was modified to divert bleed air from the engines to the trailing-edge flaps, as shown in
Fig. 1.5. The bleed air was used to add momentum to the boundary layer over the flaps, thus
increasing the maximum lift coefficient at low speeds during landing. The bleed air was
augmented by an ejector nozzle to increase the momentum of the jet. The development of
the system included a 2D transonic wind tunnel test campaign, scale model tests, and a joint
NASA-Boeing full-scale flight test program. At low approach speeds, handling qualities
were degraded when AFC was not employed. By using AFC to increase the maximum lift
coefficient, the effectiveness of the control surfaces greatly improved.
In 1957, the Cessna Aircraft Company performed full-scale flight tests of a modified
CH-1 helicopter (Fig. 1.6a) employing AFC on the rotor blades [3]. The objective of the
AFC system was to delay stall on the retreating blade and thereby increase the maximum
flight speed of the helicopter. Specially designed blades, which incorporated mid-chord
suction slots from 50% to 94% span (Fig. 1.6b), were installed on the helicopter. The slots
were connected to a suction pump in the helicopter fuselage (Fig. 1.6c) by a complex ducting
system that ran through the blade cross-section and rotor hub. The suction pump drew power
from the main engine. Suction was applied in a cyclic fashion on the retreating side of

18
 (a) CH-1 helicopter
(b) Rotor blade with suction slots

(c) Suction pump system

Figure 1.6: Helicopter rotor blade boundary layer control research by Cessna Aircraft [3]

the rotor. This significantly reduced flow separation on the rotor blades and enabled flight
speeds up to 13% above the normal stall-limited speed of the helicopter. However, the added
weight, complexity, and power consumed by the suction pump rendered the AFC system
impractical for any useful application.
During the 1970s and 1980s, circulation control concepts were applied to enhance the
capabilities of short-takeoff-and-land (STOL) vehicles. In Ref. [86], circulation control
was employed on a Navy fighter jet demonstrator, the Grumman A-6A, shown in Fig. 1.7a.
Bleed air from the jet engines was used to power trailing-edge Coandă jets, which were
capable of increasing the wing lift coefficient by a factor of two. This had implications for
reduced approach speed and reduced landing and takeoff distances when operating from a

19
 (a) Grumman A-6A circulation control demon-
strator (b) NASA Quiet Short-Haul Research Aircraft

Figure 1.7: Circulation control flight research for STOL applications [4]

naval aircraft carrier. It was suggested that this could also allow for higher aircraft payload
weights. A different form of circulation control was tested on the NASA Quiet Short-Haul
Research Aircraft [4], shown in Fig. 1.7b. This demonstrator utilized upper surface blowing
combined with trailing-edge flaps. By blowing the engine exhaust over the upper surface of
the trailing-edge flaps, this enabled the engine thrust to be redirected by more than 90◦ .
More recently, zero-net-mass-flux AFC actuators were employed on the XV-15 tiltrotor
aircraft to reduce the rotor download induced on the wings in hover [5]. In hover, the wing
in the wake of the tiltrotor is at nearly 90◦ angle of attack. The AFC actuators were used
to reduce flow separation and thereby reduce the wing download. Wind tunnel tests and
large-eddy simulations of a wing were used for an initial characterization of the flowfield
with/without AFC. Static tests on a 16% scale model of the XV-15 were performed and
compared with flight test data on the full-scale aircraft in hover, shown in Fig. 1.8. The AFC
actuators demonstrated up to 14% reduction in download measured on the full-scale aircraft.
There has been substantial research effort at the University of Manchester to develop
fixed-wing vehicles that employ AFC as the primary control system for trimming and
maneuvering the aircraft [87]. This eliminates the need for conventional control surfaces
such as ailerons, elevators, and rudders. Reference [87] provides an overview of the
development of various small-scale (<100 kg) prototype vehicles that have been flown to
demonstrate actuator control authority. The two important AFC concepts employed on the

20
 Figure 1.8: Bell XV-15 during AFC test flights and the arrangement of actuators [5]

vehicles are circulation control and thrust vectoring. Furthermore, DARPA has recently
selected Aurora Flight Sciences and Lockheed Martin to develop similar technology as part
of the Control of Revolutionary Aircraft with Novel Effectors (CRANE) program [88]. The
goal of the CRANE program is to produce a full-scale demonstrator that uses only AFC
actuation to control the aircraft.
Other recent flight tests have implemented AFC on the vertical tail of a commercial
transport aircraft. In Ref. [83], an array of sweeping jet actuators was installed on the vertical
tail of a Boeing 757 demonstrator. The purpose of the AFC system was to reduce flow
separation encountered at high rudder deflection and sideslip angles, thereby enhancing the
rudder control authority. A rudder with higher control authority implies that the aircraft can
be designed with a smaller vertical tail and still satisfy the same control requirements, e.g.,
one-engine-inoperable during takeoff. The potential benefit of a smaller vertical tail is the
reduction in overall drag. In a different study, laminar flow control was used to demonstrate
drag reduction in a more direct manner on the vertical tail of an Airbus A320 demonstrator
[89]. This involved installing a perforated leading-edge section on the vertical tail where
suction was employed to laminarize the boundary layer.

21
1.3.5 Numerical Simulations

A wide variety of AFC actuators has been developed in recent years through detailed
experimental and numerical studies [50]. Computational fluid dynamics (CFD) simulations
have been instrumental in characterizing the internal and external flow physics of the
actuators. A survey of modeling approaches was presented in Ref. [90] with commentary
on the applicability of various levels of fidelity in CFD codes, including unsteady Reynolds-
averaged Navier-Stokes (URANS) simulations, large eddy simulations (LES), and direct
numerical simulations. Simulations of fluidic oscillators, a type of actuator that exploits
naturally unstable fluid dynamics within the actuator to produce a sweeping jet, have
indicated that URANS simulations are inadequate for modeling this type of actuator [91, 92].
These studies have shown that higher-fidelity detached eddy simulations (DES, a hybrid of
URANS and LES) are necessary to accurately reproduce experimental observations, due to
the highly unsteady nature of the actuators. However, numerical studies involving simpler
actuator designs that employ pulsed blowing jets have shown that URANS simulations are
sufficiently accurate in these situations [46, 61]. A drawback of DES is the need for finer
grids and smaller time steps when flow control is applied. Reference [93] noted that DES
are typically applied to massively separated flows and that, by using AFC to suppress flow
separation, the requirements on grid resolution become even more stringent. This greatly
affects the computational cost of the simulations. Furthermore, URANS simulations provide
a balance of fidelity and computational cost that is appropriate for engineering applications.
Control of static stall and dynamic stall on 2D airfoil geometries have been examined in
several computational studies. For example, CFD simulations have been used to characterize
the operation of synthetic jet actuators [94–98]. In Ref. [94], URANS simulations were
compared with higher-fidelity hybrid URANS/LES simulations in terms of their ability to
predict stall alleviation on a static airfoil. The study demonstrated that URANS simulations
were sufficient to reproduce experimental results for low actuation frequencies. However,
hybrid URANS/LES simulations were needed to accurately predict higher actuation fre-

22
quencies. Furthermore, full LES simulations were employed in Ref. [95] and showed a high
degree of accuracy in the prediction of both the stalled and controlled airfoil performance.
In Ref. [96], URANS simulations were incorporated into an optimization framework to
determine the actuator location which maximized the lift coefficient in static stall. Dynamic
stall control using synthetic jets was considered in Refs. [97, 98]. Both studies employed
URANS simulations in order to examine the sensitivity of the dynamic stall performance to
parameters such as actuation frequency, jet strength, jet location, and jet angle on the airfoil.
A combined computational and experimental study on the use of combustion-based
actuators for dynamic stall control on a helicopter rotor was presented in Refs. [61, 99].
This included development of URANS simulations to model the pulsed operating mode
of the actuators in 2D and 3D airfoil simulations. Simulations of pulsed actuation were
correlated with high-speed wind tunnel experiments performed at 0.3 ≤ 𝑀 ≤ 0.5 for
static stall and dynamic stall of an airfoil. Additionally, a notional system-level model of
a rotor in high-speed forward flight was employed to illustrate the effect of AFC on rotor
performance. The system-level model was relatively simplistic since it assumed rigid blades
(no structural model), linear quasi-steady aerodynamics based on blade-element theory (no
unsteady aerodynamic model), and uniform inflow (no wake model). The effect of AFC was
incorporated by adding a simple model to reproduce the unsteady sectional aerodynamic
loads predicted during 2D simulations of dynamic stall with and without AFC actuation. The
system-level simulations indicated that AFC had the potential to improve rotor performance
at high advance ratios (up to 𝜇 = 0.5), including higher rotor lift-to-drag and greater power
available.
Other notional studies of the system-level performance benefits of employing AFC on
helicopter rotor blades were presented in Refs. [100, 101]. In Ref. [100], the design of
a unique stopped-rotor configuration, called the “X-Wing,” was described. The design
employed circulation control on the rotor, which involved elliptically-shaped airfoil sections
with air blowing slots on the leading and trailing edges. The intended mode of operation was

23
to employ blowing in a cyclic fashion so as to redistribute lift over the rotor at high advance
ratios (𝜇 > 0.5). A simple aerodynamic analysis based on blade element theory was used to
illustrate this concept. After accelerating to an advance ratio of 𝜇 = 0.7, the intent was to
stop the rotor completely so that the vehicle would function as a fixed-wing aircraft capable
of high forward flight speeds. A full test of this concept was never performed. In Ref. [101],
the potential of AFC to alleviate BVI noise was examined. The rotor concept employed
leading-edge blowing and suction from 60% to 100% span on the blades. A relatively
simple rotor model was employed, which assumed rigid blades and did not incorporate
viscous effects into the calculation of aerodynamic loads. Unsteady rotor wake calculations
were performed using the CAMRAD/JA code [102] and the blade pressure distribution was
calculated using a 3D potential flow solver. The simulations demonstrated that leading edge
blowing could alleviate the impulsive leading edge pressure spikes responsible for BVI
noise (although no acoustic calculations were performed) whereas leading edge suction had
an unfavorable effect.
A computational study on the passive RSB concept for helicopter rotor dynamic stall
alleviation was performed in Ref. [103] in parallel with the wind tunnel test performed by
Sikorsky in Ref. [67]. A full rotor geometry was modeled using 3D URANS simulations,
assuming rigid blades. The simulations showed good correlation with experimental data
in terms of the rotor performance, sectional aerodynamic loading, and properties of the
centrifugally-driven flow through the internal ducts within the rotor blades. Furthermore,
the simulations indicated a similar conclusion as the experiments: the passive RSB concept
produced no net benefit towards dynamic stall alleviation or rotor performance enhancement.
That is, the minor benefit of stall alleviation on the retreating side of the rotor was offset by
the additional drag penalty induced on the advancing side. The simulations also investigated
using additional pumping through the duct to increase the blowing jet strength. This was
not studied in the experiments. The powered blowing concept showed more promising
results than passive blowing. Powered blowing was more effective for stall alleviation on

24
the retreating side and also reduced the drag penalty induced on the advancing side.

1.4 Objectives of the Dissertation

The literature review provided in the previous sections indicates considerable interest in
employing active flow control on a helicopter rotor to improve aerodynamic performance.
However, AFC has not been studied in a comprehensive manner. Furthermore, its potential
for vibratory load reduction on helicopter rotors has been examined only for the case
of dynamic stall. Vibration reduction at normal flight conditions usually implies cruise
or descent conditions. Conducting such a comprehensive assessment of AFC for this
flight regime is the overall objective of this dissertation. This goal is accomplished by
incorporating an unsteady aerodynamic model representing the effects of fluidic actuation
into a comprehensive aeroelastic rotor simulation framework. The aeroelastic simulations
are performed using an extensively modified version of the AVINOR code [33, 34].
The specific objectives of this dissertation are:

1. Development of CFD simulations for modeling fluidic actuation on a 2D airfoil.

2. Validation of the CFD simulations by comparison with experimental data from recent
wind tunnel experiments performed at Georgia Tech.

3. Characterization of the unsteady aerodynamic phenomena associated with fluidic

actuation in conditions resembling the helicopter rotor operating environment.

4. Construction of a CFD-based ROM for calculating the unsteady aerodynamic loads

induced by flow control at reduced computational cost compared to direct CFD
simulations.

5. Integration of the flow control ROM into a comprehensive aeroelastic rotor simulation
code.

25
 6. Development of a higher-harmonic control (HHC) approach for active vibration
reduction that is compatible with the discrete operating characteristics of the fluidic
actuators, including treatment of the actuator saturation problem.

7. Examination of the vibratory load reduction produced by flow control when applied
in open-loop and closed-loop modes of actuation.

8. Determination of the power penalty associated with implementing flow control for
vibration reduction.

9. Comparison of the vibratory load reduction due to flow control with that obtained by
employing a conventional electromechanically-operated microflap [7].

1.5 Key Contributions of the Dissertation

The work presented in this dissertation has produced several new contributions towards
the modeling and implementation of active flow control for rotorcraft vibration reduction.
Furthermore, this is the first study in which active flow control has been integrated into a
comprehensive aeroelastic rotor analysis code for simulations aimed at reducing vibratory
hub loads. The key contributions of this dissertation are:

1. Development and validation of a CFD simulation model for a novel trailing-edge

fluidic actuator concept intended for control of aerodynamic loads at low and moderate
angles of attack.

2. Construction of a CFD-based ROM for simulating the unsteady aerodynamic loads

induced by fluidic actuation in a manner that is compatible with comprehensive
aeroelastic rotor simulations.

3. Development of a closed-loop simulation method which produces discrete-valued

control signals that are compatible the discrete operating modes of the fluidic actuators.

26
 4. An assessment of the potential of active flow control for vibration reduction on
a representative rotor configuration in forward flight and its comparison with the
vibration reduction achieved by a microflap.

5. Determination of the rotor performance penalty associated with vibration reduction,

including the rotor torque due to changes in the in-plane drag induced by AFC
combined with the power required for fluidic actuation.

1.6 Dissertation Outline

The dissertation consists of seven chapters. Chapter 1 provides a brief introduction, a review
of the relevant literature, and the objectives and key contributions of this dissertation. Chap-
ter 2 describes the CFD approach for modeling fluidic actuators. The unsteady aerodynamic
loads induced on an airfoil are characterized for conditions resembling the helicopter rotor
operating environment. The CFD simulations are validated by comparison with wind tunnel
experimental data. Chapter 3 presents the development of the CFD-based ROM for calcu-
lating unsteady aerodynamic loads induced by fluidic actuation at reduced computational
cost. The selection and generation of CFD training data for ROM construction are also
addressed. Chapter 4 describes the various components of the comprehensive aeroelastic
rotor simulation framework AVINOR. The implementation of the flow control ROM within
the AVINOR code is also described. Chapter 5 presents the control approaches employed
during closed-loop simulations of vibration reduction. The treatment of discrete control
modes associated with the fluidic actuators is addressed. Chapter 6 presents the results of
the computational simulations. The vibratory hub loads obtained from open-loop and closed-
loop control simulations employing AFC are examined at different actuation strengths and
flight conditions. The vibration reduction capability of AFC and the associated performance
penalty are compared with a microflap. Finally, Chapter 7 presents the conclusions and new
contributions of the dissertation, concluding with recommendations for future work.

27
 CHAPTER 2

CFD Simulations of Fluidic Actuation and Their

Experimental Validation

2.1 Bi-Directional Pulsed Fluidic Actuation

The flow control device considered is depicted in Fig. 2.1. The device consists of two
pulsed fluidic actuators, which are installed near the trailing edge of a VR-12 airfoil model
on the suction side (SS) and pressure side (PS). The SS and PS actuators are controlled
independently, and each have two operating modes: tangential blowing and normal blowing.
The “normal” blowing mode refers to a jet oriented upstream at an angle of 60◦ relative to
the airfoil chord line. The tangential and normal blowing outlets on the airfoil surface are
located 0.84𝑐 and 0.88𝑐 from the leading edge of the airfoil, respectively.
The tangential and normal blowing modes are switched fluidically through the interaction

Control Jets
Supply Jet
Bypass Tangential Output

Normal Output
SS

PS

Figure 2.1: Trailing edge pulsed fluidic actuators for SS and PS blowing on a tabbed VR-12
airfoil

28
of a continuous supply jet and two weaker control jets within each actuator. While the supply
jet is constantly on, only one of the control jets is active at a time. The control jets merge
and interact with the supply jet in a mixing region, and hence the merged jet is diverted to
one of the two output channels. By alternating the flow of air through the two control jets,
the actuator output alternates between tangential and normal blowing. The diverging nozzle
section downstream of the mixing region exploits the tendency of the combined jet to attach
itself to the nozzle walls (i.e., the Coandă effect). Furthermore, the actuators are classified
as a type of bi-stable wall-attachment fluidic amplifier [104]. The actuators are designed to
switch between the operating modes while operating at a fixed jet strength since they are
powered by a steady supply of compressed air. Thus, the actuators operate in a discrete
fashion. For consistency with the experiments, the discrete operation of the actuators is also
simulated in CFD.
The actuators can be modified to replace either the tangential or normal blowing modes
with a bypass outlet. When connected, the bypass internally diverts the supply jet through
tubes inside the airfoil body and vents air from the airfoil’s spanwise edges (perpendicular
to the plane of Fig. 2.1), instead of allowing the air to exit through the airfoil surface. This
represents an off condition, since the vented air has no effect on the sectional aerodynamic
loads. For example, when the bypass is connected to the tangential output channel, the
actuator will switch between “off” and normal blowing.
The actuators were designed and fabricated at Georgia Tech, and have been tested in a
series of low speed wind tunnel experiments [105]. The test section of the open return wind
tunnel facility at Georgia Tech is shown in Fig. 2.2a. The flow conditions provided by the
tunnel correspond to 𝑀∞ = 0.06 and 𝑅𝑒 𝑐 = 535, 000 (𝑈∞ = 20 m/s). A modular VR-12
airfoil (𝑐 = 381 mm with a 0.05𝑐 trailing edge tab) was constructed from 2D interconnected
sections of aluminum shells supported by an internal spar. The model spans the full width
of the test section (910 mm) and includes end plates at its spanwise edges. The model is
mounted on the tunnel walls with high-resolution 6-component load cells that measure the

29
 Supply
Normal
Tangential

Bypass

(a) Wind tunnel test section (b) Schematic of actuator module

Figure 2.2: Array of trailing edge actuators installed on VR-12 airfoil model

lift, drag, and pitching moment on the airfoil. An array of actuator modules, depicted in
Fig. 2.2b, is installed across the center 22% of the airfoil model span. This constitutes
the active center section of the airfoil. Two chordwise fences were placed on the airfoil in
order to isolate the active center section of the model from the outer sections. The fences
reduce 3D spanwise effects and provide an approximately 3D flow over the airfoil and
actuators. Furthermore, the flow is tripped near the airfoil leading edge by a seam in the
airfoil construction, yielding a fully turbulent boundary layer.
A simplified representation of the experimental apparatus is employed in the CFD
simulations. The first simplification modifies the internal geometry. Rather than fully
simulating the internal flow within the actuators, the tangential and normal output channels
are truncated and the internal geometry is retained only near the actuators outlets at the
airfoil surface, as shown in Fig. 2.3. This simplification is discussed further in Section
2.2. The second simplification is associated with the external flow conditions, assuming
that a two-dimensional (2D) simulation is sufficiently accurate. This assumption does not
account for spanwise variations in the flow, nor does it account for the finite width of the
actuator modules. A few cases were computed using 3D simulations in order to examine the
differences between 2D and 3D flow solutions.

30
 Internal details removed Internal details
from simulations retained

Figure 2.3: Simplification of internal actuator geometry for CFD simulations

2.2 CFD Model of the Flow Control Device

The pulsed fluidic actuators described in the previous section are modeled using CFD
simulations in order to evaluate the performance of the flow control device at different flow
conditions. Two- and three-dimensional simulations at the wind tunnel conditions are used
to validate the simulation approach at low speeds. Subsequently, the simulations are used to
examine more realistic flow conditions that can be encountered by actuators employed on a
helicopter rotor blade. These simulations are conducted at higher freestream Mach numbers
that could not be examined in the experiments, due to limitations of the wind tunnel facility.
The CFD simulation method is described next.

2.2.1 Description of the Flow Solver

The simulations are performed using CFD++ [106], a commercially available code produced
and marketed by Metacomp Technologies. The code solves the compressible unsteady
Reynolds-averaged Navier-Stokes (URANS) equations using a finite volume formulation.
The CFD++ solver is a general-purpose code capable of modeling a variety of mesh types,
including structured, unstructured, hybrid, patched, and overset grids through a unified grid
methodology. Spatial discretization is based on a second-order total-variation-diminishing
scheme. For time-dependent simulations, a first-order implicit algorithm is implemented,
with second-order accuracy achieved by dual time-stepping. Additionally, multi-grid conver-
gence is used in the simulations. Several different turbulence models are available in CFD++.

31
 SS Tangential
SS Normal
PS Normal
PS Tangential

Figure 2.4: VR-12 airfoil and details of simplified trailing edge fluidic actuator geometry

The Menter 𝑘-𝜔 shear stress transport (SST) model was selected for the simulations, and a
fully-turbulent boundary layer was assumed.

2.2.2 Geometry and Boundary Conditions

The airfoil and integrated actuator module are simplified in order to reduce computational
cost. The internal flow within the actuators is only modeled near the actuator outlets,
as shown in Fig. 2.4. The fluidic switching mechanism is not modeled. Instead, the
tangential and normal blowing modes are approximated by prescribing inflow conditions at
the boundaries shown in Fig. 2.4.
The strength of actuation is characterized by a jet momentum coefficient, defined as:

2 𝑤
𝜌jet𝑈jet jet
𝐶𝜇 = 2𝑐
(2.1)
𝜌 ∞ 𝑈∞

The coefficient 𝐶 𝜇 is a measure of the relative jet strength, and it represents the ratio of the
momentum injected by the jet to the momentum of the freestream flow. For a given 𝐶 𝜇 ,
the inflow boundary conditions in the simulations are prescribed based on the maximum
velocity of the jet issuing from the airfoil surface, 𝑈jet . Specifically, the jet Mach number,

32
𝑀jet , is used to determine the stagnation pressure of the jet, 𝑝 𝑜,jet , from isentropic relations:

  𝛾/(𝛾−1)
𝑝 𝑜,jet 𝛾−1 2
= 1+ 𝑀jet (2.2)
𝑝∞ 2

where 𝑝 ∞ is the freestream static pressure and 𝛾 = 1.4 for air. When one of the jets is
active, the stagnation pressure at the corresponding inlet boundary is prescribed as 𝑝 𝑜,jet .
Otherwise, the inlet boundary is treated as a wall boundary.
Jet strengths of up to 𝐶 𝜇 = 0.03 are possible in the low-speed wind tunnel at Georgia
Tech. For 𝑀∞ = 0.06, this translates to 𝑀jet ≈ 0.2. Since the CFD simulations are also used
to explore more realistic flow conditions for a helicopter rotor, such as 𝑀∞ = 0.7, a weaker
relative jet strength of 𝐶 𝜇 = 0.005 was selected for the higher-speed simulations, so as to
avoid conditions close to 𝑀jet = 1. At 𝑀∞ = 0.5, a jet strength of 𝐶 𝜇 = 0.005 translates to
𝑀jet ≈ 0.75.

2.2.3 Grids Employed for Modeling the Airfoil/Actuator Combination

The computational domain for the 2D airfoil simulations is shown in Fig. 2.5a. The domain
is a structured C-grid composed of quadrilateral elements. The grid has a farfield boundary
50 chord-lengths from the airfoil. Figure 2.5b shows the details of the grid near the airfoil
and Fig. 2.5c shows greater detail in the vicinity of the trailing edge and fluidic actuators.
The appropriate spatial resolution for the nominal 2D grid was determined by generating
several grids and comparing the convergence of the aerodynamic loads obtained from
simulations. Increasingly refined grids were used to demonstrate convergence of the lift,
moment, and drag coefficients. The coarse grid has approximately 54,000 cells, the medium
grid has 108,000 cells, and the fine grid has 212,000 cells. Each level of grid refinement adds
40% more points in both the streamwise and normal directions for approximately twice the
total number of cells. For each grid, the size of the first cell from the airfoil wall boundary
was prescribed such that 𝑦 + ≤ 1 for the most restrictive flow conditions considered in the

33
 (a) Entire domain (b) Near airfoil (c) Near actuators

Figure 2.5: Structured computational grid for 2D CFD simulations of VR-12 airfoil with
pulsed fluidic actuators

simulations (i.e., 𝑀∞ = 0.7). Within the actuator cavities, the size of the first cell from the
cavity wall boundary was prescribed such that 𝑦 + ≤ 1 for jet velocities up to 𝑀jet = 1.
The steady state aerodynamic loads were computed for the wind tunnel conditions
(𝑀∞ = 0.06, 𝑅𝑒 𝑐 = 535, 000) at 4◦ angle of attack. Table 2.1 illustrates the differences in
the lift, moment, and drag coefficients as a percentage of the loads obtained on a coarser grid.
The differences in 𝐶𝑙 and 𝐶𝑑 indicate that the CFD solutions are consistent and converge
with finer grid resolution. The difference in 𝐶𝑚 also appears to converge, with less than 3%
difference between the medium and fine grids. Consequently, the medium grid with 108,000
cells was selected as the acceptable nominal 2D grid for the simulations.
The nominal 2D grid was also used to generate 3D grids with different levels of refine-
ment in the spanwise direction. In reality, the fluidic actuators have finite spanwise width
and there are small gaps between the actuator modules, as shown in Fig. 2.2b. In order
to reduce the total size of the 3D grids and approximately capture some of the spanwise
effects, only half of an actuator module is represented in the simulations, as indicated in
Fig. 2.6. This also includes half of the gap that separates two actuator modules. The gap is
part of the airfoil surface, and is thus treated as a solid wall boundary. Symmetry boundary
conditions are enforced at the spanwise edges of the computational domain. This mimics
an actuator located near the center of a spanwise array of other identical actuators, since

34
 (a) Perspective view (b) Top view

Figure 2.6: Structured computational grid for 3D CFD simulations of VR-12 airfoil with
pulsed fluidic actuators

Table 2.1: Convergence of steady lift, moment, and drag coefficients obtained from 2D and
3D CFD simulations

Grid # cells % diff 𝐶𝑙 % diff 𝐶𝑚 % diff 𝐶𝑑

2D coarse 0.05 M – – –
2D medium 0.1 M 0.9% 7.7% 0.7%
2D fine 0.2 M 0.3% 2.5% 0.4%
3D coarse 2.6 M – – –
3D medium 5.2 M 1.3% 0.05% 4.1%
3D fine 10.5 M 0.6% 0.01% 2.3%

the flow solution is mirrored across the symmetry planes. This geometry yields an approxi-
mately 2D flow. Despite this limitation, this configuration is useful for determining the grid
spacing needed to model the actuators in 3D. Three grids were generated in this manner
with increasing level of refinement in the spanwise direction: coarse (2.6 million cells),
medium (5.2 million cells), and fine (10.5 million cells). Table 2.1 indicates convergence of
lift, moment, and drag coefficients obtained on the 3D grids for a steady flow condition at
𝑀∞ = 0.06, 𝑅𝑒 𝑐 = 535, 000, and 𝛼 = 4◦ .
A second aspect of the 2D and 3D grid refinement study is the convergence of the
aerodynamic loads obtained from simulations of steady blowing actuation. The freestream
flow conditions are identical to the steady state simulations (𝑀∞ = 0.06, 𝑅𝑒 𝑐 = 535, 000,

35
Table 2.2: Convergence of mean lift, moment, and drag coefficients obtained from 2D and
3D CFD simulations of PS normal blowing

Grid # cells % diff 𝐶𝑙 % diff 𝐶𝑚 % diff 𝐶𝑑

2D coarse 0.05 M – – –
2D medium 0.1 M 1.2% 3.7% 0.5%
2D fine 0.2 M 0.6% 1.6% 0.4%
3D coarse 2.6 M – – –
3D medium 5.2 M 1.0% 0.8% 0.4%

and 𝛼 = 4◦ ), but with PS normal blowing applied at a jet strength of 𝐶 𝜇 = 0.03. These
simulations were performed in a time-dependent mode until a new steady state condition
was reached, such that the mean values of 𝐶𝑙 , 𝐶𝑚 , and 𝐶𝑑 were invariant with time. The
time step size was selected as Δ𝑡 = 0.004𝑇conv , where 𝑇conv is the convective time scale
of the airfoil. Similar to the unactuated simulations, Table 2.2 indicates convergence of
the aerodynamic loads in the 2D and 3D cases. Because the differences in lift, moment,
and drag between the coarse and medium 3D grids are less than 1%, it was decided that
simulations with the fine 3D grid are not warranted, due to the high computational cost of
such simulations. Based on these convergence results, the 2D medium and 3D coarse grids
were selected as the nominal 2D and 3D CFD grids for the simulations.
Due to the limited spanwise extent of the simulation domain, the 3D geometry yields
results that resemble an approximately 2D flow solution. A more realistic 3D scenario is
also considered. The nominal 3D grid was extended in both spanwise directions to create a
3D wing, where the actuator spans 0.1𝑐 and the wing extends an additional 0.5𝑐 past the
actuator, as illustrated in Fig. 2.7. The total size of the 3D-wing grid is 6.4 million cells.
Symmetry conditions are applied at the spanwise edges of the domain, to mirror the flow
solution across the end planes. This resembles a semi-infinite wing. Although this geometry
is not necessarily representative of the experiments, since fences are not modeled in the
simulations, it is a useful hypothetical scenario to consider. The actuators are intended to
be installed across a segment of a rotor blade, near the 75% span location, where 3D flow

36
Figure 2.7: Computational grid for 3D-wing geometry with fluidic actuators installed across
0.1𝑐 of the span

effects may be significant. Furthermore, the 3D-wing geometry allows sufficient space for
3D flow effects induced by actuation to develop. A comparison of the solutions obtained
from 2D and 3D simulations is provided in the next section.

2.3 Characterization of Flow Control Performance on a

Static Airfoil

The ability of the flow control device to modify the aerodynamic loads on a static airfoil at
low and moderate angles of attack is considered first. Low speed simulations representing
the wind tunnel experiments in Refs. [105, 107] are first performed in order to characterize
the unsteady aerodynamic phenomena associated with actuation and validate the simulation
approach. The solutions obtained using 2D and 3D simulations are briefly compared. Then,
the simulations are used to examine the effect of the increased freestream Mach number, up
to 𝑀∞ = 0.5, representing the conditions encountered in a practical helicopter rotor setting.

37
2.3.1 Low Mach Simulations and Comparison with Experiments

2.3.1.1 Step Response

The unsteady aerodynamic response due to PS normal blowing applied as a step input is
examined first. A 2D simulation starting from a steady flow solution at 𝛼 = 0◦ with both
jets off is carried out. At 𝑡¯ = 0 (where time is nondimensionalized as 𝑡¯ = 𝑡/𝑇conv ), the PS jet
is activated with a strength of 𝐶 𝜇 = 0.03 and remains on until 𝑡¯ = 10. After 𝑡¯ = 10, the jet is
deactivated and the flow is allowed to return to its steady baseline conditions. The unsteady
evolution of the lift, moment, and drag coefficients due to the input sequence is shown in
Fig. 2.8.
Several aspects of the step response are interesting. First, an initial transient starts after
the jet is activated, from 0 < 𝑡¯ < 0.5. The sharp decrease in lift and increase in pitching
moment are a result of the formation of a recirculating flow region on the PS downstream of
the jet. Figure 2.9 illustrates this behavior in terms of the instantaneous spanwise vorticity.
The time of minimum lift and maximum pitching moment coincides with the time when the
recirculating region reaches the trailing edge (Fig. 2.9c). This is followed by a rapid reversal
of the lift and moment from 0.5 < 𝑡¯ < 0.9. The reversal is caused by flow that is entrained
around the trailing edge from the SS to the PS of the airfoil, thus forming a trailing edge
vortex (Figs. 2.9d-e).
Subsequently, the aerodynamic loads begin to asymptotically approach a new steady
state characterized by neutrally stable oscillations about a time-invariant mean. The mean
and root mean square (RMS) values of the lift, moment, and drag coefficients are shown in
Fig. 2.8. The rise time (to within 5% of the asymptotic value) for 𝐶𝑙 is 5.9𝑇conv , the settling
time for 𝐶𝑚 is 2.0𝑇conv , and the settling sime for 𝐶𝑑 is 6.7𝑇conv .
The oscillations in aerodynamic loads are due to a progression of strengthening and
shedding of vorticity concentrations from the trailing edge, depicted in Fig. 2.10. Vorticity is
shed into the wake in an alternating fashion of clockwise (CW, blue) and counterclockwise

38
 0.6

0.4 CL = 0.44 0.09

Lift Coef.
0.2

-0.2 Unsteady Mean Mean RMS

0 5 10 15 20

Moment Coef.
0.1

0
CM = -0.06 0.04
-0.1

0 5 10 15 20
0.1
Drag Coef.

0.05
CD= 0.03 0.01
0

-0.05
0 5 10 15 20
On
Input

Off
0 5 10 15 20
Non-Dimensional Time

Figure 2.8: Unsteady aerodynamic response due to PS normal blowing (𝐶 𝜇 = 0.03); changes
in lift, moment, and drag coefficients are relative to jet off condition at 𝛼 = 0◦ , time
normalized as 𝑡¯ = 𝑡/𝑇conv

(CCW, red) concentrations. While the CW trailing edge vortex is strengthened, vorticity is
shed from the CCW recirculating region (Figs. 2.10a-b). Similarly, while the recirculating
region is strengthened, the trailing edge vortex is shed into the wake (Figs. 2.10c-d). With
sustained steady blowing, this pattern repeats with a non-dimensional frequency of 𝑓¯ = 2.9
cycles per convective time scale 𝑇conv . Note that the oscillations in 𝐶𝑙 , 𝐶𝑚 , and 𝐶𝑑 also
occur at this frequency.
After the jet is deactivated at 𝑡¯ = 10, the aerodynamic loads gradually return to their
baseline steady state values. Furthermore, the relaxation times for the lift, moment, and drag

39
 (a) 𝑡¯ = 0

(b) 𝑡¯ = 0.25

(c) 𝑡¯ = 0.50

(d) 𝑡¯ = 0.75

(e) 𝑡¯ = 1.00

Figure 2.9: Initial transient due to step input PS normal blowing (𝐶 𝜇 = 0.03), depicted by
contours of instantaneous normalized spanwise vorticity; time normalized as 𝑡¯ = 𝑡/𝑇conv

40
 (a) 𝑡¯ = 0

(b) 𝑡¯ = 0.08

(c) 𝑡¯ = 0.18

(d) 𝑡¯ = 0.28

(e) 𝑡¯ = 0.36

Figure 2.10: Periodic shedding due to steady PS normal blowing (𝐶 𝜇 = 0.03), depicted by
contours of instantaneous normalized spanwise vorticity; time normalized as 𝑡¯ = 𝑡/𝑇conv

41
coefficients are similar to the respective rise time/settling times due to jet activation.

2.3.1.2 Sensitivity to Jet Strength

The effect of the relative jet strength on the resulting aerodynamic loads is examined
next. Steady normal blowing is applied on the pressure and suction sides of the airfoil for
several different values of 𝐶 𝜇 , ranging from 0.005 to 0.04. The 2D CFD simulations are
compared with wind tunnel data from Georgia Tech [107] for a similar range of 𝐶 𝜇 . Figure
2.11 shows the changes in lift, moment, and drag coefficients relative to the baseline jet
off condition. The simulation data is provided in terms of the steady mean values of 𝐶𝑙 ,
𝐶𝑚 , and 𝐶𝑑 computed after 10𝑇conv of sustained blowing. Additionally, the RMS values
of the aerodynamic loads are indicated above and below the mean values obtained from
simulations.
The lift coefficient data show good correlation between simulations and experiments for
both PS and SS blowing. The trends are consistent with previous observations mentioned in
Refs. [46, 77, 79], indicating that the change in lift due to a perpendicular jet is proportional
p
to the square root of the jet momentum coefficient: Δ𝐶𝑙 ∝ 𝐶 𝜇 . The PS and SS jets produce
nearly symmetric changes in the lift and moment. As 𝐶 𝜇 increases, the magnitude of the
oscillations in lift, moment, and drag obtained from simulations also increases, as indicated
by the RMS values. In fact, the RMS magnitude of the moment coefficient reaches up to
75% of the mean value when 𝐶 𝜇 = 0.04. For most cases, the experimental lift and moment
data differ from the mean values from simulations by less than the RMS magnitude. The
agreement between simulations and experiments for the drag coefficient is not as good as
for the lift and moment. The simulations indicate that the drag penalty due to PS blowing
is closer to the drag penalty due to SS blowing than was obtained experimentally. The
simulation data for SS blowing reveals some agreement with the experimental data for
𝐶 𝜇 > 0.01. Furthermore, both simulations and experiments indicate linear trends in Δ𝐶𝑑 as
𝐶 𝜇 increases.

42
 0.1
0.5

Moment Coefficient
0.05

Lift Coefficient 0 0

-0.05
-0.5
-0.1
Exp. PS Exp. SS Exp. PS Exp. SS
CFD PS (mean) CFD SS (mean) -0.15 CFD PS (mean) CFD SS (mean)
-1
CFD PS ( rms) CFD SS ( rms) CFD PS ( rms) CFD SS ( rms)
-0.2
0 0.01 0.02 0.03 0.04 0.05 0 0.01 0.02 0.03 0.04 0.05
Momentum Coefficient Momentum Coefficient

(a) (b)

0.04
Drag Coefficient

0.03

0.02

0.01

0
Exp. PS Exp. SS
CFD PS (mean) CFD SS (mean)
-0.01 CFD PS ( rms) CFD SS ( rms)

0 0.01 0.02 0.03 0.04 0.05

Momentum Coefficient

(c)

Figure 2.11: Steady state aerodynamic response due to PS and SS normal blowing at various
jet strengths; changes in lift, moment, and drag coefficients are relative to jet off condition
at 𝛼 = 0◦

The oscillatory components of the aerodynamic loads obtained from simulations depend
on the jet strength, as illustrated in Fig. 2.12, in terms of the RMS magnitude of the lift
and the frequency of the oscillations. As 𝐶 𝜇 increases, the RMS magnitude increases
and the shedding frequency decreases. This behavior is similar for PS and SS actuation.
As mentioned in the previous section, the oscillatory aerodynamic loads are connected to
the strengthening and shedding of two vorticity concentrations of opposite sense (i.e., the
recirculating region and the trailing edge vortex) which are present downstream of the active
jet. Figure 2.13 depicts the time-averaged flow fields obtained from a limited number of

43
 0.2 6

Non-Dim. Shedding Freq.

5

RMS Lift Coefficient

0.15
4

0.1 3

2
0.05
1

0 0
0 0.01 0.02 0.03 0.04 0.05
Momentum Coefficient

Figure 2.12: RMS magnitude and frequency of lift oscillations due to PS (solid lines) and SS
(dashed lines) normal blowing at various jet strengths; frequency normalized as 𝑓¯ = 𝑓 𝑐/𝑈∞

simulations. Streamlines are overlaid to illustrate the effect of 𝐶 𝜇 on the relative sizes of the
recirculating region and the trailing edge vortex. As 𝐶 𝜇 increases, the jet penetrates further
into the cross flow, thus creating a larger recirculating region. This also produces a stronger
trailing edge vortex, since more flow is entrained from the opposite side of the airfoil around
the trailing edge. Consequently, the increased strength of the recirculating region, combined
with the trailing edge vortex, produce larger oscillations in the aerodynamic loads, due to
shedding. This explains the increase in RMS lift as 𝐶 𝜇 increases. Furthermore, due to the
stronger vorticity concentrations, additional time is required for each to reach full strength
after shedding. Thus, the shedding frequency is reduced for higher jet strengths.

2.3.1.3 Sensitivity to Angle of Attack

The 2D simulations are compared with wind tunnel data from Ref. [105] for various angles
of attack in the range of 0◦ ≤ 𝛼 ≤ 12◦ . Several different actuation modes are examined: 1)
baseline with both jets off, 2) PS normal blowing, 3) SS normal blowing, 4) PS tangential
blowing, and 5) SS tangential blowing. For the cases where blowing is active, a constant
jet strength of 𝐶 𝜇 = 0.03 is used. The computational and experimental results for the

44
 (a) PS, 𝐶 𝜇 = 0 (b) SS, 𝐶 𝜇 = 0

(c) PS, 𝐶 𝜇 = 0.01 (d) SS, 𝐶 𝜇 = 0.01

(e) PS, 𝐶 𝜇 = 0.04 (f) SS, 𝐶 𝜇 = 0.04

Figure 2.13: Time-averaged flow fields due to steady normal blowing at different jet
strengths; streamlines are plotted over contours of normalized spanwise vorticity

lift and moment coefficients are summarized in Fig. 2.14. Because the simulations of
normal blowing modes produce an oscillatory component in the aerodynamic loads, the
computational results for PS and SS normal blowing in Figs. 2.14a and 2.14c are provided
in terms of the mean and RMS values. Tangential blowing does not cause oscillations in the
aerodynamic loads.
The baseline data for 0◦ ≤ 𝛼 ≤ 8◦ indicate good agreement between simulations and

45
 2 2

1.5 1.5

Lift Coefficient

Lift Coefficient
1 Exp. Baseline (off) 1
CFD Baseline (off)
Exp. PS Exp. Baseline (off)
0.5 CFD PS (mean) 0.5 CFD Baseline (off)
CFD PS ( rms) Exp. PS
0 Exp. SS 0 CFD PS
CFD SS (mean) Exp. SS
CFD SS ( rms) CFD SS
-0.5 -0.5
0 5 10 15 20 25 0 5 10 15 20 25
Angle of Attack (°) Angle of Attack (°)

(a) Normal Blowing (b) Tangential Blowing

0.1 0.1
Moment Coefficient

Moment Coefficient
0.05 0.05
Exp. Baseline (off)
0 CFD Baseline (off) 0
Exp. PS Exp. Baseline (off)
CFD PS (mean) CFD Baseline (off)
-0.05 -0.05
CFD PS ( rms) Exp. PS
Exp. SS CFD PS
-0.1 -0.1
CFD SS (mean) Exp. SS
CFD SS ( rms) CFD SS
-0.15 -0.15
0 5 10 15 20 25 0 5 10 15 20 25
Angle of Attack (°) Angle of Attack (°)

(c) Normal Blowing (d) Tangential Blowing

Figure 2.14: Steady state lift and moment coefficients due to normal and tangential blowing
(𝐶 𝜇 = 0.03) applied on the pressure side and suction side

experiments. In this range, the lift curve is linear and the pitching moment is nearly constant.
Above 8◦ the simulations predict some pre-stall flow separation on the suction side of the
airfoil, as is evident from the non-linear lift curve and increase in pitching moment. However,
the experiments indicate a lesser degree of separation above 8◦ . Note that the experimental
results from Ref. [105] are reported as the values measured directly by the load cells. That
is, wind tunnel corrections were not applied to these values. Using an approximate method
from Ref. [108, Ch. 9.4], the experimental values can be adjusted to compensate for the size
of the airfoil model. The correction factors vary approximately linearly for 0◦ ≤ 𝛼 ≤ 12◦ .
The correction factor varies from 0.936 to 0.924 for 𝐶𝑙 , from 0.972 to 0.960 for 𝐶𝑚 (with

46
an additional offset term that varies from 0.001 to 0.012), and from 0.961 to 0.949 for 𝐶𝑑 .
With these corrections, the experimental data are somewhat closer to the simulation data for
the higher angles of attack.
The simulations of PS and SS normal blowing agree with experiments across the range
of 𝛼, despite the differences in the baseline data above 8◦ . For most cases, the experimental
data lie within the RMS values obtained from simulations, and often differ from the mean
values by less than 10%. Both simulations and experiments indicate that the lift and moment
induced by SS blowing is relatively insensitive to angle of attack. Conversely, the effect of
PS blowing tends to increase in magnitude as 𝛼 increases. It is also interesting to note that
the RMS magnitudes of the lift and moment due to PS blowing decrease with increasing 𝛼.
Figures 2.14b and 2.14d show that tangential blowing as a weaker effect on the lift
and moment when compared to normal blowing. For 𝛼 < 8◦ , the changes in 𝐶𝑙 and 𝐶𝑚
due to tangential blowing are negligible. For 𝛼 ≥ 8◦ , tangential blowing alleviates the
adverse effects of flow separation, as indicated by the extended linear regime of the lift
curve. Suction side blowing is slightly more effective than pressure side blowing at the
higher angles of attack. This is due to the fact that in the absence of tangential blowing,
flow separation occurs on the suction side of the airfoil. Furthermore, the simulations and
experiments show an excellent level of agreement.
The effects of normal and tangential blowing on the drag coefficient are examined next.
Figure 2.15 illustrates the changes in drag coefficient relative to the baseline jet off condition.
The baseline drag data are not compared here because the 2D simulations consistently
predict a drag coefficient that is 60-65% lower than the reported experimental values from
Ref. [105]. The larger value of baseline drag in the experiments is attributed to the presence
of the two chordwise fences that isolate the actuator array from the outer sections of the wind
tunnel model. Therefore, in each case the baseline data are subtracted from the actuated data
and only the drag penalty/reduction is considered.
The drag penalty due to normal blowing does not show a similar level of agreement

47
 0.05 0.05

Drag Coefficient

Drag Coefficient
0 0
Exp. PS
CFD PS (mean)
CFD PS ( rms) Exp. PS
Exp. SS CFD PS
CFD SS (mean) Exp. SS
CFD SS ( rms) CFD SS
-0.05 -0.05
0 5 10 15 20 25 0 5 10 15 20 25
Angle of Attack (°) Angle of Attack (°)

(a) Normal Blowing (b) Tangential Blowing

Figure 2.15: Steady state changes in drag coefficient, relative to jet off condition, due to
normal and tangential blowing (𝐶 𝜇 = 0.03) applied on the pressure side and suction side

with experiments to that obtained for the lift and moment coefficients. The simulations
indicate that the mean drag due to PS and SS blowing is approximately the same for all
angles of attack, although the RMS magnitudes vary significantly with 𝛼. The simulations
provide a reasonable approximation of the drag due to PS blowing up to 𝛼 = 8◦ . However,
the simulations of SS blowing consistently predict a drag penalty that is higher than was
obtained experimentally.
Tangential blowing reduces the airfoil drag, primarily due to the forward thrust of the
jets. The simulations of SS blowing show an excellent level of agreement with experiments.
Similar to the case of normal blowing, the simulations predict that PS and SS blowing have
an approximately equal effect on the drag. This observation contradicts the experimental
results that show that PS blowing produces a 35% greater drag reduction than SS blowing.
The dissimilarities in drag due to PS and SS blowing were not expected, particularly at the
low angles of attack. This may indicate some systematic error in the experimental data, such
as an inadvertently higher jet strength on the PS. Furthermore, the effect of the chordwise
fences on the measured aerodynamic loads in the wind tunnel has not been fully resolved.

48
2.3.1.4 Comparison of 2D and 3D Simulations

The differences between 2D and 3D simulations are examined next. The 2D simulations
have shown reasonable agreement with wind tunnel data for the lift and pitching moment
induced by flow control. The 3D simulations are employed in order to determine whether the
finite spanwise width of the actuators has a significant effect on the prediction of unsteady
aerodynamic loads. Two different 3D geometries and their corresponding flow solutions are
examined. The first configuration is the nominal 3D geometry (Fig. 2.6), which contains
half of an actuator module and half of the gap that separates two actuator modules. This
represents an actuator located near the center of a spanwise array of other identical actuators,
producing an approximately 2D flow. The second configuration is a 3D wing (Fig. 2.7),
which has an actuated segment that spans 0.1𝑐 and an unactuated segment that spans 0.5𝑐.
This represents a more realistic scenario where 3D effects can develop near the edge of the
actuator.
Simulations of PS normal blowing at 𝐶 𝜇 = 0.03 were performed at 4◦ angle of attack
using the 2D, nominal 3D, and 3D-wing geometries. The unsteady aerodynamic loads
due to a step input are shown in Fig. 2.16 for each case. The loads are also summarized
in Table 2.3 in terms of the baseline, mean, and RMS values of 𝐶𝑙 , 𝐶𝑚 , and 𝐶𝑑 , as well
as the non-dimensional shedding frequency ( 𝑓¯shed ) associated with the oscillations. Note
that, for consistency with the 2D calculations, the lift, moment, and drag coefficients of
the 3D wing were calculated for the actuated segment of the span only. The baseline loads
without actuation are in close agreement, with a 0.3% difference in 𝐶𝑙 , 0.7% in 𝐶𝑚 , and
1.4% in 𝐶𝑑 . When actuation is applied, the nominal 3D simulation yields a similar unsteady
response as the 2D simulation. The nominal 3D simulation predicts slightly lower mean
(3% to 8%) and RMS values (1% to 3% of the mean, relative to the baseline) compared
to 2D. Additionally, the nominal 3D simulation predicts a 9% higher shedding frequency.
The 3D-wing simulation, however, yields significantly different results. Whereas the 2D
simulation predicts an increase in 𝐶𝑙 of 0.44 (mean), the 3D-wing simulation predicts an

49
 1.2

Lift Coef.
0.8

0.6

0.4 2D 3D 3D-wing
0 2 4 6

Moment Coef.
0.1

-0.1

0 2 4 6
0.1
Drag Coef.

0.05

-0.05
0 2 4 6
On
Input

Off
0 2 4 6
Non-Dimensional Time

Figure 2.16: Unsteady aerodynamic response obtained from 2D and 3D CFD simulations of
PS normal blowing (𝛼 = 4◦ , 𝐶 𝜇 = 0.03); time normalized as 𝑡¯ = 𝑡/𝑇conv

increase in 𝐶𝑙 of 0.13, which is 70% lower than the 2D case. The 3D-wing simulation also
predicts a 50% smaller change in 𝐶𝑚 , compared to 2D. The change in 𝐶𝑑 is only 9% smaller
than the 2D simulation. The most remarkable differences between the 3D-wing and 2D
simulations are in terms of the RMS values. The RMS values of 𝐶𝑙 , 𝐶𝑚 , and 𝐶𝑑 obtained
from the 3D-wing simulation are an order of magnitude smaller than the 2D values. This is
accompanied by a 27% lower shedding frequency, indicating that the alternating shedding
of trailing-edge vorticity observed in 2D simulations (Fig. 2.10) is significantly reduced by
3D flow effects.
The spanwise variation of the sectional lift coefficient from the 3D-wing simulation is

50
Table 2.3: Comparison of aerodynamic loads and shedding frequencies obtained from 2D
and 3D CFD simulations of PS normal blowing (𝛼 = 4◦ , 𝐶 𝜇 = 0.03)

2D 3D 3D-wing
𝐶𝑙 base 0.602 0.600 0.603
𝐶𝑙 mean 1.048 1.015 0.736
𝐶𝑙 rms 0.080 0.069 0.004
𝐶𝑚 base -0.0132 -0.0133 -0.0133
𝐶𝑚 mean -0.0861 -0.0819 -0.0482
𝐶𝑚 rms 0.0361 0.0319 0.0020
𝐶𝑑 base 0.0142 0.0144 0.0140
𝐶𝑑 mean 0.0442 0.0407 0.0414
𝐶𝑑 rms 0.0115 0.0092 0.0007
𝑓¯shed 2.63 2.87 1.91

illustrated in Fig. 2.17. Also shown are the values of 𝐶𝑙 obtained from the 2D and nominal
3D simulations. As mentioned in the previous section, symmetry conditions were applied at
the spanwise edges of the 3D-wing to mirror the flow solution across the boundaries, such
that 𝑧¯ = 0.1 corresponds to the actuator centerline. On the actuated section, 𝐶𝑙 is greatest at
the centerline and decreases toward the actuator edge (𝑧¯ = 0). The lift drops sharply to the
left of the actuator and reaches a global minimum at 𝑧¯ = −0.02. Continuing to the left, 𝐶𝑙
increases slightly, then gradually decreases toward an asymptotic value of 0.67. This value
is 11% higher than the baseline without actuation. The results demonstrate that, although
the change in lift on the actuated segment of the 3D wing is significantly lower than the 2D
predictions, the actuation has a global effect on the 3D flow.
The similarities between the time-averaged flow fields obtained from 2D and 3D simu-
lations are illustrated in Fig. 2.18. The 3D solutions are taken as a slice over the actuator
centerline. The 2D and nominal 3D solutions are nearly indistinguishable, although a
minor difference in the 3D solution is that the upstream shear layer emanating from the jet
(indicated in blue) shows reduced penetration into the cross flow when compared to the
2D simulation. The 3D-wing solution exhibits a few notable differences. The downstream
shear layer emanating from the jet (indicated in red) penetrates further into the cross flow

51
 1.1
Baseline (off)
1 PS on (2D)
PS on (3D)

Lift Coefficient
PS on (3D-wing)
0.9

0.8 Actuated section

0.7

0.6

0.5
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1
Non-Dimensional Span

Figure 2.17: Spanwise distribution of sectional lift coefficient due to PS normal blowing
(𝛼 = 4◦ , 𝐶 𝜇 = 0.03); span normalized as 𝑧¯ = 𝑧/𝑐

than predicted in the 2D simulations. Thus, the center of the recirculating region is shifted
downwards, away from the airfoil surface. As a result, the shear layer from the suction side
of the airfoil (indicated in blue) is not entrained around the trailing edge to the same extent
as in the 2D simulations. Instead, the shear layer extends further downstream into the wake.
This contributes to the reduced change in sectional lift, compared to 2D simulations.
Figure 2.19 illustrates the spanwise variation of the 3D flow field obtained from the
3D-wing simulation. The images were generated by plotting isosurfaces of the Q-criterion –
a metric based on the velocity gradient tensor [109] commonly used for identifying coherent
vortical structures – colored by contours of the normalized spanwise vorticity. This primarily
illustrates the influence of the streamwise flow near the actuator edge. The upstream vortical
structures generated by the jet extend beyond the actuator edge and are deflected downstream,
indicating a significant departure from a 2D flow solution. Near the actuator centerline,
however, the features are approximately uniform in the spanwise direction.
The difference in computational cost for 3D simulations compared to 2D simulations is
substantial. The 2D simulation required approximately 6 hours to run on 8 parallel 3.0 GHz
Intel Xeon Gold 6154 cores (48 CPU hours) using 0.4 GB of memory. By comparison, the
nominal 3D simulation required 42 hours to run on 20 cores (840 CPU hours) using 10.4

52
 (a) 2D solution

(b) 3D solution, slice evaluated at actuator centerline

(c) 3D-wing solution, slice evaluated at actuator centerline

Figure 2.18: Time-averaged flow fields obtained from 2D and 3D simulations of steady PS
normal blowing (𝐶 𝜇 = 0.03); streamlines are plotted over contours of normalized spanwise
vorticity

GB of memory. The 3D-wing simulation required 112 hours to run on 20 cores (2,240 CPU
hours) using 25.6 GB of memory. Clearly, the 3D simulation approach is not a feasible option
for performing helicopter rotor vibration control studies. Such simulations require a more
computationally efficient reduced-order model (ROM) for calculating unsteady aerodynamic
loads [8]. The 2D simulation approach is considered adequate for constructing such a ROM
based on CFD simulations. The nominal 3D simulation indicated that, near the center of
a spanwise array of actuators, a 3D simulation yields an approximately 2D flow solution.

53
 (a) Bottom view (b) Perspective view

Figure 2.19: Time-averaged 3D flow field due to steady PS normal blowing (𝐶 𝜇 = 0.03),
depicted by isosurfaces of Q-criterion colored by normalized spanwise vorticity; freestream
flow direction is along the positive 𝑥 axis

Moreover, the 3D-wing simulation illustrated that 3D flow effects are manifested at the
actuator edges and reduced the effectiveness of actuation, compared to 2D predictions. These
observations provide some indication of how the results based on the 2D flow assumption
will differ from the 3D conditions present on an actual rotor.

2.3.2 Simulations at Higher Mach Numbers

The effect of the freestream Mach number on flow control performance is examined. The
typical actuator location for control of helicopter rotor vibrations using trailing edge flaps or
microflaps is near 75% blade span [8]. For a four-blade rotor resembling the Messerschmitt-
Bölkow-Blohm (MBB) Bo-105, the average Mach number in this region of the blade is
approximately 0.45 and the effective angle of attack is typically in the range of 0◦ < 𝛼 < 10◦ .
Thus, to assess the potential of AFC actuation on a rotor blade, simulations are performed
on a static airfoil at various Mach numbers, up to 𝑀∞ = 0.5. The Reynolds number
is also increased with increasing Mach number, such that 𝑀∞ = 0.06 corresponds to
𝑅𝑒 𝑐 = 5.35 × 105 and 𝑀∞ = 0.5 corresponds to 𝑅𝑒 𝑐 = 4.46 × 106 .

54
 0.4 0.02

0.3 0

Moment Coefficient
Lift Coefficient
0.2
-0.02
0.1

0 -0.04
Baseline (off) Baseline (off)
-0.1 PS (mean) PS (mean)
-0.06
PS ( rms) PS ( rms)
-0.2
SS (mean) SS (mean)
-0.3 SS ( rms) -0.08 SS ( rms)

0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6
Mach Number Mach Number

(a) (b)

0.02
Drag Coefficient

0.015

0.01
Baseline (off)
PS (mean)
0.005 PS ( rms)
SS (mean)
SS ( rms)
0
0 0.1 0.2 0.3 0.4 0.5 0.6
Mach Number

(c)

Figure 2.20: Steady state aerodynamic response due to PS and SS normal blowing (𝐶 𝜇 =
0.005) at various freestream Mach numbers

The steady state values of the lift, moment, and drag coefficients obtained from simula-
tions of PS and SS normal blowing on an airfoil at 𝛼 = 0◦ are shown in Fig. 2.20. For each
simulation, the jet velocity is scaled with the freestream flow velocity to achieve the same
relative jet strength, 𝐶 𝜇 = 0.005. The variation of the baseline aerodynamic coefficients with
both jets off is also shown for reference. Additionally, the RMS values are shown above and
below the mean values due to PS and SS blowing.
Although the same relative jet strength was used at each flow condition, there are
noticeable differences between the aerodynamic loads computed at the low speed wind
tunnel conditions (𝑀∞ = 0.06) and the higher speeds. From 𝑀∞ = 0.06 to 𝑀∞ = 0.5, the

55
baseline lift without actuation increases by 18%, the baseline drag decreases by 30%, and
the baseline pitching moment remains approximately the same. These trends are attributed
to the higher Reynolds numbers present at the higher Mach number simulations, causing a
reduction in the thickness of the viscous boundary layers. When actuation is applied, the
changes in lift relative to the baseline also increase: 38% for PS blowing and 34% for SS
blowing. The changes in pitching moment also increase significantly: 49% for PS blowing
and 48% for SS blowing. These imply an enhanced control authority at the higher Mach
number/Reynolds number conditions. However, with the improved control authority there is
also an associated increase in the drag penalty. The drag penalty due to actuation increases
by nearly the same margin as the lift: 39% for PS blowing and 33% for SS blowing.
The ability to verify the accuracy of these computational results is limited, since the
present wind tunnel experiments were only performed at low-speed conditions (𝑀∞ = 0.06).
CFD simulations involving combustion-powered AFC, a different form of pulsed actuation
located near the blade leading edge, were extensively validated against high-speed (up
to 𝑀∞ = 0.4) wind tunnel experiments in Ref. [110]. The CFD simulation approach
employed in Ref. [110] resembles the current simulation approach, where the flow control
jets are modeled in 2D URANS simulations using a pressure boundary condition within
the actuators. The simulations displayed a reasonable degree of accuracy for several cases,
including separation control on a static airfoil at high angles of attack as well as dynamic
stall control on a pitching airfoil. Although the mode of actuation differs from the current
approach, the results from Ref. [110] establish some confidence in the ability of 2D URANS
simulations to predict the influence of AFC at high Mach numbers relevant for rotorcraft
applications.

56
2.4 Flow Control Performance on a Moving Airfoil

During normal operating conditions, a helicopter rotor blade is subject to dynamic conditions.
Therefore is is important to examine the effect of the AFC actuator under dynamic conditions.
The flow conditions encountered on a rotor blade are complex and cannot be fully captured
in a 2D simulation. Therefore, simplifying assumptions are introduced so as to examine
the effectiveness of the flow control device in a 2D setting. The first case considered is a
simple sinusoidal pitching motion. The second case is more complicated, and consists of a
pitching/plunging motion extracted from an aeroelastic rotor simulation in AVINOR [33, 34]
coupled with a sinsoidally varying freestream Mach number. The conditions simulated are
representative of a blade section at 75% blade span. This is the intended location of a flow
control device for vibration reduction.

2.4.1 Sinusoidal Pitching Motion

The ability of AFC to modify the aerodynamic loads on a pitching airfoil is considered. At
75% blade span on a rotor, the typical range of angle of attack is 0◦ < 𝛼 < 10◦ . Hence, the
pitching motion for the simulation is prescribed as

𝛼(𝑡) = 5◦ + 5◦ sin (𝜔𝛼 𝑡) . (2.3)

For convenience, the freestream Mach number for the simulation is selected as 𝑀∞ = 0.06
for comparison with wind tunnel experiments planned for the future. The pitching frequency,
𝜔𝛼 , is based on the 1/rev rotational frequency of a rotor resembling the MBB Bo-105 rotor:
𝜔𝛼 = 0.08/𝑇conv . Alternating PS/SS normal blowing is applied at the 4/rev frequency
(0.32/𝑇conv ) with a jet strength of 𝐶 𝜇 = 0.005. The typical range of actuation frequencies
used for control of 4/rev helicopter vibrations is between 2/rev and 5/rev [8].
The unsteady lift, moment, and drag coefficients due to the combination of the airfoil
motion and the alternating PS/SS blowing is shown in Fig. 2.21. Also shown is the baseline

57
 12
Unsteady (off)
10 Quasi-Steady (off)
1.5 Unsteady (PS/SS)

Angle of Attack (°)

Lift Coefficient
8 Quasi-Steady (PS on)
SS on
Quasi-Steady (SS on)
6

Input
1
4
PS on
2 0.5

-2 0
0 20 40 60 80 0 20 40 60 80
Non-Dimensional Time Non-Dimensional Time

(a) (b)

0.05
Unsteady (off) Unsteady (off)
Quasi-Steady (off) Quasi-Steady (off)
Moment Coefficient

0.06 Unsteady (PS/SS) 0.04 Unsteady (PS/SS)

Drag Coefficient
Quasi-Steady (PS on) Quasi-Steady (PS on)
Quasi-Steady (SS on) Quasi-Steady (SS on)
0.03 0.03

0 0.02

-0.03
0.01
-0.06
0
0 20 40 60 80 0 20 40 60 80
Non-Dimensional Time Non-Dimensional Time

(c) (d)

Figure 2.21: Unsteady aerodynamic response due to sinusoidal pitching motion with 4/rev
alternating PS/SS normal blowing (𝐶 𝜇 = 0.005); time normalized as 𝑡¯ = 𝑡/𝑇conv

response with both jets off, as well as the quasi-steady responses that would be obtained
by constant PS or SS blowing on a static airfoil. The quasi-steady responses were gener-
ated using a table look-up procedure based on static airfoil data for the VR-12 with and
without steady blowing at various angles of attack. These data provide an indication of
the baseline level of control authority in the absence of any unsteady effects. In each case,
two consecutive cycles of the pitching motion were simulated to allow initial transients to
subside.
The unsteadiness associated with the pitching motion has a noticeable effect on the
predicted aerodynamic loads. First, the baseline loads with both jets off are compared. There

58
are significant phase differences between the unsteady and quasi-steady predictions of 𝐶𝑚
and 𝐶𝑑 , although the phase difference for 𝐶𝑙 is only minor. Additionally, the amplitude of
the unsteady drag is significantly larger than the quasi-steady prediction. Next, the loads
due to PS/SS blowing are compared. During the first half of the pitch cycle (0 < 𝑡¯ < 40
and 𝛼 > 5◦ ), the unsteady lift is shifted downward compared to the quasi-steady values.
There is also an upward shift in the unsteady pitching moment. Unlike the quasi-steady
drag predictions, the unsteady prediction indicates that PS and SS blowing have opposite
effects on the drag. Whereas PS blowing increases the drag penalty, SS blowing reduces
the drag below the baseline. During the second half of the pitch cycle (40 < 𝑡¯ < 80 and
𝛼 < 5◦ ), the unsteady loads induced by actuation resemble the quasi-steady predictions.
This simulation is a preliminary indication that the flow control device remains effective in
an unsteady environment, but that quasi-steady aerodynamic modeling is inadequate for the
accurate prediction of the changes in aerodynamic loads. This is an important consideration
when implementing the flow control model in a comprehensive aeroelastic rotor simulation
for vibration control.

2.4.2 Rotor Blade Motion in an Unsteady Freestream

The unsteady aerodynamic environment associated with helicopter rotor vibrations is further
complicated by periodically varying approaching flow velocity as well as coupled flap, lag,
and torsional deflections of the blades. A representative case is examined using the blade
response extracted from an aeroelastic simulation of a rotor in forward flight. The AVINOR
code [33, 34] was used to obtain the blade motion and local flow velocity at 75% blade
span for a four-bladed rotor resembling the MBB Bo-105 operating at an advance ratio
of 0.3. This information was used to determine the time-varying angle of attack, vertical
displacement, and local Mach number, which were then implemented on the VR-12 airfoil
section using a 2D CFD simulation.
First, the airfoil motion was simulated without flow control actuation, to establish the

59
baseline unsteady aerodynamic response. Next, the airfoil motion was simulated with
alternating PS/SS normal blowing, to illustrate the level of control authority and drag penalty
when compared to a static airfoil. The absolute strength of the jets is fixed in the simulation
and corresponds to 𝐶 𝜇 = 0.005 at 𝑀∞ = 0.5. Since the freestream Mach number varies
during the simulation, the relative strength of the jets (𝐶 𝜇 ) also varies. Similar to the
simulation of the simple pitching airfoil, two consecutive cycles of the blade motion were
simulated to allow the initial transient behavior to subside.
Figures 2.22a-c show the variation of the angle of attack, vertical displacement, and
Mach number as a function of the blade azimuth angle. Azimuth angles of 0◦ < 𝜓 < 180◦
represent the advancing blade and angles of 180◦ < 𝜓 < 360◦ represent the retreating blade.
The vertical displacement is normalized by the span of the rotor blade. Figures 2.22d-f show
the unsteady and quasi-steady predictions of the lift, moment, and drag coefficients due to
the combination of the blade motion and 4/rev alternating PS/SS actuation. The quasi-steady
approximations were generated in a similar manner. That is, a table look-up procedure based
on static airfoil data for various angles of attack and Mach numbers was employed.
The unsteady simulations display a level of control authority similar to the quasi-
steady approximations, although there are noticeable differences in the predictions due
to unsteadiness. On the advancing blade, the unsteady response closely resembles the
quasi-steady response. A minor difference is that the unsteady lift is shifted upward near
𝜓 = 90◦ compared to the quasi-steady lift. On the retreating blade, the differences between
the unsteady and quasi-steady predictions increase. These differences are present with
and without actuation, and are mainly evident in the moment and drag responses between
180◦ < 𝜓 < 300◦ . Additionally, the magnitude of the oscillations in 𝐶𝑙 , 𝐶𝑚 , and 𝐶𝑑 induced
by PS/SS blowing increases on the retreating blade, due to the lower freestream speed and
higher relative jet strength. These simulations provide further evidence of the importance of
unsteady effects in the prediction of aerodynamic loads in a rotary wing environment.

60
 0.005
10

Non-Dim. Vertical Disp.

Angle of Attack (°)
8 0
SS on

Input
6

-0.005
4
PS on

2
-0.01
0
0 90 180 270 360 0 90 180 270 360
Azimuth Angle (°) Azimuth Angle (°)

(a) (b)

0.7
Unsteady (off)
Quasi-steady (off)
0.6 Unsteady (on)
1.5

Lift Coefficient
Mach Number

Quasi-Steady (PS on)

0.5 Quasi-Steady (SS on)

1
0.4

0.3 0.5

0.2
0 90 180 270 360 0 90 180 270 360
Azimuth Angle (°) Azimuth Angle (°)

(c) (d)

0.15 0.08
Unsteady (off)
Quasi-steady (off) 0.06
0.1
Moment Coefficient

Unsteady (on)
Drag Coefficient

0.04
Quasi-Steady (PS on)
0.05 Quasi-Steady (SS on) 0.02

0
0
-0.02 Unsteady (off)
Quasi-steady (off)
-0.05
-0.04 Unsteady (on)
Quasi-Steady (PS on)
-0.1 -0.06
Quasi-Steady (SS on)

0 90 180 270 360 0 90 180 270 360

Azimuth Angle (°) Azimuth Angle (°)

(e) (f)

Figure 2.22: Unsteady aerodynamic response due to coupled pitching/plunging motion in an

unsteady freestream representing helicopter blade motion at 75% span with 4/rev alternating
PS/SS normal blowing

61
 CHAPTER 3

Reduced Order Modeling of Unsteady

Aerodynamic Loads Induced by Flow Control

The intended application of the flow control actuators is the modification of unsteady aero-
dynamic loads acting on a helicopter rotor so as to reduce vibratory loads. One of the
main objectives of this study is to incorporate the effects of flow control actuation into an
aeroelastic simulation framework, such as AVINOR [33]. However, the prohibitive compu-
tational cost of CFD simulations is a barrier when considering active control. Therefore,
a ROM developed in this chapter is required. The purpose of the ROM is to compute the
unsteady changes in 2D sectional aerodynamic loads due to AFC at significantly lower
computational cost compared to direct CFD simulations. The ROM is subsequently linked
to the comprehensive unsteady aerodynamic model present in the AVINOR code, which
accounts for complex phenomena such as dynamic stall and 3D flow effects. A complete
description of the computation of the unsteady, compressible aerodynamic loading on the
rotor, including dynamic stall, is presented in Chapter 4 of this dissertation. This mod-
eling approach represents a simplification of the aerodynamic interactions between the
fluidic actuators and the rotor. Nonetheless, it is consistent with previous comprehensive
simulations in AVINOR where microflaps were employed for vibration reduction [7, 31].
The ROM is constructed using training data obtained from a limited number of full-order
CFD simulations representing the broad range of flow conditions encountered on a rotor
in forward flight. Additionally, the ROM is formulated such that the resulting ordinary

62
differential equations are compatible with the existing simulation framework in AVINOR.
The development of the ROM is described next.

3.1 Training Data Generation

The selection and generation of unsteady aerodynamic training data have to be considered
carefully when constructing a ROM based on CFD simulations. This is often formulated as
a system identification problem, where specially designed control input signals are applied
to extract information about the system response. References [34–36] simulated simple
harmonic motions at various reduced frequencies for a large number of training cases
(several hundred) in CFD so as to represent the wide range of flow conditions encountered
on a rotor. Alternatively, broadband excitation signals can be used to represent a range of
reduced frequencies in a relatively short time span, thus reducing the offline computational
cost of generating training data. Examples of broadband signals that have been applied to
CFD simulations include frequency sweeps [37, 111], multi-sine signals [111, 112], filtered
white Gaussian noise [113], and binary multi-step signals [114]. Pseudorandom binary
signals are particularly useful for situations in which an input variable can only be toggled
on and off, such as pulsed-jet flow control [115]. In this section, the properties of the training
dataset needed to construct a sufficiently accurate ROM based on CFD simulations of AFC
are described.

3.1.1 Freestream Conditions

Helicopter rotor blades operate in a complex, unsteady environment that encompasses a

wide range of Mach numbers and angles of attack. To determine the range of flow conditions
needed for ROM construction, aeroelastic simulations of a rotor in forward flight were first
performed in AVINOR, without actuation. A detailed description of the code is provided in
Chapter 4, but it is also described briefly here. The AVINOR code simulates the coupled

63
 Aero loads
neglected 0.75R
Blade
root
0.20R 0.12R 1.00R
Aerodynamic stations Actuator
location

Figure 3.1: AFC-integrated rotor blade schematic for aeroelastic simulations

system of equations representing the free flight of a helicopter with flexible rotor blades.
It includes a comprehensive unsteady aerodynamic model, which employs a CFD-based
rational function approximation (RFA) for 2D calculations in attached flow, the ONERA
dynamic stall model for 2D calculations in separated flow, and a free-wake model for 3D
inflow calculations. For a given steady flight condition, the code evaluates the rotor trim state
using an iterative propulsive trim procedure. The code is also capable of modeling partial-
span control surfaces on the blades, such as trailing-edge flaps and microflaps. Furthermore,
a higher harmonic control algorithm is employed within the code when the objective is to
actively reduce vibratory loads or acoustic noise.
A four-bladed hingeless rotor resembling the MBB Bo-105 rotor is considered. The
location of the flow control actuators is based on previous experience using trailing-edge
flaps and microflaps for vibration reduction (e.g., Refs. [7, 23]). The actuators span 12%
of the blade and are centered at the 75% span location, shown in Fig. 3.1. Although the
location and length of the actuated blade segment have not been optimized, the selected
configuration is convenient for comparing the vibration reduction potential of AFC with
microflaps. Figure 3.1 also shows the locations of the 19 blade stations where aerodynamic
loads are calculated in AVINOR. Note that aerodynamic loads are neglected in the inner
blade region, from the root up to 20% span.
Active vibration control in forward flight is usually intended for normal cruise conditions
in steady level flight (0.2 ≤ 𝜇 ≤ 0.3). Figure 3.2 shows the combinations of 𝑀∞ and
angles of attack evaluated at the 19 aerodynamic stations along the blade, obtained from

64
 15
AVINOR data (all stations)

Local Angle of Attack (°)

AVINOR data (0.6 x 0.9)

10

0 0.2 0.4 0.6 0.8 1

Local Mach Number

Figure 3.2: Local Mach number and angle of attack calculated at aerodynamic stations along
the rotor blades, obtained from forward flight simulations in AVINOR at 𝜇 = 0.2, 0.3 for
𝐶𝑊 = 0.005

simulations at 𝜇 = 0.2 and 𝜇 = 0.3 when the rotor is trimmed at 𝐶𝑊 = 0.005. The region of
the blade between 60 to 90% span is highlighted to show the relatively narrow range of flow
conditions encountered in the vicinity of the actuators. Figure 3.2 illustrates the fact that
not all combinations of 𝑀∞ and 𝛼 are necessary for an accurate representation of the rotor
environment. On the advancing blade, high Mach numbers are typically combined with low
angles of attack. On the retreating blade, low Mach numbers are typically combined with
moderate angles of attack.
The ranges of 𝑀∞ and 𝛼 shown in Fig. 3.2 imply that the ROM dataset does not need
to include every possible combination of Mach number and angle of attack. Instead, a
significant computational expense can be avoided by generating training data in the narrow
region representing the intended actuator location. Figure 3.3 shows the selected training data
points. The points were generated by first expanding the boundary denoting 0.6 ≤ 𝑥 ≤ 0.9
by ±2◦ angle of attack. This was done to ensure the ROM is sufficiently robust to small
variations in 𝛼. Next, the region was filled with uniformly spaced training points using a
simple geometric approach. This approach involved generating a uniform grid of points
along the two axes, 𝑀∞ and 𝛼. Then, all grid points outside the training data boundary
shown in Fig. 3.3 were removed. This produced 34 distinct combinations of 𝑀∞ and 𝛼 in

65
 15 AVINOR data (all stations)
AVINOR data (0.6 x 0.9)

Local Angle of Attack (°)

ROM training data range
ROM training points
10

0 0.2 0.4 0.6 0.8 1

Local Mach Number

Figure 3.3: Expanded training data range (AVINOR data ±2◦ ) filled with 34 evenly-spaced
ROM training points

the ranges of 0.2 ≤ 𝑀∞ ≤ 0.7 and −1.2◦ ≤ 𝛼 ≤ 10.2◦ .

Other methods for selecting training points include random Latin hypercube and optimal
Latin hypercube sampling (e.g., Ref. [36]). For a prescribed number of sample points,
these approaches produce a random distribution of points in a multi-dimensional space. A
drawback of the random Latin hypercube approach is that it can produce clusters of points
in some regions of the training space while sparsely filling other regions. Optimal Latin
hypercube sampling prevents this by maximizing a criterion for the minimum distance
between the sample points [116]. This ensures an approximately uniform sampling of
the training space. The main disadvantage of the random and optimal Latin hypercube
approaches is that they can only be used to fill rectangular regions in space. It is difficult
to incorporate complex boundaries, like the boundaries illustrated in Fig. 3.3, into these
methods. Therefore, for the present study, the sample points were generated using the simple
geometric sampling approach described.

3.1.2 Unsteady Aerodynamic Training Data

For each training point shown in Fig. 3.3, a full-order CFD simulation is performed to
generate unsteady lift, moment, and drag coefficient data due to AFC actuation at various

66
frequencies. The primary modes of flow control considered are the PS and SS normal
blowing modes. To simplify the current analysis, the tangential blowing modes are not
included. Whereas the normal blowing modes affect the lift and moment in a similar manner
as a microflap, the tangential blowing modes primarily affect the drag, due to the thrust
produced by the jets. Various combinations of normal and tangential blowing were examined
experimentally in Ref. [105]. By combining normal blowing on one side of the airfoil with
tangential blowing on the opposite side, the drag penalty induced by normal blowing can be
reduced. However, tangential blowing also reduces the effectiveness of normal blowing. For
rotor blade vibration reduction, the primary focus is control of the lift and moment rather
than the drag.
Active control studies have indicated that employing a combination of actuation fre-
quencies in the 2/rev to 5/rev range (e.g., using higher-harmonic control) is an effective
strategy for reducing the 4/rev vibratory hub loads in a helicopter rotor [8]. Hence, the ROM
used to simulate AFC actuation must be applicable for this range of operating frequencies.
Broadband excitation signals are commonly used in system identification problems since
these types of signals excite a range of frequencies over a relatively short measurement
period. Obtaining the system response at individual frequencies is a straightforward ap-
proach for generating spectral information, but it can be time-consuming [117]. Thus,
broadband signals are useful for reducing the computation time (and total cost) associated
with generating model training data.
The discrete nature of the flow control jets presents a unique problem for system
identification. Because each jet (PS, SS) is either on or off, and the strength of each jet is
fixed, discrete-valued input signals must be considered. Pseudorandom multilevel (PRML)
signals are discrete, periodic signals that have a prescribed frequency spectrum associated
with them. A PRML signal typically resembles a random sequence of switches in the input
level of the signal. However, this type of signal is always generated using a deterministic
algorithm, and it is thus repeatable. The most easily implemented types of PRML signals

67
 10
Clock Period

Power Spectral Density (dB)

SS on 0

-10

Off -20

-30

PS on -40 Clock Frequency

-50
0 5 10 15 20 25 10 -2 10 -1 10 0 10 1 10 2
Non-Dimensional Time Non-Dimensional Frequency

(a) Time domain (b) Frequency domain

Figure 3.4: Time and frequency domain representations of a pseudorandom ternary signal;
time normalized by clock period 𝑇𝑐 and frequency normalized by clock frequency 𝑓𝑐

are those based on maximum-length sequences [118], which can be generated using linear
feedback shift registers which are described in Appendix A.
A pseudorandom ternary (three-level) signal is used to represent three possible states
of actuation: 1) SS normal jet on, 2) PS normal jet on, and 3) both jets off. The time and
frequency domain representations of a typical ternary signal are shown in Fig. 3.4. The clock
frequency 𝑓𝑐 is used to scale the frequency spectrum to the desired frequency range for model
identification. This is related to the clock period 𝑇𝑐 = 1/ 𝑓𝑐 , which is the minimum time
interval over which the discrete signal must maintain a constant value before switching from
one value to another. This terminology originates from the field of electrical engineering
because digital circuits operate in discrete time according to a prescribed frequency, i.e.,
the “clock.” A rule of thumb in using multilevel signals for identification is to set the clock
frequency to 𝑓𝑐 = 𝑓max /0.4, where 𝑓max is the maximum frequency of interest [117, 118].
The power spectrum is relatively flat for frequencies below 0.4 𝑓𝑐 , as shown in Fig. 3.4b;
and thus most of the signal power is concentrated in this range.
In this study, the clock frequency is selected such that 0.4 𝑓𝑐 = 5/rev, since the frequency
range of interest for control is 2/rev to 5/rev. This is equivalent to 𝑓𝑐 = 12.5/rev. The signal
period 𝑇 is selected such that the frequency resolution of the signal is 𝑓0 = 1/𝑇 = 1/rev.

68
That is, the signal period corresponds to one rotor revolution.
The strength of the actuators is selected such that the jet exit velocity 𝑈jet is constant
in all simulations. Because each training case corresponds to a different freestream Mach
number, this implies that the relative jet strength 𝐶 𝜇 is different for each simulation. The
reference condition used to designate the actuator strength is 𝐶 𝜇 = 0.005 at 𝑀∞ = 0.5. This
corresponds to 𝑀jet ≈ 0.75. The actuators are stronger (in a relative sense) when 𝑀∞ < 0.5,
as implied by Eq. (2.1). The actuators are weaker when 𝑀∞ > 0.5.
The pseudorandom signal is applied to the PS/SS jets for each of the 34 training cases that
vary in terms of Mach number and angle of attack. Although the switching intervals between
PS and SS blowing are identical for each case, the time-step size in each CFD simulation is
based on the freestream velocity: Δ𝑡 = 0.004𝑐/𝑈∞ . Therefore, the higher-Mach-number
simulations require more time steps than the lower-Mach-number simulations, and are thus
computationally more expensive. Each simulation was performed using eight 3.0 GHz Intel
Xeon Gold 6154 cores on the University of Michigan high-performance computing cluster:
Great Lakes. The combined run time for all simulations was approximately 16,000 CPU
hours.

3.2 ROM Formulation

The ROM is formulated by treating the CFD simulation framework as a black box and
applying a system identification procedure. From a high-level perspective, the black box
maps the flow control input variables ufc (𝑡) to the unsteady changes in aerodynamic loads
Δffc (𝑡), where
 
Δ𝐶𝑙fc 

 
 
 𝑢 SS 

  
 

 
 
 

ufc = and Δffc = Δ𝐶𝑚fc (3.1)
 𝑢 PS 
  





 Δ𝐶𝑑fc 
  
 

 

69
The operating states of the jets are given by 𝑢 SS and 𝑢 PS . For each jet, a value of 0 is
used to designate the jet-off condition and a value of 1 is used to designate the jet-on
condition at full jet strength. As mentioned in the previous section, full jet strength is
defined as 𝐶 𝜇 = 0.005 at a freestream Mach number of 𝑀 = 0.5. Positive values of 𝑢 SS and
𝑢 PS between 0 and 1 are used to designate intermediate jet strengths. In the wind tunnel
experiments [105], the actuators were designed to switch each jet on and off at a fixed jet
strength. For consistency with the experiments, the discrete operation of the actuators is also
simulated when using the ROM. However, the ROM is not strictly limited to discrete-valued
input signals. The ROM can also simulate continuous-valued input signals, representing the
case of variable-jet-strength actuation.
An approximate mapping from inputs to outputs is sought in the Laplace domain:

L [𝑈 ( 𝑡¯)Δffc ( 𝑡¯)] = Q( 𝑠¯)L [ufc ( 𝑡¯)] (3.2)

∫
1 𝑡
where reduced time 𝑡¯ = 𝑏 0 𝑈 (𝜏)d𝜏, interpreted as the distance traveled by a rotor blade
measured in semi-chords, is used in order to properly account for unsteady freestream effects
[119]. For rotary-wing applications, this definition of reduced time is more appropriate than
the traditional definition used in fixed-wing applications; 𝑡¯ = (𝑈𝑡/𝑏).
The aerodynamic transfer function Q( 𝑠¯) is a 3 × 2 matrix:

 SS
𝑄 ( 𝑠¯) 𝑄 PS ( 𝑠¯) 

 𝑙 𝑙 
 
Q( 𝑠¯) = 𝑄 SS
𝑚 ( ¯
𝑠 ) 𝑄 PS ( 𝑠¯) 
𝑚  (3.3)
 
 SS PS
𝑄 𝑑 ( 𝑠¯) 𝑄 𝑑 ( 𝑠¯) 

 

Each entry of the transfer matrix is approximated using a process model of the form

˜ 𝑠¯) = 𝐾
𝑄( (3.4)
1 + 𝜏 𝑠¯

The static gain 𝐾 is interpreted in the time domain as the steady-state value of the output

70
due to sustained steady blowing. The time constant 𝜏 is associated with the rise time.
The process model form can be transformed to the time domain using the inverse Laplace
transform, which yields the state-space model

 
dxfc 𝑈 (𝑡)
= A xfc (𝑡) + B ufc (𝑡) (3.5)
d𝑡 𝑏
1
Δffc (𝑡) = C xfc (𝑡) (3.6)
𝑈 (𝑡)

where

 
A = diag −1 −1 −1 −1 −1 −1 (3.7)
𝜏𝑙SS
, 𝜏𝑙PS
, SS ,
𝜏𝑚 PS ,
𝜏𝑚 𝜏𝑑SS
, 𝜏𝑑PS
 
I2 
 
 
B = I2  (3.8)
 
I2 
 
 
 𝐾𝑙SS 𝐾𝑙PS 
 SS
𝜏𝑙PS
0 0 0 0 
 𝜏𝑙 
 SS
𝐾𝑚 PS
𝐾𝑚

C =  0 0 SS PS 0 0  (3.9)
𝜏𝑚 𝜏𝑚
𝐾 𝑑SS 𝐾 𝑑PS 
 
 0 0 0 0

𝜏𝑑SS 𝜏𝑑PS 



d 𝑏 d
and I2 denotes a 2 × 2 identity matrix. Note that the relation d𝑡¯ = 𝑈 (𝑡) d𝑡 can be used to
switch between the independent variables 𝑡 and 𝑡¯. The aerodynamic loads Δffc (𝑡) are given
by Eq. (3.6), and are a function of the aerodynamic states associated with flow control xfc (𝑡).
These states are governed by the set of first order differential equations given in Eq. (3.5).
The static gains and time constants are estimated from the CFD data. Consider one
training case at 𝑀 = 0.7 and 𝛼 = 0.6◦ . In Fig. 3.5, the unsteady aerodynamic response due
to the pseudorandom training signal is plotted in terms of the changes in lift, moment, and
drag coefficients multiplied by the freestream velocity. The MATLAB System Identification
Toolbox is employed in order to estimate the model parameters. More specifically, the
procest function in MATLAB is used to estimate the multi-input multi-output transfer

71
 CFD ROM
50

-50
0 100 200 300 400 500
10

-10
0 100 200 300 400 500
4
2
0
-2
0 100 200 300 400 500

SS on
Input

Off
PS on
0 100 200 300 400 500
Reduced Time

Figure 3.5: Process model estimation from CFD simulation of pseudorandom ternary signal,
𝑀 = 0.7, 𝛼 = 0.6◦

function Q̃( 𝑠¯) from the time-history data. As shown in Fig. 3.5, the estimated model
reproduces the CFD response accurately. This procedure is repeated to obtain a process
model, given by Eqs. (3.5) and (3.6), for each of the 34 training cases. Furthermore, a family
of process models representing the broad range of Mach numbers and angles of attack is
obtained.

3.3 Model Scheduling and Interpolation

To account for intermediate flow conditions that occur between the training cases, a model
scheduling approach resembling the CFD-based RFA [34] is used. In this formulation,
Eqs. (3.5) and (3.6) are modified by allowing the coefficient matrices, A and C, to vary with

72
𝑀 and 𝛼. The resulting state-space model in the time domain can be written as

 
dxfc 𝑈 (𝑡)
= A(𝑀, 𝛼) xfc (𝑡) + B ufc (𝑡) (3.10)
d𝑡 𝑏
1
Δffc (𝑡) = C(𝑀, 𝛼) xfc (𝑡) (3.11)
𝑈 (𝑡)

The model parameters are evaluated at intermediate flow conditions using Kriging
interpolation [116]. Kriging is a nonlinear approach capable of handling complex multidi-
mensional functions based on limited training datasets. It is also nonparametric, meaning
that is has no fixed structure and all training points are used in the prediction. Note that
the “training data” now refer to the model parameters 𝐾𝑙SS , 𝐾𝑙PS , etc., and 𝜏𝑙SS , 𝜏𝑙PS , etc.,
that were identified from the 34 training cases simulated in CFD. In this setting, the model
parameters are treated as unknown nonlinear functions of 𝑀 and 𝛼.
In Kriging, the training data are assumed to be taken from a statistical distribution. An
advantage of this approach is that by treating the response as a realization of a stochastic
(random) process, the uncertainty of the Kriging predictions can be estimated with respect to
the assumed distribution [120]. Kriging is based on the assumption that errors are correlated.
This is in contrast to typical least-squares regression, where errors are independent and
uncorrelated. The assumption of uncorrelated errors is valid when the sources of error are
random (e.g., measurement error or noise). However, it is more appropriate to assume that
the errors from a deterministic computer code are correlated, since there is no source of
random error.
The output 𝑦 is used to represent the dependent variable that is to be interpolated, e.g.,
𝑦 ≡ 𝐾𝑙SS , 𝜏𝑙SS , etc. It is represented as a random variable of the form

𝑦 = Φ(x) = 𝑓 (x) + 𝑍 (x) (3.12)

where the independent variables are x = [𝑀, 𝛼]. The regression function 𝑓 (x) represents a

73
global approximation of Φ(x), whereas the Gaussian process 𝑍 (x) compensates for the error
in the regression model such that Φ(x) reproduces the training data exactly. The Gaussian
2 .
process model has zero mean and variance 𝜎var
The regression model is composed of a set of basis functions 𝑓𝑖 (x) and regression
coefficients 𝛽𝑖 :
𝑁Õ
basis

𝑓 (x) = 𝑓𝑖 (x) 𝛽𝑖 = f T (x) 𝜷 (3.13)

𝑖=1

where the 𝑁basis functions are typically low-order polynomials. The basis functions can be
compactly expressed as an 𝑁sp × 𝑁basis matrix F(x), where the 𝑖th row corresponds to the
evaluation of the 𝑁basis functions at the 𝑖th sample point x (𝑖) in the training data. Previous
studies (e.g., Ref. [121]) have shown that the regression function can be replaced with a
constant value, i.e., 𝑓 (x) = 𝛽, with little loss of accuracy in the Kriging predictor. Other
typical regression functions are first- or second-order polynomials.
The regression coefficients 𝜷 are determined for the supplied set of training data
using maximum likelihood estimation. This is a statistical approach to parameter esti-
mation where the objective is to maximize the probability that the observed responses
y = [𝑦 (1) , 𝑦 (2) , . . . , 𝑦 (𝑁sp ) ] T were drawn from the stochastic process Φ(x) given the model
parameters 𝜷. This assumes that the covariance of the Gaussian process 𝑍 (x) is known. The
covariance matrix provides a measure of how strongly correlated each pair of sample points
are with each other:
2
Cov[𝑍 (x), 𝑍 (x)] = 𝜎var Rkrg (3.14)

where the 𝑁sp × 𝑁sp correlation matrix Rkrg is

 𝑅 krg (x (1) , x (1) ) . . . 𝑅 krg (x (1) , x (𝑁sp ) )

 

 
 .. ... .. 
Rkrg =  . . 
 (3.15)
 
 𝑅 krg (x (𝑁sp ) , x (1) ) . . . 𝑅 krg (x (𝑁 ) ,x (𝑁 ) )
 sp sp

 

and 𝑅krg is a user-defined spatial correlation function (SCF). The SCF controls the smooth-

74
ness and differentiability of the resulting Kriging model as well as the influence of nearby
points. There are several classes of SCFs that can be used, but the most commonly used
SCF for Kriging is the Gaussian form [120, 122]:

𝑁𝑥
Ö h i
( 𝑗)
(𝑖)
𝑅 krg (x , x ( 𝑗)
)= exp −𝜃 𝑘 |𝑥 𝑘(𝑖) − 𝑥 𝑘 |2 (3.16)
𝑘=1

The correlation function depends on the separation distance between two points. As two
( 𝑗)
points move closer together, |𝑥 𝑘(𝑖) − 𝑥 𝑘 | → 0 and Eq. (3.16) approaches a maximum
correlation of one. The fitting parameters 𝜃 𝑘 are additional unknown correlation parameters
that affect how far each sample point’s influence extends in the interpolation. A low 𝜃 𝑘
means that all sample points are highly correlated and Φ(𝑥 𝑘 ) is nearly constant across the
training data. A high 𝜃 𝑘 means that Φ(𝑥 𝑘 ) varies significantly across the training data. Any
values of 𝜃 𝑘 will produce a surrogate model that interpolates the sample points, but the “best”
Kriging surrogate is found by optimizing the values of 𝜃 𝑘 [120].
Based on the user’s selections for the regression and correlation functions, the unknown
parameters in the model are determined using maximum likelihood estimation. The unknown
2 , and
parameters are the regression coefficients 𝜷, the variance of the Gaussian process 𝜎var
the correlation parameters 𝜽 = [𝜃 1 , . . . , 𝜃 𝑁 𝑥 ]. The likelihood function for a Kriging model
is
−(y − F𝜷) T R−1
" #
1 krg (y − F𝜷)
𝐿= 2 ) 𝑁sp /2 |R | 1/2
exp 2
(3.17)
(2𝜋𝜎var krg 2𝜎var

This represents the objective function to be maximized. An equivalent problem is to

maximize the natural logarithm of Eq. (3.17),

𝑁sp 𝑁sp 1 (y − F𝜷) T R−1

krg (y − F𝜷)
2
ln(𝐿) = − ln(2𝜋) − ln(𝜎var ) − ln |Rkrg | − 2
(3.18)
2 2 2 2𝜎var

2 and 𝜷 as well as setting the deriva-

Taking the derivatives of Eq. (3.18) with respect to 𝜎var
tives equal to zero yield the maximum likelihood solutions for a fixed set of correlation

75
parameters 𝜽:

𝜷ˆ = ( FT R−1 −1 T −1
krg F ) F Rkrg y (3.19)
( y − F 𝜷ˆ ) T R−1 ˆ
krg ( y − F 𝜷 )
2
ˆ var
𝜎 = (3.20)
𝑁sp

The correlation parameters 𝜽 are assumed to be fixed in this solution, since taking derivatives
of Eq. (3.18) with respect to each 𝜃 𝑘 for 𝑘 = 1, . . . , 𝑁𝑥 yields a system of 𝑁𝑥 coupled
nonlinear equations that have no closed-form solution. Therefore, a numerical optimization
algorithm must be employed to determine 𝜽, which maximizes Eq. (3.18). Reference [116]
suggests using a global search method to maximize the likelihood function. In this study,
the MATLAB Kriging toolbox [123] is used to perform the optimization and construct the
Kriging surrogates.
After the unknown parameters have been optimized, the Kriging approximation of Φ(x)
is given as
𝑦ˆ = Φ̂(x) = f T (x) 𝜷ˆ + rkrg
T
(x) R−1 ˆ
krg ( y − F 𝜷 ) (3.21)

where
rkrg (x) = [𝑅 krg (x, x (1) ) , 𝑅 krg (x, x (2) ) , . . . , 𝑅 krg (x, x (𝑁sp ) )] T (3.22)

The column vector r(x) is the correlation vector between an arbitrary prediction point x and
the sample data points x (1) , . . . , x (𝑁sp ) that constitute the training data. The solution given
by Eq. (3.21) minimizes the mean square error of the Kriging prediction with respect to the
assumed stochastic process Φ(x).
In total, 12 Kriging models are generated from the training data: one for each parameter
in the ROM (𝐾𝑙SS , 𝐾𝑙PS , 𝐾𝑚SS , 𝐾𝑚PS , 𝐾 𝑑SS , 𝐾 𝑑PS , 𝜏𝑙SS , 𝜏𝑙PS , 𝜏𝑚SS , 𝜏𝑚PS , 𝜏𝑑SS , and 𝜏𝑑PS ). The
interpolation of the parameters 𝐾𝑙PS and 𝜏𝑙PS is illustrated in Fig. 3.6. The Kriging interpolant
passes through all training points exactly and is a smooth function at the intermediate values.
Note that the interpolant is only used for values within the training data range, and it is not

76
 Kriging Interp. Kriging Interp.
Training Data Training Data
60 10

50

40 5

30

20 0
10 10
0.8 0.8
5 0.6 5 0.6
Angle of Attack (°) 0.4 Angle of Attack (°) 0.4
0 0
0.2 Mach Number 0.2 Mach Number

(a) Static gain (b) Time constant

Figure 3.6: Kriging interpolation of ROM parameters for lift coefficient due to PS blowing,
based on Mach number and angle of attack

used to extrapolate beyond the scope of the training data.

3.4 ROM Validation

The accuracy of the ROM is compared with full-order CFD simulations representing rotor
blade vibrations. As a validation case, the coupled pitch, plunge, and freestream velocity
at 75% span on a rotor operating at 𝜇 = 0.3 is considered. The AVINOR code is used
to obtain the blade motion and local flow velocity at this location. This information is
then applied to the VR-12 airfoil section in a 2D CFD simulation to obtain the baseline
response without actuation as well as the response due to a 4/rev square wave signal of
alternating PS/SS blowing. The absolute strength of the jets is fixed and corresponds to
𝐶 𝜇 = 0.005 at 𝑀∞ = 0.5. Since the freestream Mach number varies within the range of
0.26 ≤ 𝑀∞ ≤ 0.62, the relative strength of the jets 𝐶 𝜇 also varies throughout the simulation.
The time-step size in CFD was selected as Δ𝑡 = 0.004𝑐/𝑈∞ for the largest freestream
velocity encountered during the simulation (𝑀∞ = 0.62). This is consistent with the step
size used when generating the ROM training data. Furthermore, the step size is small enough
that further reduction of Δ𝑡 yields negligible differences in the aerodynamic loads when

77
simulating the coupled blade motion. Moreover, two consecutive cycles of the blade motion
were simulated to allow the initial transient behavior to subside.
The CFD and ROM predictions of the combined response are shown in Fig. 3.7. The lift
and moment coefficient data show a good level of agreement. The ROM prediction of the
drag, shown in Fig. 3.7f, is somewhat inaccurate on the retreating side (within the range of
180◦ < 𝜓 < 300◦ ) when compared to CFD. The relative error of the ROM predictions is
calculated as
1 Í𝑁
𝑁 𝑛=1 | 𝑦ˆ (𝑡 𝑛 ) − 𝑦(𝑡 𝑛 )|
𝜖 ( 𝑦ˆ , 𝑦) = 100 × (3.23)
𝑦 max − 𝑦 min

This metric is the mean absolute error of 𝑦ˆ (𝑡) relative to 𝑦(𝑡), where 𝑦(𝑡) represents the
CFD response and 𝑦ˆ (𝑡) represents the approximate response of the ROM. This yields 3.3%
error in the prediction of 𝐶𝑙 , 5.4% error in 𝐶𝑚 , and 5.4% error in 𝐶𝑑 . Overall, the ROM
produces a reasonable approximation of the full-order CFD results. The advantage of using
the ROM is the substantial reduction in computer run time. A CFD simulation requires
approximately 63 hours to run on eight high-performance Intel Xeon Gold cores. However,
the ROM requires only 5.5 seconds to run on a conventional laptop PC equipped with a
dual-core seventh-generation Intel i7 processor. This is more than four orders of magnitude
faster than CFD. This reduction in computational cost is required so as to enable active
control simulations in a comprehensive code such as AVINOR.

78
 0.7 10

Angle of Attack (°)

Mach Number 0.6 8

0.5 6

0.4 4

0.3 2

0.2 0
0 90 180 270 360 0 90 180 270 360
Azimuth Angle (°) Azimuth Angle (°)
(a) (b)

Baseline (off)
SS 1.5 CFD
ROM

Lift Coefficient
Input

1
Off

0.5
PS

0
0 90 180 270 360 0 90 180 270 360
Azimuth Angle (°) Azimuth Angle (°)
(c) (d)
0.1 0.1
Baseline (off) Baseline (off)
CFD CFD
Moment Coefficient

0.05
Drag Coefficient

ROM ROM
0.05
0

-0.05 0

-0.1
-0.05
0 90 180 270 360 0 90 180 270 360
Azimuth Angle (°) Azimuth Angle (°)
(e) (f)

Figure 3.7: Unsteady aerodynamic response to rotor blade motion with 4/rev alternating
PS/SS normal blowing, predicted using CFD and ROM

79
 CHAPTER 4

Aeroelastic Rotor Simulations Using the

AVINOR Code

Active control simulations are performed using the comprehensive rotorcraft aeroelastic
analysis code AVINOR (which stands for Active Vibration and Noise Reduction). The
code has been developed at the University of Michigan to analyze the aeroelastic response
and stability of helicopter rotors in forward flight, and it has been validated in numerous
previous studies [6, 23, 24, 34]. The code includes the capability to model actively controlled
trailing edge flaps and microflaps integrated into the rotor blades. The code also features
a higher-harmonic control (HHC) algorithm, which can be used to perform closed-loop
simulations aimed at vibratory load or acoustic noise reduction. The ROM described in
Chapter 3 is incorporated into AVINOR to model the unsteady aerodynamic effects induced
by flow control on the rotor blades. The AVINOR code and the ROM implementation are
described next.

4.1 Structural Dynamics Model

The structural dynamic model used for the present study consists of a four-bladed hingeless
rotor undergoing moderate deflections with fully coupled flap-lag-torsional dynamics. The
structural equations of motion are discretized using the global Galerkin method, based on
the free vibration modes of the rotating blade. The dynamics of the blade are represented by

80
three flap, two lead-lag, and two torsional modes. The structural dynamics model used in this
study is identical to that employed in Ref. [34], which is based on the formulation presented
in Ref. [124]. The model is described in Appendix B with the modeling assumptions,
coordinate systems, and coordinate transformations used in the aeroelastic analysis.

4.2 Aerodynamic Model

Unsteady sectional aerodynamic loads in attached flow are calculated at several stations along
the blades using a CFD-based rational function approximation (RFA) model [7]. The CFD-
based RFA approach is a 2D unsteady time-domain model that accounts for compressibility
as well as variations in the oncoming flow velocity. This 2D aerodynamic model is linked to
a 3D free-wake model, developed in Ref. [23], which produces a spanwise and azimuthally
varying inflow velocity distribution over the rotor. The spanwise lift distribution is used
in the circulation distribution calculation for the free-wake analysis, while the free-wake
influences the sectional aerodynamic loads by changing the local flow velocities. For the
separated flow regime, the 2D loads are modified using the ONERA dynamic stall model,
which was implemented in Ref. [125]. The aerodynamic states associated with the CFD-
based RFA attached flow and ONERA separated flow models are combined to produce the
time-domain, state space aerodynamic model.

4.2.1 CFD-Based RFA Model for Attached Flow

The CFD-based RFA model developed in Ref. [7] is based on the Rogers approximation for
representing aerodynamic loads in the Laplace domain:

G( 𝑠¯) = Q( 𝑠¯)H( 𝑠¯) (4.1)

81
 Figure 4.1: Normal velocity distributions corresponding to generalized airfoil motions

y
b/2 b
x
α(t) h(t)
A
U(t) B

Figure 4.2: Airfoil undergoing pitching and plunging motion

where G( 𝑠¯) and H( 𝑠¯) represent Laplace transforms of the generalized aerodynamic load
and generalized motion vectors, respectively. The aerodynamic transfer matrix Q( 𝑠¯) is
approximated using the least-squares approach with a rational expression of the form

𝑛𝐿
Õ 𝑠¯
Q̃( 𝑠¯) = C0 + C1 𝑠¯ + C𝑛+1 (4.2)
𝑛=1
𝑠¯ + 𝛾𝑛

Equation (4.2) is usually denoted as the Rogers approximation. The poles 𝛾1 , 𝛾2 , . . . , 𝛾𝑛 𝐿

are assumed to be positive valued to produce stable open-loop roots, but are noncritical for
the approximation. The arbitrary motions of the airfoil are represented by two generalized
motions shown in Fig. 4.1. The generalized airfoil motions are assigned magnitudes equal to
the normal velocity at the 3/4-chord location and can be expressed in terms of the classical
pitch and plunge modes 𝛼 and ℎ, depicted in Fig. 4.2, as

𝑊0 = 𝑈𝛼 + ℎ¤ (4.3)

𝑊1 = 𝑏 𝛼¤ (4.4)

82
 In helicopter applications, 𝛼, 𝑈, and ℎ¤ are interpreted as

𝛼 = 𝜃𝐺 + 𝜙 (4.5)

𝑈 = 𝑈𝑇 (4.6)

− ℎ¤ = 𝑈𝑃 (4.7)

where 𝜃 𝐺 is the geometric pitch angle, 𝜙 is the elastic torsional deformation of the blade,
and 𝑈𝑇 and 𝑈𝑃 are the components of the oncoming flow velocity approximately tangent
and perpendicular to the hub plane.
In order to find the least-squares approximations for the coefficient matrices C0 , C1 , . . . ,
C𝑛 𝐿 +1 in Eq. (4.2), aerodynamic frequency response data Q(𝑘 𝑚 ) corresponding to each of
the two generalized motions must be obtained. The frequency-domain solutions required
for the construction of the RFA model are obtained from the compressible URANS solver,
CFD++.
To derive a state space representation of the RFA aerodynamic model, a generalized
motion vector h and a generalized load vector f are defined as:


 

 


 𝐶 𝑙



 𝑊0 

 
 
 
 

h= and f = 𝐶𝑚 (4.8)

 𝑊1 
 
 

  
 

 𝐶𝑑 
 
 

These vectors are related to the generalized motions and the generalized forces in the Laplace
transform representation, given in Eq. (4.1), as follows:

   
G( 𝑠¯) = L f( 𝑡¯)𝑈 ( 𝑡¯) and H( 𝑠¯) = L h( 𝑡¯) (4.9)

83
where reduced time is expressed as

∫ 𝑡
1
𝑡¯ = 𝑈 (𝜏)d𝜏 (4.10)
𝑏 0

in order to properly account for unsteady freestream effects [119]. The rational function
Q̃ in Eq. (4.2) can be transformed to the time domain using the inverse Laplace transform,
which yields the state space model:

dx 𝑈 (𝑡) dh
= Rx(𝑡) + E (4.11)
d𝑡 𝑏  d𝑡 
1 𝑏 dh
f(𝑡) = C0 h(𝑡) + C1 + Dx(𝑡) (4.12)
𝑈 (𝑡) 𝑈 (𝑡) d𝑡

where

h i
D = I3 I3 . . . I3 (4.13)
 
𝛾1 I3 
 
 
 𝛾 2 I3

R=− (4.14)
 

 ... 
 
 
 
 𝛾 𝑛 𝐿 I3 
 
 
 C2 
 
 
 C3 
E=  .  (4.15)
 
 .. 
 
 
 
C𝑛 𝐿 +1 
 

and I3 is a 3 × 3 identity matrix.

In order to account for flow nonlinearities, the coefficient matrices R, E, C0 , and C1 are

84
allowed to vary with Mach number and angle of attack:

dx 𝑈 (𝑡) dh
= R(𝑀, 𝛼)x(𝑡) + E(𝑀, 𝛼) (4.16)
d𝑡 𝑏  d𝑡 
1 𝑏 dh
f(𝑡) = C0 (𝑀, 𝛼)h(𝑡) + C1 (𝑀, 𝛼) + Dx(𝑡) (4.17)
𝑈 (𝑡) 𝑈 (𝑡) d𝑡

In this approach, different sets of RFA coefficients are generated at appropriate combinations
of 𝑀 and 𝛼. A two-dimensional shape-preserving piecewise cubic Hermite polynomial
interpolation scheme is then used to evaluate the coefficients at intermediate flow conditions.

4.2.2 ONERA Dynamic Stall Model for Separated Flow

Dynamic stall effects due to flow separation are modeled using a semi-empirical dynamic
stall model based on a modified version of the ONERA dynamic stall model [125]. The
modified aerodynamic state vector for each blade section consists of the CFD-based RFA
attached flow states and the ONERA separated flow states. Dynamic stall is an important
contributor to vibration levels and control loads for the high advance ratios (𝜇 ≥ 0.30)
considered in this study.
In the ONERA model developed by Petot [126], the three second-order differential
equations governing the separated flow states are:

 2 "   #
2
𝑈 𝑈 𝑈 𝑈
Γ¥ 𝑗 + 𝑎 𝑗 Γ¤ 𝑗 + 𝑟 𝑗 Γ𝑗 = − 𝑟 𝑗 𝑉Δ𝐶 𝑗ds + 𝐸 𝑗 𝑊¤ 0 (4.18)
𝑏 𝑏 𝑏 𝑏

where 𝑗 = 𝑙, 𝑚, 𝑑 represent lift, moment, and drag, respectively. The coefficients 𝑎 𝑗 , 𝑟 𝑗 , 𝐸 𝑗

are obtained empirically. The complete two-dimensional sectional airloads are given by:

𝐿 = 𝐿 𝐴 + 𝐿𝑆 , 𝑀 = 𝑀 𝐴 + 𝑀𝑆 , 𝐷 = 𝐷 𝐴 + 𝐷𝑆 (4.19)

where 𝐿 𝐴 , 𝑀 𝐴 , and 𝐷 𝐴 are the attached-flow lift, moment, and drag, respectively, calculated
using the CFD-based RFA model. The lift, moment, and drag due to the separated flow are

85
given by:

𝐿 𝑆 = 𝜌𝑏𝑈Γ𝑙 (4.20)

𝑀𝑆 = 2𝜌𝑏 2𝑈Γ𝑚 (4.21)

𝐷 𝑆 = 𝜌𝑏𝑈Γ𝑑 (4.22)

The flow separation and reattachment criterion is based on the angle of attack and a correction
similar to Prandtl-Glauert to account for compressibility. The critical angle of attack for
separation and reattachment is 𝛼𝑐𝑟 = 15◦ (1 − 𝑀 2 ). Contributions to the three sectional
airloads from the dynamic stall model are denoted by Δ𝐶𝑙ds , Δ𝐶𝑚ds , and Δ𝐶𝑑ds . They can be
either zero:
Δ𝐶𝑙ds = Δ𝐶𝑚ds = Δ𝐶𝑑ds = 0 (4.23)

or take the following values if the flow has separated [125]:

h 2
i
Δ𝐶𝑙ds = ( 𝑝 0 − 0.1𝑀 4 )(𝛼 − 𝛼𝑐𝑟 ) − 0.7(1 − 𝑀) 𝑒 (−0.5+(1.5−𝑀)𝑀 ) (𝛼−𝛼𝑐𝑟 ) − 1 (4.24)
 2
h i
Δ𝐶𝑚ds = −0.11 − 0.19𝑒 −40(𝑀−0.6) 𝑒 (−0.4−0.21 arctan[22(0.45−𝑀)]) (𝛼−𝛼𝑐𝑟 ) − 1 (4.25)
"   25−𝛼𝑐𝑟 −𝛼 #
25 − 𝛼 18−2 arctan(4𝑀 ) 𝑐𝑟
Δ𝐶𝑑ds = (0.008 − 0.3) 1 − (4.26)
25 − 𝛼𝑐𝑟

where
1 − 𝑀8
𝑝 0 = 0.1 √ (4.27)
1 − 𝑀2
The separation criteria based on the angle of attack is given by,

1. Case 1: If 𝛼 < 𝛼𝑐𝑟 = 15◦ (1 − 𝑀 2 ), then Δ𝐶𝑙ds = Δ𝐶𝑚ds = Δ𝐶𝑑ds = 0.

2. Case 2: Assume that at time 𝑡 = 𝑡0 , 𝛼 = 𝛼𝑐𝑟 and 𝛼¤ > 0. Then, for 𝑡 > 𝑡0 + 𝜏 𝑗ds , Δ𝐶𝑙ds ,
Δ𝐶𝑚ds and Δ𝐶𝑑ds are given by Eqs. (4.24)-(4.26). The nondimensional time delay for
the lift is 𝜏𝑙ds = 8 and the time delays for the moment and drag are 𝜏𝑚ds = 𝜏𝑑ds = 2.

86
 3. Case 3: When 𝛼 < 𝛼𝑐𝑟 , the flow is reattached and Δ𝐶𝑙ds , Δ𝐶𝑚ds , and Δ𝐶𝑑ds are set to
zero again.

The parameters 𝑎 𝑗 , 𝑟 𝑗 , 𝐸 𝑗 are calculated based on empirical relations presented in Ref. [126]:

𝑎 𝑗 = 𝑎 𝑗0 + 𝑎 𝑗2 (Δ𝐶𝑙ds ) 2 (4.28)
2
𝑟 𝑗 = 𝑟 𝑗0 + 𝑟 𝑗2 (Δ𝐶𝑙ds ) 2

(4.29)

𝐸 𝑗 = 𝐸 𝑗2 (Δ𝐶𝑙ds ) 2 (4.30)

Furthermore, the ONERA model features 18 empirical coefficients, 6 each (𝑟 𝑗0 , 𝑟 𝑗2 , 𝑎 𝑗0 , 𝑎 𝑗2 ,

𝐸 𝑗2 ) associated with the lift ( 𝑗 = 𝑙), moment ( 𝑗 = 𝑚), and drag ( 𝑗 = 𝑑). These quantities
can be found in Ref. [125].

4.2.3 Free-Wake Model for 3D Wake Calculation

The free-wake model used in this study is based on the CAMRAD/JA [102] wake analysis
and was incorporated into the AVINOR code in Ref. [127]. The model was later modified
in Ref. [23] to provide a finer level of wake resolution needed for modeling BVI noise and
vibratory loads.
The wake analysis consists of two elements: (1) a wake geometry calculation pro-
cedure including a free-wake analysis developed by Scully [128], which determines the
position of the vortices, and (2) an induced velocity calculation procedure as implemented in
CAMRAD/JA, which calculates the nonuniform induced velocity distribution at the blades.

4.2.3.1 Wake Geometry

The rotor wake is composed of two main elements: the tip vortex, which is a strong,
concentrated vorticity filament generated at the tip of the blade; and the near wake, which
is an inboard sheet of trailed vorticity. The near wake is much weaker and more diffused
than the tip vortex. The wake vorticity is created in the flow field as the blade rotates, and

87
then convected with the local velocity of the fluid. The local velocity of the fluid consists
of the free stream velocity, and the wake self induced velocity. Thus, the wake geometry
calculation proceeds as follows: (1) the position of the blade generating the wake element
is calculated, this is the point at which the wake vorticity is created; (2) the undistorted
wake geometry is computed as wake elements are convected downstream from the rotor
by the free stream velocity; (3) distortion of wake due to the wake self-induced velocity is
computed and added to the undistorted geometry. The position of a generic wake element is
identified by its current azimuth position 𝜓 and its age 𝜙𝑤 . Age is the nondimensional time
that has elapsed since the wake element’s creation. Thus, the position of a generic wake
element is written as:

r𝑤 (𝜓, 𝜙𝑤 ) = r𝑏 (𝜓 − 𝜙𝑤 ) + 𝜙𝑤 V 𝐴 + D(𝜓, 𝜙𝑤 ) (4.31)

where r𝑏 (𝜓 − 𝜙𝑤 ) is the position of the blade when it generates the wake element, V 𝐴 is the
free stream velocity, and D(𝜓, 𝜙𝑤 ) is the wake distortion.
To evaluate the wake self-induced distortion D(𝜓, 𝜙𝑤 ), a free wake procedure developed
by Scully [128] is employed. This procedure is used only to calculate the distorted geometry
of the tip vortices, which are the dominant feature of the rotor wake. The inboard vorticity
is determined by a prescribed wake model [102] to reduce computational cost.
In the free wake geometry calculation, the distortion D is obtained by integrating in
time the induced velocity at each wake element due to all the other wake elements. As the
wake age increases by Δ𝜓, the distortion at 𝜓 is obtained by adding the contribution of the
induced velocity to the distortion at previous azimuthal step:

D(𝜓, 𝜙𝑤 ) = D(𝜓, 𝜙𝑤 − Δ𝜓) + Δ𝜓q(𝜓) (4.32)

88
 Ω
bound vortex

γt
γs

near wake

rolling up wake

tip vortex

far wake

Figure 4.3: Vortex-lattice approximation for rotor wake model

The distortion in the wake at the time of its creation is zero. Hence,

D(𝜓, 0) = 0 (4.33)

4.2.3.2 Induced Velocity Calculation

The induced velocity calculation procedure, developed by Johnson [102], is based on a

vortex-lattice approximation for the wake. The tip vortex elements are modeled by line
segments with a small viscous core radius, while the near wake can be represented by
vortex sheet elements or by line segments with a large core radius to eliminate large induced
velocities. The near wake vorticity is generally retained for only a number 𝐾 𝑁𝑊 of azimuth
steps behind the blade. The wake structure is illustrated in Fig. 4.3.
Conservation of vorticity on a three-dimensional blade requires the bound circulation to
be trailed into the wake from the blade tip and root. The lift and circulation are concentrated
at the tip of the blade, since larger dynamic pressures are present in the tip region. Therefore,
a strong, concentrated tip vortex is generated. The vorticity in the tip vortex is distributed

89
Figure 4.4: Single peak circulation distribution model and the resulting far wake approxima-
tion

Figure 4.5: CAMRAD/JA dual peak model and the resulting far wake approximation

over a small but finite region, called the vortex core. The accuracy of a wake model is
sensitive to the value of the strength of the tip vortex prescribed. Two different approaches
are used, depending on the spanwise distribution of the bound circulation. For helicopters
in low speed forward flight, the bound circulation is positive along the entire span of the
blade (Fig. 4.4). The distribution of the bound circulation has only one peak and is referred
to as the single peak model. In the single peak model, the maximum value of the bound
circulation over the blade span, Γmax , is selected for the tip vortex strength. For helicopters
in high speed forward flight or under some means of active control, a spanwise circulation
distribution with two peaks of opposite sign can be encountered. A large positive peak
is generally located inboard and a smaller negative peak on the outboard section of blade
(Fig. 4.5). The dual peak model represents such a situation. The inboard and outboard peaks
Γ 𝐼 and Γ𝑂 , respectively, are identified, and the tip vortex strength assumes the value of the
outboard peak.
Given the blade displacements and circulation distribution, the wake geometry is calcu-
lated. Once the wake geometry has been determined, the influence coefficients are calculated
and stored in the influence coefficient matrix. The induced velocity distribution is obtained

90
by conveniently multiplying the influence coefficient matrix with the circulation distribution:

𝐽
Õ 𝐽
Õ 𝐾Õ 𝑀
𝑁𝑊 Õ

q= Γ𝑂 𝑗 C𝑂 𝑗 + Γ𝐼 𝑗 C𝐼 𝑗 + Γ𝑖 𝑗 C𝑁𝑊𝑖 𝑗 , (4.34)
𝑗=1 𝑗=1 𝑗=1 𝑖=1

where Γ 𝐼 𝑗 , Γ𝑂 𝑗 are the inboard and outboard peaks, respectively, at the azimuth 𝑗, 𝐽 and
𝑀 are the numbers of azimuth and spanwise stations, respectively; 𝐾 𝑁𝑊 is the number
of azimuth stations on which the near wake extends; C𝑂 𝑗 , C 𝐼 𝑗 and C𝑁𝑊𝑖 𝑗 are terms of the
influence coefficient matrix. For the single peak model, Γ𝑂 𝑗 = Γmax 𝑗 and Γ 𝐼 𝑗 = 0.

4.2.4 Reverse Flow Model

In forward flight, there is a region on the retreating side of the rotor disk, represented in
Fig. 4.6, where the airflow encountered by the blade is reversed, flowing from the trailing
edge to the leading edge. The boundary of this reverse flow region is described by the locus
of points on the rotor disk where the velocity of the airflow parallel to the blade is zero.
The approximate boundary of this region on the blade span as a function of azimuth 𝜓 and
advance ratio 𝜇 is given by

𝑥 rev (𝜓) = −(𝑒 + 𝜇𝑅 sin 𝜓). (4.35)

In the present analysis, it is assumed that the aerodynamic lift and moment are zero
within the reverse flow region. This is accomplished by multiplying the aerodynamic lift
and moment expressions by the reverse flow parameter 𝑅 𝐿 𝑀 , defined as follows:

 0 for 0 ≤ 𝑥 ≤ 𝑥 rev (𝜓)




𝑅𝐿 𝑀 =
 1 for 𝑥 > 𝑥 rev (𝜓)



91
 Figure 4.6: Position of the reverse flow region in forward flight

4.2.5 Sectional Aerodynamic Loads

The sectional lift, moment, and drag are obtained by combining the CFD-based RFA,
ONERA dynamic stall, and reverse flow models:

𝐿 = 𝜌𝑈 2 𝑏 𝐶𝑙 𝐴 + 𝐶𝑙 𝑆 𝑅 𝐿 𝑀


(4.36)

𝑀 = 2𝜌𝑈 2 𝑏 2 𝐶𝑚 𝐴 + 𝐶𝑚 𝑆 𝑅 𝐿 𝑀


(4.37)

𝐷 = 𝜌𝑈 2 𝑏 𝐶𝑑 𝐴 + 𝐶𝑑𝑆


(4.38)

where 𝐶𝑙 𝐴 , 𝐶𝑚 𝐴 , and 𝐶𝑑 𝐴 are obtained from Eq. (4.12) and 𝐶𝑙 𝑆 , 𝐶𝑚 𝑆 , and 𝐶𝑑𝑆 are based on
Eqs. (4.20)-(4.22).

4.2.5.1 Blade Velocity Relative to Air

The velocity components 𝑈𝑇 and 𝑈𝑃 correspond to the components of the total air velocity
V 𝐴 taken in the −ê𝑦5 and −ê𝑧5 directions, respectively, as illustrated in Fig. 4.7. The total

92
 Z5
Lift
ine
en terl
foil c
Air
α
φin -UP
Drag Y5
-UT
Moment

Figure 4.7: Orientation of tangential and perpendicular air velocities and aerodynamic loads

air velocity V 𝐴 is calculated as the sum of the airflow velocity due to forward flight, blade
rotation, and induced inflow V 𝐴1 and the airflow velocity due to blade dynamics V 𝐴2 :

V 𝐴 = V 𝐴1 − V 𝐴2 (4.39)

where
 
V 𝐴1 = Ω𝑅 (𝜇 + 𝜆 𝑥 ) ê𝑥1 + 𝜆 𝑦 ê𝑦1 + 𝜆 𝑧 ê𝑧1 (4.40)

and
V 𝐴2 = r¤ 𝐸 𝐴 + Ωê𝑧2 × r𝐸 𝐴 (4.41)

The advance ratio 𝜇 and inflow ratio 𝜆 are given by

𝑉𝐹 cos 𝛼 𝑅
𝜇= (4.42)
Ω𝑅
𝑉𝐹 sin 𝛼 𝑅 + 𝜈
𝜆= (4.43)
Ω𝑅

where 𝑉𝐹 is the freestream velocity, 𝛼 𝑅 is the rotor shaft angle, and 𝜈 is the induced flow
velocity. The position vector for a point on the elastic axis r𝐸 𝐴 is given by:

r𝐸 𝐴 = 𝑒ê𝑥2 + (𝑥 + 𝑢) ê𝑥3 + 𝑣 ê𝑦3 + 𝑤 ê𝑧3 (4.44)

93
The resultant velocity V 𝐴 is expressed in the (ê𝑥5 , ê𝑦5 , ê𝑧5 ) coordinate system using ap-
propriate coordinate transformations, described in Appendix B. Furthermore, the velocity
components in this frame are given by Eqs. (5.176)-(5.177) from Ref. [129]:

𝑈𝑇 = Ω(𝑥 + 𝑒 + 𝑢) + Ω𝑣𝑣,𝑥 −Ω𝑤𝛽 𝑝 + 𝑣¤

+ (𝜇 + 𝜆 𝑥 )Ω𝑅𝑣,𝑥 cos 𝜓 + 𝜆 𝑦 Ω𝑅𝑣,𝑥 sin 𝜓

+ (𝜇 + 𝜆 𝑥 )Ω𝑅 sin 𝜓 − 𝜆 𝑦 Ω𝑅 cos 𝜓 (4.45)

𝑈𝑃 = 𝑤¤ + Ω𝑣(𝑤,𝑥 +𝛽 𝑝 ) − 𝜆 𝑧 Ω𝑅 + (𝜇 + 𝜆 𝑥 )Ω𝑅(𝑤,𝑥 +𝛽 𝑝 ) cos 𝜓

− 𝜆 𝑦 Ω𝑅(𝑤,𝑥 +𝛽 𝑝 ) cos 𝜓 (4.46)

These expressions for 𝑈𝑇 and 𝑈𝑃 are then substituted into Eqs. (4.6) and (4.7) to yield
¤
explicit expressions for 𝑈 and ℎ.

4.2.5.2 Distributed Aerodynamic Loads

Expressions for sectional lift and drag were shown in Eqs. (4.36) and (4.38), respectively.
Lift is assumed to act normal to the total air velocity and drag is assumed to act parallel to it.
Furthermore, all aerodynamic forces act at the quarter chord, which is assumed to coincide
with the elastic axis. Thus, 𝑋 𝐴 = 0. Following these assumptions, the spanwise distributed
aerodynamic force in the “5” system is given by:

p 𝐴 = 𝑝 𝐴𝑦5 ê𝑦5 + 𝑝 𝐴𝑧5 ê𝑧5 (4.47)

where

𝑝 𝐴𝑦5 = −𝐷 cos 𝜙inflow − 𝐿 sin 𝜙inflow (4.48)

𝑝 𝐴𝑧5 = −𝐷 sin 𝜙inflow + 𝐿 cos 𝜙inflow (4.49)

94
The inflow angle 𝜙inflow is the angle between 𝑈𝑇 and the resultant air velocity, given by:

𝑈𝑃
tan 𝜙inflow = (4.50)
𝑈𝑇

The distributed aerodynamic moment in Eq. (4.37) is assumed to act about the elastic axis
of the blade. It can be expressed in the “5” system as:

q 𝐴 = 𝑞 𝐴𝑥5 ê𝑥5 (4.51)

where
𝑞 𝐴𝑥5 = 𝑀 (4.52)

Using appropriate transformations, the distributed aerodynamic force in the “3” system
is given by:
p 𝐴 = 𝑝 𝐴𝑥3 ê𝑥3 + 𝑝 𝐴𝑧3 ê𝑧3 + 𝑝 𝐴𝑧3 ê𝑧3 (4.53)

where

𝑝 𝐴𝑥3 = −𝑣,𝑥 𝑝 𝐴𝑦5 − 𝑤,𝑥 𝑝 𝐴𝑧5 (4.54)

𝑝 𝐴𝑦3 = 𝑝 𝐴𝑦5 (4.55)

𝑝 𝐴𝑧3 = −𝑣,𝑥 𝑤,𝑥 𝑝 𝐴𝑦5 + 𝑝 𝐴𝑧5 (4.56)

Similarly, the distributed aerodynamic moment is expressed in the “3” system as:

q 𝐴 = 𝑞 𝐴𝑥3 ê𝑥3 + 𝑞 𝐴𝑧3 ê𝑧3 + 𝑞 𝐴𝑧3 ê𝑧3 (4.57)

95
where

𝑞 𝐴𝑥3 = 𝑞 𝐴𝑥5 (4.58)

𝑞 𝐴𝑦3 = 𝑣,𝑥 𝑞 𝐴𝑥5 (4.59)

𝑞 𝐴𝑧3 = 𝑤,𝑥 𝑞 𝐴𝑥5 (4.60)

4.3 Coupled Aeroelastic Response and Rotor Trim Solu-

tion

The combined structural and aerodynamic equations form a system of coupled differential
equations that are written in a state-variable form. These equations are further coupled
with the rotor trim equations. The trim procedure used is based on a propulsive trim with
three force equations (longitudinal, lateral, and vertical) and three moment equations (roll,
pitch, and yaw) corresponding to a helicopter in free flight. A simplified tail rotor model,
based on uniform inflow and blade element theory, is used. A detailed description of the
procedure used to solve the combined system of equations is provided in Appendix C. Spatial
discretization of the aeroelastic equations of motion is performed using Galerkin’s method
of weighted residuals. This procedure is similar to the one developed in Refs. [23, 34, 129].
The coupled trim/aeroelastic equations are integrated in time using the DDEABM solver
[130], which employs a variable-order, predictor-corrector-based Adams-Bashforth scheme.

4.4 Previous Validation Studies

Previous validation studies of the AVINOR code [6, 23, 24, 34] have used experimental
data from the Higher Harmonic Control Aeroacoustic Rotor Test (HART) [131] to confirm
the accuracy of the aeroelastic simulation framework. The HART wind tunnel experiments
involved a detailed examination of higher harmonic blade pitch control for acoustic noise

96
 0.2

0.1

C N M2
0 Experiment

DL+RFA
CFD+RFA
-0.1
0 90 180 270 360
Azimuth (deg)

Figure 4.8: Validation of unsteady aerodynamic loads obtained from AVINOR with HART
experimental data at 87% span [6]

and vibration reduction on a scaled version of the MBB Bo-105 rotor. The numerical/experi-
mental comparisons have included correlation of unsteady aerodynamic blade loads as well
as acoustic noise levels measured on a carpet plane below the rotor. Acoustic calculations
are not performed in the present study. Nonetheless, the previous acoustic validation results
have demonstrated the accuracy of the free-wake model employed in AVINOR, which is also
employed during active vibration control. The most recent validation study [6] also used
the HART dataset to compare the accuracy of two different RFA models when calculating
2D sectional aerodynamic loads. The simulations illustrated that the RFA model based on
high-fidelity CFD data is more accurate than the RFA model based on doublet-lattice (DL)
simulations used in earlier studies (i.e., Refs. [23, 24]). A comparison of the normal force at
87% blade span from Ref. [6] is shown in Fig. 4.8.

4.5 Flow Control ROM Implementation

The flow control ROM is incorporated into AVINOR such that it is directly compatible
with the existing aerodynamic model. The ROM generates the unsteady changes in 2D
lift, moment, and drag coefficients on the blade sections where the actuators are present.

97
The expressions for the sectional lift, moment, and drag provided in Eqs. (4.36)-(4.38) are
modified as follows:

𝐿 = 𝜌𝑈 2 𝑏 𝐶𝑙 𝐴 + 𝐶𝑙 𝑆 + Δ𝐶𝑙fc 𝑅 𝐿 𝑀


(4.61)

𝑀 = 2𝜌𝑈 2 𝑏 2 𝐶𝑚 𝐴 + 𝐶𝑚 𝑆 + Δ𝐶𝑚fc 𝑅 𝐿 𝑀


(4.62)

𝐷 = 𝜌𝑈 2 𝑏 𝐶𝑑 𝐴 + 𝐶𝑑𝑆 + Δ𝐶𝑑fc


(4.63)

where 𝐶𝑙 𝐴 , 𝐶𝑚 𝐴 , and 𝐶𝑑 𝐴 are obtained from Eq. (4.12); 𝐶𝑙 𝑆 , 𝐶𝑚 𝑆 , and 𝐶𝑑𝑆 are based on
Eqs. (4.20)-(4.22); and Δ𝐶𝑙fc , Δ𝐶𝑚fc , and Δ𝐶𝑑fc are obtained from Eq. (3.11).
The flow control ROM, given by Eqs. (3.10) and (3.11), closely resembles the structure
of the CFD-based RFA model, given by Eqs. (4.16) and (4.17). This is advantageous for
combining the ROM with the AVINOR solver. Furthermore, the differential equations
representing the flow control ROM can be integrated in time using the same integration
scheme used for solving the coupled system of equations representing the aeroelastic blade
response.

4.5.1 Comment on Code Timing

In the baseline version of the AVINOR code, a single rotor revolution requires approxi-
mately 34 seconds of computation time when utilizing 8 computer cores. When the flow
control ROM is combined with AVINOR, the computation time increases to 50 seconds
per revolution. Therefore, an open-loop control simulation consisting of 10 revolutions
requires 8.3 minutes. A closed-loop control simulation aimed at vibration reduction typically
requires 200 revolutions, or 2.8 hours of computation time, to minimize vibratory hub loads.
Furthermore, the trim procedure employed in AVINOR typically requires 300 revolutions, or
4.2 hours of computation time, to reach a converged trim state. Closed-loop control and trim
calculations using the flow control ROM in AVINOR represent a significant computational
expense. However, the computational cost of the ROM is still a substantial improvement

98
over the cost of CFD, which becomes intractable for simulations on the order of hundreds
of rotor revolutions.

99
 CHAPTER 5

Control Approaches

5.1 The Higher Harmonic Control Algorithm

The higher-harmonic control (HHC) algorithm has been used extensively for active control
of vibrations and acoustic noise in rotorcraft [8]. The algorithm is a frequency-domain
approach based on the assumption that the helicopter response can be represented by a linear
model relating the output z to the control inputs u. The approach is applied to disturbances
(vibrations or noise) having known frequency content that is predominantly 𝑁 𝑏 /rev, where
𝑁 𝑏 is the number of blades. When performing vibration reduction on a four-bladed rotor,
the output z consists of the 4/rev vibratory loads and the input vector u includes harmonic
control inputs at the 2-, 3-, 4-, and 5/rev frequencies.
The HHC algorithm was examined from a control theory perspective in Ref. [132].
Convergence and stability characteristics were analyzed for three versions of the algorithm,
including the classical invariant version, an adaptive version, and a relaxed version. The
HHC algorithm has been implemented in the AVINOR code for closed-loop, on-blade control
simulations involving actively-controlled trailing edge flaps [23, 24, 132] and microflaps
[7, 31] for vibration reduction. The classical and adaptive versions employed in the present
study are described in this section.

100
5.1.1 Classical HHC

The HHC algorithm is designed for steady trimmed flight conditions. In the HHC algorithm,
the measurement of the plant output and update of the control input are performed at specific
times, 𝑡 = 𝑘𝜏, where 𝜏 is the time interval between updates during which the plant output
reaches a steady state. This time interval can be one or more rotor revolutions. For the
present simulations, six revolutions are sufficient to achieve a steady-state (converged)
condition after applying a control update.
With the introduction of a disturbance w, representing the helicopter operating condition,
the plant output at the 𝑘th time step is given by

z 𝑘 = Tu 𝑘 + Ww (5.1)

where the sensitivity matrix T represents the linear approximation of the helicopter response
to the control and is given by
𝜕z
T= (5.2)
𝜕u

At the initial condition, 𝑘 = 0:

z0 = Tu0 + Ww (5.3)

Subtracting Eq. (5.3) from Eq. (5.1) to eliminate the unknown w yields

z 𝑘 = z0 + T (u 𝑘 − u0 ) (5.4)

The controller is based on the minimization of a quadratic cost function:

𝐽 (z 𝑘 , u 𝑘 ) = zT𝑘 Qz z 𝑘 + uT𝑘 Ru 𝑘 (5.5)

101
The optimal control law is determined from the requirement

𝜕𝐽 (z 𝑘 , u 𝑘 )
=0 (5.6)
𝜕u 𝑘

which yields the optimal control input u 𝑘,opt , given by:

  −1  
u 𝑘,opt = − TT Qz T + R TT Qz (z0 − Tu0 ) (5.7)

This is the classical version of the HHC algorithm, which yields an explicit relation for the
optimal control input. In this formulation, the sensitivity matrix T is assumed to be invariant.
Furthermore, the T matrix can be identified offline, using open-loop control signals, before
performing closed-loop control.
It is also useful to consider another, recursive form of Eq. (5.4) where subsequent control
updates are written as
z 𝑘+1 = z 𝑘 + T (u 𝑘+1 − u 𝑘 ) (5.8)

and the optimal control law is given by

  −1  
T T

u 𝑘+1,opt = − T Qz T + R T Qz z 𝑘,opt − Tu 𝑘,opt (5.9)

In this equation, the index 𝑘 refers to the controller update number, an integer corresponding
to each time the control algorithm updates the values of the input vector u 𝑘 which results
in a new output vector z 𝑘 . Note that Eq. (5.4) is a special case of Eq. (5.8) where 𝑘 = 0.
It can be shown that u 𝑘,opt is independent of 𝑘 and remains constant for all future control
updates 𝑘 ≥ 1 [132]. Thus, the algorithm converges to the optimum value in a single step,
for a well-identified linear system.

102
5.1.2 Adaptive HHC

When nonlinearities in the helicopter response are significant, such as during heavy BVI
or dynamic stall conditions, an invariant T matrix produces a poor approximation of the
helicopter response. Under such conditions, it is more appropriate to use an adaptive version
of the HHC algorithm, where T is updated using least-squares methods after every control
update [132]. To describe the adaptive HHC algorithm, relative output and input vectors are
defined, Δz 𝑘 with length 2𝑝 and Δu 𝑘 with length 2𝑚 as

Δz 𝑘 = z 𝑘 − z 𝑘−1 , Δu 𝑘 = u 𝑘 − u 𝑘−1 (5.10)

and ΔZ 𝑘 of size 2𝑝 × 𝑘 and ΔU 𝑘 of size 2𝑚 × 𝑘 as

ΔZ 𝑘 = [Δz1 . . . Δz 𝑘 ] , ΔU 𝑘 = [Δu1 . . . Δu 𝑘 ] (5.11)

Hence,
ΔZ 𝑘 = TΔU 𝑘 (5.12)

Assuming ΔU 𝑘 ΔUT𝑘 is non-singular, one can define the matrix P 𝑘 :

  −1
P𝑘 = ΔU 𝑘 ΔUT𝑘 (5.13)

From Eq. 5.12, the least-squares estimate T̂LS𝑘 of T is given by

T̂LS𝑘 = ΔZ 𝑘 ΔUT𝑘 P 𝑘 (5.14)

The recursive least-squares method is used to iteratively update T̂LS𝑘 based on the past and
current values of Δz 𝑘 and Δu 𝑘 . The updated estimate of T̂LS𝑘 is used at each control update

103
step to calculate the optimal control input u 𝑘,opt as follows:

  −1   
u 𝑘,opt = − T̂TLS𝑘 Qz T̂LS𝑘 + R T̂TLS𝑘 Qz z0 − T̂LS𝑘 u0 (5.15)

The adaptive HHC algorithm has been shown to perform better than the classical
HHC when reducing vibrations encountered during heavy BVI conditions [132]. The BVI
conditions correspond to descending flight at an advance ratio of 𝜇 = 0.15 and a descent
angle of 6.5◦ . Since most of the conditions examined in the present study are not severely
impacted by nonlinear phenomena like BVI or dynamic stall, the classical version of the
HHC algorithm can be used. The classical HHC is used for vibration reduction in the range
of 0.20 ≤ 𝜇 ≤ 0.30. An additional high-speed case is considered at 𝜇 = 0.35. In this case,
dynamic stall affects the vibratory loads and introduces nonlinear effects into the helicopter
response. Therefore, the adaptive HHC is employed for the case of 𝜇 = 0.35. The classical
and adaptive versions of HHC yield nearly identical results for the more benign cases of
0.20 ≤ 𝜇 ≤ 0.30. In all cases, after the optimal control input is calculated, the rotor is
re-trimmed in AVINOR to maintain a steady flight condition.

5.2 Actuator Saturation

In a practical setting, there are limitations on any actuation system. For example, the
deflection of a trailing edge flap or a microflap cannot exceed a certain maximum value.
For an AFC system, the jet strength produced by the fluidic actuators is limited by the
power available from the air compression system. This imposes a constraint on the optimal
control input calculated by the HHC algorithm, and therefore has an effect on the vibration
reduction capability of the control system.
An optimization approach, developed in Ref. [31], is employed to ensure the AFC jet
strength does not exceed this saturation limit when performing closed-loop control. The
optimization approach restates the HHC algorithm as a constrained nonlinear optimization

104
problem:

min 𝐽 (z 𝑘 , u 𝑘 ) = zT𝑘 Qz z 𝑘 + uT𝑘 Ru 𝑘 (5.16)

u𝑘

subject to |𝛿(𝜓, u 𝑘 )| ≤ 𝛿limit (5.17)

The optimization problem is solved using a sequential quadratic programming routine

available in the fmincon tool in MATLAB. The AVINOR code, which is written in Fortran,
calls the MATLAB application as an external routine to calculate the updated control input
u 𝑘 . The nonlinear constraint on the control amplitude 𝛿(𝜓, u 𝑘 ), given by Eq. (5.17), is
enforced at every integer value of 𝜓 over the range [0◦ , 360◦ ]. After calculating u 𝑘 , six
rotor revolutions are simulated within the AVINOR code, to reach a steady-state condition,
before performing the next control update u 𝑘+1 .

5.3 Implementation for a Microflap

The implementation of the HHC algorithm for vibration control using a microflap is dis-
cussed first, since this has been studied previously and represents a conventional imple-
mentation of HHC. The implementation of the HHC algorithm for AFC is based on the
implementation for the microflap. The modifications required for AFC actuation are dis-
cussed later, and provide a useful means for comparing the approaches.
The microflap, shown in Fig. 5.1, is a deployable Gurney flap with a size of 1.5% of
the blade chord. It slides in and out of a cavity located 6%𝑐 from the trailing edge and has
a continuous range of motion from fully-retracted (𝛿 = 0) to fully-deployed (𝛿 = 1.5%𝑐).
For HHC, the total microflap deflection is expressed as a superposition of the 2-, 3-, 4-, and
5/rev harmonics:

5
Õ
𝛿(𝜓, u 𝑘 ) = 𝛿0 + [𝛿 𝑁𝑐 cos (𝑁𝜓) + 𝛿 𝑁 𝑠 sin (𝑁𝜓)] (5.18)
𝑁=2

105
 6%c

0.6%c
0.3%c

δf 1.5%c

Figure 5.1: The microflap [7]

where the quantities 𝛿 𝑁𝑐 and 𝛿 𝑁 𝑠 correspond to the cosine and sine components of the
𝑁/rev control harmonic. The static component of the microflap deflection, 𝛿0 = 0.75%𝑐, is
included so that 𝛿 ≥ 0. The control input vector u 𝑘 is given by the amplitudes of the control
harmonics:
n oT
u 𝑘 = 𝛿2𝑐 𝛿2𝑠 𝛿3𝑐 𝛿3𝑠 𝛿4𝑐 𝛿4𝑠 𝛿5𝑐 𝛿5𝑠 (5.19)

For vibration reduction on a four-blade rotor, the output vector z 𝑘 consists of the cosine

106
and sine components of the 4/rev vibratory shear forces and moments at the rotor hub:


 



 𝐹𝑥4𝑐




 


 



 𝐹𝑥4𝑠 



 

 


 𝐹𝑦4𝑐 



 


 



 𝐹𝑦4𝑠




 


 



 𝐹𝑧4𝑐 



 

 
 𝐹𝑧4𝑠 

 

z𝑘 = (5.20)
 


 𝑀𝑥4𝑐 


 


 



 𝑀 𝑥4𝑠 



 

 


 𝑀 
𝑦4𝑐 


 


 



 𝑀 𝑦4𝑠 



 


 



 𝑀 𝑧4𝑐 



 

 
 𝑀𝑧4𝑠 
 
 

The weighting matrix Qz in the cost function in Eq. (5.5) is a diagonal matrix. The
twelve diagonal terms of Qz are weights corresponding to the various components of the
vibratory hub shears (longitudinal, lateral, and vertical) and vibratory hub moments (roll,
pitch, and yaw). All components are weighted equally in the present study, implying Qz = I.

5.4 Implementation for Discrete AFC Operation

The discrete operation of the AFC actuators presents a unique challenge when implementing
higher-harmonic control. Because each actuator (PS and SS) can only be on or off at a fixed
jet strength, superposition of multiple harmonics in the sense of Eq. 5.18, like the microflap,
is not possible. To overcome this limitation, the fixed jet strength requirement is temporarily
relaxed when using the HHC algorithm.
A continuous AFC input signal, 𝑢 cont , is defined. The signal is allowed to vary con-

107
tinuously between two actuation limits, which are full-strength PS actuation (𝑢 cont = −1)
and full-strength SS actuation (𝑢 cont = 1) where 𝑢 cont = 0 indicates both jets are off. The
continuous signal is mapped to the two ROM input variables 𝑢 SS and 𝑢 PS in Eq. (3.1) as

 |𝑢 cont | , 0.01 ≤ 𝑢 cont ≤ 1



SS

𝑢 = (5.21)
 0,

𝑢 cont < 0.01

𝑢 cont > −0.01

 0,


𝑢 PS = (5.22)
 |𝑢 cont | , −1 ≤ 𝑢 cont ≤ −0.01



Note that while one jet is on, the other jet is off. Both jets are turned off below 1% jet
strength, since AFC at this strength has a negligible effect on aerodynamic loads. This
threshold ensures that it is possible for both jets to be off during the ROM simulation. A
threshold of 1% jet strength produces nearly identical results compared to a threshold of
0.1%. However, a threshold of 5% or higher produces significant changes in the computed
unsteady loads. Thus, a 1% threshold was deemed appropriate.
By using a continuous actuation signal, superposition of the 2-, 3-, 4-, and 5/rev harmon-
ics for HHC can be expressed as

5
Õ
𝑢 cont (𝜓, u 𝑘 ) = [𝑢 𝑁𝑐 cos(𝑁𝜓) + 𝑢 𝑁 𝑠 sin(𝑁𝜓)] (5.23)
𝑁=2

The control input vector u 𝑘 determined by the HHC algorithm is given by the cosine and
sine amplitudes of the control harmonics:

n oT
u 𝑘 = 𝑢 2𝑐 𝑢 2𝑠 𝑢 3𝑐 𝑢 3𝑠 𝑢 4𝑐 𝑢 4𝑠 𝑢 5𝑐 𝑢 5𝑠 (5.24)

This continuous formulation is directly compatible with the existing HHC algorithm.
First the optimal control harmonic amplitudes are obtained from the HHC algorithm,
and subsequently the continuous-valued signal 𝑢 cont is converted to a discrete-valued signal

108
using delta modulation [133]. Delta modulation is commonly used in digital circuits for
encoding analog signals into binary pulse signals. A typical application of this approach
is speech encoding in telecommunications systems, for efficient data transmission. There
are several forms of delta modulation available, with varying degrees of sophistication. A
simple linear delta modulation approach is used in the present study.
The goal of the continuous-to-discrete conversion is to obtain a discrete input signal that
mimics the unsteady aerodynamic loads generated by the continuous input signal. This is
performed as a post-processing step using a standalone version of the flow control ROM,
which is not connected to the AVINOR code. A schematic of the approach is illustrated in
Fig. 5.2. The unsteady change in lift coefficient due to the continuous signal is calculated
assuming a static airfoil operating at 𝑀 = 0.45 and 𝛼 = 5◦ . These conditions represent the
average flow conditions encountered at 75% blade span. Simulation of the continuous input
signal 𝑢 cont yields the continuous output signal Δ𝐶𝑙,cont (𝜓) shown in Fig. 5.3. Next, an error
tolerance of ±10% is defined using the maximum and minimum values of Δ𝐶𝑙,cont (𝜓) over
the interval 0◦ ≤ 𝜓 ≤ 360◦ . The tolerance is defined so as to bound the error of the discrete
output Δ𝐶𝑙,disc relative to the continuous output, calculated as

𝜖dc (𝜓) = Δ𝐶𝑙,disc (𝜓) − Δ𝐶𝑙,cont (𝜓) (5.25)

The discrete input signal 𝑢 disc is subsequently computed by performing a closed-loop

simulation where the relative error of the discrete output 𝜖dc (𝜓) is calculated at each time
step, 𝜓𝑖 . The error is fed back to a simple controller, depicted in Fig. 5.2, that switches the
input signal level when the error exceeds ±10%. The discrete signal has three possible input
levels, which correspond to PS on, SS on, or both jets off. The jet strength is fixed in the
simulation, and is selected as the maximum value of |𝑢 cont (𝜓)|. If the error is less than 10%,
then the controller maintains the current input value: 𝑢 disc (𝜓𝑖+1 ) = 𝑢 disc (𝜓𝑖 ). Otherwise, the
controller switches the input level based on the sign of 𝜖dc (𝜓𝑖 ) and the sign of Δ𝐶𝑙,cont (𝜓𝑖 ).

109
For example, if 𝜖dc < −10% (i.e., if Δ𝐶𝑙,disc is less than Δ𝐶𝑙,cont by more than 10%), then
the controller changes the input level so as to increase Δ𝐶𝑙,disc . When the sign of Δ𝐶𝑙,cont is
positive, activating the PS jet will increase Δ𝐶𝑙,disc . When the sign of Δ𝐶𝑙,cont is negative,
turning both jets off will increase Δ𝐶𝑙,disc . Similar logic is applied when 𝜖 dc > +10%. In
this case, the controller changes the input level so as to decrease Δ𝐶𝑙,disc . When the sign of
Δ𝐶𝑙,cont is negative, activating the SS jet will decrease Δ𝐶𝑙,disc . When the sign of Δ𝐶𝑙,cont is
positive, turning both jets off will decrease Δ𝐶𝑙,disc .
For the example illustrated in Fig. 5.3, the jet strength is 91% of the full jet strength. The
closed-loop simulation yields the discrete input signal 𝑢 disc (𝜓) and the discrete output signal
Δ𝐶𝑙,disc (𝜓) shown in the figure. The discrete input is a sequence of square wave segments
where the PS/SS actuators are rapidly switched on and off. Furthermore, the discrete output
is a reasonable approximation of the continuous output, since the error was constrained to
be within ±10%.
Note that, as mentioned, the continuous signal shown in Fig. 5.3 was converted to a
discrete signal assuming a static airfoil operating at 𝑀 = 0.45 and 𝛼 = 5◦ . The accuracy of
this assumption is not critical. This is confirmed by examining the sectional aerodynamic
loads computed in AVINOR when the continuous and discrete control signals are applied in
an open-loop mode. Figure 5.4 illustrates the normal force coefficient 𝐶𝑛 multiplied by the
local Mach number squared, evaluated at 73% blade span. The baseline corresponds to a
rotor operating at advance ratio 𝜇 = 0.3 without AFC. The response due to discrete AFC
closely resembles the response due to continuous AFC. High-frequency perturbations are
present in the discrete case, but the amplitude of these perturbations is small. The effect of
discrete actuation on the 4/rev vibratory hub loads is illustrated in Fig. 5.5. Compared to
continuous actuation, the differences in the vibratory loads are minor. Both the continuous
and discrete control signals produce a substantial reduction in vibratory loads.
The vibration reduction results illustrated in Fig. 5.5 were obtained using a saturation
limit of 100% of the full jet strength. This limit is based on the jet strength used for training

110
Figure 5.2: Schematic of closed-loop ROM simulation for converting a continuous signal to
a discrete-valued signal using delta modulation

the ROM, which corresponds to 𝐶 𝜇 = 0.005 at 𝑀 = 0.5. Equivalently, the full jet strength
corresponds to 𝐶 𝜇,avg = 0.0054 when it is computed at 𝑀 = 0.45, the average Mach number
at 75% blade span. This definition of full jet strength was selected in an ad hoc manner, and
therefore may not represent a realistic value for implementation on a helicopter rotor.
In a practical setting, the flow control actuators are powered by a supply of compressed
air. Therefore, it is more appropriate to limit the jet strength based on the power required
for actuation rather than the momentum coefficient. The power is estimated using the ideal
power of a fluid jet:
1 2
𝑃jet = 𝑚¤ jet𝑈jet (5.26)
2

The power required to achieve full jet strength is 3.6 kW, or 4.9 hp, per blade. Furthermore,
by limiting the power available for actuation, the saturation limit can be directly related to
the overall rotor performance. The sensitivity of the vibratory load reduction to the available

111
 Continuous Continuous 10% Discrete
0.2

Lift Coefficient
0.1

-0.1

-0.2
0 90 180 270 360
1

Control Input
0.5

-0.5

-1
0 90 180 270 360
Azimuth Angle (°)

Figure 5.3: Conversion of a continuous control input signal to a discrete-valued signal using
delta modulation

power is examined in the next chapter.

112
 0.25
Baseline

Non-Dim. Normal Force

Continuous AFC
0.2 Discrete AFC

0.15

0.1

0.05

0
0 90 180 270 360
Azimuth Angle (°)

Figure 5.4: Non-dimensional normal force (𝑀 2𝐶𝑛 ) evaluated at 73% blade span due to
continuous and discrete AFC operation; 𝜇 = 0.3, 𝐶𝑇 = 0.005

10-4
2
Non-Dim. 4/rev Hub Loads

Baseline
Continuous AFC
1.5 Discrete AFC

0.5

he
ar
he
ar
he
ar m. m. om
.
.S . S . S l l Mo h Mo M
ng La
t rt Ro c Ya
w
Lo Ve Pit

Figure 5.5: Vibratory hub load reduction due to continuous and discrete AFC operation;
𝜇 = 0.3, 𝐶𝑇 = 0.005

113
 CHAPTER 6

Results and Discussion

This chapter presents results demonstrating the effectiveness of AFC for reducing rotor
vibrations in forward flight. The effects of fluidic actuation are examined in open-loop and
closed-loop simulation modes. The control sensitivity parameters obtained from open-loop
simulations are needed in order to gain additional insight into the types of actuation signals
that are most effective for vibration reduction. Furthermore, the closed-loop vibratory load
reduction is examined as a function of the actuation strength and the rotor advance ratio.
The vibration reduction capability of AFC is also compared with that of a microflap. The
performance penalty associated with reducing vibrations is calculated so as to illustrate the
fundamental trade-off that exists in rotorcraft applications.

6.1 Helicopter Configuration

The results presented in this chapter are obtained for a helicopter configuration resembling
the MBB Bo-105, a four-bladed hingeless rotor. The rotor parameters are summarized in
Table 6.1 and are nondimensionalized using 𝑀𝑏 , 𝑅, and 1/Ω for mass, length, and time,
respectively. The mass and stiffness distributions are assumed to be constant along the
span of the blade. The built-in twist distribution of the blade is linear and decreases by 8◦
from blade root to blade tip, as indicated by the parameter 𝜃 tw . Additional information on
nondimensionalization is provided in Appendix B.4.

114
 Table 6.1: Rotor parameters used in computations

Dimensional Rotor Data

𝑅 = 4.91 m
𝑀𝑏 = 27.35 kg
Ω = 425 RPM
Non-Dimensional Main Rotor Data
𝑁𝑏 = 4 𝑐 = 0.05498𝑅
𝐿 𝑏 = 1.0 𝑒=0
𝑋𝐴 = 0 𝛽 𝑝 = 2.5◦
𝑋Ib = 0 𝑋IIb = 0
𝐼MB2 = 0 𝐼MB3 = 0.0004
𝐸 𝐼 𝜂𝜂 = 0.0302 𝐸 𝐼 𝜁 𝜁 = 0.0105
𝐺𝐽 𝑏 = 0.0015 𝜔 𝐹1 = 1.124
𝜔F2 = 3.404 𝜔F3 = 7.606
𝜔L1 = 0.732 𝜔L2 = 4.458
𝜔T1 = 3.170 𝜔T2 = 9.079
𝛾 = 5.5 𝜎 = 0.07
𝜃 tw = −8◦
Non-Dimensional Tail Rotor Data
𝑋𝑡 = 1.20 𝑍𝑡 = 0
Helicopter Data
𝐶𝑊 = 0.005 𝑓 𝑓 𝐶𝑑𝑓 = 0.031
𝑋FA = 0 𝑍FA = 0.3
𝑋FC = 0 𝑍FC = 0.3

6.2 Baseline Results Without Control

The baseline vibratory loads of the rotor without AFC are examined first. Vibratory loads
at the rotor hub are calculated by integrating the distributed loads over the entire blade
span in the rotating frame. Subsequently, the loads from each blade are transformed from
the rotating frame to the hub-fixed non-rotating frame. The contributions from each blade
are then combined to yield the total hub shear forces and moments. For a rotor with
identical blades, the 𝑁 𝑏 /rev frequency component is the most significant. Therefore, the
4/rev component of vibrations is the primary focus of this study.
The various components of the 4/rev hub loads are illustrated in Fig. 6.1 for advance
ratios of 𝜇 = 0.20, 0.25, 0.30, and 0.35 during trimmed level flight with thrust coefficient

115
 10-4
2
= 0.20
= 0.25

4/Rev Hub Loads

1.5 = 0.30
= 0.35

0.5

0
ar ar ar . . .
he he he Mom Mom Mom
. S t . S t . S ll h w
Lo
ng La Ve
r Ro Pit
c Ya

Figure 6.1: Baseline uncontrolled 4/rev hub loads during trimmed level flight at various
advance ratios; results are non-dimensional

𝐶𝑇 = 0.005, which represents a moderate to lightly loaded rotor. The low-speed cases of
𝜇 = 0.20 and 𝜇 = 0.25 are relatively benign cases to consider, since dynamic stall does not
affect blade vibrations at these speeds. At 𝜇 = 0.35, dynamic stall has a strong effect on
the in-plane shear forces, as illustrated by the increased longitudinal and lateral shears. For
the intermediate case of 𝜇 = 0.30, dynamic stall is present, but to a lesser extent compared
to 𝜇 = 0.35. Moreover, the range of advance ratios considered encompasses a variety of
different flow conditions for testing the vibration reduction potential of AFC.
The propulsive trim procedure in AVINOR is used to determine the trim variable settings
needed for steady level flight at each advance ratio. The effect of the advance ratio on
each of the six trim variables is shown in Fig. 6.2. The rotor shaft angle increases linearly
from 𝛼 𝑅 = 2.3◦ to 𝛼 𝑅 = 6.7◦ as the advance ratio increases from 𝜇 = 0.20 to 𝜇 = 0.35,
since the forward thrust provided by the rotor must also increase. The collective pitch also
increases from 𝜃 0 = 7◦ to 𝜃 0 = 11◦ to provide an increasing level of forward thrust. The
longitudinal cyclic is also related, since it is responsible for tilting the rotor disk fore and aft.
The longitudinal cyclic decreases from 𝜃 1𝑠 = −3.7◦ to 𝜃 1𝑠 = −7◦ . The other trim variables
display only minor changes as the advance ratio increases: the lateral cyclic 𝜃 1𝑐 decreases

116
 20
Collective Tail rotor pitch
15 Lat. cyclic Rotor shaft angle
Long. cyclic Roll angle

Trim Setting (°)

10

-5

-10
0.2 0.25 0.3 0.35
Advance Ratio

Figure 6.2: Trim variable settings for level flight at various advance ratios

by 0.6◦ , the tail rotor pitch 𝜃 0𝑡 increases by 0.6◦ , and the fuselage roll angle 𝜙 𝑅 increases by
1.8◦ .

6.3 Open-Loop Control Studies

6.3.1 Vibratory Loads

The vibration reduction capability of the AFC actuators is examined first in an open-loop
mode. For alternating PS/SS actuation, 𝑁/rev actuation is defined using a square wave signal
with phase offset angle 𝜙offset :

𝑢 disc (𝜓) = 0.5sgn [cos (𝑁𝜓 + 𝜙offset )] (6.1)

where sgn[ · ] denotes the sign function. The leading coefficient of 0.5 is used to produce
half-strength actuation in the open-loop simulations. Half-strength actuation corresponds to
a momentum coefficient of 𝐶 𝜇,avg = 0.0013 (since the momentum coefficient scales with the
jet velocity squared) and a jet power of 𝑃jet = 0.45 kW (0.61 hp) per blade. Simulations
using full-strength actuation (𝐶 𝜇,avg = 0.0054) produced undesirably high vibration levels in

117
the rotor, indicating overactuation, since vibrations were amplified rather than reduced.
The non-dimensional vibratory hub loads due to 𝑁/rev actuation for 𝑁 = 2, 3, 4, 5 at
a range of phase offset angles are illustrated in Fig. 6.3. In Figs. 6.3a-e, the 4/rev sine
and cosine components of the hub shears and moments are plotted. For each actuation
frequency, the vibratory load data form a circle around the baseline point as the phase
offset is swept from 0◦ to 360◦ in 45◦ increments. The size of each circle indicates the
sensitivity of the vibratory load to the actuation frequency. This can be interpreted as a
visual representation of the T matrix in Eq. (5.2). Another feature of this plotting method is
that actuation data encircling the origin indicate whether it is possible to reduce a particular
component of vibrations to zero. Increasing or decreasing the strength of actuation for a
particular harmonic controls the radius of the circle. For example, the 4/rev actuation data
in Fig. 6.3c indicate that it is possible to reduce the 4/rev vertical shear to zero by using a
slightly weaker actuation strength. Interestingly, the longitudinal shear, lateral shear, roll
moment, and pitch moment are most sensitive to 3/rev actuation, but are less sensitive to
4/rev actuation. In Fig. 6.3f, the magnitudes of the various 4/rev hub load components
are combined into a single vibration objective. The results show that 4/rev actuation at
𝜙offset = 180◦ reduces the overall vibrations by 70%. It is expected that the optimal control
input determined in closed-loop simulations will mainly utilize 4/rev actuation, since the
overall vibratory loads are most sensitive to this actuation frequency. This is confirmed by
the closed-loop simulations performed in the next section.
The variation of the rotor power due to open-loop actuation is illustrated in Fig. 6.3g,
indicating that there is a power penalty associated with AFC actuation for several cases.
This penalty can be as high as 20% of the baseline rotor power. The figure also illustrates
that there are a few cases where 2/rev or 3/rev actuation can be employed to reduce the rotor
power required. The greatest power reduction is 3% below the baseline when 2/rev actuation
is applied at a phase angle of 𝜙offset = 135◦ . In previous studies (e.g., Refs. [8, 25]), the use
of lower harmonics such as 1/rev and 2/rev actuation has been shown to be effective for

118
 0.5 2

Longitudinal Shear sin(4 )

Baseline Baseline Baseline

Vertical Shear sin(4 )

Lateral Shear sin(4 )
2/rev 0.5 2/rev 2/rev
3/rev -1
3/rev -0.5 0.5 1 3/rev
4/rev 4/rev 4/rev 1
-1.5 -1 -0.5 0.5
5/rev 10-4 5/rev 5/rev
10-4 10-4
-0.5
-0.5 -1 1 2

-1 -1
-1
10-4 10-4 10-4

Longitudinal Shear cos(4 ) Lateral Shear cos(4 ) Vertical Shear cos(4 )

(a) (b) (c)

300
Baseline Baseline 0.4 Baseline
0.2

Pitch Moment sin(4 )

Roll Moment sin(4 )

2/rev 2/rev 250 2/rev

Vibration Objective
3/rev -0.4 3/rev

(% of Baseline)
-0.6 -0.2 0.2 0.4 3/rev
0.2
4/rev 4/rev 200 4/rev
10-4 5/rev 5/rev-0.4
-0.6 -0.2 0.2 0.4 5/rev
150
-0.2
10-4
-0.2 100
-0.4
50
-0.4
10-4 10-4
0
0 90 180 270 360
Roll Moment cos(4 ) Pitch Moment cos(4 ) Phase Offset (°)

(d) (e) (f)

140
Baseline
2/rev
130
(% of Baseline)

3/rev
Rotor Power

4/rev
120 5/rev

110

100

90
0 90 180 270 360
Phase Offset (°)

(g)

Figure 6.3: Vibratory hub loads and rotor power for uncontrolled case (baseline) and open-
loop AFC (𝐶 𝜇,avg = 0.0013) applied at 2, 3, 4, and 5/rev harmonics with phase offset swept
from 0◦ to 360◦ in 45◦ increments (𝜇 = 0.30)

reducing rotor power requirements. However, vibration reduction and power reduction are
often competing objectives, as illustrated in Fig. 6.3.
Applying AFC actuation in an open-loop manner also affects the total thrust of the
rotor, i.e., the 0/rev component of the vertical force acting on the rotor. The helicopter was
originally trimmed to operate at a baseline thrust coefficient of 𝐶𝑇 = 0.005. Therefore, any
deviation from this baseline value as a result of actuation represents an untrimmed rotor.
The non-dimensional thrust coefficient for each of the actuation cases is shown in Fig. 6.4.

119
 130
Baseline
2/rev

Thrust Coefficient
120

(% of Baseline)
3/rev
4/rev
5/rev
110

100

90

0 90 180 270 360

Phase Offset (°)

Figure 6.4: Variation of thrust coefficient in forward flight (𝜇 = 0.30) due to alternating
PS/SS blowing, before retrimming the rotor

Interestingly, the thrust is most sensitive to 3/rev actuation. In the most extreme case, the
rotor thrust increased by 15%. However, for the other cases, the thrust increases or decreases
by less than 4% of the baseline.
In reality, the rotor must be retrimmed after applying AFC in order to maintain a steady
level flight condition. Obtaining the correct trim solution, however, is a computationally
intensive task in AVINOR. Therefore, the effect of AFC on the trim solution is examined
for only one case. Actuation is applied at 4/rev with 𝜙offset = 180◦ since this case produced
the largest reduction in the vibration objective. For this case, the pretrim thrust coefficient
is 5.16 × 10−3 , or 3.2% above the baseline. The rotor was then retrimmed to the original
thrust coefficient of 0.005. The updated trim variables are compared with the original trim
variables in Fig. 6.5a. The differences are minor since the changes produced are less than
1% in the rotor shaft angle, collective, and cyclic pitch settings. The roll angle and tail rotor
collective increase by about 5% each. The effect of retrimming the rotor on the 4/rev hub is
illustrated in Fig. 6.5b. Before trimming the rotor, AFC reduced the vertical shear by 85%
and reduced the other hub forces and moments by 15 to 50%. Trimming the rotor reduced
the vertical shear by an additional 1% and reduced the other components by up to 15%.

120
 10-4
3

Nondim. 4/Rev Hub Load

10
Baseline Baseline (CT =5.00x10 -3 )
AFC, trim 2.5
AFC, pre-trim (CT =5.16x10 -3 )

Trim Variable (°)

2 AFC, trim (CT =5.00x10 -3 )
5
1.5

0 1

0.5

-5 0
r r ar . . .
ea ea he om om om
. Sh t. Sh t .S ll M tch M aw M
ng La r o
R 0 1c 1s R 0t Lo Ve R Pi Y

(a) Trim state (b) Vibratory loads

Figure 6.5: Effect of re-trimming the rotor in forward flight (𝜇 = 0.30) with 4/rev alternating
PS/SS blowing

6.3.2 Angle of Attack

The effect of flow control on the angle of attack of the rotor blades is considered next. The
effective angle of attack is calculated as

𝛼eff (𝑟, 𝜓) = 𝜃 geom (𝑟, 𝜓) + 𝜙torsion (𝑟, 𝜓) − 𝜙inflow (𝑟, 𝜓) (6.2)

where the geometric angle of attack 𝜃 geom (𝑟, 𝜓) is the combination of the blade precone
angle 𝛽 𝑝 , the pretwist angle 𝜃 tw (r), and the pitch setting 𝜃 (𝜓) = 𝜃 0 + 𝜃 1𝑐 cos 𝜓 + 𝜃 1𝑠 sin 𝜓.
The elastic torsional deformation of the blade is expressed as 𝜙torsion (𝑟, 𝜓). The inflow angle
𝜙inflow (𝑟, 𝜓) is due to the contribution of the rotor wake and is expressed as

𝑈𝑃
tan 𝜙inflow = (6.3)
𝑈𝑇

where 𝑈𝑃 and 𝑈𝑇 are the components of the local flow velocity that are normal to the hub
plane and in the hub plane.
The distribution of 𝛼eff over the rotor plane is shown in Fig. 6.6a for the baseline case
with no actuation. The reverse flow region, where flow approaches the blades from the

121
 90° 90°
10° 2°
135° 45° 8° 135° 45°
1.5°
6°
1°
4°
2° 0.5°

180° 0° 0° 180° 0° 0°
Rev. Rev.
Flow -2° Flow -0.5°
-4°
-1°
-6°
-1.5°
225° -8° 225°
315° 315°
-10° -2°
270° 270°

(a) Baseline (𝛼eff ) (b) 4/rev actuation, relative to baseline (Δ𝛼eff )

Figure 6.6: Local angle of attack; oncoming flow direction is left to right (𝜇 = 0.30) and
blades rotate counter-clockwise

trailing edge to the leading edge, is indicated on the retreating side of the rotor. The region
is blanked out because aerodynamic loads in reverse flow are usually ignored. Therefore,
the lift and moment coefficients in this region are set to zero in the AVINOR code. The
change in angle of attack due to flow control, obtained after retrimming the rotor, is shown
in Fig. 6.6b. This corresponds to the case of 4/rev actuation with 𝜙offset = 180◦ that produced
the largest reduction in 4/rev vibratory loads. The angle of attack is reduced by 0.5◦ to 1◦
over several sections of the rotor as a result of actuation. There are also two small regions
near 𝜓 = 90◦ and 𝜓 = 180◦ where the angle of attack increases by about 0.5◦ . These results
indicate that the flow control actuators have a global effect on the state of the rotor that is
not strictly limited to the actuated segment of the blades.

6.3.3 Normal Force

The effect of flow control on the normal force distribution is also examined. The local
normal force coefficient at a blade section is defined using the 2D lift and drag coefficients
as well as the inflow angle:

𝐶𝑛 = 𝐶𝑙 cos 𝜙inflow − 𝐶𝑑 sin 𝜙inflow (6.4)

122
 90° 90°
0.2 0.2
135° 45° 135° 45°

0.1 0.1

180° 0° 0 180° 0° 0
Rev. Rev.
Flow Flow

-0.1 -0.1

225° 315° 225° 315°

-0.2 -0.2
270° 270°

(a) Baseline (𝑀 2𝐶𝑛 ) (b) 4/rev actuation, relative to baseline (𝑀 2 Δ𝐶𝑛 )

Figure 6.7: Local normal force; oncoming flow direction is left to right (𝜇 = 0.30) and
blades rotate counter-clockwise

where the inflow angle is usually assumed to be small. The normal force provides an
indication of the distribution of the rotor thrust over the rotor plane. Figure 6.7 shows the
baseline and controlled cases, plotted as the local Mach number squared times the normal
force coefficient. In the controlled case (Fig. 6.7b), the effect of actuation is most evident
on the actuated segment of the blade, from 69 to 81% span. The normal force changes
replicate the 4/rev actuation signal, where PS blowing increases the normal force and SS
blowing decreases the normal force. A subtler, and more global, effect is evident on the
sections of the rotor where the angle of attack is modified by actuation (Fig. 6.6b). For the
sections where the angle of attack is reduced, the normal force is also reduced. Similarly,
on the small sections of the rotor where the angle of attack increases, the normal force also
increases.

6.3.4 Rotor Torque

The effect of flow control on the rotor torque is also examined. The local torque coefficient
at a blade section is defined in a similar manner as the normal force coefficient using the 2D

123
 90° 90°
0.02 0.02
135° 45° 135° 45°

0.01 0.01

180° 0° 0 180° 0° 0
Rev. Rev.
Flow Flow

-0.01 -0.01

225° 315° 225° 315°

-0.02 -0.02
270° 270°

(a) Baseline (𝑀 2𝐶𝑄 ) (b) 4/rev actuation, relative to baseline (𝑀 2 Δ𝐶𝑄 )

Figure 6.8: Local torque; oncoming flow direction is left to right (𝜇 = 0.30) and blades
rotate counter-clockwise

lift and drag coefficients, the local inflow angle, and the distance 𝑟 from the rotor hub:

𝐶𝑄 = (𝐶𝑙 sin 𝜙inflow + 𝐶𝑑 cos 𝜙inflow ) 𝑟/𝑅 (6.5)

The torque provides an indication of the required power distribution over the rotor plane.
Figure 6.8 shows the baseline and controlled cases, plotted as the local Mach number squared
times the torque coefficient. Similar to the normal force results, the most significant effect
of flow control is displayed on the actuated segment of the blade. The sections of the rotor
where the angle of attack is modified also correspond to changes in the torque. Reduced
torque is associated with reduced angle of attack and increased torque with increased angle
of attack. The overall effect of flow control is a 5.3% increase in the total power required.
Furthermore, this increased power requirement implies a performance penalty. A potential
avenue for reducing this performance penalty is to also incorporate tangential blowing, since
this has been shown to be effective in reducing drag in 2D wind tunnel experiments [105].

124
6.3.5 Sensitivity to Actuator Strength

The sensitivity of the vibratory loads to changes in actuation strength is considered. The
previous results shown were for a jet strength of 𝐶 𝜇,avg = 0.0013, representing half-strength
actuation. In Fig. 6.9, two additional jet strengths are compared with the previous results.
The 4/rev components of the vertical shear force at the rotor hub due to actuation at
𝐶 𝜇,avg = 0.0003, 0.0013, and 0.0028 (representing 25, 50, and 75% of full jet strength)
are shown in Fig. 6.9. For each of the three jet strengths, 4/rev actuation was applied at a
range of phase offsets, within the range of 0◦ ≤ 𝜙offset ≤ 360◦ in 45◦ increments. Figure
6.9a shows the 4/rev sine and cosine components of the vertical shear force, and Fig. 6.9b
shows the 4/rev vertical shear amplitude. It is evident from Fig. 6.9a that increasing the
actuation strength yields greater sensitivity to actuation, as indicated by the larger circle
formed around the baseline point. However, this is not necessarily beneficial for reducing
vibrations since the distance from the origin of the plot (0,0) indicates the amplitude of the
4/rev vibrations. Furthermore, the optimal jet strength for reducing 4/rev vibrations to zero
appears to be slightly less than 50% at 𝜙offset = 180◦ .
The strength of actuation also has an effect on the rotor thrust and power, as illustrated
in Figs. 6.9c and 6.9d. Note that the rotor was not re-trimmed for each of the 24 actuation
cases considered since the largest increment in thrust is only 5% above the baseline. The
difference in thrust will have only a minor effect on the trim state and vibratory loads,
as indicated in Fig. 6.5, and thus the computational expense of re-trimming the rotor for
each case is not warranted. However, the increased thrust contributes to increased power
requirements. Therefore, an approximate calculation to adjust the rotor power for the
difference in thrust is employed so as to isolate the performance penalty due to applying
AFC. This was estimated by calculating the power required for a trimmed rotor operating
at 5% higher thrust coefficient, without AFC. For a 5% increase in thrust coefficient, the
rotor power increases by 2.1%. This relationship between thrust and power in the absence of
AFC is assumed to be linear in the vicinity of the baseline condition (𝐶𝑇 = 0.005). This was

125
confirmed by also computing the trim solution for a rotor operating at 3.2% higher thrust
coefficient, which produced a 1.3% increase in power. Moreover, instead of performing 24
trim calculations, only two additional trim calculations were performed. The rotor power
was adjusted by subtracting the power due to the excess thrust, as estimated by linear
interpolation between the baseline and the 5% higher thrust condition. The adjusted power
is shown in Fig. 6.9d.
The trade-off between vibratory load reduction and rotor performance is illustrated by
Figs. 6.9b and 6.9d. The actuation conditions that reduce vibrations (𝜙offset near 180◦ ) also
yield higher performance penalties, up to 11% higher power. Unlike the vibratory loads, the
performance penalty always increases with stronger actuation, due to the greater in-plane
drag induced on the rotor blades. The trade-off between vibrations and performance has
also been identified in the literature [6] where mechanical actuation was employed to reduce
vibratory loads. Vibratory load reduction has an unfavorable effect on rotor performance
and acoustic noise. Similarly, rotor performance enhancement and acoustic noise reduction
have adverse effects on the vibratory loads.
The sensitivity of the other components of the 4/rev vibratory hub loads to actuation
strength is illustrated in Fig. 6.10. Only the results corresponding to 𝜙offset = 180◦ are shown,
since this phase offset produced the greatest reduction in 4/rev vertical shear for each jet
strength. In all cases, the effect of actuation is to reduce the 4/rev loads below the baseline
levels. This is an encouraging result, since previous open loop simulations with microflaps
have shown that reducing the vertical shear may cause other components, such as the roll,
pitch, and yaw moments, to increase in amplitude if closed loop control with a combined
vibration objective is not employed [7]. The vertical shear force is clearly the most sensitive
to the jet strength.

126
 Nondim. 4/Rev Vert. Shear Amp.
10-4

Nondim. Vert. Shear sin(4 )

5
5
4
=90° 4
3
2 =0°
3
1
10-4 =180°
2
-7 -6 -5 -4 -3 -2 -1 1 2 3 4
-1
-2 1
10-4 =270°
0
Nondim. Vert. Shear cos(4 ) 0 90 180 270 360
Phase Offset (°)
(a) 4/rev vertical shear sine and cosine compo-
(b) 4/rev vertical shear amplitude
nents

10-3 10-4
5.8 4.5

5.6
Thrust Coefficient

Power Coefficient
4
5.4

5.2 3.5

5
3
4.8
0 90 180 270 360 0 90 180 270 360
Phase Offset (°) Phase Offset (°)
(c) Rotor thrust (d) Rotor power, adjusted for difference in thrust

Figure 6.9: Sensitivity of vertical shear force vibrations and rotor power to actuation strength
(4/rev actuation)

6.4 Examination of the Control Sensitivity Matrix for HHC

The control sensitivity matrix T used in the HHC algorithm is examined next. The matrix is
defined in Eq. (5.2) and relates the output variables contained in z (the 4/rev vibratory hub
load components) to the input variables contained in u (the harmonic control inputs). Since
there are 8 input variables and 12 output variables, T is an 8 × 12 matrix. The sensitivity
matrix is constructed by perturbing the control inputs Δu and calculating the changes in
vibratory loads Δz compared to the baseline without control. A period of 6 rotor revolutions
is simulated after each perturbation, to ensure a converged solution.

127
 10-4
2

Nondim. 4/Rev Hub Load

1.5

0.5

0
ar r r . . .
he ea ea om Mom Mom
.S t . Sh t . Sh l lM h w
ng La r o c Ya
Lo Ve R Pit

Figure 6.10: Sensitivity of 4/rev hub loads to actuation strength (4/rev actuation with
𝜙offset = 180◦ )

There are a total of 96 unique entries in the T matrix. It is infeasible to examine how
each of these entries varies with advance ratio 𝜇. However, it is convenient to express T as
an array of several smaller 2 × 2 submatrices:

 
T2,𝐹𝑥4 T2,𝐹𝑦4 T2,𝐹𝑧4 T2,𝑀𝑥4 T2,𝑀𝑦4 T2,𝑀𝑧4 

 
T3,𝐹 T3,𝐹𝑦4 T3,𝐹𝑧4 T3,𝑀𝑥4 T3,𝑀𝑦4 T3,𝑀𝑧4 
𝑥4
T=  (6.6)


T T4,𝐹𝑦4 T4,𝐹𝑧4 T4,𝑀𝑥4 T4,𝑀𝑦4 T4,𝑀𝑧4 
 4,𝐹𝑥4
 
 
T5,𝐹𝑥4 T5,𝐹𝑦4 T5,𝐹𝑧4 T5,𝑀𝑥4 T5,𝑀𝑦4 T5,𝑀𝑧4 
 

Each submatrix contains information relating the sine and cosine components of one input
harmonic to the sine and cosine components of one hub load. For example, the submatrix
relating 2/rev actuation (consisting of the 𝑢 2𝑐 and 𝑢 2𝑠 inputs) to the 4/rev longitudinal shear
(consisting of the 𝐹𝑥4𝑐 and 𝐹𝑥4𝑠 outputs) is expressed as

 𝜕𝐹 𝜕𝐹𝑥4𝑠 

 𝑥4𝑐
 𝜕𝑢 𝜕𝑢 2𝑐 
T2,𝐹𝑥4 =  2𝑐  (6.7)
 𝜕𝐹𝑥4𝑐 𝜕𝐹𝑥4𝑠 
 𝜕𝑢2𝑠 𝜕𝑢 2𝑠 


For the present analysis, the determinant of each 2×2 submatrix is calculated. This yields
24 pieces of information instead of 96. Although the individual sine and cosine components

128
of the 2 × 2 submatrices provide information about the phase differences between inputs and
outputs, the determinant of each submatrix is easier to interpret. Each determinant provides
an overall measure of the sensitivity between one input harmonic and the magnitude of one
hub load component.
The effect of the advance ratio 𝜇 on each sensitivity determinant is illustrated in Fig. 6.11.
Each plot shows the sensitivity of one hub load component as a function of 𝜇, where each
line on the plot represents one control harmonic. The 4/rev longitudinal and lateral shears
(Figs. 6.11a and 6.11b) are most sensitive to 3/rev actuation and become more sensitive as
𝜇 increases. The 3/rev harmonic has the highest sensitivity because in-plane blade loads
occurring at 𝑁/rev are transmitted to the rotor hub as (𝑁 − 1)/rev and (𝑁 + 1)/rev harmonics.
Since the four rotor blades are assumed to be identical in the simulations, only the 4/rev
harmonic of the hub loads is non-zero. Thus, the blade loads generated by 3/rev actuation
are transmitted to the rotor hub as 4/rev in-plane vibrations. By contrast, vertical blade loads
occurring at 𝑁/rev are directly transmitted to the rotor hub at the 𝑁/rev harmonic. Hence,
for low advance ratios, the 4/rev vertical shear force (Fig. 6.11c) is most sensitive to 4/rev
actuation. At higher advance ratios (𝜇 ≥ 0.30), where dynamic stall is present, the vertical
shear becomes more sensitive to 3/rev actuation. The roll and pitching moments are most
sensitive to 3/rev actuation and become less sensitive as 𝜇 increases. The yaw moment
vibrations are relatively insensitive to actuation at any frequency.
The control sensitivities provide some insight into the types of input signals that will be
most effective for reducing vibrations when HHC is employed. Since the vertical shear force
is the largest component of vibrations (e.g., Fig. 6.1), it is expected that the optimal control
signals will be predominantly 4/rev for lower advance ratios or 3/rev for higher advance
ratios.

129
 -6-6-6 -6-6-6 -6-6-6
101010 101010 101010

Shear Sensitivity
Shear Sensitivity
Shear Sensitivity
202020 202020 202020

Shear Sensitivity
Shear Sensitivity
Shear Sensitivity
Shear Sensitivity
Shear Sensitivity
Shear Sensitivity
151515 -6-6-6 151515 -6-6-6 151515 -6-6-6
101010 101010 101010

Long.Sensitivity
Long.Sensitivity
Long.Sensitivity
202020 202020 202020

Vert.Sensitivity
Vert.Sensitivity
Vert.Sensitivity
Lat.Sensitivity
Lat.Sensitivity
Lat.Sensitivity
101010 101010 101010
151515 151515 151515
555 555 555
101010 101010 101010
Mom. Sensitivity Long. Shear
Mom. Sensitivity Long. Shear
Mom. Sensitivity Long. Shear

Mom. Sensitivity Vert. Shear

Mom. Sensitivity Vert. Shear
Mom. Sensitivity Vert. Shear
Mom. Sensitivity Lat. Shear
Mom. Sensitivity Lat. Shear
Mom. Sensitivity Lat. Shear
000 000 000
5 5 50.2
0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 5 5 50.2
0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 5 5 50.2
0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35
Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio
0 0 0 -6-6-6 0 0 0 -6-6-6 0 0 0 -6-6-6
101010 101010 101010
2020200.2
0.2 (a)
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 2020200.2
0.2 (b)
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 2020200.2
0.2
0.2 0.25
0.25(c) 0.30.30.3 0.35
0.25 0.35
0.35
Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance Ratio
Ratio
Ratio2/Rev
2/Rev
2/Rev
151515 -6-6-6 151515 -6-6-6 151515 -6-6-6 3/Rev
3/Rev
3/Rev
101010 101010 101010 4/Rev
4/Rev
4/Rev
202020 202020 202020

PitchSensitivity
PitchSensitivity
PitchSensitivity

YawSensitivity
YawSensitivity
YawSensitivity
RollSensitivity
RollSensitivity
RollSensitivity

5/Rev
5/Rev
5/Rev
2/Rev
2/Rev
2/Rev
101010 101010 101010
151515 151515 151515 3/Rev
3/Rev
3/Rev
4/Rev
4/Rev
4/Rev
555 555 555
5/Rev
5/Rev
5/Rev
101010 101010 101010
Pitch Mom.
Pitch Mom.
Pitch Mom.

Yaw Mom.
Yaw Mom.
Yaw Mom.
Roll Mom.
Roll Mom.
Roll Mom.

000 000 000

5 5 50.2
0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 5 5 50.2
0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 5 5 50.2
0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35
Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio
000 000 000
0.2
0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 0.2
0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 0.2
0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35
Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio
(d) (e) (f)

Figure 6.11: Variation of 4/rev hub load sensitivities to AFC actuation applied at 2, 3, 4, and
5/rev harmonics; results are non-dimensional

6.5 Closed-Loop Control of Vibratory Loads

The HHC approach is employed for closed-loop vibration control using AFC. The HHC
algorithm is based on the minimization of the objective function 𝐽 in Eq. (5.5), which
is a combination of the 4/rev hub shears and moments. The optimal control signal is
determined using the control sensitivity matrix T, subject to actuator saturation constraints.
As mentioned, the classical HHC algorithm uses an invariant T matrix that is identified
offline prior to the closed-loop simulations. The adaptive HHC algorithm, however, uses
a T matrix that is updated using a least-squares approach after each control update in the
simulation. Adaptive HHC is needed for the high-speed case (𝜇 = 0.35) where dynamic stall
significantly affects the prediction of vibratory loads. For the lower advance ratios, adaptive
HHC produces nearly identical results as classical HHC, although at greater computational

130
cost. Therefore, the classical HHC approach is used for 0.20 ≤ 𝜇 ≤ 0.30 and the adaptive
HHC approach is used for 𝜇 = 0.35.

6.5.1 Effect of the Actuator Saturation Limit

Figure 6.12 illustrates the reduction in the vibration objective as a function of the saturation
limit. The saturation limit is shown on the horizontal axis in terms of the jet momentum coef-
ficient. The corresponding power required for actuation, calculated according to Eq. (5.26),
is shown on the right vertical axis. Alternatively, the power required for actuation can be
interpreted as the power available for actuation, since it also represents the saturation limit
enforced in the HHC algorithm. The slope of the vibration objective curve is initially quite
steep, indicating that a small increase in the saturation limit yields a large reduction in
vibrations. As the saturation limit further increases, the vibration objective curve begins
to level out. The maximum vibration reduction achieved is 83% below the baseline. Note
that the closed-loop results using a saturation limit of 𝐶 𝜇,avg = 0.0054 (𝑃jet = 3.6 kW/blade)
are identical to the results using a limit of 𝐶 𝜇,avg = 0.0044 (𝑃jet = 2.7 kW/blade). Hence,
the power available for actuation does not need to exceed 𝑃jet = 2.7 kW/blade for vibration
reduction, since the HHC algorithm does not utilize all the available power when the satu-
ration limit is above this value. A saturation limit of 1.2 kW/blade yields 80% reduction
of vibrations, which is a reasonable compromise between vibration reduction and power
penalty.
The optimal control input signals computed by the HHC algorithm for various saturation
limits are illustrated in Fig. 6.13. Both the continuous and discrete versions of each signal
are shown. The closed-loop control signals are predominantly 4/rev. This is consistent
with the open-loop results, which indicated that the vibratory loads are most sensitive to
4/rev actuation. In Figs. 6.13a and 6.13b, the power available for actuation is low, and
the HHC algorithm appears to utilize only 4/rev actuation. When the available power
is increased, the HHC algorithm employs a combination of all the 2-, 3-, 4-, and 5/rev

131
 % Change in Vibration Objective

Actuation Power per Blade (kW)

0 4

-20
3

-40
2
-60

1
-80

-100 0
0 1 2 3 4 5 6
Momentum Coefficient 10-3

Figure 6.12: Sensitivity of closed-loop vibratory load reduction to actuator saturation limit
(𝐶 𝜇,avg ) enforced during HHC; variation of actuation power required is also shown

harmonics. Consequently, the actuators must be switched more often, to reproduce the four
superimposed harmonics.

6.5.2 Comparison with the Microflap

The closed-loop vibration reduction capability of AFC is compared with that of the microflap.
Both devices cover 12% blade span and are centered at the 75% span location on each blade.
The saturation limit for the microflap deflection is 1.5%𝑐, and the saturation limit for AFC
is selected as 𝑃jet = 1.2 kW/blade. The vibratory hub loads are compared in Fig. 6.14. The
reduction of 4/rev vertical shear is similar for both devices, since the microflap produces an
88% reduction and AFC produces a 92% reduction. The AFC actuators are less effective
than the microflap for reducing the lateral shear, roll moment, and yaw moment vibrations.
Overall, the microflaps reduce the vibratory loads by 86% and AFC reduces vibratory loads
by 80%.
The effects of the microflap and AFC on the sectional normal force at 73% blade span
are illustrated in Fig. 6.15. The control input signals obtained from the HHC algorithm
are also shown. The results show that despite the differences in the actuation signals used

132
 1 Continuous AFC Discrete AFC 1

Control Input

Control Input
0 0

-1 -1
0 90 180 270 360 0 90 180 270 360
Azimuth Angle (°) Azimuth Angle (°)

(a) 𝑃jet = 0.1 kW/blade (b) 𝑃jet = 0.4 kW/blade

1 1
Control Input

Control Input
0 0

-1 -1
0 90 180 270 360 0 90 180 270 360
Azimuth Angle (°) Azimuth Angle (°)

(c) 𝑃jet = 1.2 kW/blade (d) 𝑃jet = 2.7 kW/blade

Figure 6.13: Continuous and discrete AFC input signals computed by HHC with various
saturation limits on the actuation power 𝑃jet

by each device, the unsteady aerodynamic responses are remarkably similar. The normal
force induced by the microflap is shifted upward compared to AFC since the microflap is
only deployed in one direction on the PS of the blade, whereas the AFC actuators have
bi-directional control authority. Also, the AFC response contains some additional high-
frequency content, due to the discrete nature of the input signal. Otherwise, the responses
are nearly identical. The distribution of the normal force coefficient over the rotor plane is
illustrated in Fig. 6.16 for the baseline and controlled cases. On the actuated segment of
the blade, the normal force changes replicate the predominantly 4/rev closed-loop actuation
signal, where PS actuation increases the normal force and SS actuation decreases the normal
force. The normal force is reduced on several other sections of the rotor due to a reduction
in the blade angle of attack from re-trimming the rotor. The normal force is reduced by the
greatest margin near the tip on the advancing blade.
The effects of the microflap and AFC on the rotor torque are also examined. Figure 6.17
shows the distribution of the local torque coefficient over the rotor plane for the baseline

133
 10-4
2

Non-Dim. 4/rev Hub Loads

Baseline
Microflap
1.5 AFC

0.5

0
ar ar ar . . .
he he he Mom Mom Mom
. S t . S t . S ll h w
Lo
ng La Ve
r Ro Pit
c Ya

Figure 6.14: Comparison of 4/rev vibratory hub load reduction using the microflap and AFC
with a saturation limit of 𝑃jet = 1.2 kW/blade

and controlled cases. The effects of actuation are most evident on the actuated segment
of the blade, from 69 span to 81% span. Both the microflap and the AFC actuators show
some evidence of the predominantly 4/rev actuation signals employed for control. The
microflap induces a greater torque (shown in red) on the actuated segment than the AFC
actuators, due to the drag penalty associated with deploying the microflap. However, the
microflap also reduces the torque (shown in light blue) on the other sections of the rotor,
due to a slight reduction in the blade angle of attack. The AFC actuators also reduce the
torque on these sections of the rotor, but to a lesser extent than the microflap. Overall, the
microflaps increase the total power required to overcome in-plane drag forces by 4.6%,
whereas the AFC actuators increase the power required by 5.6%. The increased power
requirements imply a performance penalty. Previously, the calculation of the performance
penalty induced by microflaps during vibration and noise reduction [7, 31] has not included
the power required for actuation. The total performance penalty induced by AFC, including
the actuation power required, is examined in the next section.

134
 0.25
Baseline

Non-Dim. Normal Force

Microflap
0.2 AFC

0.15

0.1

0.05

0
0 90 180 270 360
Azimuth Angle (°)
(a) Normal force (𝑀 2𝐶𝑛 )
Microflap Deflection (%c)

1.5

0.5

-0.5
0 90 180 270 360
Azimuth Angle (°)

(b) Microflap input signal

1
AFC Input

-1
0 90 180 270 360
Azimuth Angle (°)

(c) AFC input signal

Figure 6.15: Non-dimensional normal force evaluated at 73% blade span from closed-loop
simulations using the microflap and AFC with a saturation limit of 𝑃jet = 1.2 kW/blade

6.5.3 Rotor Performance Penalty

Reducing vibrations in the rotor has an adverse effect on the rotor power required. This
issue has been identified in the literature for applications involving vibration reduction using
trailing edge flaps and microflaps [8]. The total performance penalty associated with AFC

135
 90° 90°
0.2 0.2
135° 45° 135° 45°

0.1 0.1

180° 0° 0 180° 0° 0
Rev. Rev.
Flow Flow

-0.1 -0.1

225° 315° 225° 315°

-0.2 -0.2
270° 270°

(a) Baseline (𝑀 2𝐶𝑛 ) (b) Microflap, relative to baseline (𝑀 2 Δ𝐶𝑛 )

90°
0.2
135° 45°

0.1

180° 0° 0
Rev.
Flow

-0.1

225° 315°
-0.2
270°

(c) AFC, relative to baseline (𝑀 2 Δ𝐶𝑛 )

Figure 6.16: Non-dimensional normal force coefficient for uncontrolled case (baseline)
and closed-loop control using the microflap and AFC with a saturation limit of 𝑃jet = 1.2
kW/blade; oncoming flow direction is left to right (𝜇 = 0.30) and blades rotate counter-
clockwise

actuation is estimated by combining the power due to the in-plane drag penalty with the
power required for actuation. Previous studies involving rotor vibration reduction have not
included the power required for actuation. Including this additional power requirement
provides a more realistic estimate of the overall cost to rotor performance. Figure 6.18
illustrates the increments in rotor power due to the drag penalty, the actuation power, and the
combined performance penalty associated with vibration reduction. The power increment,
shown in Fig. 6.18b, is calculated as a percentage of the baseline power required to overcome
in-plane drag forces on the rotor without AFC. When the available actuation power is small,

136
 90° 90°
0.02 0.02
135° 45° 135° 45°

0.01 0.01

180° 0° 0 180° 0° 0
Rev. Rev.
Flow Flow

-0.01 -0.01

225° 315° 225° 315°

-0.02 -0.02
270° 270°

(a) Baseline (𝑀 2𝐶𝑄 ) (b) Microflap, relative to baseline (𝑀 2 Δ𝐶𝑄 )

90°
0.02
135° 45°

0.01

180° 0° 0
Rev.
Flow

-0.01

225° 315°
-0.02
270°

(c) AFC, relative to baseline (𝑀 2 Δ𝐶𝑄 )

Figure 6.17: Non-dimensional torque coefficient for uncontrolled case (baseline) and closed-
loop control using the microflap and AFC with a saturation limit of 𝑃jet = 1.2 kW/blade;
oncoming flow direction is left to right (𝜇 = 0.30) and blades rotate counter-clockwise

the performance penalty is mostly due to the drag penalty. As the actuation power increases,
it represents a larger contribution to the overall performance penalty. Furthermore, Fig. 6.18
illustrates that, by using smaller jet strengths, it is possible to achieve moderate vibration
reduction levels while keeping the performance penalty low. The penalty associated with
increasing the actuation power from 1.2 kW/blade to 2.7 kW/blade far outweighs the
benefit, since vibrations are reduced from 80% to 83% below the baseline, but the overall
performance penalty increases from 7.2% to 11.8% above the baseline.
The component of the performance penalty due to actuation power required is a dif-

137
 % Change in Vibration Objective
0 15
Drag Penalty
Actuation Power

(% of Baseline Power)
-20 Total = Drag + Actuation

Power Increment
10
-40

-60
5

-80

-100 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Actuation Power per Blade (kW) Actuation Power per Blade (kW)
(a) Vibration reduction (b) Performance penalty

Figure 6.18: Performance penalty associated with vibration reduction using closed-loop
AFC, including contributions from in-plane drag and power required for actuation

ficult quantity to find in the active rotor control literature. It is generally noted, however,
that higher-harmonic control of the swashplate and individual blade pitch control require
substantial control effort compared to on-blade actuation methods such as the active twist
rotor, partial-span trailing edge flaps, and microflaps [8]. The component of the performance
penalty due to in-plane drag (i.e., rotor torque) is more easily compared with other vibration
control approaches. In the previous section, microflaps displayed a similar level of drag
penalty to AFC when reducing vibrations to approximately 80% below baseline levels. The
microflaps induced a 4.6% penalty on the rotor power and AFC induced a 5.6% penalty.
The performance penalty of the NASA/Army/MIT Active Twist Rotor (ATR) [21] is also
compared with the AFC results, since vibration reduction was performed at a similar flight
condition (𝜇 = 0.333, 𝐶𝑇 /𝜎 = 0.068). In wind tunnel experiments, the ATR produced a
torque penalty of 2.3% when 4/rev hub load vibrations were reduced by approximately 60%.
This torque penalty is comparable to AFC when vibrations are reduced by a similar margin.
Interpolation of the AFC results in Fig. 6.18 indicates a 2.5% penalty when vibrations are
reduced by 60%.

138
6.5.4 Effect of the Rotor Advance Ratio

Figure 6.19 illustrates the vibration reduction due to closed-loop AFC as a function of
the advance ratio. For all simulations performed, the saturation limit for the actuators
was selected as 𝑃jet = 1.2 kW/blade since this jet strength yields a good balance between
vibration reduction and performance penalty. The baseline and controlled levels of the
vibratory hub load components are compared. The vertical shear force is reduced by the
greatest margin among the various hub load components: it is reduced by 81 to 86% below
the baseline levels. Also, the controlled level of vertical shear vibrations is nearly constant
across the range of advance ratios. This is an encouraging result which demonstrates
the effectiveness of the AFC approach in a variety of flow conditions. Furthermore, the
reduction of longitudinal and lateral shear vibrations improves as 𝜇 increases, indicating
that the adverse effects of dynamic stall are alleviated by AFC.
The optimal control input signals computed by the HHC algorithm are compared in
Fig. 6.20 for the four advance ratios. The continuous and discrete versions of each signal are
shown. The signals are all predominantly 4/rev, with minor variations due to superposition
with other harmonics. In Fig. 6.21, the signals are decomposed in terms of the amplitude
and phase of each harmonic (2-, 3-, 4-, and 5/rev) as a function of 𝜇. The 4/rev harmonic
is consistently the largest component of the signals. The 4/rev phase increases gradually
from 136◦ to 178◦ as 𝜇 increases. At low advance ratios (𝜇 < 0.30), the 5/rev harmonic is
the second-largest component of the signal. This is somewhat unexpected since the control
sensitivity matrices indicated that vibratory loads are relatively insensitive to 5/rev actuation
(Fig. 6.11). Therefore, the 5/rev harmonic is likely only needed to satisfy the saturation limit
constraint. The contribution of the 3/rev harmonic is only significant when 𝜇 = 0.35. This is
necessary to reduce the higher in-plane vibratory loads generated by dynamic stall.
Figure 6.22 illustrates the rotor performance penalty associated with active vibration
reduction. Figure 6.22a shows the overall change in the vibration objective, which is
a combination of all the 4/rev hub load components. The power increment, shown in

139
 -4-4-4 -4-4-4 -4-4-4
101010 101010 101010
222 222 222

Long. Shear
Long. Shear
Long. Shear
1.5
1.5
1.51010
-4-4-4 1.5
1.5
1.51010
-4-4-4 1.5
1.5
1.51010
-4-4-4

Vert. Shear
Vert. Shear
Vert. Shear
10 10 10

Lat. Shear
Lat. Shear
Lat. Shear
222 222 222

111 111 111

Long. Shear
Long. Shear
Long. Shear
1.5
1.5
1.5 1.5
1.5
1.5 1.5
1.5
1.5

Vert. Shear
Vert. Shear
Vert. Shear
Lat. Shear
Lat. Shear
Lat. Shear
0.5
0.5
0.5 0.5
0.5
0.5 0.5
0.5
0.5
111 111 111

000 000 000

0.50.50.2
0.5 0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 0.50.50.2
0.5 0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 0.50.50.2
0.5 0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35
Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio
-4-4-4
0 0 01010 -4-4-4
0 0 01010 -4-4-4
0 0 01010
10 10 10
2 2 20.2
0.2 (a)
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 2 2 20.2
0.2 (b)
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 2 2 20.2
0.2
0.2 0.25 (c) 0.30.30.3 0.35
0.25
0.25 0.35
0.35
Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance Baseline
Baseline
Ratio
Ratio Baseline
Ratio
1.5
1.5
1.51010
-4-4-4 1.5
1.5
1.51010
-4-4-4 1.5
1.5
1.51010-4-4-4 Controlled
Controlled
Controlled

Pitch Mom.
Pitch Mom.
Pitch Mom.

Yaw Mom.
Yaw Mom.
Yaw Mom.
10 10 10
Roll Mom.
Roll Mom.
Roll Mom.

222 222 222

Baseline
Baseline
Baseline
111 111 111
1.5
1.5
1.5 1.5
Pitch Mom.1.5
1.5 1.5
1.5
1.5 Controlled
Controlled
Controlled
Pitch Mom.
Pitch Mom.

Yaw Mom.
Yaw Mom.
Yaw Mom.
Roll Mom.
Roll Mom.
Roll Mom.

0.5
0.5
0.5 0.5
0.5
0.5 0.5
0.5
0.5
111 111 111
000 000 000
0.50.50.2
0.5 0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 0.50.50.2
0.5 0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 0.50.50.2
0.5 0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35
Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio
000 000 000
0.2
0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 0.2
0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35 0.2
0.2
0.2 0.25
0.25
0.25 0.3
0.3
0.3 0.35
0.35
0.35
Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio Advance
Advance
Advance
Ratio
Ratio
Ratio
(d) (e) (f)

Figure 6.19: Baseline and controlled 4/rev hub loads obtained using closed-loop AFC at
various advance ratios; results are non-dimensional

Fig. 6.22b, is calculated as a percentage of the baseline power required to overcome in-plane
drag forces on the rotor without AFC. Note that the baseline conditions are different for each
advance ratio. The baseline power required increases with 𝜇 since the rotor must produce an
increasing level of forward thrust. Therefore, the power increment is calculated relative to
the baseline conditions at each advance ratio.
The level of vibration reduction is consistent across the range of advance ratios, varying
between 70% and 80% below the baseline. The total power increment, however, grows as a
function of the advance ratio. The component due to drag increases rapidly from 2.5% to
8.8% as the advance ratio increases from 𝜇 = 0.20 to 𝜇 = 0.35. As 𝜇 increases, the mean
effective angle of attack on the rotor blades increases. This is indicated by the collective
pitch setting required for trim (Fig. 6.2). The amplitude of the 1/rev angle of attack variations
also increases, as indicated by the cyclic pitch settings required for trim. Furthermore, when

140
 1 Continuous AFC Discrete AFC 1

Control Input

Control Input
0 0

-1 -1
0 90 180 270 360 0 90 180 270 360
Azimuth Angle (°) Azimuth Angle (°)

(a) 𝜇 = 0.20 (b) 𝜇 = 0.25

1 1
Control Input

Control Input
0 0

-1 -1
0 90 180 270 360 0 90 180 270 360
Azimuth Angle (°) Azimuth Angle (°)

(c) 𝜇 = 0.30 (d) 𝜇 = 0.35

Figure 6.20: Continuous and discrete AFC input signals computed by HHC algorithm for
vibration reduction at various advance ratios

AFC is employed at high angles of attack, the drag penalty increases significantly. Note that
the power required for actuation is constant among the four cases considered since the same
actuator saturation limit is used. The relative contribution of the actuation power, however,
decreases as 𝜇 increases since the baseline power increases with increasing 𝜇.

141
 Input Amplitude
2/Rev 3/Rev 4/Rev 5/Rev
0.6

0.4

0.2

0
0.2 0.25 0.3 0.35
Advance Ratio

(a)
180°
Input Phase (°)
90°

0°

-90°

-180°
0.2 0.25 0.3 0.35
Advance Ratio

(b)

Figure 6.21: Amplitude and phase of 2, 3, 4, and 5/rev harmonics computed by HHC
algorithm for closed-loop vibration reduction using AFC at various advance ratios
% Change in Vibration Objective

0 15
Drag Penalty
Actuation Power
(% of Baseline Power)

-20 Total = Drag + Actuation

Power Increment

10
-40

-60
5

-80

-100 0
0.2 0.25 0.3 0.35 0.2 0.25 0.3 0.35
Advance Ratio Advance Ratio
(a) Vibration reduction (b) Performance penalty

Figure 6.22: Vibration reduction and associated performance penalty using closed-loop
AFC at various advance ratios

142
 CHAPTER 7

Concluding Remarks

This study has demonstrated through computational simulations the potential of active flow
control for vibratory load reduction on a helicopter rotor. CFD simulations were employed
to model and characterize a novel trailing-edge actuator concept that employs fluid blowing
on the pressure side and suction side of the rotor blades. Furthermore, the CFD simulations
were validated using data from recent wind tunnel experiments performed at Georgia Tech.
A CFD-based ROM that captures the unsteady aerodynamic effects introduced by AFC was
constructed and integrated into a comprehensive aeroelastic rotor simulation code. Closed-
loop simulations based on the HHC algorithm which account for the discrete operation of
the AFC actuators were developed to actively reduce vibratory hub loads. The vibration
reduction and unsteady aerodynamic loads produced by AFC were also compared with a
microflap. Finally, the effect of AFC on the overall rotor performance, including the power
required for actuation, was calculated to illustrate the performance penalty associated with
vibration reduction.

7.1 Conclusions and New Contributions

The main conclusions and contributions of this dissertation are summarized next:

1. The PS/SS normal jets produce bi-directional changes in the airfoil lift and moment
that are proportional to the relative strength of the jets, as indicated by the square root

143
 p
of the momentum coefficient, 𝐶 𝜇 .

2. At low angles of attack, the lift and moment induced by SS normal blowing are
relatively insensitive to angle of attack, but the effect of PS normal blowing tends to
increase in magnitude as 𝛼 increases.

3. Tangential blowing has a minor effect on the lift and moment at moderate angles of
attack where flow separation is present, and has a beneficial effect on the drag.

4. The 2D CFD simulations show good agreement with experimental data in terms of
the lift and moment coefficients induced by normal and tangential blowing.

5. Results obtained from more costly 3D CFD simulations indicate that 3D flow effects
develop near the actuator edges and reduce the effectiveness of actuation, compared
to 2D predictions.

6. The actuator control authority improves at higher freestream Mach numbers when an
equal relative jet strength is applied; however, there is an associated increase in drag
penalty.

7. The actuators retain their effectiveness under dynamic conditions mimicking heli-
copter rotor vibrations. However, the unsteadiness associated with blade motion
produces significant variations in the loads which cannot be predicted using a quasi-
steady approximation.

8. Applying a discrete pseudorandom actuation signal in CFD simulations is an efficient

method for generating unsteady aerodynamic training data for ROM construction.
Discrete pseudorandom signals are compatible with the discrete operating characteris-
tics of the AFC actuators and excite a broadband aerodynamic response in a relatively
short time period.

144
 9. The ROM, based on unsteady aerodynamic training data obtained from full-order CFD
simulations across a broad range of flow conditions, produces a close approximation
of the lift, moment, and drag predicted in CFD simulations at a fraction of the
computational cost. For a validation case consisting of coupled pitch/plunge motion
in an unsteady freestream with alternating PS/SS actuation, the ROM run time is
5.5 s on a conventional laptop PC, compared to a CFD run time of 63 h on eight
high-performance cores. This represents a reduction in computer run time by more
than four orders of magnitude.

10. Open-loop control simulations involving alternating PS/SS actuation indicated that
when the 4/rev components of the vibratory hub loads are combined as a single
vibration objective, 4/rev actuation is the most effective for reducing the overall level
of vibrations.

11. Secondary effects associated with vibratory load reduction include reduced angle of
attack on the blades, increased rotor thrust, and increased rotor power requirements.

12. The entries of the control sensitivity matrix T used in the HHC algorithm vary
significantly depending on the rotor advance ratio. The 4/rev in-plane hub shear forces
are most sensitive to 3/rev actuation and become more sensitive as 𝜇 increases. The
4/rev vertical hub shear force is most sensitive to 4/rev actuation at low advance ratios
but becomes more sensitive to 3/rev actuation at higher advance ratios where dynamic
stall is present.

13. The discrete operating characteristics of the AFC actuators were incorporated into
closed-loop control simulations by temporarily relaxing the requirement that each
actuator (PS/SS) be on or off at a fixed jet strength. The HHC algorithm yields
a continuous-valued control input signal. Converting this continuous signal into a
discrete signal using delta modulation yields only minor differences in the resulting
aerodynamic and vibratory loads.

145
 14. Selection of a saturation limit for the AFC actuators when performing closed-loop
control should be based on the power available for actuation. The sensitivity of the
vibration reduction to the saturation limit was examined. A maximum vibration
reduction of 83% below the baseline was achieved using an actuation power of 2.7
kW/blade. Therefore, the actuation power does not need to be greater than this value.

15. The vibration reduction capability of the AFC actuators is similar to that of a microflap.
Both devices are capable of reducing vibrations by more than 80%. When employed
for vibration reduction, the AFC actuators and microflaps increase the power required
to overcome in-plane drag forces on the rotor by 5.6% and 4.6%, respectively.

16. The total performance penalty associated with AFC when reducing vibrations was
calculated by combining the drag penalty and the power required for actuation. Mod-
erate levels of vibration reduction can be achieved using low actuation power with
minimal impact on the overall performance. However, the improvement in vibration
reduction diminishes when stronger actuation is employed, and the cost to overall
performance grows substantially.

17. The closed-loop AFC approach is effective for reducing vibrations at all advance
ratios considered (0.20 ≤ 𝜇 ≤ 0.35). The 4/rev vertical shear was reduced by more
than 80% in all cases. The reduction of the 4/rev in-plane shear force vibrations
improved with increasing advance ratio, indicating that the adverse effects of dynamic
stall were alleviated.

18. Although the overall level of vibration reduction was consistent around 70% to 80%
across the range of advance ratios, the associated performance penalty increases with
increasing advance ratio, due to higher drag penalty associated with fluidic actuation.

The open-loop and closed-loop simulations performed in this study have shed a new
light on the potential of AFC for vibration reduction on a helicopter rotor. Furthermore,

146
the simulations have illustrated that AFC is a potentially viable option for vibration control
that is comparable to the microflap. This is the first study to demonstrate the potential of
closed-loop AFC in forward flight when the rotor is at moderate angles of attack encountered
in level cruising flight.

7.2 Recommendations for Future Research

The computational simulations performed in this dissertation represent the first comprehen-
sive study of AFC for rotorcraft vibration reduction. Examination of the control authority
and associated performance penalty has illustrated the potential of the AFC concept. Fur-
thermore, there are several avenues for future research.
The aeroelastic computations performed in the AVINOR code have been shown in
previous studies (e.g., Refs. [6, 23]) to be sufficiently accurate in representing experimental
data from the HART experiments [131]. The AVINOR code represents a mid-fidelity
analysis which has a level of computational efficiency suitable for closed-loop control
simulations. Potential enhancements to the aerodynamic load calculations include the
viscous vortex particle method [134] for improved unsteady wake modeling, full-rotor
CFD calculations [30], and coupled CFD-CSD aeroelastic calculations [32]. Use of such
high-fidelity modeling tools may provide additional insight about the complex aerodynamic
interactions that occur between the rotor, the AFC actuators, and the rotor wake.
The simulations performed in this study have demonstrated the ability of AFC to reduce
vibratory loads when the actuators are placed near the 75% span location on the blades.
Optimization of the actuator placement and utilization of multiple actuators on each blade
can potentially improve the level of vibration reduction and/or reduce the performance
penalty.
Finally, verification of the computational results through experiments is an important
area for future research. Given the novelty of the fluidic actuators employed in this study,

147
experimental data has been limited to wind tunnel experiments on a 2D airfoil. Further
testing on a rotating blade is needed in order to examine actuator performance in the presence
of 3D unsteady aerodynamic effects and centrifugal forces.

148
 APPENDIX A

Pseudorandom Signal Generation

The method for generating a discrete-valued pseudorandom signal using a maximum-length

sequence is described.
A maximum-length sequence is a pseudorandom periodic signal on a discrete-time
grid that resembles random noise, but is generated from a deterministic algorithm. For a
𝑞-level sequence, the sample values are taken only from a finite set of values representing
the 𝑞 possible levels of the signal. For example, the three levels of a ternary signal are
usually assumed to be −1, 0, and +1. A discrete sequence is typically converted into a
continuous-time signal for system identification by applying a zero-order hold (ZOH) to the
sequence, which produces a piecewise-continuous signal. The ZOH “holds” each sample
value from the sequence for one clock period 𝑇𝑐 , such that the value of the continuous-time
signal is constant over that sampling interval. The method for generating maximum-length
sequences is described in this Appendix, with emphasis on ternary signals.
A 𝑞-level sequence is generated using a linear feedback shift register, which consists of
𝑛 stages, or bins, that each hold a logical value in the range of 0, 1, 2, . . . , (𝑞 − 1). This set
of logical values constitutes a finite field of order 𝑞. A finite field refers to a collection of 𝑞
elements for which addition and multiplication are accomplished by modulo-𝑞 arithmetic.
For example, a ternary finite field has three logical elements (0, 1, and 2) with addition
and multiplication operations defined according to Table A.1. These operations represent
modulo-3 arithmetic.

149
 Table A.1: Sum and product tables for ternary finite field

+ 0 1 2 × 0 1 2
0 0 1 2 0 0 0 0
1 1 2 0 1 0 1 2
2 2 0 1 2 0 2 1

To generate a sequence, the 𝑛 stages of the shift register are initialized with any combi-
nation of logical values (except 𝑛 zeros). The register state at time 𝑡 𝑘 is denoted as

r(𝑡 𝑘 ) = [𝑟 1 (𝑡 𝑘 ) 𝑟 2 (𝑡 𝑘 ) ... 𝑟 𝑛 (𝑡 𝑘 )] T , for 𝑡 𝑘 = 𝑘𝑇𝑐 , 𝑘 = 0, 1, 2, . . . (A.1)

and the output of the register 𝑦(𝑡 𝑘 ) is taken as the value of the final stage 𝑟 𝑛 (𝑡 𝑘 ). The register
will advance through all possible combinations of register states before returning to the
initial state of the register r(0), thus generating a periodic sequence as the output 𝑦(𝑡).
The discrete-time equations that govern the evolution of the register state are expressed
compactly in state-space form:

r(𝑡 𝑘+1 ) = A r(𝑡 𝑘 ) (A.2)

𝑦(𝑡 𝑘 ) = C r(𝑡 𝑘 ) (A.3)

where

𝑎 1 𝑎 2 · · · 𝑎 𝑛−1 𝑎 𝑛 
 
 
 
1 0 ··· 0 0 
 
 
A =  0 1 · · · 0 0  (A.4)
 
 .. .. . . .. .. 
. . . . .
 
 
0 0 ··· 1 0 
 
h i
C = 0 0 ... 0 1 (A.5)

150
 The selected values for the feedback coefficients 𝑎 1 , . . . , 𝑎 𝑛 determine whether the
resulting sequence is a maximum-length sequence. A maximum-length sequence is the
longest possible periodic sequence that can be generated from an 𝑛-stage feedback shift
register, and has a length of 𝑞 𝑛 − 1. In general, most combinations of feedback coefficients
do not produce maximum-length sequences. Reference [118] provides several tabulated
values of the appropriate feedback coefficients for 𝑞-level 𝑛-stage registers. For example,
consider a three-stage register (𝑛 = 3). A ternary maximum-length sequence can be formed
by using {𝑎 1 , 𝑎 2 , 𝑎 3 } = {1, 2, 2}. The matrix-vector multiplication in Eq. (A.2) in this case is
performed using modulo-3 arithmetic. As a result, the register generates a periodic sequence
as the output repeats itself every 33 − 1 = 26 time steps. Note that the sequence shown in
Fig. 3.4 was generated in this manner.
In order to convert a sequence into a continuous-time signal for system identification,
the logical values from the output sequence 𝑦(𝑡) are first mapped to the desired signal
levels. This is a fairly simple operation. For a ternary signal, the logical values are either
0, 1, or 2; and they map to 0, +1, and −1, respectively. After this mapping is performed,
the zero-order hold operation is applied to the discrete-time sequence, which produces a
piecewise-continuous pseudorandom signal.

151
 APPENDIX B

Structural Dynamic Model and Generalized

Loads

B.1 Modeling Assumptions

The basic assumptions used to develop the aeroelastic analysis model for the rotor blade are
as follows:

1. The rotor blade is cantilevered at the hub, with a root offset 𝑒 from the axis of rotation
(see Fig. B.2).

2. The blade has a precone angle 𝛽 𝑝 (see Fig. B.2) and it has built-in pretwist distribution
𝜃 𝑡𝑤 about the elastic axis of the blade.

3. The blade has no sweep, droop or torque offset.

4. The blade cross section is assumed to be symmetric with respect to its major principal
axes.

5. The blade feathering axis is coincident with the elastic axis.

6. The deflections in the blade are assumed to be moderate and the strains to be small.

7. The blade has coupled flap, lead-lag, torsional and axial dynamics.

152
 8. The blade is assumed to be inextensible.

9. The rotor shaft is assumed to be rigid and body degrees of freedom are suppressed.

10. The structural effects of the AFC actuators and the microflap are neglected.

11. The distributed aerodynamic loads are obtained using CFD-based RFA aerodynamic
model.

12. The induced inflow is nonuniform and is obtained by a free wake analysis included in
the aeroelastic model.

13. Reverse flow effects are included by setting the lift and moment equal to zero inside
the reverse flow region.

14. The speed of rotation Ω of the rotor is constant.

15. The helicopter is in trimmed, steady, level flight. Either propulsive or wind tunnel
trim can be implemented.

B.2 Coordinate Systems

The following six coordinate systems are used to formulate the aeroelastic model:

“0” System: This is an inertial reference frame with origin at the hub center 𝑂 𝐻 oriented
such that the gravitational vector is aligned along the negative 𝑧 0 axis. The tail of the
helicopter is assumed to lie in the 𝑥 0 𝑧0 plane in the direction of the positive 𝑥0 axis.

“1” System: This is an inertial reference frame with origin at 𝑂 𝐻 . The 𝑦 1 axis is coincident
with the 𝑦 0 axis, and the 𝑧1 axis pitched forward at an angle 𝛼 𝑅 about the 𝑦 0 axis
so that it is aligned with the rotor axis of rotation. The “1” system provides the
non-rotating reference frame. The “0” and “1” systems are depicted in Fig. B.1.

153
 THRUST TR

DEFORMING BLADE Z1 Z0

X1
OH
X0

NE
L HUB PLA ZFC
NOMINA
DRAG
VELOCITY (V)

aR

GRAVITY (g) WEIGHT

XFC

Figure B.1: Transformation from the “0” system to the “1” system

“2” System: This system has its origin at 𝑂 𝐻 . The 𝑧 2 axis is coincident with the 𝑧1 axis
but rotates with the blades about the 𝑧1 axis. The “2” system is the rotating reference
frame.

“3” System: This system rotates with the blades and has its origin at the blade root located
at a distance 𝑒 away from the axis of rotation along the 𝑥 2 axis, as shown in Fig. B.2.
The 𝑥 3 axis is “preconed" by an angle 𝛽 𝑝 around the 𝑦 2 axis such that the 𝑥 3 axis lies
along the undeformed elastic axis of the blade. The principal axes of the undeformed
blade cross-section at any point along the span lie in a plane parallel to the 𝑦 3 𝑧3 plane
and are oriented at an angle 𝜃 𝐺 (𝑥) about the 𝑥3 axis. Angle 𝜃 𝐺 (𝑥) is the sum of
collective and cyclic pitch inputs at the root and geometric pretwist of the blade at the
particular point along the span. The “3” system provides the undeformed reference
frame.

“4” System: This is a blade attached system. Before deformation, the “3” and “4” systems
are parallel. After deformation, the “4” system is translated and rotated such that the
𝑥 4 axis is tangent to the elastic axis of the blade at each blade cross-section along the

154
 Z2 Z4
DEFORMED BLADE
X4
UNDEFORMED BLADE
( FEATHERING AXIS )
A’
W w
Z3 u X3
A
bP
X2
e
x

REAR VIEW

Y2
Y4
DEFORMED BLADE

Y3 A’ X4

v
W
X3
UNDEFORMED BLADE A
e
TOP VIEW

Figure B.2: The transformation from the “2” system to the “4” system

span. The principal axes of the blade cross-section lie in the 𝑦 4 𝑧4 plane, rotated at an
angle 𝜃 𝐺 (𝑥) about the 𝑥 4 axis. The “4” system provides the deformed reference frame.
The relationships between the “2”, “3” and “4” systems are depicted in Fig. B.2.

“5” System: This is also a blade attached system and represents the “4” system with the
torsional deformation removed, as shown in Fig. B.3. Thus, the principal axes of
the blade cross section are rotated at an angle 𝜃 𝐺 (𝑥) + 𝜙(𝑥) about the 𝑥5 axis where
𝜙(𝑥) is the elastic twist. This reference frame is convenient in the derivation of the
aerodynamic loads. The relationships between the “3”, “4” and “5” systems are

155
 Z4 Z5 hb

qG
Y4 f
Z3
Y5
hb
DEFORMED BLADE
qG
CROSS SECTION
Y3

UNDEFORMED BLADE
CROSS SECTION

Figure B.3: The transformation from the “3” system to the “5” system

depicted on Fig. B.3.

B.3 Coordinate Transformations

The set of coordinate transformation matrices that were used to move between the various
systems listed above were presented in Ref. [129], Eqs (4.1)-(4.20):

“0” system to “1” system

    
ê𝑥1  cos 𝛼 𝑅 0 sin 𝛼 𝑅  ê𝑥0

 
  


 
   



 

  
 


ê𝑦1 =  0 1 0  ê
  𝑦0 (B.1)

 
   

   
 ê𝑧1   − sin 𝛼 𝑅 0 cos 𝛼 𝑅  ê

 
   
 
  𝑧0

   

“1” system to “2” system

    
ê𝑥2  cos 𝜓 sin 𝜓 0    ê𝑥1

 
 


 
   



 

  
 


ê𝑦2 =  − sin 𝜓 cos 𝜓 0  ê𝑦1 (B.2)

 
   

   
 ê𝑧2 0 0 1   ê

 
   

  𝑧1
  
   

156
“2” system to “3” system

    
ê𝑥3  ê𝑥2
 1 0 𝛽𝑝  

 
 


 
   



 

  
 


ê𝑦3 =  0 1 0  ê𝑦2
  (B.3)

 
   

   
 ê𝑧3   −𝛽 𝑝 0 1  
 ê

 
   

  𝑧2

   

“3” system to “4” system The coordinate transformation from the undeformed “3” system
to the deformed “4” system is obtained using a sequence of angular rotations. The
sequence used in this study is flap-lag-torsion, and consists of 1) a flap rotation by
the angle 𝑤,𝑥 clockwise about the 𝑦 3 axis, 2) a lead-lag rotation by the angle 𝑣,𝑥
counterclockwise about the 𝑧3 axis, and 3) a torsional rotation given by the twist
angle 𝜙 counterclockwise about the 𝑥4 axis, in that order. Hence, the coordinate
transformation from the undeformed “3” system to the deformed “4” system is given
by the matrix product:

       
ê𝑥4  1 0 0  1 𝑣,𝑥 0   1 0  ê𝑥3
   





 

   𝑤, 𝑥 





 

     
 


ê𝑦4 =  0 cos 𝜙 sin 𝜙   −𝑣,
 𝑥 1 0   0 1 0   ê𝑦3

 
      

   
 ê𝑧4   0 − sin 𝜙 cos 𝜙   0 0 1   −𝑤,𝑥 0 1    ê

 
      

  𝑧3

      
(B.4)
Note that the small angle assumptions cos 𝑣,𝑥  1, cos 𝑤,𝑥  1, sin 𝑣,𝑥  𝑣,𝑥 , and
sin 𝑤,𝑥  𝑤,𝑥 have been made. Performing the matrix multiplication yields:

    
ê𝑥4 1  ê𝑥3
   







 𝑣,𝑥 𝑤,𝑥 





 

  
 


ê𝑦4 =  −𝑣 𝑥 cos 𝜙 − 𝑤,𝑥 sin 𝜙 cos 𝜙 sin 𝜙 − 𝑣,𝑥 𝑤,𝑥 cos 𝜙  ê
  𝑦3 (B.5)

 
   

   
 ê𝑧4   𝑣,𝑥 sin 𝜙 − 𝑤,𝑥 cos 𝜙 − sin 𝜙 cos 𝜙 + 𝑣,𝑥 𝑤,𝑥 sin 𝜙  ê

 
   
 
  𝑧3

   

157
“3” system to “5” system

    
ê𝑥5  1  ê𝑥3
   





  𝑣,𝑥 𝑤,𝑥 





 

  
 


ê𝑦5 =  −𝑣,𝑥 1 −𝑤,𝑥 𝑣,𝑥  ê
  𝑦3 (B.6)

 
   

   
 ê𝑧5   −𝑤,𝑥 0 1  ê

 
   
 
  𝑧3

   

“4” system to “5” system

    
ê𝑥5  1 0 0  ê𝑥4

 
  


 
   



 

  
 


ê𝑦5 =  0 cos 𝜙 − sin 𝜙  ê
  𝑦4 (B.7)

 
   

   
 ê𝑧5   0 sin 𝜙 cos 𝜙  ê

 
   
 
  𝑧4

   

B.4 Ordering Scheme

An ordering scheme is applied to eliminate the higher order nonlinear terms in the structural
equations of motion, in a consistent manner. This is accomplished by assigning orders
of magnitude to various commonly encountered nondimensional physical terms and then
neglecting terms with an order higher than a predetermined threshold value.
It is assumed the slopes of the deformed rotor blades are of the order 𝜖 (0.10 ≤ 𝜖 ≤ 0.20),
based on the moderate deflection assumption. The ordering scheme assumes the terms of
order 𝜖 2 or higher can be neglected with respect to terms of order 1, i.e.,

𝑂 (1) + 𝑂 (𝜖 2 )  1 (B.8)

A careful and systematic application of this ordering scheme yields expressions of manage-
able size and with sufficient accuracy for rotor stability and response calculations.
To assign orders of magnitude to individual terms, they must first be expressed in

158
nondimensional form. This is performed using the following reference quantities:

[length] = 𝑅, rotor radius,

[mass] = 𝑀𝑏 , mass of one blade,
1
[time] = , inverse of the rotor speed.
Ω

Orders of magnitude have been assigned to common nondimensional quantities as

follows:

𝑥 𝐿𝑏 𝑚𝑏 𝜌𝑏
𝑂 (1) : , , , , 𝜇, 𝜓, cos 𝜓, sin 𝜓,
𝑅 𝑅 (𝑀𝑏 /𝑅) (𝑀𝑏 /𝑅 3 )
𝜕 1 𝜕 𝜕
𝑎𝑜 , 𝑅 , ,
𝜕𝑥 Ω 𝜕𝑡 𝜕𝜓
𝐿 𝑐𝑠
𝑂 (𝜖 1/2 ) : , 𝜃𝐺 , 𝛿
𝑅
𝑐 𝑏 𝑒 𝑐 𝑐𝑠 𝑚𝑐 𝑋ℎ
𝑂 (𝜖) : , , , , , 𝜃 𝑝𝑡 , 𝜆, 𝛼 𝑅 , 𝛽 𝑝 ,
𝑅 𝑅 𝑅 (𝑀𝑏 /𝑅) 𝑅
𝑣 𝑤
, , 𝑣,𝑥 , 𝑤,𝑥 , 𝜙
𝑅 𝑅
𝑀𝑐 𝑋 𝐴 𝑋 𝐼𝑏 𝑋 𝐼𝑐 𝑋 𝐼 𝐼𝑏 𝑋 𝐼 𝐼𝑐
𝑂 (𝜖 3/2 ) : , , , , , , 𝐶𝑑0
𝑀𝑏 𝑅 𝑅 𝑅 𝑅 𝑅
𝑢 𝐸 𝐼𝜁 𝜁 𝐸 𝐼𝜂𝜂 𝑓 𝐶𝑑𝑓
𝑂 (𝜖 2 ) : , 3 2
, 3 2
, 2 ,
𝑅 𝑀𝑏 𝑅 Ω 𝑀𝑏 𝑅 Ω 𝑅
𝐼 𝑀 𝐵2 𝐼 𝑀 𝐵3
𝑂 (𝜖 5/2 ) : ,
𝑀𝑏 𝑅 𝑀𝑏 𝑅
𝑔 𝐺𝐽𝑏
𝑂 (𝜖 3 ) : ,
Ω2 𝑅 𝑀𝑏 𝑅 3 Ω2
𝐼 𝑀𝐶2 𝐼 𝑀𝐶3
𝑂 (𝜖 7/2 ) : ,
𝑀𝑏 𝑅 𝑀𝑏 𝑅
The orders of magnitude presented here are consistent with those used in Refs. [23, 34,
124].

B.5 Equations of Motion for the Elastic Blade

The hingeless blades are modeled as slender rods of linearly elastic, homogeneous material,
cantilevered at an offset 𝑒 from the rotor hub, using a blade model taken from Ref. [124].
This blade model describes the fully coupled flap-lag-torsional dynamics of an isotropic

159
blade. Note that a composite blade formulation is also available in Ref. [135]. The blade
model described in Ref. [124] was derived to reflect the deformation sequence flap-lag-
torsion.
The equations of motion for the elastic blade consist of a set of nonlinear partial
differential equations of motion, with the distributed loads left in general symbolic form.
The distributed loads on the blade, not including control surface loads, can be expressed in
the “3” system as:

p𝑏 = 𝑝 𝑏𝑥3 ê𝑥3 + 𝑝 𝑏𝑦3 ê𝑦3 + 𝑝 𝑏𝑧3 ê𝑧3 , (B.9)

q 𝑀 𝑏 = 𝑞 𝑏𝑥3 ê𝑥3 + 𝑞 𝑏𝑦3 ê𝑦3 + 𝑞 𝑏𝑧3 ê𝑧3 , (B.10)

where p𝑏 and q 𝑀 𝑏 represent the total distributed spanwise force and moment, respectively.
The equations of motion for the elastic blade, derived in Ref. [129], Eqs. (4.23)-(4.25), are
then given by:
Flap Equation

−[(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 ) sin 𝜃 𝐺 cos 𝜃 𝐺 (𝑣,𝑥𝑥 +2𝜙𝑤,𝑥𝑥 )

+(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 )𝜙𝑣,𝑥𝑥 cos 2𝜃 𝐺 + (𝐸 𝐼 𝜁 𝜁 sin2 𝜃 𝐺 + 𝐸 𝐼𝜂𝜂 cos2 𝜃 𝐺 )𝑤,𝑥𝑥

−𝑇 𝑋 𝐼 𝐼𝑏 (sin 𝜃 𝐺 + 𝜙 cos 𝜃 𝐺 )],𝑥𝑥 +(𝐺𝐽𝑏 𝜙,𝑥 𝑣,𝑥𝑥 ),𝑥 +(𝑤,𝑥 𝑇),𝑥

−(𝑣,𝑥 𝑞 𝑏𝑥3 ),𝑥 +𝑞 𝑏𝑦3 ,𝑥 +𝑝 𝑏𝑧3 = 0 (B.11)

Lag Equation

−[(𝐸 𝐼 𝜁 𝜁 cos2 𝜃 𝐺 + 𝐸 𝐼𝜂𝜂 sin2 𝜃 𝐺 )𝑣,𝑥𝑥 +(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 )𝜙𝑤,𝑥𝑥 cos 2𝜃 𝐺

+(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 ) sin 𝜃 𝐺 cos 𝜃 𝐺 (𝑤,𝑥𝑥 −2𝜙𝑣,𝑥𝑥 )

−𝑇 𝑋 𝐼 𝐼𝑏 (cos 𝜃 𝐺 − 𝜙 sin 𝜃 𝐺 )],𝑥𝑥 −(𝐺𝐽𝑏 𝜙,𝑥 𝑤,𝑥𝑥 ),𝑥 +(𝑣,𝑥 𝑇),𝑥

+(𝑤,𝑥 𝑞 𝑏𝑥3 ),𝑥 −𝑞 𝑏𝑧3 ,𝑥 +𝑝 𝑏𝑦3 = 0 (B.12)

160
Torsional Equation

[𝐺𝐽𝑏 (𝜙,𝑥 −𝑣,𝑥 𝑤,𝑥𝑥 )],𝑥

+(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 ) [(𝑣,𝑥𝑥 2 − 𝑤,𝑥𝑥 2 ) sin 𝜃 𝐺 cos 𝜃 𝐺 − 𝑣,𝑥𝑥 𝑤,𝑥𝑥 cos 2𝜃 𝐺 ]

+𝑇 𝑋 𝐼 𝐼𝑏 (𝑤,𝑥𝑥 cos 𝜃 𝐺 − 𝑣,𝑥𝑥 sin 𝜃 𝐺 ) + 𝑞 𝑏𝑥3 + 𝑣,𝑥 𝑞 𝑏𝑦3 + 𝑤,𝑥 𝑞 𝑏𝑧3 = 0 (B.13)

where 𝑇 is the axial tension.

B.6 Incorporation of the On-Blade Control Devices in the

Blade Equations of Motion

The inertial and aerodynamic effects of the on-blade control devices are included in the
model. The effect of the microflap/AFC on the structural properties of the blade is assumed
to be negligible. Thus, the control surfaces influence the behavior of the blade only through
their contribution to the spanwise distributed loads on the blade.
The distributed force and moment on the blade due to the control devices can be
represented in the “3” system by Ref. [129], Eqs. (4.26)-(4.27):

p𝑐 = 𝑝 𝑐𝑥3 ê𝑥3 + 𝑝 𝑐𝑦3 ê𝑦3 + 𝑝 𝑐𝑧3 ê𝑧3 (B.14)

q𝑐 = 𝑞 𝑐𝑥3 ê𝑥3 + 𝑞 𝑐𝑦3 ê𝑦3 + 𝑞 𝑐𝑧3 ê𝑧3 (B.15)

For a single microflap/AFC configuration, with the microflap/AFC inboard edge located at a
1 from the blade root, the distributed loads are given by:
distance 𝑥 𝑐𝑠

 1 − 𝐿 𝑐𝑠
0 for



 𝑥 < 𝑥 𝑐𝑠 2



p𝑐 , q𝑐 =
 p1𝑐 , q1𝑐 for 𝑥 𝑐𝑠
1 − 𝐿 𝑐𝑠 ≤ 𝑥 ≤ 𝑥 1 +
2 𝑐𝑠
𝐿 𝑐𝑠
2


 0

 for 1 + 𝐿 𝑐𝑠
𝑥 > 𝑥 𝑐𝑠
 2

161
where p1𝑐 and q1𝑐 represent the distributed loads due to the microflap/AFC, and are described
in the next section.
The effect of the control surfaces is included in the blade equations of motion by adding
the distributed loads due to the control surfaces, given in Eqs. (B.14)-(B.15), to the distributed
loads for the blade alone. The equations of motion for the blade, Eqs. (B.11)-(B.12), can be
rewritten to reflect this change as:
Flap Equation

−[(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 ) sin 𝜃 𝐺 cos 𝜃 𝐺 (𝑣,𝑥𝑥 +2𝜙𝑤,𝑥𝑥 )

+(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 )𝜙𝑣,𝑥𝑥 cos 2𝜃 𝐺 + (𝐸 𝐼 𝜁 𝜁 sin2 𝜃 𝐺 + 𝐸 𝐼𝜂𝜂 cos2 𝜃 𝐺 )𝑤,𝑥𝑥

−𝑇 𝑋 𝐼 𝐼𝑏 (sin 𝜃 𝐺 + 𝜙 cos 𝜃 𝐺 )],𝑥𝑥 +(𝐺𝐽𝑏 𝜙,𝑥 𝑣,𝑥𝑥 ),𝑥 +(𝑤,𝑥 𝑇),𝑥

−(𝑣,𝑥 (𝑞 𝑏𝑥3 + 𝑞 𝑐𝑥3 )),𝑥 +(𝑞 𝑏𝑦3 + 𝑞 𝑐𝑦3 ),𝑥 +( 𝑝 𝑏𝑧3 + 𝑝 𝑐𝑧3 ) = 0 (B.16)

Lag Equation

−[(𝐸 𝐼 𝜁 𝜁 cos2 𝜃 𝐺 + 𝐸 𝐼𝜂𝜂 sin2 𝜃 𝐺 )𝑣,𝑥𝑥 +(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 )𝜙𝑤,𝑥𝑥 cos 2𝜃 𝐺

+(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 ) sin 𝜃 𝐺 cos 𝜃 𝐺 (𝑤,𝑥𝑥 −2𝜙𝑣,𝑥𝑥 )

−𝑇 𝑋 𝐼 𝐼𝑏 (cos 𝜃 𝐺 − 𝜙 sin 𝜃 𝐺 )],𝑥𝑥

−(𝐺𝐽𝑏 𝜙,𝑥 𝑤,𝑥𝑥 ),𝑥 +(𝑣,𝑥 𝑇),𝑥 +(𝑤,𝑥 (𝑞 𝑏𝑥3 + 𝑞 𝑐𝑥3 )),𝑥

−(𝑞 𝑏𝑧3 + 𝑞 𝑐𝑧3 ),𝑥 +( 𝑝 𝑏𝑦3 + 𝑝 𝑐𝑦3 ) = 0 (B.17)

Torsion Equation

[𝐺𝐽𝑏 (𝜙,𝑥 −𝑣,𝑥 𝑤,𝑥𝑥 )],𝑥 +(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 ) [(𝑣,𝑥𝑥 2 − 𝑤,𝑥𝑥 2 ) sin 𝜃 𝐺 cos 𝜃 𝐺

−𝑣,𝑥𝑥 𝑤,𝑥𝑥 cos 2𝜃 𝐺 ] + 𝑇 𝑋 𝐼 𝐼𝑏 (𝑤,𝑥𝑥 cos 𝜃 𝐺 − 𝑣,𝑥𝑥 sin 𝜃 𝐺 )

+(𝑞 𝑏𝑥3 + 𝑞 𝑐𝑥3 ) + 𝑣,𝑥 (𝑞 𝑏𝑦3 + 𝑞 𝑐𝑦3 ) + 𝑤,𝑥 (𝑞 𝑏𝑧3 + 𝑞 𝑐𝑧3 ) = 0 (B.18)

162
B.7 Distributed Loads

A complete description of the aeroelastic equations governing the motion of the rotor blade
requires a derivation of the distributed inertial, aerodynamic, gravitational, and the structural
damping loads.
Distributed inertial, gravitational, and damping loads on a flapped rotor blade were de-
rived as explicit expressions of blade displacement in Ref. [124]. Expressions for distributed
aerodynamic loads were derived in Ref. [129]. These expressions have been used in the
present analysis. The purpose of this section is to show how the complete equations of
motion are formulated. This will be accomplished by establishing the blade kinematics first
and subsequently the distributed loads.

B.7.1 Blade Kinematics

To formulate explicit expressions of the distributed loads acting on the blade, the position
of an arbitrary point on the blade or control surface must be defined in terms of the blade
degrees of freedom. The approach described in this chapter is taken from Ref. [129]. The
kinematic description of the blade used in the derivation of the distributed loads is based on
the assumptions of Euler-Bernoulli beam theory: plane sections normal to the elastic axis
of the beam before deformation remain plane after deformation, and strains within cross-
sections are neglected. Accordingly, an arbitrary point on the beam before deformation,
represented by the vector

R 𝑝 = 𝑒ê𝑥2 + 𝑥 ê𝑥3 + 𝑦 𝑜 ê𝑦3 + 𝑧 𝑜 ê𝑧3 (B.19)

is described after deformation by the vector

r 𝑝 = 𝑒ê𝑥2 + (𝑥 + 𝑢) ê𝑥3 + 𝑣 ê𝑦3 + 𝑤 ê𝑧3 + 𝑦 𝑜 ê𝑦4 + 𝑧 𝑜 ê𝑧4 (B.20)

163
where 𝑢, 𝑣, and 𝑤 represent the displacement of a point on the elastic axis of the blade as
illustrated in Fig. B.2. If the coordinate pair ( 𝑦¯ 𝑜 , 𝑧¯𝑜 ) can be interpreted as the pair (𝑦 𝑜 , 𝑧 𝑜 )
expressed in the “5" coordinate system, i.e.

𝑦 𝑜 ê𝑦4 + 𝑧 𝑜 ê𝑧4 = 𝑦¯ 𝑜 ê𝑦5 + 𝑧¯𝑜 ê𝑧5 (B.21)

then
r 𝑝 = 𝑒ê𝑥2 + (𝑥 + 𝑢) ê𝑥3 + 𝑣 ê𝑦3 + 𝑤 ê𝑧3 + 𝑦¯ 𝑜 ê𝑦5 + 𝑧¯𝑜 ê𝑧5 (B.22)

The velocity and acceleration of a point in a reference frame that is translating and rotating
relative to an inertial frame can be found using the classical relations:

¤ 𝑜 + r¤ + 𝝎 × r
v=R (B.23)

¥ 𝑜 + r¥ + 2𝝎 × r¤ + 𝝎
a=R ¤ × r + 𝝎 × (𝝎 × r) (B.24)

where R𝑜 is the position of the origin of the moving reference frame in the inertial frame,
and 𝜔 the vector of angular velocity. The time derivatives of R𝑜 are taken in the inertial
frame, while those for r are taken in the rotating frame. For the rotor case, the inertial frame
is that of the hub, described by the “1” system. The rotating frame rotates with the blades
and corresponds to the “2” system. The origin of the rotating frame is assumed to coincide
with the that of the non-rotating frame. Thus:

¥𝑜 = R
R ¤ 𝑜 = R𝑜 = 0 (B.25)

164
 ¤ = 0. Hence, the velocity and acceleration of any
Also, 𝝎 = Ωê𝑧2 , and since Ω is constant, 𝝎
point in the rotating reference frame (“2” system) are given by:

v 𝑝 = r¤ 𝑝 + Ωê𝑧2 × r 𝑝 (B.26)

a 𝑝 = r¥ 𝑝 + 2Ωê𝑧2 × r¤ 𝑝 + Ωê𝑧2 × (Ωê𝑧2 × r 𝑝 ) (B.27)

Equations (B.26) and (B.27), taken from Ref. [129], are the fundamental kinematic relations
used in the derivation of the distributed loads.

B.7.2 Inertial Loads

The inertial loads are obtained using D’Alembert’s principle. Expressions for the inertial
loads will first be formulated in the “2” system, and then transformed to the “3” system to
be compatible with the blade elastic equations of motion. Given an arbitrary point in the
rotating frame (“2” system), represented by the vector:

r 𝑝 = 𝑟 𝑝𝑥2 ê𝑥2 + 𝑟 𝑝𝑦2 ê𝑦2 + 𝑟 𝑝𝑧2 ê𝑧2 (B.28)

the acceleration of this point can be found using Eq. (B.27). Expressed in the “2” system,
this is given by:
a 𝑝 = 𝑎 𝑝𝑥2 ê𝑥2 + 𝑎 𝑝𝑦2 ê𝑦2 + 𝑎 𝑝𝑧2 ê𝑧2 (B.29)

where:

𝑎 𝑝𝑥2 = 𝑟¥𝑝𝑥2 − 2Ω𝑟¤𝑝𝑦2 − Ω2𝑟 𝑝𝑥2 (B.30)

𝑎 𝑝𝑦2 = 𝑟¥𝑝𝑦2 + 2Ω𝑟¤𝑝𝑥2 − Ω2𝑟 𝑝𝑦2 (B.31)

𝑎 𝑝𝑧2 = 𝑟¥𝑝𝑧2 (B.32)

with time derivatives of r 𝑝 taken in the “2” system.

165
 From Eq. (B.22), a point on the deformed blade can be expressed as:

r𝑏 = 𝑒ê𝑥2 + (𝑥 + 𝑢) ê𝑥3 + 𝑣 ê𝑦3 + 𝑤 ê𝑧3 + 𝑦¯ 𝑜𝑏 ê𝑦5 + 𝑧¯𝑜𝑏 ê𝑧5 (B.33)

The inertial forces and moments taken about the elastic axis of the blade at a given spanwise
location are given by:

∫
p 𝐼𝑏 = − 𝜌 𝑏 ab d𝐴 (B.34)
𝐴𝑏
∫
q 𝐼𝑏 = − r𝑜𝑏 × 𝜌 𝑏 a𝑏 d𝐴 (B.35)
𝐴𝑏

where:
r𝑜𝑏 = 𝑦¯ 𝑜𝑏 ê𝑦5 + 𝑧¯𝑜𝑏 ê𝑧5 (B.36)

The resulting spanwise distributed inertia force is expressed in the “2” system as
Ref. [129], Eqs. (5.36)-(5.39):

p 𝐼𝑏 = 𝑝 𝐼𝑏𝑥2 ê𝑥2 + 𝑝 𝐼𝑏𝑦2 ê𝑦2 + 𝑝 𝐼𝑏𝑧2 ê𝑧2 (B.37)

166
where the components of p 𝐼𝑏 are given by:

𝑝 𝐼𝑏𝑥2 = 𝑚 𝑏 Ω2 (𝑥 + 𝑒) + 2𝑚 𝑏 Ω𝑣¤ + 𝑚 𝑏 𝛽 𝑝 ( 𝑤¥ − 𝑤Ω2 )

¥ − 2𝑚 𝑏 𝑋 𝐼𝑏 Ω sin(𝜃 𝐺 + 𝜙)( 𝜃¤𝐺 + 𝜙)

+ 𝑚 𝑏 (𝑢Ω2 − 𝑢) ¤ (B.38)

¤ 𝑝 + 𝑚 𝑏 (𝑣Ω2 − 𝑣¥ ) − 2𝑚 𝑏 Ω𝑢¤
𝑝 𝐼𝑏𝑦2 = 2𝑚 𝑏 Ω𝑤𝛽

+ 𝑚 𝑏 𝑋 𝐼𝑏 cos(𝜃 𝐺 + 𝜙)(( 𝜃¤𝐺 + 𝜙)

¤ 2 + Ω(Ω + 2𝑣¤ ,𝑥 ) + 2Ω( 𝜃¤𝐺 + 𝜙)(𝑤,
¤ 𝑥 +𝛽 𝑝 ))

+ 𝑚 𝑏 𝑋 𝐼𝑏 sin(𝜃 𝐺 + 𝜙)(( 𝜃¥𝐺 + 𝜙) ¤ 𝑥 −2Ω( 𝜃¤𝐺 + 𝜙)𝑣,

¥ + 2Ω𝑤, ¤ 𝑥) (B.39)

𝑝 𝐼𝑏𝑧2 = −𝑚 𝑏 𝑢𝛽
¥ 𝑝 − 𝑚 𝑏 𝑤¥

+ 𝑚 𝑏 𝑋 𝐼𝑏 cos(𝜃 𝐺 + 𝜙)(−( 𝜃¥𝐺 + 𝜙)

¥ + 𝑣¥ ,𝑥 (𝑤,𝑥 +𝛽 𝑝 ) + 2𝑤,
¤ 𝑥 𝑣¤ ,𝑥 +𝑤,
¥ 𝑥 𝑣,𝑥 )

+ 𝑚 𝑏 𝑋 𝐼𝑏 sin(𝜃 𝐺 + 𝜙)(( 𝜃¤𝐺 + 𝜙)

¤ 2 − ( 𝜃¥𝐺 + 𝜙)(𝑤,
¥ 𝑥 +𝛽 𝑝 )𝑣,𝑥 ) (B.40)

Similarly, distributed spanwise moment is expressed in the “2” system as Ref. [129],
Eqs. (5.40)-(5.43):
q 𝐼𝑏 = 𝑞 𝐼𝑏𝑥2 ê𝑥2 + 𝑞 𝐼𝑏𝑦2 ê𝑦2 + 𝑞 𝐼𝑏𝑧2 ê𝑧2 (B.41)

167
where:

𝑞 𝐼𝑏𝑥2 = 𝑚 𝑏 𝑋 𝐼𝑏 cos(𝜃 𝐺 + 𝜙)((𝑣Ω2 − 𝑣¥ )(𝑤,𝑥 +𝛽 𝑝 )𝑣,𝑥 −𝑤¥ − 𝑢𝛽

¥ 𝑝)

𝑚 𝑏 𝑋 𝐼𝑏 sin(𝜃 𝐺 + 𝜙)(( 𝑣¥ − 𝑣Ω2 ) + 2Ω𝑢¤ − 2Ω𝑤𝛽

¤ 𝑝)

− (𝐼 𝑀 𝐵2 + 𝐼 𝑀 𝐵3 )( 𝜃¥𝐺 + 𝜙)
¥

+ (𝐼 𝑀 𝐵2 − 𝐼 𝑀 𝐵3 ) cos(𝜃 𝐺 + 𝜙) sin(𝜃 𝐺 + 𝜙)Ω((Ω + 2𝑣¤ ,𝑥 ) + 2( 𝜃¤𝐺 + 𝜙)(𝑤,

¤ 𝑥 +𝛽 𝑝 )

+ 2(𝐼 𝑀 𝐵2 cos2 (𝜃 𝐺 + 𝜙) + 𝐼 𝑀 𝐵3 sin2 (𝜃 𝐺 + 𝜙))Ω(( 𝜃¤𝐺 + 𝜙)𝑣,

¤ 𝑥 −𝑤,
¤ 𝑥)

+ (𝐼 𝑀 𝐵2 sin2 (𝜃 𝐺 + 𝜙) + 𝐼 𝑀 𝐵3 cos2 (𝜃 𝐺 + 𝜙))

(2𝑣¤ ,𝑥 𝑤, ¥ 𝑥 +(𝑤,𝑥 +𝛽 𝑝 )(Ω2 𝑣,𝑥 +¥𝑣 ,𝑥 ))

¤ 𝑥 +𝑣,𝑥 𝑤, (B.42)

𝑞 𝐼𝑏𝑦2 = −𝑚 𝑏 𝑋 𝐼𝑏 cos(𝜃 𝐺 + 𝜙)(Ω2 𝑥(𝑤,𝑥 +𝛽 𝑝 ) + 𝑤)𝑣,

¥ 𝑥

+ 𝑚 𝑏 𝑋 𝐼𝑏 sin(𝜃 𝐺 + 𝜙)(Ω2 (𝑥 + 𝑒) − (Ω2 𝑤𝛽 𝑝 + 𝑤𝑤,

¥ 𝑥 ) + 2Ω𝑣¤ + (𝑢Ω2 − 𝑢))
¥

− (𝐼 𝑀 𝐵2 + 𝐼 𝑀 𝐵3 )( 𝜃¥𝐺 + 𝜙)𝑣,
¥ 𝑥

+ (𝐼 𝑀 𝐵2 − 𝐼 𝑀 𝐵3 ) cos(𝜃 𝐺 + 𝜙) sin(𝜃 𝐺 + 𝜙)((𝑣,𝑥 Ω2 − 𝑣¥ ,𝑥 ) − 2( 𝜃¤𝐺 + 𝜙)

¤ 𝑤,
¤ 𝑥)

+ (𝐼 𝑀 𝐵2 cos2 (𝜃 𝐺 + 𝜙) + 𝐼 𝑀 𝐵3 sin2 (𝜃 𝐺 + 𝜙))

(𝑤,𝑥 −Ω2 (𝑤,𝑥 +𝛽 𝑝 ) − 2( 𝜃¤𝐺 + 𝜙)(Ω

¤ + 𝑣¤ ,𝑥 )) (B.43)

𝑞 𝐼𝑏𝑧2 = 𝑚 𝑏 𝑋 𝐼𝑏 cos(𝜃 𝐺 + 𝜙)(−Ω2 (𝑥 + 𝑒) − 2Ω𝑣¤ + ( 𝑢¥ − 𝑢Ω2 )

+ (𝑤Ω2 − 𝑤)
¥ 𝛽 𝑝 + ( 𝑣¥ − 𝑣Ω2 )𝑣,𝑥 )

+ 𝑚 𝑏 𝑋 𝐼𝑏 sin(𝜃 𝐺 + 𝜙)( 𝑣¥ − 𝑣Ω2 )(𝑤,𝑥 +𝛽 𝑝 )

− (𝐼 𝑀 𝐵2 + 𝐼 𝑀 𝐵3 )( 𝜃¥𝐺 + 𝜙)(𝑤,
¥ 𝑥 +𝛽 𝑝 )

¥ 𝑥 −2( 𝜃¤𝐺 + 𝜙)(Ω

+ (𝐼 𝑀 𝐵2 − 𝐼 𝑀 𝐵3 ) cos(𝜃 𝐺 + 𝜙) sin(𝜃 𝐺 + 𝜙)( 𝑤, ¤ + 𝑣¤ ,𝑥 ))

− (𝐼 𝑀 𝐵2 sin2 (𝜃 𝐺 + 𝜙) + 𝐼 𝑀 𝐵3 cos2 (𝜃 𝐺 + 𝜙))( 𝑣¥ ,𝑥 +2( 𝜃¤𝐺 + 𝜙)

¤ 𝑤,
¤ 𝑥) (B.44)

Using the coordinate transformations described in the previous sections of this Appendix,

168
the distributed spanwise inertial force can be expressed in the “3” system as:

p 𝐼𝑏 = 𝑝 𝐼𝑏𝑥3 ê𝑥3 + 𝑝 𝐼𝑏𝑦3 ê𝑦3 + 𝑝 𝐼𝑏𝑧3 ê𝑧3 (B.45)

where:

𝑝 𝐼𝑏𝑥3 = 𝑝 𝐼𝑏𝑥2 + 𝛽 𝑝 𝑝 𝐼𝑏𝑧2 (B.46)

𝑝 𝐼𝑏𝑥3 = 𝑝 𝐼𝑏𝑦2 (B.47)

𝑝 𝐼𝑏𝑥3 = −𝛽 𝑝 𝑝 𝐼𝑏𝑥2 + 𝑝 𝐼𝑏𝑧2 (B.48)

Similarly, the distributed spanwise moment can be expressed in the “3” system as:

q 𝐼𝑏 = 𝑞 𝐼𝑏𝑥3 ê𝑥3 + 𝑞 𝐼𝑏𝑦3 ê𝑦3 + 𝑞 𝐼𝑏𝑧3 ê𝑧3 , (B.49)

where:

𝑞 𝐼𝑏𝑥3 = 𝑞 𝐼𝑏𝑥2 + 𝛽 𝑝 𝑞 𝐼𝑏𝑧2 (B.50)

𝑞 𝐼𝑏𝑥3 = 𝑞 𝐼𝑏𝑦2 (B.51)

𝑞 𝐼𝑏𝑥3 = −𝛽 𝑝 𝑞 𝐼𝑏𝑥2 + 𝑞 𝐼𝑏𝑧2 (B.52)

The derivation of the inertia loads due to a control surface is identical to that for the
blade and can be found in Ref. 1291998Myrtle.

B.7.3 Gravitational Loads

The distributed gravitational loads can be derived by integrating the gravitational force
and moment per unit volume over the blade cross-section. Gravitational acceleration 𝑔 is
directed along the negative 𝑧0 axis:
g = −𝑔ê𝑧0 (B.53)

169
Expressed in the “2” system, this becomes:

g = 𝑔𝑥2 ê𝑥2 + 𝑔 𝑦2 ê𝑦2 + 𝑔𝑧2 ê𝑧2 (B.54)

where:

𝑔𝑥2 = −𝑔 sin 𝛼 𝑅 cos 𝜓 (B.55)

𝑔 𝑦2 = 𝑔 sin 𝛼 𝑅 sin 𝜓 (B.56)

𝑔𝑧2 = −𝑔 cos 𝛼 𝑅 (B.57)

These expressions are then used to derive the distributed force and moment. The distributed
gravitational force is derived by integrating the gravitational force per unit volume over the
blade cross-section:
∫
p𝐺𝑏 = 𝜌 𝑏 g d𝐴 (B.58)
𝐴𝑏

This can be expressed in the “2” system as Ref. [129], Eqs. (5.103)-(5.106):

p𝐺𝑏 = 𝑝 𝐺𝑏𝑥2 ê𝑥2 + 𝑝 𝐺𝑏𝑦2 ê𝑦2 + 𝑝 𝐺𝑏𝑧2 ê𝑧2 (B.59)

where:

∫
𝑝 𝐺𝑏𝑥2 = 𝜌 𝑏 𝑔𝑥2 d𝐴 = −𝑚 𝑏 𝑔 sin 𝛼 𝑅 cos 𝜓 (B.60)
𝐴𝑏
∫
𝑝 𝐺𝑏𝑦2 = 𝜌 𝑏 𝑔 𝑦2 d𝐴 = 𝑚 𝑏 𝑔 sin 𝛼 𝑅 sin 𝜓 (B.61)
𝐴𝑏
∫
𝑝 𝐺𝑏𝑧2 = 𝜌 𝑏 𝑔𝑧2 d𝐴 = −𝑚 𝑏 𝑔 cos 𝛼 𝑅 (B.62)
𝐴𝑏

170
Similarly, the distributed gravitational moment about the elastic axis is derived by integrating
the gravitational moment per unit volume over the blade cross-section:

∫
q𝐺𝑏 = ( 𝑦¯ 0𝑏 ê𝑦5 + 𝑧¯0𝑏 ê𝑧5 ) × 𝜌 𝑏 g d𝐴 (B.63)
𝐴𝑏

Expressed in the “2” system, this becomes Ref. [129], Eqs. (5.108)-(5.111):

q𝐺𝑏 = 𝑞 𝐺𝑏𝑥2 ê𝑥2 + 𝑞 𝐺𝑏𝑦2 ê𝑦2 + 𝑞 𝐺𝑏𝑧2 ê𝑧2 (B.64)

where:

∫
𝑞 𝐺𝑏𝑥2 = 𝜌 𝑏 ( 𝑦¯ 0𝑏 (𝑔𝑧2 + (𝑤,𝑥 +𝛽 𝑝 )𝑣,𝑥 𝑔 𝑦2 ) − 𝑧¯0𝑏 𝑔 𝑦2 ) d𝐴 (B.65)
𝐴𝑏
∫
𝑞 𝐺𝑏𝑦2 = 𝜌 𝑏 ( 𝑦¯ 0𝑏 (𝑔𝑧2 − (𝑤,𝑥 +𝛽 𝑝 )𝑣,𝑥 +𝑧¯0𝑏 ((𝑤,𝑥 +𝛽 𝑝 )𝑔𝑧2 + 𝑔𝑥2 )) d𝐴 (B.66)
𝐴𝑏
∫
𝑞 𝐺𝑏𝑧2 = 𝜌 𝑏 (− 𝑦¯ 0𝑏 (𝑣,𝑥 𝑔 𝑦2 + 𝑔𝑥2 ) − 𝑧¯0𝑏 (𝑤,𝑥 +𝛽 𝑝 )𝑔 𝑦2 ) d𝐴 (B.67)
𝐴𝑏

Substituting (B.55)-(B.57) into (B.65)-(B.67) leads to:

𝑞 𝐺𝑏𝑥2 = −𝑚 𝑏 𝑔𝑋𝐼𝑏 cos(𝜃 𝐺 + 𝜙)(cos 𝛼 𝑅 − (𝑤,𝑥 +𝛽 𝑝 )𝑣,𝑥 sin 𝛼 𝑅 sin 𝜓)

− 𝑚 𝑏 𝑔𝑋𝐼𝑏 sin(𝜃 𝐺 + 𝜙) sin 𝛼 𝑅 sin 𝜙 (B.68)

𝑞 𝐺𝑏𝑦2 = −𝑚 𝑏 𝑔𝑋𝐼𝑏 cos(𝜃 𝐺 + 𝜙)(cos 𝛼 𝑅 − (𝑤,𝑥 +𝛽 𝑝 ) sin 𝛼 𝑅 cos 𝜓)𝑣,𝑥

− 𝑚 𝑏 𝑔𝑋𝐼𝑏 sin(𝜃 𝐺 + 𝜙)((𝑤,𝑥 +𝛽 𝑝 ) cos 𝛼 𝑅 + sin 𝛼 𝑅 cos 𝜙) (B.69)

𝑞 𝐺𝑏𝑧2 = −𝑚 𝑏 𝑔𝑋 𝐼𝑏 cos(𝜃 𝐺 + 𝜙) sin 𝛼 𝑅 (𝑣,𝑥 sin 𝜙 − cos 𝜙)

− 𝑚 𝑏 𝑔𝑋𝐼𝑏 sin(𝜃 𝐺 + 𝜙) sin 𝛼 𝑅 sin 𝜙) (B.70)

These expressions are then transformed to the “3” system to be compatible with the blade
equations of motion. The derivation of the distributed gravitational loads due to a control
surface is identical to that described for the blade and can be found in Ref. [129].

171
B.7.4 Damping Loads

Distributed structural damping loads are assumed to be of viscous type, and act only on the
blade. The distributed damping force is defined as:

p𝐷 = −𝑔𝑆 𝐿 𝑣¤ ê𝑦3 − 𝑔𝑆 𝐹 𝑤¤ ê𝑧3 (B.71)

The distributed damping moment is given by:

¤ 𝑥4
q𝐷 = −𝑔𝑆𝑇 𝜙ê (B.72)

which can be expressed in the “3” system as Ref. [129], Eq. (5.196):

¤ 𝑥3 − 𝑣,𝑥 ê𝑦3 − 𝑤,𝑥 ê𝑧3 )

q𝐷 = −𝑔𝑆𝑇 𝜙(ê (B.73)

𝑔𝑆 𝐿 , 𝑔𝑆 𝐹 , and 𝑔𝑆𝑇 are the distributed structural damping factors in lag, flap and torsion,
respectively.

B.7.5 Total Distributed Loads

The total distributed loads are found by summing the inertial, gravitational, aerodynamic,
and damping contributions. The distributed aerodynamic loads are derived in Chapter 4. For
the total distributed load per unit length on the blade:

p𝑏 = p 𝐼𝑏 + p𝐺𝑏 + p 𝐴 + p𝐷 (B.74)

where the subscripts 𝐼, 𝐺, 𝐴, and 𝐷 correspond to the inertial, gravitational, aerodynamic,

and damping loads respectively. For the distributed moment per unit length about the elastic
axis:
q 𝑀 𝑏 = q 𝐼𝑏 + q𝐺𝑏 + q 𝐴 + q𝐷 (B.75)

172
 APPENDIX C

Solution Procedure

C.1 Coupled Trim/Aeroelastic Response Solutions

C.1.1 Solution of the Blade Equations of Motion

The spatial dependence of the equations of motion is removed using Galerkin’s method of
weighted residuals. Three flap, two lead-lag, and two torsional free vibration modes of a
rotating beam are used to represent the flexibility of the blade. Each free vibration mode
was calculated using the first nine exact nonrotating modes of a uniform cantilevered beam.
The displacements 𝑣 and 𝑤 and twist 𝜙 are thus represented by:

3
Õ
𝑤≈ 𝑞 𝑤𝑖 (𝜓)𝑊𝑖 (𝑥) (C.1)
𝑖=1
Õ2
𝑣≈ 𝑞 𝑣𝑖 (𝜓)𝑉𝑖 (𝑥) (C.2)
𝑖=1
2
Õ
𝜙≈ 𝑞 𝜙𝑖 (𝜓)Φ𝑖 (𝑥) (C.3)
𝑖=1

where 𝑊𝑖 , 𝑉𝑖 , and Φ𝑖 are the i-th rotating flap, lead-lag, and torsional uncoupled mode
shapes, respectively, with participation coefficients 𝑞 𝑤𝑖 , 𝑞 𝑣𝑖 , and 𝑞 𝜙𝑖 . These mode shapes

173
satisfy the boundary conditions of a hingeless blade cantilevered to the hub, which implies:

𝑤(𝜓, 0) = 𝑣(𝜓, 0) = 𝜙(𝜓, 0) = 0 (C.4)

𝑤,𝑥 (𝜓, 0) = 𝑣,𝑥 (𝜓, 0) = 0 (C.5)

𝑤,𝑥𝑥 (𝜓, 𝑅) = 𝑣,𝑥𝑥 (𝜓, 𝑅) = 0 (C.6)

𝑤,𝑥𝑥𝑥 (𝜓, 𝑅) = 𝑣,𝑥𝑥𝑥 (𝜓, 𝑅) = 0 (C.7)

Galerkin’s method is applied by substituting (C.1)-(C.3) in (B.16)-(B.18). The error residuals

are then multiplied by the appropriate mode shape and integrated over the span of the blade.
After having introduced the appropriate boundary conditions, seven equations of motion are
obtained Ref. [129], Eqs. (6.12)-(6.14):
Flap Equations (𝑖 = 1, 2, 3):

∫ 𝐿𝑏
{−[(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 ) sin 𝜃 𝐺 cos 𝜃 𝐺 (𝑣,𝑥𝑥 +2𝜙𝑤,𝑥𝑥 )
0

+(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 )𝜙𝑣,𝑥𝑥 cos 2𝜃 𝐺

+(𝐸 𝐼 𝜁 𝜁 sin2 𝜃 𝐺 + 𝐸 𝐼𝜂𝜂 cos2 𝜃 𝐺 )𝑤,𝑥𝑥 −𝑇 𝑋 𝐼 𝐼𝑏 (sin 𝜃 𝐺 + 𝜙 cos 𝜃 𝐺 )]𝑊𝑖 ,𝑥𝑥

+(𝐺𝐽𝑏 𝜙,𝑥 𝑣,𝑥𝑥 +𝑤,𝑥 𝑇 − 𝑣,𝑥 (𝑞 𝑏𝑥3 + 𝑞 𝑐𝑥3 ) + (𝑞 𝑏𝑦3 + 𝑞 𝑐𝑦3 ))𝑊𝑖 ,𝑥

+( 𝑝 𝑏𝑧3 + 𝑝 𝑐𝑧3 )𝑊𝑖 }d𝑥 = 0 (C.8)

Lag Equations (𝑖 = 1, 2):

∫ 𝐿𝑏
{−[(𝐸 𝐼 𝜁 𝜁 cos2 𝜃 𝐺 + 𝐸 𝐼𝜂𝜂 sin2 𝜃 𝐺 )𝑣,𝑥𝑥 +(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 )𝜙𝑤,𝑥𝑥 cos 2𝜃 𝐺
0

+(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 ) sin 𝜃 𝐺 cos 𝜃 𝐺 (𝑤,𝑥𝑥 −2𝜙𝑣,𝑥𝑥 ) − 𝑇 𝑋 𝐼 𝐼𝑏 (cos 𝜃 𝐺 − 𝜙 sin 𝜃 𝐺 )]𝑉𝑖 ,𝑥𝑥

+(−𝐺𝐽𝑏 𝜙,𝑥 𝑤,𝑥𝑥 +𝑣,𝑥 𝑇 + 𝑤,𝑥 (𝑞 𝑏𝑥3 + 𝑞 𝑐𝑥3 )

−(𝑞 𝑏𝑧3 + 𝑞 𝑐𝑧3 ))𝑉𝑖 ,𝑥 +( 𝑝 𝑏𝑦3 + 𝑝 𝑐𝑦3 )𝑉𝑖 }d𝑥 = 0 (C.9)

174
Torsional Equations (𝑖 = 1, 2):

∫ 𝐿𝑏
{[𝐺𝐽𝑏 (𝜙,𝑥 −𝑣,𝑥 𝑤,𝑥𝑥 )]Φ𝑖 ,𝑥 +(𝐸 𝐼 𝜁 𝜁 − 𝐸 𝐼𝜂𝜂 ) [(𝑣,𝑥𝑥 2 − 𝑤,𝑥𝑥 2 ) sin 𝜃 𝐺 cos 𝜃 𝐺
0

−𝑣,𝑥𝑥 𝑤,𝑥𝑥 cos 2𝜃 𝐺 ]Φ𝑖 + 𝑇 𝑋 𝐼 𝐼𝑏 (𝑤,𝑥𝑥 cos 𝜃 𝐺 − 𝑣,𝑥𝑥 sin 𝜃 𝐺 )Φ𝑖

+((𝑞 𝑏𝑥3 + 𝑞 𝑐𝑥3 ) + 𝑣,𝑥 (𝑞 𝑏𝑦3 + 𝑞 𝑐𝑦3 ) + 𝑤,𝑥 (𝑞 𝑏𝑧3 + 𝑞 𝑐𝑧3 ))Φ𝑖 }d𝑥 = 0
(C.10)

The spatial dependence of these equations is eliminated by substituting the mode shapes and
carrying out the required integration. The spanwise integrations are carried out numerically
in the simulation, using Gaussian quadrature [136]. This process produces a set of seven
nonlinear ordinary differential equations in terms of the generalized degrees of freedom,
𝑞 𝑤1 , 𝑞 𝑤2 , 𝑞 𝑤3 , 𝑞 𝑣1 , 𝑞 𝑣2 , 𝑞 𝜙1 , and 𝑞 𝜙2 . They can be expressed in state variable form, where
the vector q𝑏 of blade degrees of freedom is:

n oT
q𝑏 = 𝑞 𝑤1 𝑞 𝑤2 𝑞 𝑤3 𝑞 𝑣1 𝑞 𝑣2 𝑞 𝜙1 𝑞 𝜙2 (C.11)

For integration using Gaussian quadrature, the integrand is evaluated at a set number of
stations along the span of the blade at locations corresponding to predefined Gaussian points.
At each station, the sectional airloads are provided by the CFD based RFA model which
requires solving a set of aerodynamic state equations. These aerodynamic state equations
are fully coupled with the blade equations of motion given in Eqs. (C.8)-(C.10) through
the blade degrees of freedom and the aerodynamic loads. The structural and aerodynamic
equations include a set of trim parameters. The trim parameters are obtained by solving a
set of trim equations which enforce force and moment equilibrium at the hub.

175
C.1.2 Propulsive Trim Procedure

The primary trim procedure is based on propulsive trim analysis, modeling actual free-flight
conditions. A helicopter in free flight has six degrees of freedom; thus, six equilibrium
equations must be satisfied. The trim procedure, taken from Ref. [127], enforces these
equilibrium equations in straight and level flight conditions. A modified version of this
procedure developed in Refs. [23, 34] is used for descending flight conditions.
In the case of actual flight, the pilot inputs consist of collective and cyclic inputs (𝜃 0 ,
𝜃 1𝑠 ,𝜃 1𝑐 ) and the tail rotor pitch (𝜃 0𝑡 ). For a given flight condition, the quantities 𝐶𝑊 and 𝜇
are specified, and the trim procedure solves for the equilibrium values of 𝜃 0 , 𝜃 1𝑠 , 𝜃 1𝑐 , 𝜙 𝑅 ,
𝛼 𝑅 , and 𝜃 0𝑡 . These variables comprise the six-component helicopter trim vector q𝑡 .
Only the average values of the rotor hub forces and moments, identified by overbars,
are required. Since non-uniform inflow is used in this study, the trim procedure does not
require an explicit inflow relation. The complete six equilibrium equations are enforced in
the present trim calculation. A simplified model for the tail rotor, developed in Ref. [127], is
used.
The vector q𝑡 of trim variables is defined as

n oT
q𝑡 = 𝛼 𝑅 , 𝜃 0 , 𝜃 1𝑐 , 𝜃 1𝑠 , 𝜃 0𝑡 , 𝜙 𝑅 (C.12)

A schematic of a helicopter in descending flight is depicted in Fig. C.1. The equilibrium

equations are formulated in the nonrotating, hub-fixed system (𝑥 1 , 𝑦 1 , 𝑧1 ).
The helicopter weight 𝑊 acts at the center of gravity of the fuselage, which is offset from
the hub center by the distances 𝑋𝐹𝐶 and 𝑍 𝐹𝐶 in the −ê𝑥1 and −ê𝑧1 directions respectively.
The trim procedure has a provision for accommodating the aerodynamic drag that acts at a
location (the aerodynamic center) that is different from the center of gravity. However, in
all the cases considered in this study, the flat plate drag always acts at the center of gravity
of the fuselage, and 𝑋𝐹 𝐴 = 𝑋𝐹𝐶 , 𝑌𝐹 𝐴 = 𝑌𝐹𝐶 , and 𝑍 𝐹 𝐴 = 𝑍 𝐹𝐶 . The lateral center of gravity

176
 THRUST TR

Z1 Z0

X1
OH
X0 Zt
ZFA
ZFC
NE
L HUB PLA
NOMINA
AC
αD DRAG (DF)
αR CG
VELOCITY
(VA)
Xt
XFA
WEIGHT
(W)
XFC

Figure C.1: A schematic of the helicopter in descending flight

offset 𝑌𝐹𝐶 is also set to zero for cases considered in this study. The flat-plate drag is given
by:
1
𝐷𝑓 = 𝜌 𝐴 |V 𝐴 | 2𝐶𝑑𝑓 𝜋𝑅 2 (C.13)
2

A typical value for the flat-plate drag coefficient is

𝐶𝑑𝑓 = 0.01 (C.14)

In descending flight, a constant angle 𝛼𝐷 is defined as the angle between the forward
flight velocity V 𝐴 and the horizontal plane, and is shown in Fig. C.1. The descent angle
can be specified, in combination with 𝑊 and 𝜇. For the cases considered, the descent angle
was zero. Note that the drag force 𝐷 𝑓 will continue to act parallel to the direction of the
resultant velocity V 𝐴 . Setting the descent angle 𝛼𝐷 = 0, trim equations for a level flight
condition similar to those found in Ref. [127] can be recovered. The trim equations are:
a. Pitching Moment

177
Moment equilibrium about the 𝑦 1 axis requires:

𝑝𝑡
𝑀 + 𝑊 [−𝑋𝐹𝐶 cos 𝜙 𝑅 cos 𝛼 𝑅 + 𝑍 𝐹𝐶 cos 𝜙 𝑅 sin 𝛼 𝑅 ]

−𝐷 𝑓 [−𝑋𝐹 𝐴 sin(𝛼 𝑅 − 𝛼𝐷 ) + 𝑍 𝐹 𝐴 cos(𝛼 𝑅 − 𝛼𝐷 )] − 𝑄 𝑡 = 0 (C.15)

b. Rolling Moment
Moment equilibrium about the 𝑥 1 axis requires:

𝑟𝑙
𝑀 − 𝑍 𝐹𝐶 𝑊 sin 𝜙 𝑅 + 𝑇𝑡 𝑍𝑡 = 0 (C.16)

where 𝑇𝑡 is the tail rotor thrust and 𝑍𝑡 is the vertical distance between the hub axis and the
center of the tail rotor.
c. Yawing Moment
Moment equilibrium about the 𝑧1 axis requires:

𝑦𝑤
𝑀 − 𝑋𝐹𝐶 𝑊 sin 𝜙 𝑅 + 𝑇𝑡 𝑋𝑡 = 0 (C.17)

where 𝑇𝑡 is the tail rotor thrust and 𝑋𝑡 is the horizontal distance between the hub axis and
the center of the tail rotor (Fig. C.1).
d. Vertical Force
Force equilibrium in the 𝑧1 direction requires:

𝐹𝑇 − 𝑊 cos 𝛼 𝑅 cos 𝜙 𝑅 − 𝐷 𝑓 sin(𝛼 𝑅 − 𝛼𝐷 ) = 0 (C.18)

e. Longitudinal Force
Force equilibrium in the 𝑥1 direction requires:

𝐻 − 𝑊 sin 𝛼 𝑅 cos 𝜙 𝑅 + 𝐷 𝑓 cos(𝛼 𝑅 − 𝛼𝐷 ) = 0 (C.19)

178
f. Lateral Force
Force equilibrium in the 𝑦 1 direction requires:

𝑌 − 𝑊 sin 𝛼 𝑅 cos 𝜙 𝑅 + 𝑇𝑡 = 0 (C.20)

C.2 Time Integration of the Coupled Trim/Aeroelastic

Equations

The complete aeroelastic model for the blade and control surface consists of three sets of
equations. These are sets of nonlinear differential equations that describe the structural,
aerodynamic, and the trim components of the model. The blade equations of motion
(C.8)-(C.10) required for the coupled trim/aeroelastic response, can be written in the vector
form:
f𝑏 (q𝑏 , q¤ 𝑏 , q¥ 𝑏 , x𝑎 , q𝑡 ; 𝜓) = 0 (C.21)

where q𝑏 represents the vector of blade degrees of freedom, described in Eq. (C.11), x𝑎
represents the vector of aerodynamic states, and q𝑡 represents the trim vector. To convert
Eq. (C.21) to first order form, define a mass matrix given by:

𝜕f𝑏
M𝑏 = (C.22)
𝜕 q¥ 𝑏

This allows Eq. (C.21) to be decomposed into the form:

f𝑏 = g𝑏 (q𝑏 , q¤ 𝑏 , x𝑎 , q𝑡 ; 𝜓) + M𝑏 (q𝑏 , q𝑡 ; 𝜓) q¥ 𝑏 = 0 (C.23)

Solving for q¥ 𝑏 yields

q¥ 𝑏 = −M−1
𝑏 g𝑏 (C.24)

179
Explicit expressions for M𝑏 and g𝑏 can be found in Ref. [124]. The second order system in
Eq. (C.24) can be written in the following state variable form

   
0 I   0 
x¤ 𝑏 =  x + (C.25)
   
0 0 𝑏
 
−M−1 g 





 𝑏 𝑏 

where x𝑏 is the state vector of blade degrees of freedom:


 q𝑏 
 
 

x𝑏 = (C.26)
 q¤ 𝑏 
 
 

Similarly, the complete set of aerodynamic state equations is represented by the vector
expression
f𝑎 (q𝑏 , q¤ 𝑏 , q¥ 𝑏 , x𝑎 , x¤ 𝑎 , q𝑡 ; 𝜓) = 0 (C.27)

which can be written in the first order form

x¤ 𝑎 = g𝑎 (q𝑏 , q¤ 𝑏 , q¥ 𝑏 , x𝑎 , q𝑡 ; 𝜓) (C.28)

The dependence on q¥ 𝑏 is eliminated by substituting Eq. (C.24) into Eq. (C.28), producing
the reduced set of equations
x¤ 𝑎 = g𝑎𝑅 (x𝑏 , x𝑎 , q𝑡 ; 𝜓) (C.29)

Define a state vector y as


 x𝑏 
 
 

y= (C.30)
 x𝑎 
 
 
then, combining Eqs. (C.25) and (C.29) yields a system of coupled first-order state variable
equations of the form:
y¤ = F(y; 𝑡) (C.31)

180
This system is solved numerically using the ODE solver DDEABM, which is a general-
purpose predictor-corrector Adams-Bashforth differential system solver [130].

C.2.1 Solution of the Trim Equations

The dependence of the trim equations (C.15)-(C.20) on blade degrees of freedom q𝑏 and
the aerodynamic states x𝑎 in the trim equations occurs through terms representing the rotor
hub loads. However, only the average values of the hub loads are used in the trim equations.
When only the steady state response of the system is considered, the average values of the
hub loads will depend only on the trim variables q𝑡 defined in Eq. (C.12). The trim equations
are solved using an iterative procedure referred to as the autopilot trim procedure. The trim
solution presented here is identical to that of Ref. [125]. The trim equations can be written
in the form:
f𝑡 (q𝑡 ) = 0 (C.32)

Let R𝑡𝑖 be the vector of trim residuals at the trim condition q𝑡𝑖 at iteration 𝑖:

f𝑡 (q𝑡𝑖 ) = R𝑡𝑖 (C.33)

An iterative optimal control strategy is then used to reduce the value of R𝑡𝑖 based on the
minimization of the performance index:

𝐽 = R𝑡T𝑖 R𝑡𝑖 (C.34)

This algorithm resembles a feedback controller used for vibration reduction. The trim
parameters at the 𝑖th iteration are then given by:

q𝑡𝑖 = −T𝑖−1 R𝑡𝑖−1 + q𝑡𝑖−1 (C.35)

181
where T𝑖 is a transfer matrix describing the sensitivities of trim residuals to changes in the
trim variables:
𝜕R𝑡𝑖
T𝑖 = (C.36)
𝜕q𝑡𝑖

where T𝑖 is computed using a finite difference scheme. Under certain complex flight
conditions, convergence of this procedure can be improved using a relaxation approach:

q𝑡𝑖 = −𝛼T𝑖−1 R𝑡𝑖−1 + q𝑡𝑖−1 (C.37)

where 𝛼 is a relaxation parameter less than unity. The use of this relaxation parameter was
pioneered by Depailler [125].

C.3 Vibratory Hub Shears and Moments

The resultant force and moment at the root of the 𝑘th blade is found by integrating the
distributed inertial, gravitational, aerodynamic, and damping loads p𝑏 , p𝑐 , q𝑏 , and q𝑐 given
in Eqs. (B.74)-(B.75), over the blade span. Following the procedure used in Ref. [125],
Eqs. (8.48)-(8.50), the resultant shears and moments of the 𝑘th blade, at azimuth 𝜓 𝑘 , may
be expressed in the rotating “2” frame as:

∫ 𝐿𝑏
F 𝑅𝑘 (𝜓 𝑘 ) = (p𝑏 + p𝑐 ) d𝑟 (C.38)
0
∫ 𝐿𝑏
M 𝑅𝑘 (𝜓 𝑘 ) = (q𝑏 + q𝑐 ) d𝑟 (C.39)
0

where:
2𝜋(𝑘 − 1)
𝜓𝑘 = 𝜓 + (C.40)
𝑁𝑏

Then, rotor hub shears and moments in the nonrotating “1” frame 𝐹𝐻 (𝜓), 𝑀𝐻 (𝜓) are
computed by summing the contribution of each blade 𝐹𝑅𝑘 (𝜓 𝑘 ), 𝑀𝑅𝑘 (𝜓 𝑘 ) and by converting
them from the “2” frame to the “1” frame using a coordinate transformation described in

182
Eq. (B.2).
In an 𝑁 𝑏 -bladed helicopter, 𝑁 𝑏 /rev is the dominant harmonic of vibratory loads trans-
ferred to the hub. The quantities F𝐻4𝑐 , F𝐻4𝑠 , M𝐻4𝑐 , and M𝐻4𝑠 represent the sin and cos
components of the 4/rev hub shears and moments, and are found using

1 2𝜋
∫
F𝐻4𝑐 = F𝐻 (𝜓) cos 4𝜓 d𝜓 (C.41)
𝜋 0
1 2𝜋
∫
F𝐻4𝑠 = F𝐻 (𝜓) sin 4𝜓 d𝜓 (C.42)
𝜋 0
1 2𝜋
∫
M𝐻4𝑐 = M𝐻 (𝜓) cos 4𝜓 d𝜓 (C.43)
𝜋 0
1 2𝜋
∫
M𝐻4𝑠 = M𝐻 (𝜓) sin 4𝜓 d𝜓 (C.44)
𝜋 0

183
 BIBLIOGRAPHY

[1] Braslow, A. L., A History of Suction-Type Laminar-Flow Control with Emphasis on

Flight Research, No. 13 in Monographs in Aerospace History, NASA, Washington,
DC, 1999.
[2] Gratzer, L. B., and O'Donnell, T. J., “Development of a BLC High-Lift System
for High-Speed Airplanes,” Journal of Aircraft, Vol. 2, No. 6, 1965, pp. 477–484.
https://doi.org/10.2514/3.43686.
[3] Hinton, S. H., “Application of Boundary Layer Control to Rotor Blades,” Journal of
the American Helicopter Society, Vol. 2, No. 2, 1957, pp. 36–64. https://doi.org/10.
4050/jahs.2.36.
[4] Shovlin, M. D., Englar, R. J., Eppel, J. C., and Nichols, J. H., “Large-Scale Static In-
vestigation of Circulation-Control-Wing Concepts Applied to Upper-Surface-Blowing
Aircraft,” NASA Technical Paper 2684, Jan. 1987.
[5] McVeigh, M. A., Nagib, H., Wood, T., and Wygnanski, I., “Full-Scale Flight Tests
of Active Flow Control to Reduce Tiltrotor Aircraft Download,” Journal of Aircraft,
Vol. 48, No. 3, 2011, pp. 786–796. https://doi.org/10.2514/1.46956.
[6] Padthe, A. K., Friedmann, P. P., Chia, M. H., and Liu, L., “Comprehensive Numerical
Assessment of Rotorcraft Vibration and Noise Control Using Microflaps,” Journal of
Aircraft, Vol. 53, No. 4, 2016, pp. 1113–1130. https://doi.org/10.2514/1.c033448.
[7] Liu, L., Padthe, A. K., and Friedmann, P. P., “Computational Study of Microflaps
with Application to Vibration Reduction in Helicopter Rotors,” AIAA Journal, Vol. 49,
No. 7, 2011, pp. 1450–1465. https://doi.org/10.2514/1.j050829.
[8] Friedmann, P. P., “On-Blade Control of Rotor Vibration, Noise, and Performance:
Just Around the Corner? The 33rd Alexander Nikolsky Honorary Lecture,” Journal
of the American Helicopter Society, Vol. 59, No. 4, 2014, pp. 1–37. https://doi.org/10.
4050/jahs.59.041001.
[9] Loewy, R. G., “Helicopter Vibrations: A Technological Perspective,” Journal of the
American Helicopter Society, Vol. 29, No. 4, 1984, pp. 4–30.
[10] Friedmann, P. P., “Helicopter Vibration Reduction Using Structural Optimization
with Aeroelastic/Multidisciplinary Constraints,” Journal of Aircraft, Vol. 28, No. 1,
1991, pp. 8–21.

184
[11] Celi, R., “Recent Applications of Design Optimization to Rotorcraft – A Survey,”
Journal of Aircraft, Vol. 36, No. 1, 1999, pp. 176–189.

[12] Glaz, B., Friedmann, P. P., and Liu, L., “Helicopter Vibration Reduction throughout
the Entire Flight Envelope Using Surrogate-Based Optimization,” Journal of the
American Helicopter Society, Vol. 54, No. 1, 2009, pp. 1–15. https://doi.org/10.4050/
jahs.54.012007.

[13] Johnson, W., “Self-Tuning Regulators for Multicyclic Control of Helicopter Vibra-
tions,” NASA Technical Paper 1996, Mar. 1982.

[14] Hammond, C. E., “Higher Harmonic Control: A Historical Perspective,” Journal of

the American Helicopter Society, Vol. 66, No. 2, 2021, pp. 1–14. https://doi.org/10.
4050/jahs.66.022001.

[15] Wood, E. R., Powers, R. W., Cline, J. H., and Hammond, C. E., “On Developing
and Flight Testing a Higher Harmonic Control System,” Journal of the American
Helicopter Society, Vol. 30, No. 1, 1985, pp. 3–20. https://doi.org/10.4050/jahs.30.1.3.

[16] Norman, T. R., Theodore, C., Shinoda, P. M., Fuerst, D., Arnold, U. T. P., Makinen,
S., Lorber, P., and O'Neill, J., “Full-Scale Wind Tunnel Test of a UH-60 Individual
Blade Control System for Performance Improvement and Vibration, Loads, and Noise
Control,” 65th AHS Annual Forum, American Helicopter Society, Grapevine, TX,
May 27-29, 2009.

[17] Yeo, H., Jain, R., and Jayaraman, B., “Investigation of Rotor Vibratory Loads of
a UH-60A Individual Blade Control System,” Journal of the American Helicopter
Society, Vol. 61, No. 3, 2016, pp. 1–16. https://doi.org/10.4050/jahs.61.032009.

[18] Roth, D., “Advanced Vibration Reduction by IBC Technology,” 30th European
Rotorcraft Forum, Marseilles, France, September 14-16, 2004.

[19] Friedmann, P. P., and Millott, T. A., “Vibration Reduction in Rotorcraft Using Active
Control - A Comparison of Various Approaches,” Journal of Guidance, Control, and
Dynamics, Vol. 18, No. 4, 1995, pp. 664–673. https://doi.org/10.2514/3.21445.

[20] Wilbur, M. L., Mirick, P. H., Yeager, Jr., W. T., Langston, C. W., Cesnik, C. E. S.,
and Shin, S. J., “Vibratory Loads Reduction Testing of the NASA/Army/MIT Active
Twist Rotor,” Journal of the American Helicopter Society, Vol. 47, No. 2, 2002, pp.
123–133.

[21] Bernhard, A. P. F., and Wong, J., “Wind-Tunnel Evaluation of a Sikorsky Active Rotor
Controller Implemented on the NASA/ARMY/MIT Active Twist Rotor,” Journal of
the American Helicopter Society, Vol. 50, No. 1, 2005, pp. 65–81. https://doi.org/10.
4050/1.3092844.

[22] Shin, S., Cesnik, C. E. S., and Hall, S. R., “Closed-Loop Control Test of the
NASA/Army/MIT Active Twist Rotor for Vibration Reduction,” Journal of the

185
 American Helicopter Society, Vol. 50, No. 2, 2005, pp. 178–194. https://doi.org/10.
4050/1.3092854.

[23] Patt, D., Liu, L., and Friedmann, P. P., “Rotorcraft Vibration Reduction and Noise
Prediction Using a Unified Aeroelastic Response Simulation,” Journal of the Ameri-
can Helicopter Society, Vol. 50, No. 1, 2005, pp. 95–106. https://doi.org/10.4050/1.
3092846.

[24] Patt, D., Liu, L., and Friedmann, P. P., “Simultaneous Vibration and Noise Reduction
in Rotorcraft Using Aeroelastic Simulation,” Journal of the American Helicopter
Society, Vol. 51, No. 2, 2006, pp. 127–140. https://doi.org/10.4050/jahs.51.127.

[25] Ravichandran, K., Chopra, I., Wake, B. E., and Hein, B., “Trailing-Edge Flaps for
Rotor Performance Enhancement and Vibration Reduction,” Journal of the American
Helicopter Society, Vol. 58, No. 2, 2013, pp. 1–13. https://doi.org/10.4050/jahs.58.
022006.

[26] Koratkar, N. A., and Chopra, I., “Wind Tunnel Testing of a Smart Rotor Model with
Trailing-Edge Flaps,” Journal of the American Helicopter Society, Vol. 47, No. 4,
2002, pp. 263–272. https://doi.org/10.4050/jahs.47.263.

[27] Roget, B., and Chopra, I., “Wind-Tunnel Testing of Rotor with Individually Con-
trolled Trailing-Edge Flaps for Vibration Reduction,” Journal of Aircraft, Vol. 45,
No. 3, 2008, pp. 868–879. https://doi.org/10.2514/1.28455.

[28] Straub, F. K., Anand, V. R., Lau, B. H., and Birchette, T. S., “Wind Tunnel Test of
the SMART Active Flap Rotor,” Journal of the American Helicopter Society, Vol. 63,
No. 1, 2018, pp. 1–16. https://doi.org/10.4050/jahs.63.012002.

[29] Rabourdin, A., Maurice, J., Dieterich, O., and Konstanzer, P., “Blue Pulse Active
Rotor Control at Airbus Helicopters – New EC145 Demonstrator and Flight Test
Results,” 70th AHS Annual Forum, American Helicopter Society, Montréal, Canada,
May 20-22, 2014.

[30] Matalanis, C. G., Wake, B. E., Opoku, D., Min, B.-Y., Yeshala, N., and Sankar,
L., “Aerodynamic Evaluation of Miniature Trailing-Edge Effectors for Active Rotor
Control,” Journal of Aircraft, Vol. 48, No. 3, 2011, pp. 995–1004. https://doi.org/10.
2514/1.c031191.

[31] Padthe, A. K., and Friedmann, P. P., “Simultaneous Blade-Vortex Interaction Noise
and Vibration Reduction in Rotorcraft Using Microflaps, Including the Effect of
Actuator Saturation,” Journal of the American Helicopter Society, Vol. 60, No. 4,
2015, pp. 1–16. https://doi.org/10.4050/jahs.60.042002.

[32] Min, B.-Y., Sankar, L. N., and Bauchau, O. A., “A CFD–CSD coupled-analysis
of HART-II rotor vibration reduction using gurney flaps,” Aerospace Science and
Technology, Vol. 48, 2016, pp. 308–321. https://doi.org/10.1016/j.ast.2015.11.024.

186
[33] Glaz, B., Friedmann, P., Liu, L., Kumar, D., and Cesnik, C., “The AVINOR Aeroelas-
tic Simulation Code and Its Application to Reduced Vibration Composite Rotor Blade
Design,” 50th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Ma-
terials Conference, Palm Springs, CA, May 4-7, 2009. https://doi.org/10.2514/6.2009-
2601, AIAA Paper 2009-2601.

[34] Padthe, A., “Active Noise and Vibration Alleviation in Rotorcraft Using Microflaps,”
Ph.D. thesis, University of Michigan, 2011.

[35] Attar, P. J., Dowell, E. H., White, J. R., and Thomas, J. P., “Reduced Order Nonlinear
System Identification Methodology,” AIAA Journal, Vol. 44, No. 8, 2006, pp. 1895–
1904. https://doi.org/10.2514/1.16221.

[36] Glaz, B., Liu, L., and Friedmann, P. P., “Reduced-Order Nonlinear Unsteady Aerody-
namic Modeling Using a Surrogate-Based Recurrence Framework,” AIAA Journal,
Vol. 48, No. 10, 2010, pp. 2418–2429. https://doi.org/10.2514/1.J050471.

[37] Ghoreyshi, M., Jirásek, A., and Cummings, R. M., “Reduced order unsteady aerody-
namic modeling for stability and control analysis using computational fluid dy-
namics,” Progress in Aerospace Sciences, Vol. 71, 2014, pp. 167–217. https:
//doi.org/10.1016/j.paerosci.2014.09.001.

[38] Cribbs, R., and Friedmann, P., “Actuator Saturation and Its Influence on Vibration
Reduction by Actively Controlled Flaps,” 42nd AIAA/ASME/ASCE/AHS/ACS Struc-
tures, Structural Dynamics, and Materials Conference, Seattle, WA, April 16-19,
2001. https://doi.org/10.2514/6.2001-1467, AIAA Paper 2001-1467.

[39] Roget, B., and Chopra, I., “Closed-Loop Test of a Rotor with Individually Controlled
Trailing-Edge Flaps for Vibration Reduction,” Journal of the American Helicopter
Society, Vol. 55, No. 1, 2010, pp. 1–12. https://doi.org/10.4050/jahs.55.012009.

[40] Gad-el-Hak, M., Flow Control: Passive, Active, and Reactive Flow Management,
Cambridge University Press, 2000. https://doi.org/10.1017/cbo9780511529535.

[41] Fiedler, H., and Fernholz, H.-H., “On management and control of turbulent shear
flows,” Progress in Aerospace Sciences, Vol. 27, No. 4, 1990, pp. 305–387. https:
//doi.org/10.1016/0376-0421(90)90002-2.

[42] Prandtl, L., “Über Flüssigkeitsbewegung bei sehr kleiner Reibung,” Verhandlungen
des III. Internationalen Mathematiker-Kongresses, Heidelberg, Germany, 1904, pp.
484–491. Source of English translation: Prandtl, L., “Motion of Fluids with Very
Little Viscosity,” NACA TM 452, March 1928.

[43] Williams, D. R., and MacMynowski, D. G., “Brief History of Flow Control,” Fun-
damentals and Applications of Modern Flow Control, Progress in Astronautics and
Aeronautics, Vol. 231, edited by R. D. Joslin and D. N. Miller, American Institute of
Aeronautics and Astronautics, Reston, VA, 2009, Chap. 1, pp. 1–20.

187
[44] Krishnan, K., Bertram, O., and Seibel, O., “Review of hybrid laminar flow control
systems,” Progress in Aerospace Sciences, Vol. 93, 2017, pp. 24–52. https://doi.org/
10.1016/j.paerosci.2017.05.005.

[45] Joslin, R. D., and Jones, G. S. (eds.), Applications of Circulation Control Technologies,
Progress in Astronautics and Aeronautics, Vol. 214, American Institute of Aeronautics
and Astronautics, Reston, VA, 2006.

[46] Blaylock, M., Chow, R., Cooperman, A., and van Dam, C. P., “Comparison of
Pneumatic Jets and Tabs for Active Aerodynamic Load Control,” Wind Energy,
Vol. 17, No. 9, 2014, pp. 1365–1384. https://doi.org/10.1002/we.1638.

[47] Dowling, A. P., and Morgans, A. S., “Feedback Control of Combustion Oscillations,”
Annual Review of Fluid Mechanics, Vol. 37, No. 1, 2005, pp. 151–182. https:
//doi.org/10.1146/annurev.fluid.36.050802.122038.

[48] Raman, G., and Raghu, S., “Cavity Resonance Suppression Using Miniature Fluidic
Oscillators,” AIAA Journal, Vol. 42, No. 12, 2004, pp. 2608–2612. https://doi.org/10.
2514/1.521.

[49] Raghu, S., “Fluidic Oscillators for Flow Control,” Experiments in Fluids, Vol. 54,
No. 2, 2013, pp. 1–11. https://doi.org/10.1007/s00348-012-1455-5.

[50] Cattafesta, L. N., and Sheplak, M., “Actuators for Active Flow Control,” Annual
Review of Fluid Mechanics, Vol. 43, No. 1, 2011, pp. 247–272. https://doi.org/10.
1146/annurev-fluid-122109-160634.

[51] Crittenden, T., Woo, G. T., and Glezer, A., “Combustion Powered Actuators for
Separation Control,” 6th AIAA Flow Control Conference, New Orleans, LA, June
25-28, 2012. https://doi.org/10.2514/6.2012-3135, AIAA Paper 2012-3135.

[52] Leishman, J. G., Principles of Helicopter Aerodynamics, 2nd ed., Cambridge Univer-
sity Press, 2016.

[53] Karim, M. A., and Acharya, M., “Suppression of Dynamic-Stall Vortices over Pitching
Airfoils by Leading-Edge Suction,” AIAA Journal, Vol. 32, No. 8, 1994, pp. 1647–
1655. https://doi.org/10.2514/3.12155.

[54] Yu, Y. H., Lee, S., McAlister, K. W., Tung, C., and Wang, C. M., “Dynamic Stall
Control for Advanced Rotorcraft Application,” AIAA Journal, Vol. 33, No. 2, 1995,
pp. 289–295. https://doi.org/10.2514/3.12496.

[55] Gardner, A. D., Richter, K., Mai, H., and Neuhaus, D., “Experimental Investigation
of Air Jets to Control Shock-Induced Dynamic Stall,” Journal of the American
Helicopter Society, Vol. 59, No. 2, 2014, pp. 1–11. https://doi.org/10.4050/jahs.59.
022003.

188
[56] Gardner, A. D., Richter, K., Mai, H., and Neuhaus, D., “Experimental investigation of
high-pressure pulsed blowing for dynamic stall control,” CEAS Aeronautical Journal,
Vol. 5, No. 2, 2014, pp. 185–198. https://doi.org/10.1007/s13272-014-0099-y.

[57] Green, R. B., Prince, S. A., Wang, Y., Khodagolian, V., and Coton, F. N., “Delay
of Dynamic Stall Using Pulsed Air-Jet Vortex Generators,” AIAA Journal, Vol. 56,
No. 5, 2018, pp. 2070–2074. https://doi.org/10.2514/1.j056560.

[58] Prince, S., Green, R., Coton, F., and Wang, Y., “The Effect of Steady and Pulsed Air
Jet Vortex Generator Blowing on an Airfoil Section Model Undergoing Sinusoidal
Pitching,” Journal of the American Helicopter Society, Vol. 64, No. 3, 2019, pp. 1–14.
https://doi.org/10.4050/jahs.64.032004.

[59] Magill, J., and McManus, K., “Control of Dynamic Stall Using Pulsed Vortex Gener-
ator Jets,” 36th AIAA Aerospace Sciences Meeting and Exhibit, Reno, NV, January
12-15, 1998. https://doi.org/10.2514/6.1998-675, AIAA Paper 98-0675.

[60] Woo, G. T. K., and Glezer, A., “Controlled transitory stall on a pitching airfoil
using pulsed actuation,” Experiments in Fluids, Vol. 54, No. 6, 2013, pp. 1–15.
https://doi.org/10.1007/s00348-013-1507-5.

[61] Matalanis, C. G., Min, B.-Y., Bowles, P. O., Jee, S., Wake, B. E., Crittenden, T. M.,
Woo, G., and Glezer, A., “Combustion-Powered Actuation for Dynamic-Stall Suppres-
sion: High-Mach Simulations and Low-Mach Experiments,” AIAA Journal, Vol. 53,
No. 8, 2015, pp. 2151–2163. https://doi.org/10.2514/1.j053641.

[62] Tan, Y., Crittenden, T. M., and Glezer, A., “Aerodynamic Control of a Dynamically
Pitching VR-12 Airfoil using Discrete Pulsed Actuation,” 54th AIAA Aerospace
Sciences Meeting, San Diego, CA, January 4-8, 2016. https://doi.org/10.2514/6.2016-
0321, AIAA Paper 2016-0321.

[63] Bons, J., Frankhouser, M., and Gregory, J., “Synchronized Flow Control of Dynamic
Stall under Coupled Pitch and Freestream Oscillations,” 73rd AHS Annual Forum,
American Helicopter Society, Fort Worth, TX, May 9-11, 2017.

[64] Rice, T. T., Taylor, K., and Amitay, M., “Wind tunnel quantification of dynamic stall
on an S817 airfoil and its control using synthetic jet actuators,” Wind Energy, Vol. 22,
No. 1, 2018, pp. 21–33. https://doi.org/10.1002/we.2266.

[65] Müller-Vahl, H. F., Nayeri, C. N., Paschereit, C. O., and Greenblatt, D., “Dynamic
Stall Control via Adaptive Blowing,” Renewable Energy, Vol. 97, No. 1, 2016, pp.
47–64. https://doi.org/10.1016/j.renene.2016.05.053.

[66] McCloud, J. L., Hall, L. P., and Brady, J. A., “Full-Scale Wind-Tunnel Tests of
Blowing Boundary-Layer Control Applied to a Helicopter Rotor,” NASA Technical
Note TN D-335, Sep. 1960.

189
[67] Lorber, P., Wallace, B., and Scott, M., “Internal and Blade Surface Pressure Measure-
ments of Centrifugal Flow Control on a Large Scale Rotor,” 73rd AHS Annual Forum,
American Helicopter Society, Fort Worth, TX, May 9-11, 2017.

[68] Le Pape, A., Lienard, C., Verbeke, C., Pruvost, M., and De Coninck, J.-L., “He-
licopter Fuselage Drag Reduction Using Active Flow Control: A Comprehensive
Experimental Investigation,” Journal of the American Helicopter Society, Vol. 60,
No. 3, 2015, pp. 1–12. https://doi.org/10.4050/jahs.60.032003.

[69] De Gregorio, F., “Helicopter Fuselage Model Drag Reduction by Active Flow Control
Systems,” Journal of the American Helicopter Society, Vol. 64, No. 2, 2019, pp. 1–15.
https://doi.org/10.4050/jahs.64.022001.

[70] Woo, G. T. K., Glezer, A., Yorish, S., and Crittenden, T. M., “Pulsed Actuation
Control of Flow Separation on a ROBIN Rotorcraft Fuselage,” AIAA Journal, Vol. 54,
No. 10, 2016, pp. 3274–3289. https://doi.org/10.2514/1.j054008.

[71] Seifert, A., Darabi, A., and Wygnanski, I., “Delay of Airfoil Stall by Periodic
Excitation,” Journal of Aircraft, Vol. 33, No. 4, 1996, pp. 691–698. https://doi.
org/10.2514/3.47003.

[72] Greenblatt, D., and Wygnanski, I. J., “The control of flow separation by periodic
excitation,” Progress in Aerospace Sciences, Vol. 36, No. 7, 2000, pp. 487–545.
https://doi.org/10.1016/s0376-0421(00)00008-7.

[73] Darabi, A., and Wygnanski, I., “Active Management of Naturally Separated Flow
Over a Solid Surface. Part 1. The Forced Reattachment Process,” Journal of Fluid Me-
chanics, Vol. 510, 2004, pp. 105–129. https://doi.org/10.1017/s0022112004009231.

[74] DeSalvo, M., Whalen, E., and Glezer, A., “High-Lift Performance Enhancement
Using Active Flow Control,” AIAA Journal, Vol. 58, No. 10, 2020, pp. 4228–4242.
https://doi.org/10.2514/1.j059143.

[75] Lin, J. C., Melton, L. P., Hannon, J. A., Andino, M. Y., Koklu, M., Paschal, K. B.,
and Vatsa, V. N., “Testing of High-Lift Common Research Model with Integrated
Active Flow Control,” Journal of Aircraft, Vol. 57, No. 6, 2020, pp. 1121–1133.
https://doi.org/10.2514/1.c035906.

[76] Chabert, T., Dandois, J., and Garnier, É., “Experimental closed-loop control of flow
separation over a plain flap using slope seeking,” Experiments in Fluids, Vol. 55,
No. 8, 2014, pp. 1–19. https://doi.org/10.1007/s00348-014-1797-2.

[77] Leopold, D., Krothapalli, A., and Tavella, D. A., “Some Observations on the Aerody-
namics of an Airfoil with a Jet Exhausting from the Lower Surface,” 21st Aerospace
Sciences Meeting, Reno, NV, January 10-13, 1983. https://doi.org/10.2514/6.1983-
173, AIAA Paper 1983-173.

190
[78] Al-Battal, N., Cleaver, D., and Gursul, I., “Aerodynamic Load Control through
Blowing,” 54th AIAA Aerospace Sciences Meeting, San Diego, CA, January 4-8,
2016. https://doi.org/10.2514/6.2016-1820, AIAA Paper 2016-1820.

[79] Traub, L. W., Miller, A. C., and Rediniotis, O., “Comparisons of a Gurney and Jet-
Flap for Hinge-Less Control,” Journal of Aircraft, Vol. 41, No. 2, 2004, pp. 420–423.
https://doi.org/10.2514/1.6023.

[80] Cooperman, A. B., “Wind Tunnel Testing of Microtabs and Microjets for Active Load
Control of Wind Turbine Blades,” Ph.D. thesis, University of California, Davis, 2012.

[81] Johnson, S. J., Baker, J. P., van Dam, C. P., and Berg, D., “An overview of active load
control techniques for wind turbines with an emphasis on microtabs,” Wind Energy,
Vol. 13, No. 2-3, 2010, pp. 239–253. https://doi.org/10.1002/we.356.

[82] Lin, J. C., Whalen, E. A., Andino, M. Y., Graff, E. C., Lacy, D. S., Washburn,
A. E., Gharib, M., and Wygnanski, I. J., “Full-Scale Testing of Active Flow Control
Applied to a Vertical Tail,” Journal of Aircraft, Vol. 56, No. 4, 2019, pp. 1376–1386.
https://doi.org/10.2514/1.c034907.

[83] Whalen, E. A., Shmilovich, A., Spoor, M., Tran, J., Vijgen, P., Lin, J. C., and Andino,
M., “Flight Test of an Active Flow Control Enhanced Vertical Tail,” AIAA Journal,
Vol. 56, No. 9, 2018, pp. 3393–3398. https://doi.org/10.2514/1.j056959.

[84] Williams, D. R., Tadmor, G., Colonius, T., Kerstens, W., Quach, V., and Buntain, S.,
“Lift Response of a Stalled Wing to Pulsatile Disturbances,” AIAA Journal, Vol. 47,
No. 12, 2009, pp. 3031–3037. https://doi.org/10.2514/1.45407.

[85] Abramson, P., Vukasinovic, B., and Glezer, A., “Fluidic Control of Aerodynamic
Forces on a Bluff Body of Revolution,” AIAA Journal, Vol. 50, No. 4, 2012, pp.
832–843. https://doi.org/10.2514/1.j051156.

[86] Englar, R. J., Hemmerly, R. A., Moore, W. H., Seredinsky, V., Valckenaere, W., and
Jackson, J. A., “Design of the Circulation Control Wing STOL Demonstrator Aircraft,”
Journal of Aircraft, Vol. 18, No. 1, 1981, pp. 51–58. https://doi.org/10.2514/3.57463.

[87] Warsop, C., and Crowther, W. J., “Fluidic Flow Control Effectors for Flight Control,”
AIAA Journal, Vol. 56, No. 10, 2018, pp. 3808–3824. https://doi.org/10.2514/1.
j056787.

[88] “DARPA Selects Teams to Develop Active Flow Control X-Plane,” DARPA news
update, 26 August 2021. URL https://www.darpa.mil/news-events/2021-08-26, ac-
cessed: 13 September 2021.

[89] Schrauf, G., and von Geyr, H., “Hybrid Laminar Flow Control on A320 Fin: Retrofit
Design and Sample Results,” Journal of Aircraft, 2021. Published online 20 May
2021.

191
 [90] Collis, S. S., Joslin, R. D., Seifert, A., and Theofilis, V., “Issues in Active Flow
Control: Theory, Control, Simulation, and Experiment,” Progress in Aerospace
Sciences, Vol. 40, No. 4-5, 2004, pp. 237–289. https://doi.org/10.1016/j.paerosci.
2004.06.001.

[91] Aram, S., Lee, Y.-T., Shan, H., and Vargas, A., “Computational Fluid Dynamic
Analysis of Fluidic Actuator for Active Flow Control Applications,” AIAA Journal,
Vol. 56, No. 1, 2018, pp. 111–120. https://doi.org/10.2514/1.j056255.

[92] Koukpaizan, N. K., Heathcote, D. J., Glezer, A., and Smith, M. J., “Numerical
Simulation of Fluidic Oscillators for Flow Control,” 75th VFS Annual Forum, Vertical
Flight Society, Philadelphia, PA, May 13-16, 2019.

[93] Höll, T., vel Job, A. K., Giacopinelli, P., and Thiele, F., “Numerical Study of Active
Flow Control on a High-Lift Configuration,” Journal of Aircraft, Vol. 49, No. 5, 2012,
pp. 1406–1422. https://doi.org/10.2514/1.c031718.

[94] Parekh, D., Palaniswamy, S., and Goldberg, U., “Numerical Simulation of Separation
Control via Synthetic Jets,” 1st Flow Control Conference, St. Louis, MO, June 24-26,
2002. https://doi.org/10.2514/6.2002-3167, AIAA Paper 2002-3167.

[95] You, D., and Moin, P., “Active Control of Flow Separation over an Airfoil using
Synthetic Jets,” Journal of Fluids and Structures, Vol. 24, No. 8, 2008, pp. 1349–1357.
https://doi.org/10.1016/j.jfluidstructs.2008.06.017.

[96] Duvigneau, R., Hay, A., and Visonneau, M., “Study on the Optimal Location of a
Synthetic Jet for Stall Control,” 3rd AIAA Flow Control Conference, San Francisco,
CA, June 5-8, 2006. https://doi.org/10.2514/6.2006-3679, AIAA Paper 2006-3679.

[97] Duraisamy, K., and Baeder, J., “Active Flow Control Concepts for Rotor Airfoils
Using Synthetic Jets,” 1st Flow Control Conference, St. Louis, MO, June 24-26, 2002.
https://doi.org/10.2514/6.2002-2835, AIAA Paper 2002-2835.

[98] Tran, S. A., Fisher, A., Corson, D. A., and Sahni, O., “Dynamic Stall Alleviation for
an SC1095 Airfoil using Synthetic Jets,” 53rd AIAA Aerospace Sciences Meeting,
Kissimmee, FL, January 5-9, 2015. https://doi.org/10.2514/6.2015-1038, AIAA
Paper 2015-1038.

[99] Matalanis, C. G., Bowles, P. O., Jee, S., Min, B.-Y., Kuczek, A. E., Croteau, P. F.,
Wake, B. E., Crittenden, T., Glezer, A., and Lorber, P. F., “Dynamic Stall Suppression
Using Combustion-Powered Actuation (COMPACT),” NASA CR-2016-219336, Sep.
2016.

[100] Williams, R. M., “Application of Circulation Control Rotor Technology to a Stopped

Rotor Aircraft Design,” 1st European Rotorcraft and Powered Lift Aircraft Forum,
Southampton, England, 1975.

192
[101] Hassan, A. A., Straub, F. K., and Charles, B. D., “Effects of Surface Blowing/Suction
on the Aerodynamics of Helicopter Rotor Blade-Vortex Interactions (BVI) — A
Numerical Simulation,” Journal of the American Helicopter Society, Vol. 42, No. 2,
1997, pp. 182–194. https://doi.org/10.4050/jahs.42.182.
[102] Johnson, W., “A Comprehensive Analytical Model of Rotorcraft Aerodynamics and
Dynamics,” Johnson Aeronautics, Palo Alto, CA, 1988.
[103] Min, B.-Y., Lorber, P. F., Berezin, C. R., Wake, B. E., and Scott, M. W., “Numerical
Study of Retreating Side Blowing Concept for a Rotor in High-Speed Flight,” Journal
of the American Helicopter Society, Vol. 63, No. 4, 2018, pp. 1–13. https://doi.org/10.
4050/jahs.63.042003.
[104] Epstein, M., “Theoretical Investigation of the Switching Mechanism in a Bistable
Wall Attachment Fluid Amplifier,” Journal of Basic Engineering, Vol. 93, No. 1,
1971, pp. 55–62. https://doi.org/10.1115/1.3425182.
[105] Tan, Y., and Glezer, A., “Transitory, Bi-Directional Control of the Aerodynamic
Loads on an Airfoil,” AIAA Scitech 2020 Forum, Orlando, FL, January 6-10, 2020.
https://doi.org/10.2514/6.2020-0823, AIAA Paper 2020-0823.
[106] Chakravarthy, S., Peroomian, O., Goldberg, U., and Palaniswamy, S., “The CFD++
Computational Fluid Dynamics Software Suite,” AIAA and SAE, 1998 World Aviation
Conference, Anaheim, CA, September 28-30, 1998. https://doi.org/10.2514/6.1998-
5564, AIAA Paper 1998-5564.
[107] Tan, Y., and Glezer, A., “Controlled Aerodynamic Loads at Low Angles of Attack
using Fluidic Actuation,” AIAA Scitech 2019 Forum, San Diego, CA, January 7-11,
2019. https://doi.org/10.2514/6.2019-0889, AIAA Paper 2019-0889.
[108] Barlow, J. B., Rae, W. H., and Pope, A., Low-Speed Wind Tunnel Testing, 3rd ed.,
John Wiley & Sons, New York, NY, 1999.
[109] Jeong, J., and Hussain, F., “On the Identification of a Vortex,” Journal of Fluid
Mechanics, Vol. 285, 1995, pp. 69–94. https://doi.org/10.1017/s0022112095000462.
[110] Jee, S., Bowles, P., Matalanis, C., Min, B.-Y., Wake, B., Crittenden, T., and Glezer, A.,
“Computations of Combustion-Powered Actuation for Dynamic Stall Suppression,”
72nd AHS Annual Forum, American Helicopter Society, West Palm Beach, FL, May
17-19, 2016.
[111] O'Neill, C. R., and Arena, A. S., “Time Domain Training Signals Comparison
for Computational Fluid Dynamics Based Aerodynamic Identification,” Journal of
Aircraft, Vol. 42, No. 2, 2005, pp. 421–428. https://doi.org/10.2514/1.6424.
[112] Wang, Q., Medeiros, R. R., Cesnik, C. E., Fidkowski, K., Brezillon, J., and Bleecke,
H. M., “Techniques for Improving Neural Network-based Aerodynamics Reduced-
order Models,” AIAA Scitech 2019 Forum, San Diego, CA, January 7-11, 2019.
https://doi.org/10.2514/6.2019-1849, AIAA Paper 2019-1849.

193
[113] Raveh, D. E., “Identification of Computational-Fluid-Dynamics Based Unsteady
Aerodynamic Models for Aeroelastic Analysis,” Journal of Aircraft, Vol. 41, No. 3,
2004, pp. 620–632. https://doi.org/10.2514/1.3149.

[114] Cowan, T. J., Arena, A. S., and Gupta, K. K., “Accelerating Computational Fluid
Dynamics Based Aeroelastic Predictions Using System Identification,” Journal of
Aircraft, Vol. 38, No. 1, 2001, pp. 81–87. https://doi.org/10.2514/2.2737.

[115] Kerstens, W., Pfeiffer, J., Williams, D., King, R., and Colonius, T., “Closed-Loop
Control of Lift for Longitudinal Gust Suppression at Low Reynolds Numbers,” AIAA
Journal, Vol. 49, No. 8, 2011, pp. 1721–1728. https://doi.org/10.2514/1.j050954.

[116] Forrester, A., Sóbester, A., and Keane, A., Engineering Design via Surrogate Model-
ing: A Practical Guide, AIAA, Reston, VA, 2008. Chapter 2.

[117] Pintelon, R., and Schoukens, J., System Identification, 2nd ed., John Wiley & Sons,
Hoboken, NJ, 2012.

[118] Godfrey, K. (ed.), Perturbation Signals for System Identification, Prentice Hall,
Englewood Cliffs, NJ, 1993.

[119] Myrtle, T. F., and Friedmann, P. P., “Application of a New Compressible Time
Domain Aerodynamic Model to Vibration Reduction in Helicopters Using an Actively
Controlled Flap,” Journal of the American Helicopter Society, Vol. 46, No. 1, 2001,
pp. 32–43. https://doi.org/10.4050/jahs.46.32.

[120] Sacks, J., Welch, W., Mitchell, T., and Wynn, H., “Design and Analysis of Computer
Experiments,” Statistical Science, Vol. 4, No. 4, 1989, pp. 409–435. https://doi.org/
10.1214/ss/1177012413.

[121] Jones, D. R., Schonlau, M., and Welch, W. J., “Efficient Global Optimization of
Expensive Black Box Functions,” Journal of Global Optimization, Vol. 13, No. 4,
1998, pp. 455–492. https://doi.org/10.1023/A:1008306431147.

[122] Simpson, T., Peplinski, J., Koch, P. N., and Allen, J., “Metamodels for Computer-
Based Engineering Design: Survey and Recommendations,” Engineering with Com-
puters, Vol. 17, No. 2, 2001, pp. 129–150. https://doi.org/10.1007/PL00007198.

[123] Lophaven, S. N., Nielsen, H. B., and Søndergaard, J., “DACE - A Matlab Kriging
Toolbox, Version 2.0,” Technical Report, Informatics and Mathematical Modelling,
Technical University of Denmark, DTU, Lyngby, Denmark, 2002. IMM-TR-2002-12.

[124] Millot, T. A., and Friedmann, P. P., “Vibration Reduction in Helicopter Rotors Using
an Actively Controlled Partial Span Trailing Edge Flap Located on the Blade,” NASA
CR 4611, Jun. 1994.

[125] Depailler, G., “Alleviation of Dynamic Stall Induced Vibrations in Helicopter Rotors
Using Actively Controlled Flaps,” Ph.D. thesis, University of Michigan, 2002.

194
[126] Petot, D., “Differential Equation Modeling of Dynamic Stall,” La Recherche Aérospa-
tiale, Vol. 5, 1989, pp. 59–71.

[127] de Terlizzi, M., “Blade Vortex Interaction and Its Alleviation Using Passive and
Active Control Approaches,” Ph.D. thesis, University of California, Los Angeles,
1999.

[128] Scully, M. P., “Computation of Helicopter Rotor Wake Geometry and Its Influence
on Rotor Harmonic Airloads,” Ph.D. thesis, Massachusetts Institute of Technology,
1975.

[129] Myrtle, T., “Development of an Improved Aeroelastic Model for the Investigation of
Vibration Reduction in Helicopter Rotors Using Trailing Edge Flaps,” Ph.D. thesis,
University of California, Los Angeles, 1998.

[130] Shampine, L. F., and Watts, H. A., “DEPAC - Design of a User Oriented Package of
ODE Solvers,” Report SAND79-2374, Sandia Laboratories, 1979.

[131] Splettstoesser, W. R., Kube, R., Wagner, W., Seelhorst, U., Boutier, A., Micheli,
F., Mercker, E., and Pengel, K., “Key Results From a Higher Harmonic Control
Aeroacoustic Rotor Test (HART),” Journal of the American Helicopter Society,
Vol. 42, No. 1, 1997, pp. 58–78. https://doi.org/10.4050/jahs.42.58.

[132] Patt, D., Liu, L., Chandrasekar, J., Bernstein, D. S., and Friedmann, P. P., “Higher-
Harmonic-Control Algorithm for Helicopter Vibration Reduction Revisited,” Journal
of Guidance, Control, and Dynamics, Vol. 28, No. 5, 2005, pp. 918–930. https:
//doi.org/10.2514/1.9345.

[133] Couch, L. W., Digital and Analog Communication Systems, 8th ed., Pearson, 2013.
Chapter 3.

[134] Singh, P., “Aeromechanics of Coaxial Rotor Helicopters using the Viscous Vortex
Particle Method,” Ph.D. thesis, University of Michigan, 2020.

[135] Friedmann, P. P., Glaz, B., and Palacios, R., “A Moderate Deflection Helicopter Rotor
Blade Model with Improved Cross-Sectional Analysis,” International Journal of
Solids and Structures, Vol. 46, 2009, pp. 2186–2200.

[136] Abramowitz, M., and Stegun, I. E. (eds.), Handbook of Mathematical Functions, with
Formulas, Graphs, and Mathematical Tables, Dover Publications, New York, 1965.
Chapter 25.

195
 ProQuest Number: 29030847

INFORMATION TO ALL USERS

The quality and completeness of this reproduction is dependent on the quality
and completeness of the copy made available to ProQuest.

Distributed by ProQuest LLC ( 2022 ).

Copyright of the Dissertation is held by the Author unless otherwise noted.

This work may be used in accordance with the terms of the Creative Commons license
or other rights statement, as indicated in the copyright statement or in the metadata
associated with this work. Unless otherwise specified in the copyright statement
or the metadata, all rights are reserved by the copyright holder.

This work is protected against unauthorized copying under Title 17,

United States Code and other applicable copyright laws.

Microform Edition where available © ProQuest LLC. No reproduction or digitization

of the Microform Edition is authorized without permission of ProQuest LLC.

ProQuest LLC
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346 USA