modeling and identification of unsteady airwake
TRANSCRIPT
The Pennsylvania State University
The Graduate School
MODELING AND IDENTIFICATION OF
UNSTEADY AIRWAKE DISTURBANCES ON ROTORCRAFT
A Thesis in
Aerospace Engineering
by
Sade M. Sparbanie
2008 Sade M. Sparbanie
Submitted in Partial Fulfillment
of the Requirements
for the Degree of
Master of Science
December 2008
ii
The thesis of Sade M. Sparbanie was reviewed and approved* by the following:
Joseph F. Horn
Associate Professor of Aerospace Engineering
Thesis Advisor
Jack W. Langelaan
Assistant Professor of Aerospace Engineering
George Lesieutre
Professor of Aerospace Engineering
Head of the Department of Aerospace Engineering
*Signatures are on file in the Graduate School
iii
ABSTRACT
Shipboard operations can be among the most difficult missions for rotorcraft
pilots. Ordinary tasks such as precision hovering and landing can become increasingly
difficult when performed in the unsteady winds that develop behind ship superstructures
and above a moving ship deck. Often, the increased pilot workload is the limiting factor
in determining whether a particular task can be performed in certain wind-over-deck
conditions. In recent years, research has been performed to identify the ship gusts that are
affecting pilots. If the unsteady winds can be identified, then gust rejection tools can be
developed to help reduce pilot workload, and to make missions safer. Additionally, from
accurate wind information, high fidelity real time simulations can be developed to test
new engineering designs and to train naval pilots.
This study further investigates gust identification methods by applying new
engineering tools to develop offline and online gust identification processes. In the
offline environment, aircraft time history data is used to develop colored noise filters that
capture the spectral properties of the ship airwake gusts. When the continuous filter is
transformed to a discrete transfer function, and excited with a white noise input, the
power spectral density of the filtered output matches the power spectral properties of the
stochastic portion of the unsteady airwake gusts.
Although the offline gust identification method is successful, it can become
computationally intensive to develop gust identification filters for every ship airwake
condition a particular helicopter may encounter. As a result, an online identification
method is examined. In this initial feasibility study, a sixth order autoregressive Burg
iv
model is used to determine colored noise filter coefficients in a real-time simulation
environment. Initial results shows that the spectral properties of the filters converge to
key peaks that were evident in the offline filter fitting method.
In this study, the gust identification principles used to identify the stochastic
airwake properties are then applied to create a real time simulation airwake model.
Previous approaches to developing airwake simulation models of the shipboard
environment employed computational fluid dynamics. Often the memory restrictions
placed on the size of data allowed for the real time simulation applications either limited
the fidelity of the CFD solution, or the length of unrepeatable data created. By relying on
a grid of stochastic filters, which are excited with a random white noise, the
computational effort is significantly reduced and guaranteed not to show noticeable
repetitions in flight simulation data. The stochastic airwake is tested by performing
simulated flight trajectories, as well as using piloted simulation testing at Sikorsky
Aircraft. Initial results suggest similar pilot workload, demonstrating the feasibility and
accuracy of the stochastic airwake simulation model.
v
TABLE OF CONTENTS
LIST OF FIGURES................................................................................................. 7
LIST OF TABLES .................................................................................................. 15
ACKNOWLEDGEMENTS..................................................................................... 16
Chapter 1 Introduction............................................................................................ 1
1.1 Motivation .................................................................................................. 1
1.2 Literature Review ....................................................................................... 2
1.2.1 Ship Airwake Modeling Through Experimental Testing .................... 2
1.2.2 Computational Fluid Dynamics Ship Airwake Models ...................... 3
1.2.3 Gust Disturbance Identification......................................................... 4
1.2.4 Stability Augmentation System Control for Gust Rejection ............... 4
1.2.5 Model Following Control Architecture for Gust Rejection ................ 5
1.2.6 Trailing Edge Gust Rejection Controller ........................................... 7
1.3 Focus of Current Research .......................................................................... 8
1.3.1 Disturbance Identification in Offline and Online Environments......... 8
1.3.2 Simulation Modeling of Airwakes using Stochastic Filters ................ 9
Chapter 2 Simulation Environment ......................................................................... 10
2.1 Genhel-PSU................................................................................................ 10
2.2 CFD Airwakes Used with Genhel-PSU ....................................................... 12
2.2.1 USS Winston S. Churchill Destroyer (DDG-81)................................ 12
2.2.2 USS Peleliu (LHA-5) Tarawa Class Solutions................................... 14
2.3 Simulation Hover Locations used with Genhel-PSU ................................... 15
2.4 Sikorsky GENHEL ..................................................................................... 16
2.4.1 MTE Definition................................................................................. 17
2.4.2 Cooper Harper Rating Scale .............................................................. 18
Chapter 3 Stochastic Filter Theory.......................................................................... 19
3.1 General Theory........................................................................................... 19
3.2 Previous Filter Design................................................................................. 20
3.3 New Design Methodology .......................................................................... 23
3.4 Offline Filter Fitting Results ....................................................................... 26
3.5 Discrete Filter Generation ........................................................................... 33
3.6 Equivalent Body Gusts Transformed to Local Velocity Disturbances.......... 36
3.7 Initial Simulation Results ............................................................................ 38
Chapter 4 Online Identification of Stochastic Filters ............................................... 42
vi
4.1 Previous Work ............................................................................................ 42
4.2 Burg Autoregressive Spectral Estimation .................................................... 43
4.3 Real Time Workshop / Implementation into Genhel-PSU ........................... 46
4.4 Results of Burg AR Model.......................................................................... 49
Chapter 5 Simulation Modeling of Airwakes Using Stochastic Filters..................... 52
5.1 Motivation .................................................................................................. 52
5.2 Stochastic Filter Grid Design ...................................................................... 53
5.3 Implementation into GENHEL-PSU ........................................................... 58
5.4 Genhel-PSU Stochastic Airwake Results..................................................... 62
5.5 Implementation into Sikorsky GENHEL and Piloted Results ...................... 71
Chapter 6 Conclusions ............................................................................................ 76
6.1 Analysis Summary...................................................................................... 76
6.2 Recommendations for Future Work ............................................................ 80
6.2.1 Equivalent Gusts ............................................................................... 81
6.2.2 Stochastic Airwake Model Limitations.............................................. 81
Bibliography............................................................................................................ 83
Appendix A Settings for Genhel-PSU..................................................................... 87
Appendix B Offline Stochastic Filter Fitting Results............................................... 89
B.1 Windowing Comparison............................................................................. 89
B.2 Filter Fit Comparison ................................................................................. 92
B.3 LHA Filter Fits: 30 knots / 0 deg ................................................................ 96
B.4 LHA Filter Fits: 30 knots / 30 deg .............................................................. 99
B.5 DDG-81 Filter Fits: 25 knots / 0 deg........................................................... 103
B.6 DDG-81 Initial Simulation Results............................................................. 106
B.7 LHA 30 kts / 0 deg Initial Simulation Results............................................. 111
B.8 LHA 30 kts / 30 deg Initial Simulation Results........................................... 116
Appendix C Online AR Filter Results ..................................................................... 122
C.1 Online Burg AR Model Fitting Results....................................................... 122
Appendix D Stochastic Simulation Modeling Results ............................................. 126
D.1 Genhel-PSU Simulation Hovering Results ................................................. 126
D.2 Genhel-PSU Simulation Trajectory Maneuver Results ............................... 141
D.3 Sikorsky GENHEL Piloted Simulation Results .......................................... 148
vii
LIST OF FIGURES
Figure 1.1: MFC Architecture.................................................................................. 6
Figure 1.2: Airwake Compensator Comparisons using a Model Following
Controller. [Horn, Bridges, 2007]. .................................................................... 7
Figure 2.1: DDG-81 stationed in Virginia Beach, VA. [Sparbanie, 2008]................ 12
Figure 2.2: Dimensions of the CFD airwake behind the DDG-81 ship...................... 13
Figure 2.3: USS Peleliu (LHA 5) ............................................................................. 14
Figure 2.4: Top View of the LHA Class Ship Outlining Landing Spot 8 .................. 15
Figure 2.5: Approximate Hover Location behind a DDG-81. ................................... 16
Figure 2.6: Visuals used with Sikorsky’s GENHEL during Piloted Simulations.
[Geiger, et al., 2008] ......................................................................................... 17
Figure 2.7: Cooper-Harper Rating Scale .................................................................. 18
Figure 3.1: PSD of the Lateral Gust, v, using the filter fitting method by [Horn,
Bridges, and Lee, 2006]. ................................................................................... 23
Figure 3.2: Longitudinal Equivalent Gust Velocity for the Windowing
Comparison ...................................................................................................... 27
Figure 3.3: Vertical Velocity Equivalent Body Gust for the Filter Design
Comparison ...................................................................................................... 28
Figure 3.4: Roll Rate Equivalent Body Gust for the Filter Design Comparison ........ 29
Figure 3.5: Vertical Velocity Equivalent Body Gust: LHA 30 kts / 0 deg................. 30
Figure 3.6: Yaw Rate Equivalent Body Gust: LHA 30 kts / 0 deg ............................ 30
Figure 3.7: Lateral Velocity Equivalent Body Gust: LHA 30 kts / 30 deg ............... 31
Figure 3.8: Pitch Component Equivalent Body Gust for LHA 30 kts / 30 deg .......... 31
Figure 3.9: Initial Simulation: Longitudinal Velocity Equivalent Body Gust for
DDG-81............................................................................................................ 40
Figure 3.10: Initial Simulation: Longitudinal Velocity PSD for DDG-81 ................. 40
viii
Figure 3.11: Initial Simulation: Vertical Velocity Equivalent Body Gust for DDG-
81 ..................................................................................................................... 41
Figure 3.12: Initial Simulation: Vertical Velocity PSD for DDG-81......................... 41
Figure 4.1: AR Filter Diagram [Orfandis, 1988] ...................................................... 44
Figure 4.2: High Level View of the Gust Filter Identification Controller.................. 47
Figure 4.3: Gust Identification using a Burg AR Estimator ...................................... 48
Figure 4.4: Online AR Filter Convergence for the Roll Rate Equivalent Gust .......... 50
Figure 4.5: Online AR Filter Convergence for the Pitch Rate Equivalent Gust ........ 50
Figure 5.1: Schematic of the x-y coordinate grid. ..................................................... 55
Figure 5.2: Schematic of the x-z coordinate grid. ..................................................... 57
Figure 5.3: Three Dimensional View of the Stochastic Filter Grid. .......................... 58
Figure 5.4: Genhel-PSU Coordinate System Diagram.............................................. 59
Figure 5.5: Aircraft Position and Nearest Eight Grid Points ..................................... 60
Figure 5.6: Lateral Stick Input for 439-0-42 Location .............................................. 64
Figure 5.7: Longitudinal Stick Input for 454-0-52.................................................... 64
Figure 5.8: Collective Stick Input for 454-0-52........................................................ 65
Figure 5.9: Pedal Inputs for 544-37-58..................................................................... 65
Figure 5.10: Longitudinal Velocity of Gust acting on the Fuselage center for 439-
0-42.................................................................................................................. 66
Figure 5.11: Lateral Velocity of Gust acting on the Fuselage center for 454-0-52 .... 67
Figure 5.12: Vertical Velocity of Gust acting on the Fuselage center for 439-0-42... 67
Figure 5.13: Pedal Input for Trajectory Path 1 ......................................................... 70
Figure 5.14: Collective Stick Input for Trajectory Path 2 ......................................... 70
Figure 5.15: Pilot HQR assessment for various Airwake conditions......................... 72
Figure 5.16: Lateral Stick Input for Pilot 4............................................................... 73
ix
Figure 5.17: Longitudinal Stick for Pilot 5............................................................... 74
Figure 5.18: Pedal Input for Pilot 6 .......................................................................... 74
Figure B.1: u - Velocity Component for the Windowing Comparison ...................... 89
Figure B.2: v - Velocity Component for the Windowing Comparison ...................... 90
Figure B.3: w - Velocity Component for the Windowing Comparison ..................... 90
Figure B.4: p - Rotational Component for the Windowing Comparison ................... 91
Figure B.5: q - Velocity Component for the Windowing Comparison ...................... 91
Figure B.6: r - Velocity Component for the Windowing Comparison....................... 92
Figure B.7: u - Velocity Component for the Filter Design Comparison .................... 93
Figure B.8: v - Velocity Component for the Filter Design Comparison .................... 93
Figure B.9: w - Velocity Component for the Filter Design Comparison ................... 94
Figure B.10: p - Roll Component for the Filter Design Comparison......................... 94
Figure B.11: q - Pitch Component for the Filter Design Comparison........................ 95
Figure B.12: r - Yaw Component for the Filter Design Comparison......................... 95
Figure B.13: u - Velocity Component for LHA 30 kts / 0 deg .................................. 96
Figure B.14: v - Velocity Component for LHA 30 kts / 0 deg .................................. 97
Figure B.15: w - Velocity Component for LHA 30 kts / 0 deg ................................. 97
Figure B.16: p - Roll Component for LHA 30 kts / 0 deg......................................... 98
Figure B.17: q - Pitch Component for LHA 30 kts / 0 deg........................................ 98
Figure B.18: r – Yaw Component for LHA 30 kts / 0 deg ........................................ 99
Figure B.19: u - Velocity Component for LHA 30 kts / 30 deg ................................ 100
Figure B.20: v - Velocity Component for LHA 30 kts / 30 deg ................................ 100
Figure B.21: w - Velocity Component for LHA 30 kts / 30 deg ............................... 101
Figure B.22: p - Roll Component for LHA 30 kts / 30 deg....................................... 101
x
Figure B.23: q - Pitch Component for LHA 30 kts / 30 deg...................................... 102
Figure B.24: r - Yaw Component for LHA 30 kts / 30 deg....................................... 102
Figure B.25: u - Velocity Component for DDG-81 25 kts / 0 deg............................. 103
Figure B.26: v - Velocity Component for DDG-81 25 kts / 0 deg............................. 104
Figure B.27: w - Velocity Component for DDG-81 25 kts / 0 deg............................ 104
Figure B.28: p - Roll Component for DDG-81 25 kts / 0 deg ................................... 105
Figure B.29: q - Pitch Component for DDG-81 25 kts / 0 deg.................................. 105
Figure B.30: r - Yaw Component for DDG-81 25 kts / 0 deg ................................... 106
Figure B.31: u - Velocity Component for DDG-81 Initial Simulation Results .......... 107
Figure B.32: v - Velocity Component for DDG-81 Initial Simulation Results .......... 107
Figure B.33: w - Velocity Component for DDG-81 Initial Simulation Results ......... 108
Figure B.34: p – Roll Component for DDG-81 Initial Simulation Results ................ 108
Figure B.35: q - Pitch Component for DDG-81 Initial Simulation Results ............... 109
Figure B.36: r - Yaw Component for DDG-81 Initial Simulation Results................. 109
Figure B.37: u - Velocity PSD of the DDG-81 Initial Simulation Results................. 110
Figure B.38: v - Velocity PSD of the DDG-81 Initial Simulation Results................. 110
Figure B.39: w - Velocity PSD of the DDG-81 Initial Simulation Results................ 111
Figure B.40: u - Velocity Component for LHA 30 kts / 0 deg Initial Simulation
Results.............................................................................................................. 112
Figure B.41: v - Velocity Component for LHA 30 kts / 0 deg Initial Simulation
Results.............................................................................................................. 112
Figure B.42: w - Velocity Component for LHA 30 kts / 0 deg Initial Simulation
Results.............................................................................................................. 113
Figure B.43: p - Roll Component for LHA 30 kts / 0 deg Initial Simulation
Results.............................................................................................................. 113
Figure B.44: q - Pitch Component for LHA 30 kts / 0 deg Initial Simulation
Results.............................................................................................................. 114
xi
Figure B.45: r - Yaw Component for LHA 30 kts / 0 deg Initial Simulation
Results.............................................................................................................. 114
Figure B.46: u - Velocity PSD of the LHA 30 kts / 0 deg Initial Simulation
Results.............................................................................................................. 115
Figure B.47: v - Velocity PSD of the LHA 30 kts / 0 deg Initial Simulation
Results.............................................................................................................. 115
Figure B.48: w - Velocity PSD of the LHA 30 kts / 0 deg Initial Simulation
Results.............................................................................................................. 116
Figure B.49: u - Velocity Component for LHA 30 kts / 30 deg Initial Simulation
Results.............................................................................................................. 117
Figure B.50: v - Velocity Component for LHA 30 kts / 30 deg Initial Simulation
Results.............................................................................................................. 117
Figure B.51: w - Velocity Component for LHA 30 kts / 30 deg Initial Simulation
Results.............................................................................................................. 118
Figure B.52: p - Roll Component for LHA 30 kts / 30 deg Initial Simulation
Results.............................................................................................................. 118
Figure B.53: q - Pitch Component for LHA 30 kts / 30 deg Initial Simulation
Results.............................................................................................................. 119
Figure B.54: r - Yaw Component for LHA 30 kts / 30 deg Initial Simulation
Results.............................................................................................................. 119
Figure B.55: u - Velocity PSD of the LHA 30 kts / 30 deg Initial Simulation
Results.............................................................................................................. 120
Figure B.56: v - Velocity PSD of the LHA 30 kts / 30 deg Initial Simulation
Results.............................................................................................................. 120
Figure B.57: w - Velocity PSD of the LHA 30 kts / 30 deg Initial Simulation
Results.............................................................................................................. 121
Figure C.1: Identified Longitudinal Gust Velocity Gust Filters ................................ 122
Figure C.2: Identified Lateral Gust Velocity Gust Filters ........................................ 123
Figure C.3: Identified Vertical Gust Velocity Gust Filters....................................... 123
Figure C.4: Identified Roll Rate Gust Filters ........................................................... 124
xii
Figure C.5: Identified Pitch Rate Gust Filters.......................................................... 124
Figure C.6: Identified Yaw Rate Gust Filters .......................................................... 125
Figure D.1: Lateral Sick Input at 439-0-42............................................................... 127
Figure D.2: Longitudinal Sick Input at 439-0-42...................................................... 127
Figure D.3: Collective Stick Input at 439-0-42......................................................... 128
Figure D.4: Pedal Input at 439-0-42......................................................................... 128
Figure D.5: Longitudinal Velocity acting on Fuselage at 439-0-42........................... 129
Figure D.6: Lateral Velocity acting on Fuselage at 439-0-42.................................... 129
Figure D.7: Vertical Velocity acting on Fuselage at 439-0-42.................................. 130
Figure D.8: Lateral Sick Input at 454-0-52............................................................... 130
Figure D.9: Longitudinal Sick Input at 454-0-52...................................................... 131
Figure D.10: Collective Stick Input at 454-0-52....................................................... 131
Figure D.11: Pedal Input at 454-0-52....................................................................... 132
Figure D.12: Longitudinal Velocity acting on Fuselage at 454-0-52......................... 132
Figure D.13: Lateral Velocity acting on Fuselage at 454-0-52.................................. 133
Figure D.14: Vertical Velocity acting on Fuselage at 454-0-52 ................................ 133
Figure D.15: Lateral Sick Input at 545-31-33........................................................... 134
Figure D.16: Longitudinal Sick Input at 545-31-33.................................................. 134
Figure D.17: Collective Stick Input at 545-31-33..................................................... 135
Figure D.18: Pedal Input at 545-31-33..................................................................... 135
Figure D.19: Longitudinal Velocity acting on Fuselage at 545-31-33....................... 136
Figure D.20: Lateral Velocity acting on Fuselage at 545-31-33................................ 136
Figure D.21: Vertical Velocity acting on Fuselage at 545-31-33 .............................. 137
Figure D.22: Lateral Sick Input at 454-37-58........................................................... 137
xiii
Figure D.23: Longitudinal Sick Input at 454-37-58.................................................. 138
Figure D.24: Collective Stick Input at 454-37-58..................................................... 138
Figure D.25: Pedal Input at 454-37-58..................................................................... 139
Figure D.26: Longitudinal Velocity acting on Fuselage at 454-37-58....................... 139
Figure D.27: Lateral Velocity acting on Fuselage at 454-37-58................................ 140
Figure D.28: Vertical Velocity acting on Fuselage at 454-37-58 .............................. 140
Figure D.29: Three Dimensional View of Trajectory Path 1..................................... 142
Figure D.30: Aircraft Position Time History of Trajectory Path 1 ............................ 143
Figure D.31: Lateral Stick Input for Trajectory Path 1 ............................................. 143
Figure D.32: Longitudinal Stick Input for Trajectory Path 1 .................................... 144
Figure D.33: Collective Stick Input for Trajectory Path 1 ........................................ 144
Figure D.34: Pedal input for Trajectory Path 1......................................................... 145
Figure D.35: Three Dimensional View of Trajectory Path 2..................................... 145
Figure D.36: Aircraft Position Time History of Trajectory Path 2 ............................ 146
Figure D.37: Lateral Stick Input for Trajectory Path 2 ............................................. 146
Figure D.38: Longitudinal Stick Input for Trajectory Path 2 .................................... 147
Figure D.39: Collective Stick Input for Trajectory Path 2 ........................................ 147
Figure D.40: Pedal input for Trajectory Path 2......................................................... 148
Figure D.41: Lateral Stick Input for Pilot 1.............................................................. 149
Figure D.42: Longitudinal Stick Input for Pilot 1..................................................... 149
Figure D.43: Collective Stick Input for Pilot 1 ......................................................... 150
Figure D.44: Pedal Input for Pilot 1 ......................................................................... 150
Figure D.45: Lateral Stick Input for Pilot 2.............................................................. 151
Figure D.46: Longitudinal Stick Input for Pilot 2..................................................... 151
xiv
Figure D.47: Collective Stick Input for Pilot 2 ......................................................... 152
Figure D.48: Pedal Input for Pilot 2 ......................................................................... 152
Figure D.49: Lateral Stick Input for Pilot 3.............................................................. 153
Figure D.50: Longitudinal Stick Input for Pilot 3..................................................... 153
Figure D.51: Collective Stick Input for Pilot 3 ......................................................... 154
Figure D.52: Pedal Input for Pilot 3 ......................................................................... 154
Figure D.53: Lateral Stick Input for Pilot 4.............................................................. 155
Figure D.54: Longitudinal Stick Input for Pilot 4..................................................... 155
Figure D.55: Collective Stick Input for Pilot 4 ......................................................... 156
Figure D.56: Pedal Input for Pilot 4 ......................................................................... 156
Figure D.57: Lateral Stick Input for Pilot 5.............................................................. 157
Figure D.58: Longitudinal Stick Input for Pilot 5..................................................... 157
Figure D.59: Collective Stick Input for Pilot 5 ......................................................... 158
Figure D.60: Pedal Input for Pilot 5 ......................................................................... 158
Figure D.61: Lateral Stick Input for Pilot 6.............................................................. 159
Figure D.62: Longitudinal Stick Input for Pilot 6..................................................... 159
Figure D.63: Collective Stick Input for Pilot 6 ......................................................... 160
Figure D.64: Pedal Input for Pilot 6 ......................................................................... 160
xv
LIST OF TABLES
Table 3.1: Gust Filters Derived for DDG81 Hover Location, WOD: 25kts / 0deg .... 32
Table 3.2: Gust Filters Derived for LHA LS 8, WOD: 30kts / 0deg ........................ 32
Table 3.3: Gust Filters Derived for LHA LS 8, WOD: 30kts / 30deg....................... 32
Table A.1: DDG-81 Hover Coordinates ................................................................... 87
Table A.2: LHA Hover Coordinates ........................................................................ 87
Table A.3: Genhel-PSU Matlab GUI Interface......................................................... 88
xvi
ACKNOWLEDGEMENTS
This research was supported by the Center for Rotorcraft Innovation (CRI) under
contract (WBS 2007-B-01-02.3-P3) with Sikorsky Aircraft Corporation and NAVAIR under
the SBIR contract (N68335-07-C00310) with Barron Associates. I would like to extend
additional thanks to Derek Geiger of Sikorsky Aircraft Corporation and Jared Cooper of
Barron Associates for their help. The thesis also uses CFD airwake data for the DDG-81
provided by NAVAIR Advanced Aerodynamics branch (AIR-4.3.2.1) under a Memorandum
of Agreement with Penn State. I would like to thank Dr. Susan Polsky for her help in
providing access and understanding of the DDG-81 airwake data.
I would also like to thank my advisor, Dr. Joseph F. Horn, for his guidance over the
past two years. From course work to research he has always taken the time to answer my
questions and teach me the fundamentals necessary to pursue a career that I will enjoy for
years to come. I am very grateful to have had the opportunity to be one of his students.
My graduate education would not have been the same if it weren’t for my coworkers.
I would like to extend my gratitude to Derek Bridges for his time and patience while teaching
me how to use the simulator and for his advice on research over the past year. Additional
thanks to Wei Guo, Conor Marr, Pamela Montanye, Sana Safraz, and Eric Tobias, I have
learned much from each of you. I am also thankful to all the coworkers and friends who have
kept graduate school fun in-between long hours at the office.
Finally, I would like to thank my parents. Without their love and support, I would not
be here today.
Chapter 1
Introduction
1.1 Motivation
Shipboard operations continue to be among the most challenging for maritime
rotorcraft pilots. The unsteady airwake behind a ship requires pilots to continuously
compensate for gust disturbances, resulting in high pilot workload. In addition to the
unsteady air flow, the ship deck is rolling and pitching, and at times the pilots have to be
prepared to fly in adverse weather conditions. With the environmental factors considered,
the ordinary tasks such as takeoff, landing, and precision hovering become increasingly
difficult and dangerous while at sea.
Often high pilot workload is the limiting factor when determining the allowable
sea state and wind-over-deck (WOD) conditions for particular rotorcraft and ship
combinations. In order to provide a quick response in critical situations, it is of high
interest of the Navy to extend these flight envelopes. Significant work has been
completed in recent years to accurately model the ship board environment and to develop
gust alleviation controls to improve pilot workload. Accurate simulations allow for pilot
training, simulation of experimental gust alleviation controllers, and partial development
of the WOD envelope [1]. With continuous research effort, flight envelopes will be
extended, and flying conditions for pilots can be made safer.
2
1.2 Literature Review
1.2.1 Ship Airwake Modeling Through Experimental Testing
Several researchers have experimented with developing ship airwakes for flight
simulations models through full-scale testing and model-scale testing. Full scale testing is
costly in terms of time and money. As a result, other methods are used in an effort to
supplement the majority of flight tests. Model-scale testing is an alternative, but can
present an obstacle if the wind tunnel can not meet the required Reynolds number for the
scaled model [2]. Early attempts to accurately model a ship airwake were conducted
through wind tunnel experiments by Healey [3]. Healey was able to make a preliminary
database, but found it difficult to accurately measure the turbulence close to the ship.
Clement, et al. also comments that wind tunnel tests are good for defining the steady
winds around the ship, but can not provided definitive results for ship-induced turbulence
[4].
When the means are available, simulations derived from wind tunnel data (or
computational efforts) can be evaluated with limited full scale flight testing. An example
of such testing was the V-22 Osprey during initial ship deck flights [5]. The pilots felt the
simulation was not representative of the disturbances they were facing in flight. As a
result the flight test data was reduced to provide insight on discrepancies between the
simulation model and actual flight.
3
1.2.2 Computational Fluid Dynamics Ship Airwake Models
A common alternative to experimental testing is computational fluid dynamics
(CFD). Polsky and Bruner examined a time-accurate ship airwake for the LHA class ship
[6]. The study used COBALT, an unstructured grid solver that relies on the Navier-
Stokes equations. Woodson and Ghee used the same solver to examine the DDG-81 ship
and compared CFD results with wind tunnel testing [7]. Polsky, et al. examined minor
discrepancies between the CFD and wind tunnel tests to build confidence in CFD airwake
solutions using the COBALT solver [8]. At Penn State, a parallel flow solver, PUMA2
was used to examine airwake solutions behind LHA class ships [9]. PUMA2 uses a finite
volume formulation of the Euler/Navier-Stokes equations.
Additional advances in CFD airwake solutions have been made by Zan [10]. He
suggested that a detailed time varying solution is important in high-fidelity simulations.
He also states that a new CFD model should be created for each wind direction because
the angle that the wind comes across the deck will affect the simulation significantly.
Forrest, et al. examined the shipboard problem in more detail by testing several WOD
conditions on two separate ships, using FLUENT CFD solvers [11]. The study
determined that geometric details that result in medium to large scale flow should be
considered to achieve an accurate airwake model. Finally, Shipman, et al. define the
modeling parameters that are most sensitive in generating a CFD ship airwake [12]. The
paper examines the tradeoff between the computing power required to produce the CFD
airwake versus the required fidelity of the simulation. The majority of the airwake
dynamics are characterized by bluff body shedding of the main ship geometries;
4
however, a high fidelity solution would require additional details to be included.
Parameters that can be modified are grid size and the resolution of boundary layers.
Depending on the intended use of the CFD solution, computing power can be reduced by
concentrating efforts on computing very detailed solutions directly downstream of critical
structures such as a large ship mast or a deck’s edge.
1.2.3 Gust Disturbance Identification
Along with flight simulation, researchers have been trying to design systems to
help reject gust disturbances. In doing so, the first step is to identify the gusts that are
creating disturbances on the aircraft. Labows, Blanken, and Tishcler presented a study
that reduced flight test data of a UH-60 BLACK HAWK to a low speed turbulence model
[13]. The study first extracted the remnant aircraft rates caused by the atmospheric
disturbances, and then fit a von Karman turbulence model to the power spectral density
(PSD) of the equivalent gusts. The idea is if white noise is fed through this filter, the
resultant noise would be representative of the actual test data. The work was continued by
Lursardi, et al. and it was determined that the parametric turbulence models are scalable
for varying levels of turbulence and wind directions [14].
1.2.4 Stability Augmentation System Control for Gust Rejection
Horn, Bridges, and Lee applied the concepts presented in [13,14] to alleviate pilot
workload during shipboard operations [15]. The study used the UH-60 BLACK HAWK
5
GENHEL model hovering in an LHA airwake. The controller design was constrained to
represent a limited authority Stability Augmentation System (SAS). The first design step
was to determine the equivalent aircraft gusts acting on the aircraft, and to model the
spectral properties with a von Karman turbulence model. The SAS was then designed
using H2 synthesis to reject gust disturbances based on the von Karman turbulence
models. The simulation indicated a significant reduction in pilot control activity and
angular motion of the aircraft when using an optimized SAS. The drawback to this design
was that the optimized SAS changed the frequency response of the vehicle to the pilot
input. In other words, if the pilot entered an input at the same frequency of a known gust,
then it too would be filtered out since the SAS controller was designed to reject
disturbances occurring at specific frequencies.
1.2.5 Model Following Control Architecture for Gust Rejection
To differentiate between desired pilot input and the gust disturbances, Horn and
Bridges developed model-following control (MFC) architecture to be used in gust
rejection [16]. The study used the same gust identification principals presented in [15].
This MFC architecture uses feed-forward compensation to track pilot inputs and when
engaged, implements feed-back compensation to reject gust disturbances. The advantage
to using this control architecture allows for the gust disturbance rejection to be turned on
or off as needed, as well as easily substituting new compensators for different airwakes.
Overall, the commands generated are based on angular rate tracking error. The controller
6
uses a simple model following design with the closed-loop attitude dynamics as the plant
model. The basic MFC architecture is presented in Fig. 1.1.
Overall, Horn and Bridges found there was an improvement in the aircraft
handling qualities. Additionally, the fluctuations in the aircraft’s angular rates and
attitudes were significantly reduced. The reduction in angular rates is visible in Fig. 1.2.
In this figure the red line represents the airwake compensator off, and the blue line
represents the airwake compensator on.
Figure 1.1: MFC Architecture
7
1.2.6 Trailing Edge Gust Rejection Controller
A recent feasibility studied was performed by Montanye [17] to use active trailing
edge flaps as a gust alleviation controller. The study implemented a model following
controller that used the swashplate and pedals to ensure that the vehicle follows the
desired response to pilot inputs. Unlike studies by Horn, the airwake compensator rejects
airwake disturbances with trailing edge flaps and a H2 controller. The study was
implemented with a UH-60 BLACK HAWK GENHEL model and an LHA airwake. The
initial results suggest that trailing edge flaps are capable of reducing the magnitude of the
vehicle angular gust response in the roll and pitch axes.
Figure 1.2: Airwake Compensator Comparisons using a Model Following Controller.
[Horn, Bridges, 2007].
8
1.3 Focus of Current Research
This research provides further investigation of ship airwake identification
principles in offline and online environments. Additionally, using the theory from the
airwake identification methods, this study examines an alternative simulation of ship
airwake disturbances.
1.3.1 Disturbance Identification in Offline and Online Environments
This first goal of this research is to accurately identify airwake characteristics.
This study modifies the principles of [13,14,15,16] by applying modern tools offered in
recent releases of MATLAB. The new methodology guarantees stable filter development
and constrains the PSD filter fit to the frequency range that affects handling qualities of
pilots. This process is done in an offline environment; relying on previously generated
time histories of the aircraft flight dynamics from a simulation performed in Genhel-PSU.
While this paper focuses using results from an aircraft simulation model, the filter design
process can also be applied to flight test data.
Offline gust identification provides useful information in rejecting gusts for a
particular ship and airwake condition; however, it would require significant effort to
analyze aircraft time histories for every helicopter, ship, and WOD condition. As a result,
it is ideal to develop an online learning algorithm that will identify the properties of the
airwake gusts during flight. This research presents an initial feasibility study of an
autoregressive model used for online gust identifications. The system presented could
9
then be used in coordination with an adaptive controller to improve gust rejection during
flight in future applications.
1.3.2 Simulation Modeling of Airwakes using Stochastic Filters
Finally, although the CFD solutions presented in Section 1.2.2 can provide an
accurate environment, the computing power required to implement detailed solutions can
be a limiting factor in real-time simulation applications. The large velocity databases
typically have to be loaded into memory prior to the start of a simulation. Consequently,
the length of the simulation is usually limited to 40-60 seconds of data in an effort to
keep the database at a manageable size. To achieve longer simulations, the airwake can
be cycled, but it produces noticeable repetitions in flight data.
Chapter 2
Simulation Environment
This study required the use of rotorcraft simulation models capable of interacting
with several ship airwake models. The majority of the simulation tests utilized Genhel-
PSU, a FORTRAN based simulation of a UH-60A BLACK HAWK. A secondary
simulation environment used was a version of GENHEL maintained by Sikorsky
Aircraft. This chapter will review the simulation environments as well as their
corresponding airwake models.
2.1 Genhel-PSU
The primary flight dynamics model used for this research is Genhel-PSU.
GENHEL (GENeral HELicopter) is a Fortran-based simulation that was initially created
at Sikorsky Aircraft Corporation and documented under contract with the U.S. Army /
NASA Ames Research Center [18]. The simulation was provided to Penn State VLRCOE
by the Rotorcraft Division of NASA Ames. Several modifications have been made to the
code, resulting in the version called Genhel-PSU [19]. These changes include network
communications to interact with other programs and computers, incorporating user-
defined controllers, a graphical user interface, generation of high order linear models, and
the addition of shipboard interface modules [20]. With these advancements, Genhel-PSU
is capable of running real-time piloted simulations of ship-based operations.
11
Although there have been updates to GENHEL to meet current research needs,
the basic mathematical models of the aerodynamics and flight dynamics is mostly
unchanged from the original version. The simulation is a nonlinear model of the UH-60A
that includes accurate models of the main rotor, fuselage aerodynamics, horizontal and
vertical tail surfaces, servo and sensor models, landing gear, and mechanical flight
controls. The main rotor model also includes blade flapping, lagging, and hub rotational
degrees of freedom. The helicopter position, velocity, and acceleration are calculated at
every time step by summing each of the six forces and moments at the center of gravity.
Additional specifics are outlined in [18].
To simulate ship-board landing missions, Genhel-PSU provides many options.
There are visuals and CFD data for the Winston Churchill Destroyer (DDG-81) with
WOD condition of 25 kts/0 deg, and the Peleliu LHA class ship with WOD conditions of
30 kts/0 deg and 30 kts/30 deg. Each of the airwake solutions can be scaled for different
wind speeds. For added accuracy of the simulation environment, ship motion has also
been added. The user can select to model a ship moving at constant speed and heading as
well as selecting added time-varying motion for a particular sea state. A semi-analytical
method that calculates the heave, roll, and pitch motion using sum of sine equations is
outlined in [21,22] and is implemented into Genhel-PSU in [23].
In the absence of experienced pilots, Genhel-PSU offers a pilot model that will fly
a prescribed trajectory without the need for human inputs [24]. The pilot model is a
closed-loop control law that was designed to emulate a human pilot. This multi-input,
multi-output (MIMO) control law is scheduled with airspeed to follow the trajectory path
that was defined by a set of position, speed, altitude, and heading terms. This pilot model
12
is used to hover at specific points within the ship airwakes to gather simulation data
throughout this study.
2.2 CFD Airwakes Used with Genhel-PSU
2.2.1 USS Winston S. Churchill Destroyer (DDG-81)
The Winston-Churchill Destroyer (DDG-81) provides the navy with multi-
mission offensive and defensive capabilities. The ship is 505 ft long and 59 ft wide and
can achieve 30 plus knots at sea [25]. The ship can carry two Sea Hawks, a variant of the
UH-60 BLACK HAWK. A visual of the DDG-81 ship deck is made available in Genhel-
PSU’s visual library and it can be flown with a CFD airwake of the DDG-81.
Figure 2.1: DDG-81 stationed in Virginia Beach, VA. [Sparbanie, 2008].
13
The CFD model used with Genhel-PSU was developed by NAVAIR using cell-
centered finite volume Navier-Stokes solver called Cobalt [26]. The CFD model was
originally developed and documented by Woodson and Ghee [7] and further investigated
by Polsky, et al. [8]. Within these papers, the CFD model is validated with wind tunnel
results. The nominal CFD WOD condition is 25 kts at 0 deg wind. The CFD data can be
scaled to simulate other wind speeds, but only for deck winds of 0 deg.
The CFD analysis used a trapezoidal shaped grid, fanning out in height and width.
The first plane is aligned with the hangar doors and measures 264.7 ft by 196.86 ft. The
final plane is 817 ft behind the hangar doors and measures 1640.5 ft by 524.96 ft. The
detailed grid shown in Fig. 2.2 has 55 x-planes with each plane consisting of 77 y-points
and 61 z-points. At each grid point there are time-varying u, v, and w velocity
components. The airwake contains data for 60 seconds, with a time step of 0.01 seconds
[27].
Figure 2.2: Dimensions of the CFD airwake behind the DDG-81 ship.
14
2.2.2 USS Peleliu (LHA-5) Tarawa Class Solutions
The USS Peleliu is a US Naval LHA Tarawa-class amphibious assault ship that
has been used in attack missions, carrying several helicopters ranging from CH-53 Super
Sea Stallions to CH-46 Sea Knights as well as AV-8B Harrier jets [28]. The deck is 820
ft long and 106 ft wide.
The LHA class ship airwake solution was calculated with PUMA 2, a modified
version of a Parallel Unstructured Maritime Aerodynamics (PUMA) CFD solver [9,
29,30]. This solver uses a finite volume formulation of the integral form of the Euler and
Navier-Stokes equations for three-dimensional, compressible, unsteady/steady solutions
for complex geometries [9].
The nominal case used for simulation is a CFD solution around landing spot 8.
The user can select either a WOD condition of 30 knots with 0 degree winds or 30 knots
with 30 degree winds. The solutions can be scaled to seek solutions at different wind
Figure 2.3: USS Peleliu (LHA 5)
15
speeds. Landing spot 8 is depicted in Fig. 2.4, and is known to be one of the most
difficult places to land on the ship in 30 deg winds due to the added wind turbulence of a
cross wind breaking around the super structure on the deck.
2.3 Simulation Hover Locations used with Genhel-PSU
The main hover location selected for the DDG-81 in this study was for the aircraft
to hover within CFD airwake coordinate system at 430 ft behind the origin, centered
along the midline of the ship, and 31.3 ft above the origin of the airwake. This relates to
roughly 205 ft behind and 28 ft above the center of the DDG-81 ship in Genhel-PSU.
The red dot in Fig. 2.5 represents the approximate hover location for the center of
fuselage.
Figure 2.4: Top View of the LHA Class Ship Outlining Landing Spot 8
16
For repeatability of these simulations, the specifics of the trajectory files and
Genhel-PSU settings used for these hover maneuvers and are outlined in Appendix A.
2.4 Sikorsky GENHEL
Sikorsky GENHEL is a modular object-oriented code that can be readily modified
to represent different rotorcraft. It also has the capability to interact with a ship flow field
and ship motion modules. For this study, a S-92 fly-by-wire (FBW) aircraft was flown in
different ship airwakes. The S-92 FBW aircraft was chosen based on the deck size of the
DDG-81 and because Sikorsky Aircraft has already validated the aircraft’s simulation
model. The ship airwakes used with Sikorsky GENHEL is the same CFD airwake for the
DDG-81 that was provided by NAVAIR to PSU for simulations with Genhel-PSU.
Figure 2.5: Approximate Hover Location behind a DDG-81.
(x,y,z) = (389.95, y ,16.40)
(x,y,z) = (389.95, y ,10.11)(x,y,z) = (470.42, y ,13.29)
(x,y,z) =
(389.38, y , 27.40)
x = 429.3 ft
17
2.4.1 MTE Definition
For evaluation of the stochastic airwake grid, trained pilots were used in a fixed
based simulator. To evaluate the pilot workload in the turbulent flow-field, the pilots
were instructed to fly the Maritime Task Element (MTE) presented in [27]. The MTE
consisted of a standard approach to the back of the ship, and a precision hover task over
the landing spot on the ship deck. The maneuver begins at a 70 ft hover approximately
400-500 ft behind the ship. The pilot moves towards the back of the ship with a closure
rate of 6-10 kts. After the pilot performs a smooth flare over the fantail, the pilot has
about 10 sec to begin the stable hover maneuver. The 30 sec hover should be directly
over the landing spot at a gear height of 9-13 ft. Marginal and desired performance
standards were defined by visual cues seen by the pilots. A snapshot of the visuals that
the pilot would see can be seen in Fig. 2.6.
Figure 2.6: Visuals used with Sikorsky’s GENHEL during Piloted Simulations. [Geiger,
et al., 2008]
18
2.4.2 Cooper Harper Rating Scale
To numerically evaluate the pilot workload during the simulation performing the
MTE above, the pilots were asked to asses the current simulation run on the Cooper-
Harper Rating Scale [31]. The scale rates pilot work from 1 to 10, where 1 is the least
amount of work, and 10 means the aircraft is uncontrollable. If the pilot maintains
desirable tolerances, as defined by the above MTE, then the rating falls between 1 and 3.
If the pilot maintains adequate performance, the HQR rating falls between 4 and 6. The
Cooper-Harper Rating Scale is in Fig. 2.7.
Figure 2.7: Cooper-Harper Rating Scale
Chapter 3
Stochastic Filter Theory
It is desirable to accurately identify the airwake characteristics of the flow field
behind a ship at sea. The method presented by Labows [13] and Lusardi [14] offers an
offline method to extract equivalent gusts and to estimate the peak characteristics through
a colored noise filter. The method modified and implemented by Horn, Bridges, and Lee
[15] and again by Horn and Bridges [16] to reject gust disturbances in a model following
control architecture. This chapter explores their work further and refines the colored
noise filter.
3.1 General Theory
The airwake flow field can be divided into deterministic and stochastic
components. The deterministic component is the mean velocity of the wind. The
stochastic component can be modeled as a set of stochastic colored noise filters. For any
given point in space, there are six gust filters; three for the translational components and
three for rotational gusts.
If one can isolate the turbulence of the air, then the stochastic component of the
airwake can be modeled as a random process. It is assumed that the airwake will repeat
itself over a range of frequencies, and as a result, the airwake can be characterized by
averaging the estimates of the power spectrum using periodograms [32]. When the
20
estimated linear system is excited with an uncorrelated white noise, the resulting signal
should have similar frequency domain properties as the stochastic signal of the original
airwake.
3.2 Previous Filter Design
The first revision of stochastic filter design produced by Horn, Bridges, and Lee
[15] used a method that resulted in a design similar to the Von Karman turbulence model.
The first step was performing simulated hovers over the ship deck using the pilot model
discussed in Section 2.1. During the simulation, the aircraft’s rigid body states, state
derivatives, and control inputs were recorded. To isolate the airwake, the aircraft open-
loop dynamics had to be accounted for. A high order linearized model was developed that
took the form of:
where x is the state vector, u is the pilot inputs, and w is the equivalent disturbance
vector. Here the state vector is defined as:
The control input vector is:
GwBuAxx ++=& Eq. 3.1
Trqpwvux ][ θφ= Eq. 3.2
21
and the equivalent disturbance vector is:
The components of the disturbance vector represent the average gust velocity over
the body of the vehicle as well as the spatial variation modeled as equivalent angular
rates.
Using the pseudo-inverse of the G matrix from a reduced 9-state linear model, the
disturbance vector can be found.
The time histories of the equivalent gusts are produced and the mean value is
subtracted for each of the six airwake components. The mean value is considered the
deterministic portion and can be included again during real time simulation.
The last step taken was to derive a transfer function, that when stimulated with the
uncorrelated white noise vector, a disturbance would be created that had similar spectral
T
pedcolllonglatu ][ δδδδ= Eq. 3.3
[ ]Tgggggg rqpwvuw = Eq. 3.4
( )BuAxxGw −−= +&
Eq. 3.5
22
properties as the airwake observed during the simulation hovers. To do this, the PSD of
the equivalent gusts were computed. Then a filter similar to the von Karman turbulence
model was fitted to the PSD. The filter is shown Eq. 3.6.
In Eq. 3.6, the intensity factor,σ , the length scale, L , and the coefficients ia , ib ,
are selected to fit the PSD over the frequency range of 0.1 to 40 rad/sec. The parameters
were fit using a nonlinear least-squares routine to reduce the error between the fitting
filter and the equivalent gusts. The parameter V in this equation is the reference speed of
the nominal wind condition.
Horn, Bridges, and Lee [15] found desirable results from this approach and are
outlined in Fig. 3.1. The red line is the PSD of the equivalent gusts, and the blue line is
the filter designed to match the spectral properties the equivalent gust.
( )3
3
2
21
2
21
1
14
+
++
++
=
sV
Las
V
Las
V
La
sV
Lbs
V
Lb
V
L
sH
σ
Eq. 3.6
23
3.3 New Design Methodology
In an effort to achieve a smoother fit, the filter fitting routines have been modified
to use newer functions included in recent releases of MATLAB. In doing so, the effort
used Welch’s Method for spectrum fitting. The Welch method is a modification of the
Barlett procedure that can directly compute the power spectrum estimate using Fast
Fourier Transforms (FFTs) [32]. It uses periodogram estimates based on splitting time
series of overlapped segments multiplied by data windows. The window is applied to the
data segments before computation of the periodograms.
Figure 3.1: PSD of the Lateral Gust, v, using the filter fitting method by [Horn, Bridges,
and Lee, 2006].
24
Another change to the filter fitting routine is that the system examines four
window sizes: 512, 1024, 2048, and 4096, where the time step was 0.01 sec. The final
PSD is an average of each PSD resulting from the four window sizes. This was done in
an effort to minimize the effect of window size selection, and to assure a consistent filter
fit through any data set. Moreover, using four different window sizes allows one to
accurately capture a wide range of frequencies. A smaller window would allow one to
capture high frequency information, whereas a larger window would highlight smaller
frequencies.
The colored filter was also modified. Previously the filter was designed after the
von Karman gust models, and now a simple third order system is designed with poles and
damping coefficients. The benefit to the change in filter fitting method is that the poles
and damping coefficients can easily be constrained. Moreover, there is one less parameter
to optimize during the filter fitting routine. The transfer function that is fitted to the
equivalent gust’s PSD now takes the form:
In Eq. 3.6, K is a gain, zζ and pζ are damping ratios, nz
ω and npω are natural
frequencies, and p is a pole. Each of these parameters were constrained and given an
initial guess for the non-linear least squares fit. Overall, constraining the parameters was
an advantage because the filters were guaranteed to be stable. Additionally, placing
( ) ( )( ) ( )psss
ssKsG
npnpp
nznzz
+++
++=
22
22
2
2
ωωζ
ωωζ Eq. 3.7
25
constraints allowed for the filter fit to avoid high frequency modes which can create
problems in numerical implementation.
Finally, different windowing methods were examined. Welch’s method was
applied using Hamming, Hann, and Blackman windows. The general form of the
windowing techniques were documented in MATLAB’s help files, and are outlined by
the Eqs. 3.8- 3.10.
Hamming Window:
Hann Windowing:
Blackman Windowing:
In all of these equations, the window length, L, is N +1, where N is the width in
samples. The parameter n is an integer between zero and N .
( ) NnN
nnw ≤≤
−= 0,
2cos46.054.0
π
Eq. 3.8
( ) NnN
nnw ≤≤
−= 0,
2cos1
21
π
Eq. 3.9
( ) NnN
n
N
nnw ≤≤
+
−= 0,
4cos08.0
2cos5.042.0
ππ
Eq. 3.10
26
3.4 Offline Filter Fitting Results
Initial filter fitting results were generated from the UH-60A performing a standard
hover for the DDG-81 ship airwake with a WOD condition of 25 kts / 0 deg. The hover
specifications were outlined in Section 2.3 and the corresponding Genhel-PSU settings
are outlined in Appendix A.
The first examination is if the various windowing techniques had an effect on the
colored filter generation while using Welch’s Method. Figure 3.2 shows the PSD of the
longitudinal equivalent gust velocity vector as well as the resulting filters generated from
the Hamming, Hann, and Blackman windowing techniques. The equivalent gust vector
was extracted from simulation hover time history data, as outlined in Eqs. 3.1-3.5. The
PSD of the equivalent gust was found using a Hamming window. The filters were
generated using the filter fit outlined in Eq. 3.7 and followed the windowing Eqs. 3.8-
3.10.
Figure 3.2 figure suggests that each of these windowing techniques will generate
very similar stochastic gust filters. Additional results for the v, w, p, q, r components can
also be found in Appendix B. Since the windowing technique does not affect our results,
this study will arbitrarily pick the Hamming Windowing for the rest of the filter fitting
results presented in this chapter.
27
Next, the new filter fitting method is compared to the filter fitting method in [16].
They both use Welch’s method and Hamming windowing, but the change in filter models
as described in Eqs. 3.6 and 3.7 are being examined. With the new filter fitting method, it
is desired that the filter fits be smoother through the frequency range of 0.4 to 10 rad/sec.
For the older filter fit, initial guesses were made for each of the filter fits. The
intensity factor,σ ,was initialized at 3. The length scale, L , was set to 10. The coefficients
in the numerator were set to 2.7478 and 0.3398. The coefficients in the denominator were
set to 2.9958, 1.9754, and 0.1539. A nonlinear least squares routine was used to
determine the filter coefficients that minimized the error between the equivalent gusts and
the filter fit.
100
101
10-2
10-1
100
101
102
103
104
Windowing Comparision for DDG81 25kt / 0 deg Hover
Freq (rad/sec)
u -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Hamming
Hann
Blackman
Figure 3.2: Longitudinal Equivalent Gust Velocity for the Windowing Comparison
28
For the new filter fit, the lower and upper bounds on the damping ratios were
constrained to be 0.1 and 2. Additionally, the natural frequencies and the pole location
were forced to be between 0.4 and 20 rad/sec. The gain K was constrained between 0 and
1000. Finally, an initial guess of 1 for each of the five parameters yielded the best results.
Again, a nonlinear least squares routine was used to determine the filter coefficients that
minimized the error between the equivalent gusts and the filter fit.
Two examples of the filter fits are seen in Figs. 3.3-3.4. One can see the new filter
design yielded better results. Additional results can be seen Section B.2.
100
101
10-2
10-1
100
101
102
103
Freq (rad/sec)
w -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Old Filter Design
New Filter Design
Figure 3.3: Vertical Velocity Equivalent Body Gust for the Filter Design Comparison
29
Finally, to demonstrate the filter design process can be used for any airwake and
WOD condition, the new filter design that used Welch’s method and Hamming
windowing was applied to two new airwake conditions. A hover, using the pilot model,
was performed in the LHA airwake for WOD conditions of 30 kts / 0 deg and 30 kts / 30
deg. The trajectory file information used to command a hover above landing spot 8, and
the settings for Genhel-PSU, is tabulated in Appendix A. A sampling of the results for the
extracted equivalent gusts and corresponding filter fits are in Figures 3.5-3.8. A figure for
every equivalent gust and filter fit can be seen in Appendix B. The final filter found for
each airwake and WOD conditions are outline in Tables 3.1-3.3.
Since this process works well, it should also be noted that the same process can be
applied to data collected during precision hovers maneuvers performed in flight testing of
full scaled aircraft.
100
101
10-4
10-3
10-2
10-1
100
101
Freq (rad/sec)
p -
Roll
Com
ponent [(
rad/s
)2]
Equivalent Gusts
Old Filter Design
New Filter Design
Figure 3.4: Roll Rate Equivalent Body Gust for the Filter Design Comparison
30
100
101
10-3
10-2
10-1
100
101
102
103
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Freq (rad/sec)
w -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure 3.5: Vertical Velocity Equivalent Body Gust: LHA 30 kts / 0 deg
100
101
10-6
10-4
10-2
100
102
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Freq (rad/sec)
r -
Yaw
Com
ponent [(
rad/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure 3.6: Yaw Rate Equivalent Body Gust: LHA 30 kts / 0 deg
31
100
101
10-1
100
101
102
103
104
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Freq (rad/sec)
v -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure 3.7: Lateral Velocity Equivalent Body Gust: LHA 30 kts / 30 deg
100
101
10-4
10-3
10-2
10-1
100
101
102
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Freq (rad/sec)
q -
Pitch C
om
ponent [(
rad/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure 3.8: Pitch Component Equivalent Body Gust for LHA 30 kts / 30 deg
32
Table 3.1: Gust Filters Derived for DDG81 Hover Location, WOD: 25kts / 0deg
17958.02034.006506.0
26.33717.1197.0:
23
2
+++
++
sss
ssu
14771.01175.003711.0
179.105297.0007704.0:
23
2
+++
++
sss
ssp
17268.03531.006054.0
62.32477.11911.0:
23
2
+++
++
sss
ssv
14829.01262.003304.0
5583.00182.000322.0:
23
2
+++
++
sss
ssq
108.13706.007942.0
21.14391.11154.0:
23
2
+++
++
sss
ssw
16784.02818.005642.0
969.10699.001152.0:
23
2
+++
++
sss
ssr
Table 3.2: Gust Filters Derived for LHA LS 8, WOD: 30kts / 0deg
15478.01558.004946.0
896.77888.005844.0:
23
2
+++
++
sss
ssu
13036.0116.002419.0
299.001723.0002257.0:
23
2
+++
++
sss
ssp
1681.27087.02971.0
59.26445.13561.0:
23
2
+++
++
sss
ssv
13388.01362.002443.0
2412.001108.0001022.0:
23
2
+++
++
sss
ssq
1952.1145.12503.0
83.14305.32155.0:
23
2
+++
++
sss
ssw
1396.46638.04856.0
883.11175.002368.0:
23
2
+++
++
sss
ssr
Table 3.3: Gust Filters Derived for LHA LS 8, WOD: 30kts / 30deg
16776.02017.006615.0
64.67146.55576.0:
23
2
+++
++
sss
ssu
13921.01277.003402.0
075.21224.00119.0:
23
2
+++
++
sss
ssp
1127.28238.02247.0
1.120468.9366.1:
23
2
+++
++
sss
ssv
16801.02663.006962.0
197.21419.001553.0:
23
2
+++
++
sss
ssq
1567.3037.3708.1
3.1051.119427.8:
23
2
+++
++
sss
ssw
1218.15853.01661.0
106.97991.008261.0:
23
2
+++
++
sss
ssr
33
3.5 Discrete Filter Generation
It is the goal of this research to not only identify the airwake spectral properties,
but to also use the spectral information to produce a real time simulation environment. In
a real time application, the continuous filters will need to be transformed into discrete
filters. In this paper, the Tustin approximation method is applied [33]. The z-transform of
a continuous transfer function is approximated as:
Where the continuous frequency variable is estimated as:
In Eq. 3.12, T is the sample period. To demonstrate implementation, let’s take a standard
4th order continuous transfer function:
Substituting the z-transformation into Eq. 3.13, one can find the discrete transfer
function:
( ) ( )'sHzH = Eq. 3.11
1
12'
+
−=
z
z
Ts Eq. 3.12
εδχβα +⋅+⋅+⋅+⋅
+⋅+⋅+⋅′=
sss
dscsbsaKsH
234
23
s)( Eq. 3.13
34
A simplification of Eq. 3.14 leads to:
When the terms are expanded, Eq. 3.15 can be expressed as:
Where the coefficients i
k are expressed as:
( )εδχβα +
+
−+
+
−+
+
−+
+
−
+
+
−+
+
−+
+
−′
=
1
12
1
12
1
12
1
12
1
12
1
12
1
12
234
23
z
z
Tz
z
Tz
z
Tz
z
T
dz
z
Tc
z
z
Tb
z
z
TaK
zH Eq. 3.14
( ) ( ) ( ) ( )( )( ) ( )
( ) ( ) ( ) ( ) ( )( )( ) ( )
+++−⋅+
+−⋅++−⋅+−⋅
+++−⋅+
+−⋅++−⋅′
=
−−−
−−−−−
−−−
−−−−
4143113
2121213141
4143113
21212131
1112
114118116
1112
114118
)(
zTzzT
zzTzzTz
zdTzzcT
zzbTzzaTK
zHd
εδ
χβα
Eq. 3.15
4
10
3
9
2
8
1
76
4
5
3
4
2
3
1
21)(−−−−
−−−−
++++
++++′=
zkzkzkzkk
zkzkzkzkkKzH Eq. 3.16
35
For numerical implementation, one can solve for the output based on the previous
input values, previous output values, and the coefficients of the continuous transfer
function. The standard discrete transfer function is the output ( )zY divided by the
input ( )zX .
Cross multiplication leads to:
Then, to find the current output based on the previous inputs and outputs, the outputi
y is:
432
10
43
9
42
8
43
7
432
6
432
5
43
4
42
3
43
2
432
1
24816
441664
6896
441664
24816
248
4416
68
4416
248
TTTTk
TTTk
TTk
TTTk
TTTTk
dTcTbTaTk
dTcTaTk
dTbTk
dTcTaTk
dTcTbTaTk
εδχβα
εδβα
εχα
εδβα
εδχβα
+−+−=
+−+−=
+−=
++−−=
++++=
+−+−=
+−=
+−=
++−=
+++=
Eq. 3.17
( )( ) 4
10
3
9
2
8
1
76
4
5
3
4
2
3
1
21)(−−−−
−−−−
++++
++++′==
zkzkzkzkk
zkzkzkzkkK
zX
zYzH Eq. 3.18
( ) ( ) ( ) ( ) ( )[ ]( ) ( ) ( ) ( ) ( )4321
4321
109876
54321
−+−+−+−+
=−+−+−+−+′
kykkykkykkykkyk
kxkkxkkxkkxkkxkK Eq. 3.19
36
For implementation, a normally distributed random number vector is applied to
Eq. 3.20. The resulting output vector should have the same disturbance output as the
continuous transfer function when excited by the same random number vector.
3.6 Equivalent Body Gusts Transformed to Local Velocity Disturbances
Since the continuous filters were derived from extracted equivalent body gusts,
then the derived filters create equivalent body gusts when excited by a white noise vector.
For non-linear simulations, the equivalent body gusts need to be transformed into local
velocity disturbances. The local longitudinal, lateral, and vertical velocities due to the
stochastic component of the airwake can be found applying the following relations:
In Eq. 3.21, Xpos, Ypos, and Zpos are the components of a position vector of the
local component compared to the fuselage center of the aircraft. Moreover, Ubwake, Vbwake,
Wbwake, Pbwake, Qbwake, and Rbwake are the equivalent body gusts found at each time step by
[ ]
6
410`392817
453423121
k
ykykykyk
xkxkxkxkxkK
y iiii
iiiii
i
−−−−
−−−−
−−−−
++++′
= Eq. 3.20
posbwakeposbwakebwakes
posbwakeposbwakebwakes
posbwakeposbwakebwakes
ypxqww
zpxrvv
zqyruu
+−=
++=
+−=
Eq. 3.21
37
exciting the corresponding discrete transfer function filter. To find the total local velocity
of the airwake acting on component of the aircraft, then a time-averaged airwake value
also needs to be included.
In this study, the time-averaged airwake value is found from averaging the CFD
airwake velocity over the span of the time history for a particular location within the
airwake. It should be clarified that in Genhel-PSU, the airwake CFD files do not include
the steady nominal wind. For the DDG-81 WOD condition of 25 kts / 0 deg, the steady
nominal wind is 25 kts. For Genhel-PSU, it should be noted that the steady nominal wind
condition is a second deterministic airwake component, and is added into the final local
velocity under another variable name.
Finally, when computing the fuselage center local velocities, the u, v, and w
velocities due to the stochastic component of the airwake are simply the u, v, and w
equivalent body gusts. This is because the position vector would be zeros. To perform an
initial simulation testing of the stochastic filters, it is ideal to apply this simplification.
TAs
TAs
TAs
www
vvv
uuu
+=
+=
+=
Eq. 3.22
38
3.7 Initial Simulation Results
It is the purpose of this study to be able to recreate similar gusts as the CFD
simulation data in terms of the spectral properties of the velocity data when the linear
filters designed above are excited with white noise. In an initial implementation effort,
the continuous filters in Tables 3.1- 3.3 were excited with white noise in the SIMULINK
environment. Additionally, the continuous filters were transformed to discrete filters
using the Tustin approximation method described in Section 3.5. The discrete filters were
excited with the same white noise input vector as the continuous filters. The output of
discrete and continuous filters was then compared to the time history of the equivalent
gust vectors that were used in the derivation of the continuous filters. Finally, the time
history of the CFD velocity vectors that acted on the fuselage center during simulation
was plotted against the offline simulation results. When using the CFD velocity data set,
the mean velocity component was subtracted out such that only the stochastic gusts were
being compared. For frequency analysis, the PSD of the velocity components of the
continuous filters, discrete filters, extract equivalent gusts, and CFD velocity vectors
were found.
A full set of figures showing the comparison of the velocity vectors for the
translational and rotational axes for the DDG-81 with WOD conditions of 25 kts / 0 deg,
LHA 30 kts / 0 deg and LHA 30 kts / 30 deg can be found in Appendix Sections B.6-B.8.
For discussion purposes, the longitudinal and vertical velocity vectors for the DDG-81
will be examined as well as the corresponding PSD plots.
39
The longitudinal velocity component of the DDG-81 airwake is presented in
Fig. 3.9. The dark blue line is the results from exciting the Tustin discrete filter. The
green line depicts the disturbance from the continuous filter that was excited in the
SIMULINK environment. The red line represents the equivalent gusts determined by
removing the aircraft’s dynamics from the simulation results. Finally, the light blue line
is the stochastic portion of the CFD gust that acted on the center of the fuselage during
simulation. Overall Fig. 3.9 shows fairly good agreement. To examine the frequency
response, the PSD of each data set is taken in Fig. 3.10. The figure demonstrates similar
spectral properties and validates the continuous and discrete filters.
Next, the vertical velocity component of the DDG-81 airwake is examined in
Fig. 3.11. There is a discrepancy in the stochastic portion of the CFD gust as compared to
the equivalent gusts, continuous filter, and discrete filter results. Since the filters match
the equivalent gusts, one can pin point the error relating to accurately developing the
equivalent gusts and not in the filter fitting routines or the method of filter excitation.
When examining Fig. 3.12, it is clear that the magnitude of the PSD of the CFD gust is
not being accurately captured in the equivalent gusts. A possible source of error could be
the accuracy of the reduced order linear model. This is a key component in accurately
developing this method for industry applications and should be examined further. This
topic is presented as a suggestion for future work and is discussed further in Section 6.2.
40
0 20 40 60 80 100 120-15
-10
-5
0
5
10
15DDG-81 Hover Location: WOD Case: 25kts / 0deg
time [s]
u -
Velo
city C
om
ponent [(f
t/s)]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure 3.9: Initial Simulation: Longitudinal Velocity Equivalent Body Gust for DDG-81
100
101
10-4
10-2
100
102
104
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Freq (rad/sec)
u -
Velo
city C
om
ponent [(f
t/s)2
]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure 3.10: Initial Simulation: Longitudinal Velocity PSD for DDG-81
41
0 20 40 60 80 100 120-15
-10
-5
0
5
10
time [s]
w -
Velo
city C
om
ponent [(f
t/s)]
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure 3.11: Initial Simulation: Vertical Velocity Equivalent Body Gust for DDG-81
100
101
10-4
10-2
100
102
104
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Freq (rad/sec)
w -
Velo
city C
om
ponent [(f
t/s)2
]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure 3.12: Initial Simulation: Vertical Velocity PSD for DDG-81
Chapter 4
Online Identification of Stochastic Filters
Developing a database of stochastic filters for every airwake condition that a
helicopter might encounter would require significant effort. As a result, it is also of
interest to develop an online learning algorithm that will learn and possibly reject gusts
during flight. This research study demonstrates an initial feasibility of an autoregressive
method used for online stochastic gust identification.
4.1 Previous Work
An initial feasibility study that was conducted by Horn, et al, [34] investigated
online learning algorithms to determine the deterministic and stochastic portions of a
CFD ship airwake. The study relied on a SIMULINK based tiltrotor simulation [35,36]
that used a composite airwake model of an LHA amphibious assault ship. The airwake
model incorporated time-averaged data provided by NAVAIR and Von Karman gust
models developed by Horn.
For identifying the stochastic airwake, the study used a Burg autogressive model.
In general, an autogressive model was selected for computational efficiency. Also, the
autogressive model is attractive because it does not require the time-domain data to be
transformed into the frequency domain. Overall, the study concluded that a 6th order
model is the minimum order model that would accurately estimate the stochastic airwake.
43
The paper then compared the online results of a 6th
order autogressive model to
benchmark filters found offline. The online filter results demonstrated convergence to
key peaks in the handling qualities frequency range when examining the magnitude plots
of the generated filters.
To continue this study, a 6th order Burg autoregressive model is implemented into
a real-time simulation environment in an effort to estimate the stochastic portion of a
CFD airwake. Successful integration can lead to future developments of an online
adaptive controller rejecting gust disturbances during flight. This goal is lucrative,
although it could be faced with opposition in industry as the new controller would have to
be proven to be robust in all flight conditions. A second motivation to this process is that
during flight, an external computer could be connected to the flight control system to
compute autogressive filters as a means to store data about the airwake behind a ship. As
a result, only the final filter coefficient values would need to be stored. This would
elevate the need to store large time histories of the aircraft dynamics during flight, as well
as minimizing the computational effort required to develop the filters in an offline
environment. The filters of the airwakes gathered during actual flights can then be used to
develop gust rejection controllers or to verify airwake simulations.
4.2 Burg Autoregressive Spectral Estimation
The Burg autoregressive model is a method to estimate the spectral properties of a
data signal by processing blocks of data at a time [37]. The Burg model offers an
advantage over other block estimation methods as it guarantees the generation of
44
minimum phase filters [38]. Additionally, Orfandis [38] states a rule of thumb that the
Burg model should work better than other autogressive models for short time histories of
data. Overall, all the methods will improve as the data records become longer.
As discussed in Chapter 3, the stochastic component of the airwake can be
estimated as a filter, that when excited by a white noise, recreates a disturbance similar to
the measured stochastic airwake in terms of the spectral properties. Orfandis presents the
signal modeling process as depicted in Fig. 4.1 where +pe is a prediction error sequence
that is approximately white, ( )zA is the prediction error filter that places poles, and n
y is
the disturbance output.
The optimal forward and backward predictors are defined by:
Where ( )ne p
+ and ( )ne p
− are found from filtering n
y through a prediction-error filter and
its reverse.
Figure 4.1: AR Filter Diagram [Orfandis, 1988]
( )[ ] ( )[ ] minmin22
== −+neEandneE pp Eq. 4.1
45
The Burg method then minimizes the least squares of both the forward and
backward prediction errors when determining the filter coefficients of ( )zA . When
solving for the filter coefficients, the Levinson-Durbin recursion must be used. The
Levinson-Durbin recursion is defined by:
Where the reflection coefficient,P
γ , is solved by:
In MATLAB, the autogressive filter solves for the Burg AR model and presents
the final filter as:
( ) pnppnpnpnp yayayayne −−−+ ++++= L2211
( ) npppnppnppnp yayayayne ++++= +−+−−−
L2211
Eq. 4.2
−
=
−
−−
−−
−−
−
−
−
1
011
1,1
2,1
1,1
,
1,1
2,1
1,1
,
1,
2,
1,
p
pp
pp
p
pp
pp
p
p
pp
pp
p
p
a
a
a
a
a
a
a
a
a
a
a
MMMγ Eq. 4.3
( ) ( )
( ) ( )[ ]∑
∑−
=
−−
+−
−−
−
=
+−
−+
−
=1
2
1
2
1
1
1
1
1
12
N
pn
pp
p
N
pn
p
p
nene
nene
γ Eq. 4.4
46
Where the order of A(z) is p, and G is a scalar gain. When the Burg estimator
block is used in the SIMULINK environment, a vector of A coefficients are returned,
along with the scalar G.
4.3 Real Time Workshop / Implementation into Genhel-PSU
To test the Burg model in a real-time simulation environment, Genhel-PSU had to
be modified. A controller was implemented into Genhel-PSU in which it accepted current
pilot commands, actuator commands, sensor outputs, trim values at initialization, and an
off or on switch from Genhel-PSU. If the switch is on, the controller will determine the
filter coefficients that best estimate the airwake properties by using a 6th
order Burg
autogressive model. The coefficients from the Burg model are written to a file.
Additionally, the controller simulates the general UH-60A BLACK HAWK mechanical
control system with SAS, so the controller passes out the actuator commands that are
used to move the swashplate in Genhel-PSU.
To develop the code for the controller, MATLAB’s Real Time Workshop (RTW)
complier was used. This is a picture to code approach that relied on SIMULINK models
of the UH-60 mechanical controls (with SAS) and of the Burg autoregressive model. The
RTW controller is developed as a dynamic link library (DLL) file that can be loaded by
Genhel-PSU during the initialization of the controller. The benefit of using a DLL file is
( )( ) ( ) p
p zpaza
GzH
−− ++++=
1211
L Eq. 4.5
47
that Genhel-PSU does not have to be recompiled every time a new DLL file is
implemented.
The UH-60A mechanical controls were validated by comparing the unaltered
Genhel-PSU model with simulation results of the RTW DLL file. The high level view of
the RTW controller can be seen in Fig. 4.2.
The gust filter identification block accepts the actuator commands as well as the
aircraft dynamics (sensor commands) at every time step of 0.01 sec. The gust
identification block first determines the equivalent body gusts as developed in Chapter 3.
Each extracted equivalent body gust is sampled at 20 Hz, or at every 0.05 sec, and is then
Figure 4.2: High Level View of the Gust Filter Identification Controller
48
passed to a buffer. The buffer collects 600 samples, or 30 sec, of equivalent gust vectors.
The buffer has an overlap of 560 samples, so that there is a smooth transition in filter
coefficients from one time step to the next. The deterministic portion of the equivalent
gusts (the mean of the signal) is then subtracted from the data set since the autoregressive
filter should only be estimating the stochastic winds. The signal is then sent to a 6th order
Burg autoregressive estimator block that is provided in the signal processing toolbox in
SIMULINK. The Burg model will determine new filter coefficients every 2 sec. The flow
chart for this process can be seen in Fig. 4.3.
The filter coefficients determined during the gust filter identification routine is
then sent to Genhel-PSU workspace, and written to a file at each time step. Since the
Burg autoregressive model will only update the filter coefficients every 2 sec, the
coefficients will be repeated in the output data file for 200 lines. For the initial feasibility
Figure 4.3: Gust Identification using a Burg AR Estimator
49
study, this did not create a problem in data processing. In future applications it may
become necessary to modify the code to only write when the filters are updated, or
perhaps only as needed as determined by other parameters. For large data collections
(hours of flight test data at several hover locations), it would become cumbersome to
have to handle and process unnecessarily large files.
4.4 Results of Burg AR Model
The autogressive model was implemented into Genhel-PSU and tested in the
DDG-81 airwake with a WOD condition of 25 kts / 0 deg. For this simulation, a new
linear model was developed to initialize the SIMULINK diagram used for RTW code
generation. The RTW was then used to generate a DLL file that was loaded into the
controls directory of Genhel-PSU. Once the DLL is loaded, Genhel-PSU is set to hover in
the selected wind condition with no additional SAS controllers turned on. Since the RTW
controller already estimates the UH-60A Mechanical system with SAS, adding additional
SAS to the controller would be adding additional SAS and skewing the results.
When the simulation is completed, the Burg AR model coefficients are examined
at every 20 seconds. To do so, the coefficients at each selected time step was used to
develop a filter. The frequency response of the transfer function was then compared to
the offline filter results found in Chapter 3. Over a short period time, one would expect
the Burg AR model to converge to the stochastic filter models determined in the offline
environment. Plots for each equivalent body gust can be found in Appendix C. Selected
figures for the DDG-81 CFD airwake and Burg AR Model are presented in Figs. 4.4-4.5.
50
100
101
-30
-25
-20
-15
-10
-5
0
5
10
15
20
pg (
dB
)
DDG81 25 kts 0 deg / AR Burg Method - 6th Order
Frequency (rad/sec)
Off-line
20 sec
40 sec
60 sec
80 sec
100 sec
120 sec
Figure 4.4: Online AR Filter Convergence for the Roll Rate Equivalent Gust
100
101
-35
-30
-25
-20
-15
-10
-5
0
5
10
qg (
dB
)
DDG81 25 kts 0 deg / AR Burg Method - 6th Order
Frequency (rad/sec)
Off-line
20 sec
40 sec
60 sec
80 sec
100 sec
120 sec
Figure 4.5: Online AR Filter Convergence for the Pitch Rate Equivalent Gust
51
Overall, all of the figures show convergence to a given filter in the online
environment; however, the convergence isn’t always to the filter found in the offline
environment. One draw back to the Burg AR model is that it analyzes the entire
spectrum, whereas the filter fitting technique in Chapter 3 was able to constrain the filter
fit to a frequency range between 0.4 and 10 rad/sec. This overall fitting method by the
Burg method may capture peaks at higher frequencies that are not of interest, and
providing a slightly worse filter fit through the frequencies that affect handling qualities
the most.
Even though the Burg method tends to capture the PSD within the spectral filter
over a larger range of frequencies, the Burg method still demonstrates convergence to
peaks at the same frequencies as the offline filter results. Sometimes, the magnitudes of
the PSDs from the Burg model are slightly lower than the offline filter results.
Chapter 5
Simulation Modeling of Airwakes Using Stochastic Filters
This chapter discusses the simulation modeling of an airwake using stochastic
filters. First, the motivation of this research effort is reviewed. Then, the details of the
simulation modeling and the code used for implementation into Genhel-PSU are
discussed. Finally, the simulation results from using this airwake in Genhel-PSU and
Sikorsky GENHEL are examined.
5.1 Motivation
In an effort to improve the simulation model of the shipboard environment, this
study examines replacing the detailed CFD-based ship airwake models presented in
Chapter 2 with a grid of stochastic filters. By relying on a CFD model, 3-axis velocity
data had to be stored into memory prior to running a simulation. Due to memory
requirements, limitations were placed on the length of the airwake and/or the fidelity of
the simulation. With these limitations, the CFD airwake is usually limited to 40-60
seconds of data. To run longer simulations, the airwake can be repeated; however, there
are noticeable repetitions in flight simulation data.
To reduce the dependence on memory requirements, this research moves towards
using a grid of stochastic filters that will be excited by random white noise input. For this
simulation environment, only the filter coefficients and time averaged CFD data for the
53
entire grid need to be stored into to memory. The benefits are that the airwake is no
longer constrained by memory requirements and that the airwake can be run indefinitely
without noticeable repetitions of data. Additionally, this approach could be used to
develop a stochastic airwake model derived from experimental data. This offers an
alternative to relying solely on a CFD model that may have uncertainties.
5.2 Stochastic Filter Grid Design
For an initial feasibility study of the stochastic airwake, a new grid design was
developed. At each grid point, the stochastic filter coefficients for the translational and
rotational axes are known. If the aircraft is in between a set of grid points, then the
stochastic filter coefficients would be interpolated. To minimize the dependence on
computing power requirements, but trying to create an accurate environment, it was
decided to design a grid around the desired landing path for a pilot performing a
shipboard landing. For this design, there is a condensed cluster of filters near the ship
deck, and sparsely placed filters near the start of the pilot’s approach. This was done
because the airwake is more turbulent behind hangar doors than 800 ft back. By selecting
a less detailed grid in areas expected to have a smaller changes velocity data, the
computational effort will be further reduced.
The grid was defined by first creating a vector of x-coordinate locations which
specified the distances behind the ship deck for each y-z grid plane. For this grid, there
are 11 equally spaced grid points with lower and upper bounds at x-airwake coordinates
of 424 and 574 ft. Then there are an additional four equally spaced coordinates behind
54
them with an upper bound of x-coordinate of 1000 ft. Overall, this would create a vector
ix , of length 15. Again, the vector ix ensured that the details of the ship airwake would
be preserved in the area with the highest turbulence near the ship deck, and allowed for
reduced computing power further out away from the deck where the ship airwake tends to
have smaller changes in disturbance characteristics.
Next, a matrix of x-y coordinates was defined using a constant change in the
parameter β . The vector of β angles is derived from the relationship of a desired y
coordinate with the x midpoint coordinate of 574 ft. The jβ vector is defined as:
In Eq. 5.1, 1Px is the midpoint x value of 574 feet. The variable 1Py is the
maximum y value at the midpoint 1Px . For this grid, yP1 was selected to be 80 ft. The
vector jβ is selected to have a length of 3. This would create angle of zero degrees at the
center line of the ship, and two angles extending in the positive direction. When creating
the full matrix x-y matrix, the y-values from the positive side of the grid will be mirrored
across the center line of the ship to create a symmetric grid. Finally, ny in Eq. 5.1 is
defined as ( )1−j .
The y-coordinate values are a function of the jβ vector and the ix vector. The x-
y matrix of size ( )ij , is found computing Eq. 5.2.
( ) ( )1tan
tan 11
1
1
11 −−
=
−−
j
n
PP
P
Pj i
y
xy
x
yβ Eq. 5.1
55
The x-y matrix is then mirrored about the center line of the ship to create a
symmetric x-y matrix of size [ ]( )ij ,12 − . The x-y coordinate grid can be seen in Fig. 5.1.
In this figure, the black circles represent the x-y coordinate location for a particular set of
translational and rotational equivalent gust body filters.
Next a matrix of x-z coordinates was defined. This used a similar approach to the
x-y coordinate matrix, but employed a constant change in a parameter α . The vector of
( )jiij xy βtan, = Eq. 5.2
500 600 700 800 900 1000
-200
-150
-100
-50
0
50
100
150
200
x (ft)
y (
ft)
Figure 5.1: Schematic of the x-y coordinate grid.
56
α angles is derived from the relationship between the difference in the upper and lower
bound of the z-axis when compared to the x axis. The kα vector is defined as:
In Eq. 5.3, the parameter min0Pz is the lower limit of the z-coordinates and is
defined as 24.2 feet. This was set from the lowest possible grid point that allowed the
aircraft to trim in Genhel-PSU while running the CFD DDG-81 airwake. The upper
bound of the z-axis at location 1x , is the parameter max0Pz . It is defined as 60 ft, which
gave adequate room to hover in the z-axis. The vector kα has a length of 5, and nz is
defined to be ( )1−k .
The z-coordinate values are a function of the kα vector and the ix vector. The x-z
matrix of size ( )ik , is found computing Eq. 5.4.
The x-z coordinate grid can be seen Fig. 5.2. In this figure, the black circles
represent the x-z coordinate location for a particular set of translational and rotational
equivalent gust body filters.
( )1
tan
tan0
min0max01
0
min0max01 −
−
−
−=
−
−k
n
P
PP
P
PP
ki
z
x
zz
x
zzα
Eq. 5.3
( ) min0, tan Pkiik zxz += α Eq. 5.4
57
Incorporating the x-y and x-z matrices together, a three dimensional matrix can be
composed. The grid concentrated around typical areas that the pilot would fly during the
approach to a landing on the back of the destroyer and also allowed significant space for
the pilot to deviate from the desired flight path. A three dimensional view of the
stochastic grid can be seen in Fig. 5.3.
400 500 600 700 800 900 10000
50
100
150
x (ft)
z (
ft)
Figure 5.2: Schematic of the x-z coordinate grid.
58
5.3 Implementation into GENHEL-PSU
To implement the stochastic airwake into Genhel-PSU, additional code had to be
written. The original method for incorporating airwake effects was that first the entire
CFD airwake was read into memory. Then, each aircraft component would call the
memory to determine the local airwake velocity by interpolating the x, y, and z velocities
in time and with a weighted position coordinate.
The new modifications will first read in stochastic and deterministic airwake
grids. The stochastic grid is the same as what was presented in Section 5.2. The
deterministic airwake is simply the time averaged solution of the CFD airwake, so the
grid used for deterministic portion is the same grid file as the CFD airwake that is
500
600
700
800
900
1000
-100
0
100
50
100
x (ft)
y (ft)
z (
ft)
Figure 5.3: Three Dimensional View of the Stochastic Filter Grid.
59
presented in Section 2.2.1. Also during initialization, a file containing the stochastic filter
coefficients for each stochastic grid point is read in, as well as a file with the u, v, and w
time averaged velocities for each deterministic grid points.
Finally, to excite the stochastic filters during flight, a white noise source is
required. During the initialization of Genhel-PSU, an arbitrarily long random number
vector for each six stochastic components is read into memory. The built in random
number generator in FORTRAN is not used because the white noise source must be
normally distributed. MATLAB offers a randn command that can be used to create a
normally distributed vector. Saving the MATLAB workspace of six normally distributed
vectors to a text file provides a quick solution for the initial implementation effort.
After Genhel-PSU is initialized, the stochastic airwake program is called at every
time step by main aerodynamic module. The stochastic airwake subroutine accepts the
fuselage location and translates it from North, East, Down (NED) coordinates to airwake
coordinates. For visualization, the aircraft, ship, and airwake coordinate system is
illustrated in Fig. 5.4.
Figure 5.4: Genhel-PSU Coordinate System Diagram
60
Once in airwake coordinates, the program determines the aircraft’s location both
in the stochastic and deterministic grids. Additionally, the nearest 8 grid points from each
grid are determined. Figure 5.5 shows the aircraft fuselage’s position within either grid
system. Note that the grid will fan out in both the y and z coordinates in both the
stochastic and deterministic airwake grids.
The airwake program then needs to determine the local stochastic filter
coefficients. The nearest eight grid points’ filter coefficients are read from the stochastic
filter matrix. A location-weighted interpolation is performed to find the stochastic filter
coefficients for the local point in the stochastic airwake grid. The local coefficients are
then passed into a Tustin subroutine with a newly selected random number. The
subroutine follows the methodology of Section 3.5. This subroutine also requires the last
four Tustin routine outputs and the last four random numbers. These values are saved into
two arrays from time step to time step, and are updated after each Tustin calculation. The
new Tustin output is the current stochastic airwake velocity component. Since the
Figure 5.5: Aircraft Position and Nearest Eight Grid Points
61
stochastic filters are derived in the aircraft’s body coordinates, the current stochastic
component is considered to be the stochastic airwake velocity acting on the center of the
fuselage in body coordinates.
Next, the deterministic velocity is determined. Similar to the stochastic method, a
location-weighted interpolation is performed to find the local time averaged velocities.
Since the local time averaged velocities are in airwake coordinates, the values are
transformed into body coordinates. The u, v, w deterministic components are then added
to the stochastic u, v, w components. The summation of the time averaged velocities and
the stochastic local components are then stored into a global ‘body wake’ variable. In
Genhel-PSU, the u component is called UBWAKE. The other five components have
similar variable names.
Each aerodynamic component (Fuselage, Main Rotor, Tail Rotor, Tail) will rely
on these ‘body wake’ variables to determine the local component velocity. Since the
global variables are actually the equivalent body gusts (plus the time averaged velocity
components), it is necessary to transform the equivalent body gusts to the local
component velocities using the Eqs. 3.21-3.22. Note that it is technically incorrect to pass
the time averaged values through Eq. 3.21; however, since these variables are only faced
with addition and subtraction and would eventually be added through Eq. 3.22, this
implementation method should be okay. Additionally, since in the Genhel-PSU airwake
environment differentiates between the mean steady wind and airwake turbulence
velocities, the mean steady wind will then need to be added to the final stochastic and
time averaged velocities to find the total wind velocity for aerodynamic calculations.
62
Overall, this program methodology should provide a good first implementation of
a stochastic airwake; however, there are known limitations of the program. First, the
aircraft must remain inside the airwake. The simulation will not store zeros into the
output and input vectors of for the Tustin subroutine when the subroutine is not called. As
a result, when the aircraft moves out of the airwake, and then reenters, it will rely on
information from a few time steps ago. Second, this program assumes the aircraft is at the
same attitude at which the helicopter had during the hover maneuver performed to
generate the continuous filters. If the aircraft nears the back of the ship with sideslip, or a
nose down attitude, then the simulation may not provide an accurate estimation of aircraft
disturbances. Finally, the length of unrepeated flight simulation data is dependent on the
length of the random number vectors read into memory at the start of simulation.
5.4 Genhel-PSU Stochastic Airwake Results
To test the stochastic airwake code, a simulated hover was performed in the
stochastic airwake and at the corresponding location within the DDG-81 CFD airwake.
The PSD of the inceptor commands that were created from the pilot model are examined
and compared for both airwakes. Additionally, the PSD of the airwake velocities acting
on the aircraft at the fuselage center are examined. For confirmation of the results, the
analysis process was performed at several locations. Four of these simulated hover
comparisons are outlined in Appendix D. The four locations presented are 544-37-58,
439-0-42, 454-0-52, and 454-31-33. The numbers present the X-Y-Z airwake coordinate
positions. For example 454-31-33 was a hover performed at 454 ft behind the origin of
63
the ship, 31 ft to the starboard side of the ship, and 33 ft above the origin of the ship. For
clarity, one can reference Fig. 5.4.
Overall, the PSDs of lateral stick inputs corresponded well at lower frequencies
and tended to develop discrepancies at higher frequencies. The longitudinal inputs
showed more consistent results across the frequency range, but in specific cases exhibited
a constant a small shift in magnitude. In the collective stick input, the PSDs were
consistent between the CFD and stochastic airwakes. When examining the pedals, the
PSDs had varying results across the frequency range but suggest a correlation between
the workloads between the CFD and stochastic airwakes. Figures 5.6- 5.9 present sample
results for each inceptor input.
64
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 439-0-42
Freq (rad/sec)
Late
ral S
tick, X
a [(%
)2]
Stochastic
CFD
Figure 5.6: Lateral Stick Input for 439-0-42 Location
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 454-0-52
Freq (rad/sec)
Longitudin
al S
tick, X
b [(%
)2]
Stochastic
CFD
Figure 5.7: Longitudinal Stick Input for 454-0-52
65
100
101
10-5
10-4
10-3
10-2
10-1
100
101
DDG-81 WOD Case: 25kts / 0deg at 454-0-52
Freq (rad/sec)
Colle
ctive S
tick, X
c [(
%)2
]
Stochastic
CFD
Figure 5.8: Collective Stick Input for 454-0-52
100
101
10-8
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 544-37-58
Freq (rad/sec)
Pedals
, X
p [(
%)2
]
Stochastic
CFD
Figure 5.9: Pedal Inputs for 544-37-58
66
When examining the PSDs of the airwake velocities acting on fuselage center,
clear trends emerged. The longitudinal axis PSD matched well where as the vertical axis
tended to show large discrepancies. Overall, the stochastic airwake tended to be less
violent in terms of the vertical axis. When comparing the CFD airwake and the stochastic
airwake for the lateral axis, the data tended to match at low and high frequencies, but
showed a weaker stochastic velocity in the middle of the handling qualities range.
Samples of each axis are shown in Figs. 5.10-5.12.
100
101
10-4
10-2
100
102
104
DDG-81 WOD Case: 25kts / 0deg at 439-0-42
Freq (rad/sec)
U -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure 5.10: Longitudinal Velocity of Gust acting on the Fuselage center for 439-0-42
67
100
101
10-2
10-1
100
101
102
103
104
DDG-81 WOD Case: 25kts / 0deg at 454-0-52
Freq (rad/sec)
V -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure 5.11: Lateral Velocity of Gust acting on the Fuselage center for 454-0-52
100
101
10-2
10-1
100
101
102
103
DDG-81 WOD Case: 25kts / 0deg at 439-0-42
Freq (rad/sec)
W -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure 5.12: Vertical Velocity of Gust acting on the Fuselage center for 439-0-42
68
The discrepancies in the vertical axis and lateral axes are not clearly explained.
One source of error could be that the stochastic filters are derived from a helicopter
trimmed in a hover. The trimmed hover maneuver tends to leave the helicopter slightly
nose up and tilted in a left-wing down position. As a result, when the equivalent gusts are
extracted, there is an error in the assumption that we are fitting a filter to the direct lateral
and vertical gusts.
Although there are differences in a direct wind comparison between the CFD and
stochastic airwake solution, the goal of this study to be able to find a stochastic airwake
solution that will induce the same pilot workload. The simulated hovers near specific grid
points produced reasonable results with some inconsistencies.
To further examine the control effort required in the stochastic airwake, two
simulated approach to landings were performed. In the first trajectory path, the aircraft
started in a hover 450 ft behind the ship deck and 60 ft to the starboard side at an altitude
of 77 ft. After 30 sec, the aircraft takes 30 sec to move the aircraft to a location near the
desired landing spot. The aircraft hovers for a full minute before the end of the
simulation. The second approach is similar; however, the starting location is about 400
feet behind the ship deck, 35 ft on the port side of the ship, and at a height of 60 feet. The
trajectories were flown in both the stochastic airwake and the CFD airwake. Additionally,
to determine how much pilot workload required by the maneuver, versus how much an
airwake would affect the pilot workload, the same trajectory paths were also flown in a
steady 25 kts wind, with no airwake turbulence.
The trajectory paths and sample flight paths for the aircraft flying both maneuvers
within the stochastic airwake can be found in Appendix D. Additionally, the PSD
69
comparison of inceptor commands required to perform the maneuver in each wind
airwake condition are presented. Overall, there is remarkable agreement between the pilot
workload required in all axes between the CFD and stochastic airwakes. Additionally,
one can see that the airwake turbulence tends to affect the lateral and pedal pilot
workload more than the longitudinal axis workload. Another interesting observation is
that most of the pilot workload in the collective came from flying the maneuver, only a
slight increase in PSD magnitude was found for the airwake conditions. Two sample
figures are presented in Figs. 5.13 and 5.14. Again, a complete set of figures are shown in
Appendix D.
70
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg
Freq (rad/sec)
Pedals
, X
p [(
%)2
]
Stochastic
CFD
No Airwake
Figure 5.13: Pedal Input for Trajectory Path 1
100
101
10-4
10-2
100
102
104
DDG-81 WOD Case: 25kts / 0deg
Freq (rad/sec)
Colle
ctive S
tick, X
c [(
%)2
]
Stochastic
CFD
No Airwake
Figure 5.14: Collective Stick Input for Trajectory Path 2
71
5.5 Implementation into Sikorsky GENHEL and Piloted Results
The stochastic filter airwake was also implemented into Sikorsky GENHEL. The
grid outlined in Section 5.2 was programmed into the real-time simulation by Sikorsky
Aircraft Corporation. Each grid point included six colored filters developed using the
mythology presented in Section 3.3, and excited with a white noise input. The simulation
was of the S-92 FBW aircraft flying in the airwake model behind a DDG-81 with a WOD
condition of 25 knots / 0 deg.
To asses the new airwake model, piloted simulations were performed. The model
was tested with six pilots; three with Navy backgrounds, three with Army backgrounds.
All pilots were new to flying the S-92 fly-by-wire type aircraft. The pilots were asked to
fly the MTE outlined in Section 2.4.1.
The piloted simulations were a smaller mission task of a larger simulation testing
program that evaluated different gust alleviation controllers as well as several airwake
models of the turbulence behind the DDG-81. The simulations presented here will only
compare the CFD airwake model provided by NAVAIR with the stochastic airwake
model, with no gust alleviation controllers active in the system. The pilots were asked to
rate the workload using the Cooper-Harper rating system for HQR values. Additionally,
the aircraft position, attitudes, rates and pilot stick/pedal input were recorded for each
simulation run.
From the piloted simulations, the pilots generally felt the stochastic airwake was
slightly more violent than the CFD airwake. In terms of HQRs, the pilots rated the
stochastic airwake case with an average of 4.75, which is nearly 0.75 higher than the
72
CFD airwake turbulence case. The consensus of the pilot comments was that the
workload was increased compared to the CFD solution; however the large 0.75 HQR
difference may be due to a HQR rating of 7 given by a single pilot. Some pilots
commented that there was a large pull on the aircraft near the hanger roof, which was not
present in their CFD case runs. A summary of HQR ratings per pilot can be seen in
Fig. 5.15.
To examine the HQRs and pilot comments more, pilot workload was analyzed by
plotting the PSD of the pilot control inputs. Figs. 5.16-5.18 show sample PSD plots of
various pilot’s input for both the CFD and stochastic filter airwakes, as well as with the
Figure 5.15: Pilot HQR assessment for various Airwake conditions
73
airwakes turned off. The data selected for this analysis is the pilots performing a 30
second hover over the landing spot. These figures show that the stochastic airwake is
fairly comparable in PSD magnitudes. When examining several of the pilots, it is clear
that once again the workload in the collective axis tends to come from the maneuver
itself. Typically in this set of data, the collective PSD for the no airwake condition
matches the CFD and stochastic airwake PSD values. Additional PSD plots can be seen
in Appendix D.3.
100
101
10-8
10-6
10-4
10-2
100
Frequency (rad/s)
PS
D o
f Late
ral S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure 5.16: Lateral Stick Input for Pilot 4
74
100
101
10-6
10-4
10-2
100
102
Frequency (rad/s)
PS
D o
f Long. S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure 5.17: Longitudinal Stick for Pilot 5
100
101
10-6
10-4
10-2
100
102
Frequency (rad/s)
PS
D o
f P
edals
, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure 5.18: Pedal Input for Pilot 6
75
The HQR ratings leave room for question on to why the stochastic airwake was
perceived as more difficult to fly. Although some PSD figures show comparable results,
there are a few figures that show the CFD airwake as more work, as well as some figures
showing the stochastic airwake as requiring more pilot work load. The HQR ratings
should be used as a reason to further explore possible complications.
One known problem with the Sikorsky GENHEL stochastic solution is that the
filters were generated 13 ft higher from the perceived location. (This error was corrected
in the Genhel-PSU based simulation.) The 13 ft error would most likely cause the
airwake to be more violent behind the hangar doors because there wind shear effect off
the back of the superstructure that is now 13 ft lower than expected. Additionally, a
trained pilot may not be expecting the wind sheer to occur at this height.
Another possible problem may be the random number generator for the Tustin
filter computation. At the time of simulation testing, the random number generator was a
uniformly distributed vector, and not a normally distributed noise vector. The resulting
stochastic model disturbance would not be as realistic with the uniformly distributed
vector.
Overall, both the Genhel-PSU and Sikorsky GENHEL testing show that
stochastic airwake could be developed in the manner presented in this paper to be used in
future simulation testing. Although some errors emerged in correlating the stochastic
airwake to the CFD airwake, the pilot workload induced by Genhel-PSU was very similar
between the airwakes. The piloted simulation HQR results suggested slightly higher
workload; however, it would need to be tested further to detect additional problems.
Chapter 6
Conclusions
Shipboard operations continue to be one of the most challenging for naval pilots.
The unsteady winds and moving ship deck make precision hovering and landing tasks
increasingly difficult. The pilot workload created during these tasks can often be a
limiting factor in the WOD envelop for a particular ship and aircraft combination. To
improve the handling qualities of aircraft flying in these adverse conditions, recent
research has been performed in industry to identify and reject gust disturbances. To test
flight control modifications and to train maritime pilots, significant effort has also been
put towards developing accurate real time simulation models.
This research study further investigated the shipboard landing environment by
developing tools to accurately identify the spectral properties of the stochastic winds in
both an offline and online environment. The principles from the wind identification were
then applied to create a real time stochastic simulation model that required less
computational effort than a full CFD solution.
6.1 Analysis Summary
For control law gust rejection development and accurate real time simulation
models, it is desirable to be able to accurately identify the spectral properties of the
stochastic winds affecting pilots during shipboard operations. Previous work suggested
77
fitting von Karman turbulence models to the power spectral density of an aircraft’s
equivalent airwake gust. This method was modified by changing the fitting routines to fit
an ordinary third order transfer function to the spectral properties. This technique was an
improvement because it guaranteed stable filters that were design to fit the equivalent
gusts through the frequency range that affects the handling qualities of an aircraft most
(0.4 to 10 rad/sec). Additionally, the filter fitting routines were explored by trying
different windowing techniques. It was found that the popular windowing techniques did
not affect the final filter fit.
As an initial simulation effort, the continuous gust identification filters were
transformed into discrete filters using a Tustin 4th
order model. The discrete and
continuous filters were excited with a normally distributed white noise and compared to
the extracted CFD gusts. The continuous and discrete filters provided almost identical
stochastic results. When comparing the filter results to the CFD model, the longitudinal
axis generally had the same PSD magnitudes. The lateral axis matched slightly less.
There was significant misalignment in terms of the PSD magnitudes of the vertical gusts
between the stochastic filters and CFD model.
Since it would be computationally intensive to collect data and generate filters for
every ship, aircraft, and WOD combination that an aircraft might encounter, it is
desirable to develop a gust identification technique that can be performed during flight or
in an online simulation environment. A previous study suggested Burg 6th order AR
models. To explore this subject more, the Burg models were integrated into Genhel-PSU
using generic controller that accepted a dynamic linked library file that contained a UH-
60A mechanical model (with SAS) as well as a Burg AR model used to identify filters of
78
the equivalent gusts acting on the aircraft. The DLL file was created using MATLAB’s
Real Time Workshop, which converted a SIMULINK diagram of the mechanical controls
and the gust identification techniques to code.
During real time simulations in Genhel-PSU, the Burg AR Models were capable
of identifying key spectral peaks of the gust. It was shown the Burg AR Models could be
used to store the filter coefficients in an online environment, and could be used in future
industry applications.
Finally, the gust identification principles were used to develop a stochastic model
of the DDG-81 CFD solution provided by NAVAIR. In the first third of the research, it
was found that three rotational and three translational gust filters could be defined based
on equivalent gusts. A grid was defined that focused on a typical flight path for a pilot
landing on the back of at a ship. For each grid point, the six filters were developed.
Additionally, a time-averaged velocity solution was found from the CFD solution for a
series of grid points.
A stochastic airwake code was implemented into Genhel-PSU that during
initialization, read in the stochastic airwake grid, stochastic airwake continuous filter
coefficients, time-averaged airwake grid, and time-averaged velocity values. During
flight, the aircraft’s position was used to perform a position weighted interpolation of
stochastic filter coefficients and time averaged values. The stochastic filter coefficients
were then passed to a Tustin subroutine that accepted a normally distributed white noise
input to excite the discrete filters. The stochastic wind components were added to the
time averaged components, and passed back to the general aerodynamic model to
calculate local body velocity coordinates.
79
To test the stochastic airwake model, hover simulations were performed within
the CFD airwake and the stochastic airwake. The PSD of the pilot inputs were compared
between the airwakes. It was found that they generally agreed with a few discrepancies.
The longitudinal, lateral, and vertical velocities acting on the fuselage center was also
examined. It was found the PSDs compared well for the longitudinal axis between the
CFD and stochastic airwake model. The lateral axes had some discrepancies in the
middle of the handling qualities frequency range, and the vertical axis did not match well.
During the simulation testing of the stochastic model, the aircraft was also
commanded to fly a trajectory path that included an initial hover, to fly towards the
landing spot, and then to hover for another 60 seconds. The PSD of the pilot inceptor
commands matched extremely well between the CFD and stochastic airwake model. The
success from this testing suggests that this method could be used for accurate real time
simulation environments. The benefit to such simulation models is that it is less
computationally demanding on computing requirements and that the airwake solution
would not have to be continuously cycled. In past work, repetitions in flight data were
found when the CFD solutions were repeated for longer real time simulations. Moreover,
this technique could be used to derive airwake simulation model from flight test data.
This is beneficial as CFD solutions can sometimes include uncertainties.
Finally, the stochastic airwake model was also implemented into Sikorsky
Genhel. Piloted simulations were performed where the pilots were asked to fly a specific
maritime Mission Task Element, and then to rate the pilot workload on the Cooper-
Harper Rating scale. Overall, the pilots commented that the stochastic airwake was more
violent than the CFD airwake solution. In particular, the pilots rated the stochastic
80
airwake to be 0.75 HQRs higher than the CFD solution. When examining the pilot
command inputs between the CFD and stochastic airwake, no clear indication was found
to what was causing the pilots to feel the stochastic airwake model was more violent. In
several cases, the PSD of the pilot stick commands were similar between the CFD and
stochastic airwakes. In a few cases the stochastic airwake did create more workload, but
in other cases the CFD provided to be more work also.
During analysis of the Sikorsky GENHEL results, it was realized that the filters
were generated from the CFD solution 13 ft higher than expected. As a result, the filters
were effectively 13 feet lower in real time simulation than what they should have been.
This would have created misalignment between the stochastic airwake model and the ship
visuals that the pilots relied on to perform the MTE tasks. This could be related to the
differences seen as the pilots would have felt a strong wind shear below the visual ship
superstructure instead of it being aligned with the superstructure.
6.2 Recommendations for Future Work
Although this research has made significant progress in creating accurate gust
identification models and a stochastic airwake model, a few points could be examined in
greater detail. The first relates to the equivalent gust extraction process, and the second
relates to known limitations of the programmed stochastic airwake grid within Genhel-
PSU.
81
6.2.1 Equivalent Gusts
Overall the gust identification in both the offline and online environments worked
well in evaluating the key spectral properties of the equivalent gusts. However, when the
initial simulation was run where the continuous filters from the offline identification
method was excited, there were significant discrepancies found the vertical axis as
compared to the CFD airwake gusts. Since the filter fits showed agreement between the
equivalent gusts and the generated gusts from the continuous filters, the error must fall
from the equivalent gust extraction method. It is suggested that this area be examined
further. A suggested area is to resolve whether the reduced linear model is accurately
representing the aircraft dynamics.
6.2.2 Stochastic Airwake Model Limitations
When programmed, the stochastic airwake subroutine for Genhel-PSU was
developed with known limitations:
First, the subroutine assumes the aircraft is at the same attitude at which the
helicopter had during the hover maneuver performed to generate the continuous filters. If
the aircraft nears the back of the ship with sideslip, or a nose down attitude, then the
simulation may not provide an accurate estimation of the stochastic aircraft disturbances.
Secondly, the time averaged airwake values are passed in through the body wake
variables (eg. UBWAKE). This assumption also requires the aircraft is aligned with the
ship axes. If the aircraft is in a side slip 90 degrees to the right, then the u-component
time averaged solution is still being added to from the nose of the aircraft, and not from
82
the side of the aircraft. It is suggested that this be reprogrammed to first assume that the
body wake variables are actually in ship airwake coordinates (instead of body
coordinates.) This might be a reasonable assumption since the body coordinates are fairly
well aligned with the ship axes with the filters are derived.
Next, the stochastic airwake grid should be expanded to include more grid points.
It is known that the PSD of the CFD airwake changes rapidly in magnitude along the
vertical axis over the ship deck. It is suggested to add more grid points to create a tighter
grid above the ship deck. Additionally, adding points to the left and right of the ship deck
would allow for simulated trajectories to approach the back of the ship at various angles
instead of just shallow sideslip maneuvers.
Finally, the length of unrepeated flight simulation data is dependent on the length
of the random number vectors read into memory at the start of simulation. The stochastic
airwake subroutine requires a set of random number vectors to be read into simulation
because the built in random number generators in FORTRAN are not normally
distributed. Future work in this area should involve developing a subroutine that
calculates a normally distributed random number vector series during real time
simulation. With this subroutine included, Genhel-PSU could run indefinitely without
noticeable repetitions in flight data.
Bibliography
1. Wilkinson, C.H, Roscoe, M.F., and VanderVliet, G.M, “Determining Fidelity
Standards for the Shipboard Launch and Recovery Task,” Proceedings of the
AIAA Modeling and Simulation Technologies Conference and Exhibit, Montreal,
Canada, August 2001.
2. Polsky, S. A., “A Computational Study of Unsteady Ship Airwake,” Proceedings
of the 40th
AIAA Aerospace Sciences Meeting & Exhibit, Reno, Nevada, AIAA
Paper 2002-1022, Jan. 2002.
3. Healey, J.V., “Estabilishing a Database for Flight in the Wakes of Structures,”
Journal of Aircraft, Vol. 29, No.4, July-Aug 1992.
4. Clement, W.F., Gorder, P.J., Jewell, W.F., “Development of a real-time
simulation of a ship-correlated airwake model interfaced with a rotorcraft
dynamic model,” Proceedings of the AIAA/AHS Flight Simulation Technologies
Conference, Hilton Head Island, SC, Aug 24-26, 1992.
5. Host, C.T., “Estimating Ship Deck Aerodynamic Influence on the V-22 using
Modified Inverse Simulation Techniques,” Proceedings of the American
Helicopter Society 56th
Forum, Virginia Beach, Virginia, May 2-4, 2000.
6. Polsky, S. A., and Bruner, C. W. S., “Time-Accurate Computational Simulations
of an LHA Ship Airwake,” AIAA Paper 2000-4126, Aug. 2000.
7. Woodson, S.H., and Ghee, T.A., “A Computational and Experimental
Determination of the Air Flow Around the Landing Deck of a US Navy Destroyer
(DDG)”, in Proceedings of the 23rd AIAA Applied Aerodynamics Conference,
Toronto, Ontario, June 2005, AIAA-2005-4958.
8. Polsky, S., Imber, R., Czerwiec, R. and Ghee, T. “A Computational and
Experimental Determination of the Air Flow Around the Landing Deck of a US
Navy Destroyer (DDG): Part II,” in Proceedings of the 37th AIAA Fluid
Dynamics Conference and Exhibit, Miami, FL, June 2007, AIAA-2007-4484.
9. Lee, D., Sezer-Uzol, N., Horn, J.F., Long, L. N., “Simulation of Helicopter
Shipboard Launch and Recovery with Time-Accurate Airwakes,” Journal of
Aircraft, Vol. 42, No. 2, March-April 2005.
10. Zan, S. J., “Computational-Fluid-Dynamics Based Advanced Ship-Airwake
Database for Helicopter Flight Simulation,” Journal of Aircraft,Vol. 40, No. 5,
2003, p. 1007.
84
11. Forrest, J.S., Hodge, S.J., Owen, I., Padfield, G.D., “Towards fully simulated
ship-helicopter operating limits: The importance of ship airwake fidelity,”
Proceedings of AHS 64th Forum, Montreal, Canada, May 2008.
12. Shipman, J., Arunajatesan, S., Menchini, C., and Sinha, N., “Ship Airwake
Sensitivities to Modeling Parameters,” AIAA 2005-1105-881.
13. Labows, S.J., Blanken, C.L., Tishcler, M.B., “Uh-60 Black Hawk Disturbance
Rejection Study for Hover/Low Speed Handling Qualities Criteria and
Turbulence Modeling.” Proceedings of the American Helicopter Society 56th
Forum, Virginia Beach, VA, May 2-4, 2000.
14. Lusardi, J.A., Tischler, M.B., Blanken, C.L., Labows, S.J., “Empirically Derived
Helicopter Response model and Control System Requirements for Flight in
Turbulence.” Journal of the American Helicopter Society, Vol. 49, No. 3, July
2004, pp. 340 - 349.
15. Horn, J.F., Bridges, D.O., Lee, D., “Flight Control Design for Alleviation of Pilot
Workload during Helicopter Shipboard Operations.” Proceedings of the American
Helicopter Society 62nd
Forum, Phoenix, AZ, May 9-11, 2006.
16. Horn, J.F. and Bridges, D.O., “A Model Following Controller Optimized for Gust
Rejection during Shipboard Operations,” Proceedings of the American Helicopter
Society 63rd
Forum, Virginia Beach, VA, May 1-3, 2007.
17. Montanye, Pamela, “Shipboard Helicopter Gust Response Alleviation Using
Active Trailing Edge Flaps,” Masters Thesis, The Aerospace Engineering
Department, The Pennsylvania State University, August 2008.
18. Howlett, J., “UH-60A BLACK HAWK Engineering Simulation Program:
Volume I – Mathematical Model,” NASA CR_177542, USAAVSCOM TR 89-A-
001, September 1989.
19. Horn, J.F., Bridges, D.O., Lopes, L.V., and Brentner, K.S., “Development of a
Low-Cost, Multi-Disciplinary Rotorcraft Simulation Facility,” Journal of
Aerospace Computing, Information, and Communication, vol. 2, no. 7, pp. 267–
284, July 2005.
20. Horn, J.F. and Lee, D., “Simulation of Pilot Workload for a Helicopter Operating
in a Turbulent Ship Wake,” Proceedings of the American Helicopter Society 63rd
Annual Forum, Virginia Beach VA, May 1-3, 2007
21. Hess, R.A., “A Simplified Technique for Modeling Piloted Rotorcraft Operations
Near Ships,” in Proceedings of the AIAA Atmospheric Flight Mechanics
Conference and Exhibit, San Francisco, CA, August 16–18, 2005, AIAA Paper
2005-6030.
85
22. Jensen, J.J., Mansour, A.E., and Olsen, A.S., “Estimations of Ship Motions Using
Closed-Form Expressions,” Ocean Engineering, vol. 31, pp. 61–85, 2004.
23. Bridges, D.O., “Flight Dynamics Modeling, Simulation, and Control of Rotorcraft
in Airwakes,” Ph.D. Thesis, Department of Aerospace Engineering, The
Pennsylvania State University, University Park, PA, pending publication.
24. Lee, D., Horn, J.F., Sezer-Uzol, N., and Long, L.N. “Simulation of Pilot Control
Activity during Helicopter Shipboard Operation,” in Proceedings of the AIAA
Atmospheric Flight Mechanics Conference and Exhibit, Austin, TX, August 11–
14 2003, AIAA Paper 2003-5306.
25. “United States Navy Fact File: Destroyers – DDG,” http://www.navy.mil, Aug
2008.
26. Strang, W.Z., Tomaro, R.F., and Grismer, M.J., “The Defining Methods of
Cobalt60: A Parallel, Implicit, Unstructured Euler/Navier-Stokes Flow Solver,”
AIAA-99-0786, January 1999.
27. Geiger, D., Sahasrabudhe, V., Horn, J.F., Bridges, D., Polsky, S., “Advanced
modeling and Flight Control Design for Gust Alleviation on Ship-Based
Helicopters,” Proceedings of the American Helicopter Society 64th
Annual
Forum, Montreal, Canada, May 2008.
28. “USS Peleliu (LHA 5)”, www.tarawa.navy.mil, August 2008.
29. Sezer-Uzol, Nilay, “Unsteady Flow Simulations Around Complex Geometries
Using Stationary or Rotating Unstructured Grids,” PhD Dissertation, The
Aerospace Engineering Department, Penn State University, Dec. 2006.
30. Sezer-Uzol, N., Sharma, A., and Long, L. N., "Computational Fluid Dynamics
Simulations of Ship Airwakes," Vol. 219, Part G, Journal of Aerospace
Engineering, 2005.
31. “Cooper-Harper Rating Scale,” http://history.nasa.gov/SP-3300/fig66.htm,
October, 2008.
32. Oppenheim, A.V., Schager, R.S., Digital Signal Processing, Prentice Hall,
Englewood Cliffs, NJ, 1975. pgs 532-571.
33. Franklin, G.F., Powell,J.D, Workman, M.L., “Digital Control of Dynamic
Systems, 2nd
Ed.”Addison-Wesley Publishing Company, Inc, Reading, MA, 1990.
34. Horn, J.F, Cooper, J., Schierman, J., Sparbanie, S.M., “Adaptive Gust Alleviation
for a Tilt-Rotor UAV Operating in Turbulent Airwakes,” Proceedings of the
86
AIAA Guaidance, Navigation and Control Conference and Exhibit, 18-21 August
2008, Honolulu, Hawaii.
35. McCormick, Anthony V., Schierman, John D., and Ward, David G., “UAV
Shipboard Operations: Simulation Model Development”, tech. report, NASA
N68335-03-D-0097 – D.O. 0004, November 2006.
36. Schierman, J.D., Cooper, J., and Horn, J.F., “Innovative Rotorcraft Control for
Shipboard Operations,” tech. report, N68335-07-C-0310, October 2007.
37. Marple, S. Lawrence, “Digital Spectral Analysis: With Applications,” Prentence-
Hall, Inc., EnglewoodCiffs, NJ, 1987.
38. Orfandis, Sophocles J., “Optimum Signal Processing: An Introduction, 2nd
Edition,” McCraw-Hill Publishing Company, New York, NY, 1988.
39. Haykin, Simon, Adaptive Filter Theory, 3rd
Edition, Prentice Hall, Inc., Upper
Saddle River, NJ, 1996.
Appendix A
Settings for Genhel-PSU
Table A.1: DDG-81 Hover Coordinates
Airwake Coordinates:
X 430
Y 0
Z 0
Trajectory File (Flight Gear Off):
X -204.8525
Y 0
Z 27.8683
Trajectory File (Flight Gear On):
X -204.8525
Y 0
Z 14.6754
Standard WOD Condition 25 kts / 0 deg
Table A.2: LHA Hover Coordinates
Trajectory File (Flight Gear Off):
X -307.9400
Y -46.2500
Z 93.7773
Trajectory File (Flight Gear On):
X -307.9400
Y -46.2500
Z 87.6999
Standard WOD Conditions: 30 knots / 30 deg
30 knots / 0 deg
88
Table A.3: Genhel-PSU Matlab GUI Interface
Initial Conditions
Location: Other
Latitude [deg N] 0.02
Longitude [deg W] 0.02
X-Offset [ft] Set by Trajectory File
Y-Offset [ft] Set by Trajectory File
Z-Offset [ft] Set by Trajectory File
Altitude [ft] Set by Trajectory File
Speed [kt] Set by Trajectory File
χ [deg] 0
γ [deg] 0
ψ [deg] 0
Weight/CG Location
Weight [lb] 16825.0
Fuse. Station CG [in] 355.0
Waterline CG [in] 248.2
Buttline CG [in] 0.0
Control Settings
Pilot Model - ON Roll SAS – ON
Analog SAS – ON Pitch SAS – ON
Digital SAS – ON Yaw SAS – ON
ECU – ON Automatic Stabilator Control – ON
Controller: UH-60 Mech (All other settings OFF)
Other Settings
Inflow Model: Pitt-Peters
Blade Segments: 30
Output Decimation: 1
Zero Sideslip Speed [kts]: 60
Ground Effect: Turned On
Time Step [s], [deg]: 0.01, 15.4699
Turbulence: Airwake
Wind Speed, Direction: Set by Airwake Selections
Appendix B
Offline Stochastic Filter Fitting Results
B.1 Windowing Comparison
The PSDs of the equivalent gusts were found using Welch’s Method and three
different types of windowing techniques to determine which windowing technique would
provide the best filter fit. The analysis used the UH-60A hovering in the DDG81 with a
WOD condition of 25kts / 0deg at the hover location described in Appendix A,
Table A.1.
100
101
10-2
10-1
100
101
102
103
104
Windowing Comparision for DDG81 25kt / 0 deg Hover
Freq (rad/sec)
u -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Hamming
Hann
Blackman
Figure B.1: u - Velocity Component for the Windowing Comparison
90
100
101
10-2
10-1
100
101
102
103
104
Windowing Comparision for DDG81 25kt / 0 deg Hover
Freq (rad/sec)
v -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Hamming
Hann
Blackman
Figure B.2: v - Velocity Component for the Windowing Comparison
100
101
10-2
10-1
100
101
102
103
Windowing Comparision for DDG81 25kt / 0 deg Hover
Freq (rad/sec)
w -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Hamming
Hann
Blackman
Figure B.3: w - Velocity Component for the Windowing Comparison
91
100
101
10-6
10-4
10-2
100
102
Windowing Comparision for DDG81 25kt / 0 deg Hover
Freq (rad/sec)
p -
Roll
Com
ponent [(
rad/s
)2]
Equivalent Gusts
Hamming
Hann
Blackman
Figure B.4: p - Rotational Component for the Windowing Comparison
100
101
10-5
10-4
10-3
10-2
10-1
100
101
Windowing Comparision for DDG81 25kt / 0 deg Hover
Freq (rad/sec)
q -
Pitch C
om
ponent [(
rad/s
)2
Equivalent Gusts
Hamming
Hann
Blackman
Figure B.5: q - Velocity Component for the Windowing Comparison
92
B.2 Filter Fit Comparison
The PSDs of the equivalent gusts were found using Welch’s Method with
Hamming windowing. Two types of filter designs were tested. The analysis used the UH-
60A hovering in the DDG81 with a WOD condition of 25kts / 0deg at the hover location
described in Appendix A, Table A.1.
100
101
10-5
10-4
10-3
10-2
10-1
100
101
Windowing Comparision for DDG81 25kt / 0 deg Hover
Freq (rad/sec)
r -
Yaw
Com
ponent [(
rad/s
)2]
Equivalent Gusts
Hamming
Hann
Blackman
Figure B.6: r - Velocity Component for the Windowing Comparison
93
100
101
10-2
10-1
100
101
102
103
104
Freq (rad/sec)
u -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Old Filter Design
New Filter Design
Figure B.7: u - Velocity Component for the Filter Design Comparison
100
101
10-2
10-1
100
101
102
103
104
Freq (rad/sec)
v -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Old Filter Design
New Filter Design
Figure B.8: v - Velocity Component for the Filter Design Comparison
94
100
101
10-2
10-1
100
101
102
103
Freq (rad/sec)
w -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Old Filter Design
New Filter Design
Figure B.9: w - Velocity Component for the Filter Design Comparison
100
101
10-4
10-3
10-2
10-1
100
101
Freq (rad/sec)
p -
Roll
Com
ponent [(
rad/s
)2]
Equivalent Gusts
Old Filter Design
New Filter Design
Figure B.10: p - Roll Component for the Filter Design Comparison
95
100
101
10-5
10-4
10-3
10-2
10-1
100
101
Freq (rad/sec)
q -
Pitch C
om
ponent [(
rad/s
)2]
Equivalent Gusts
Old Filter Design
New Filter Design
Figure B.11: q - Pitch Component for the Filter Design Comparison
100
101
10-5
10-4
10-3
10-2
10-1
100
101
Freq (rad/sec)
r -
Yaw
Com
ponent [(
rad/s
)2]
Equivalent Gusts
Old Filter Design
New Filter Design
Figure B.12: r - Yaw Component for the Filter Design Comparison
96
B.3 LHA Filter Fits: 30 knots / 0 deg
The PSDs of the equivalent gusts were found using Welch’s Method with
Hamming windowing. The analysis used the UH-60A hovering in the LHA airwake
above landing spot 8 with a WOD condition of 30 kts / 0 deg at the hover location
described in Appendix A.
100
101
10-2
10-1
100
101
102
103
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Freq (rad/sec)
u -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.13: u - Velocity Component for LHA 30 kts / 0 deg
97
100
101
10-3
10-2
10-1
100
101
102
103
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Freq (rad/sec)
v -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.14: v - Velocity Component for LHA 30 kts / 0 deg
100
101
10-3
10-2
10-1
100
101
102
103
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Freq (rad/sec)
w -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.15: w - Velocity Component for LHA 30 kts / 0 deg
98
100
101
10-5
10-4
10-3
10-2
10-1
100
101
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Freq (rad/sec)
p -
Roll
Com
ponent [(
rad/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.16: p - Roll Component for LHA 30 kts / 0 deg
100
101
10-5
10-4
10-3
10-2
10-1
100
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Freq (rad/sec)
q -
Pitch C
om
ponent [(
rad/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.17: q - Pitch Component for LHA 30 kts / 0 deg
99
B.4 LHA Filter Fits: 30 knots / 30 deg
The PSDs of the equivalent gusts were found using Welch’s Method with
Hamming windowing. The analysis used the UH-60A hovering in the LHA airwake
above landing spot 8 with a WOD condition of 30 kts / 30 deg at the hover location
described in Appendix A.
100
101
10-6
10-4
10-2
100
102
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Freq (rad/sec)
r -
Yaw
Com
ponent [(
rad/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.18: r – Yaw Component for LHA 30 kts / 0 deg
100
100
101
10-1
100
101
102
103
104
105
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Freq (rad/sec)
u -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.19: u - Velocity Component for LHA 30 kts / 30 deg
100
101
10-1
100
101
102
103
104
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Freq (rad/sec)
v -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.20: v - Velocity Component for LHA 30 kts / 30 deg
101
100
101
10-1
100
101
102
103
104
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Freq (rad/sec)
w -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.21: w - Velocity Component for LHA 30 kts / 30 deg
100
101
10-4
10-3
10-2
10-1
100
101
102
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Freq (rad/sec)
p -
Roll
Com
ponent [(
rad/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.22: p - Roll Component for LHA 30 kts / 30 deg
102
100
101
10-4
10-3
10-2
10-1
100
101
102
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Freq (rad/sec)
q -
Pitch C
om
ponent [(
rad/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.23: q - Pitch Component for LHA 30 kts / 30 deg
100
101
10-3
10-2
10-1
100
101
102
103
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Freq (rad/sec)
r -
Yaw
Com
ponent [(
rad/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.24: r - Yaw Component for LHA 30 kts / 30 deg
103
B.5 DDG-81 Filter Fits: 25 knots / 0 deg
The PSDs of the equivalent gusts were found using Welch’s Method with
Hamming windowing. The analysis used the UH-60A hovering in the DDG-81 airwake
with a WOD condition of 25 kts / 0 deg at the hover location described in Appendix A.
100
101
10-2
10-1
100
101
102
103
104
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Freq (rad/sec)
u -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.25: u - Velocity Component for DDG-81 25 kts / 0 deg
104
100
101
10-2
10-1
100
101
102
103
104
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Freq (rad/sec)
v -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.26: v - Velocity Component for DDG-81 25 kts / 0 deg
100
101
10-2
10-1
100
101
102
103
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Freq (rad/sec)
w -
Velo
city C
om
ponent [(
ft/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.27: w - Velocity Component for DDG-81 25 kts / 0 deg
105
100
101
10-4
10-3
10-2
10-1
100
101
102
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Freq (rad/sec)
p -
Roll
Com
ponent [(
rad/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.28: p - Roll Component for DDG-81 25 kts / 0 deg
100
101
10-5
10-4
10-3
10-2
10-1
100
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Freq (rad/sec)
q -
Pitch C
om
ponent [(
rad/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.29: q - Pitch Component for DDG-81 25 kts / 0 deg
106
B.6 DDG-81 Initial Simulation Results
The analysis used the UH-60A hovering in the DDG-81 airwake with a WOD
condition of 25kts / 0deg at the hover location described in Appendix A. The dark blue
line is the results from the Tustin discrete filter. The green line depicts the disturbance
from the continuous filter that was excited in the SIMULINK environment. The red line
depicts the equivalent gusts determined by removing the aircraft’s dynamics from the
simulation results. Finally, the light blue line is the stochastic portion of the CFD gust
that acted on the center of the fuselage during simulation.
100
101
10-5
10-4
10-3
10-2
10-1
100
101
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Freq (rad/sec)
r -
Yaw
Com
ponent [(
rad/s
)2]
Equivalent Gusts
Welch / Hamming Filter
Figure B.30: r - Yaw Component for DDG-81 25 kts / 0 deg
107
0 20 40 60 80 100 120-15
-10
-5
0
5
10
15DDG-81 Hover Location: WOD Case: 25kts / 0deg
time [s]
u -
Velo
city C
om
ponent [(f
t/s)]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.31: u - Velocity Component for DDG-81 Initial Simulation Results
0 20 40 60 80 100 120-15
-10
-5
0
5
10
15DDG-81 Hover Location: WOD Case: 25kts / 0deg
time [s]
v -
Velo
city C
om
ponent [(f
t/s)]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.32: v - Velocity Component for DDG-81 Initial Simulation Results
108
0 20 40 60 80 100 120-15
-10
-5
0
5
10
time [s]
w -
Velo
city C
om
ponent [(f
t/s)]
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.33: w - Velocity Component for DDG-81 Initial Simulation Results
0 20 40 60 80 100 120-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
time [s]
p -
Velo
city C
om
ponent [r
ad/s
]
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Tustin
Simulink
Equivalent Gust
Figure B.34: p – Roll Component for DDG-81 Initial Simulation Results
109
0 20 40 60 80 100 120-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
time [s]
q -
Velo
city C
om
ponent [r
ad/s
]
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Tustin
Simulink
Equivalent Gust
Figure B.35: q - Pitch Component for DDG-81 Initial Simulation Results
0 20 40 60 80 100 120-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
time [s]
r -
Velo
city C
om
ponent [r
ad/s
]
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Tustin
Simulink
Equivalent Gust
Figure B.36: r - Yaw Component for DDG-81 Initial Simulation Results
110
100
101
10-4
10-2
100
102
104
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Freq (rad/sec)
u -
Velo
city C
om
ponent [(f
t/s)2
]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.37: u - Velocity PSD of the DDG-81 Initial Simulation Results
100
101
10-4
10-2
100
102
104
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Freq (rad/sec)
v -
Velo
city C
om
ponent [(
ft/s
)2]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.38: v - Velocity PSD of the DDG-81 Initial Simulation Results
111
B.7 LHA 30 kts / 0 deg Initial Simulation Results
The analysis used the UH-60A hovering in the LHA airwake with a WOD
condition of 30 kts / 0 deg at the hover location described in Appendix A. The dark blue
line is the results from the Tustin discrete filter. The green line depicts the disturbance
from the continuous filter that was excited in the SIMULINK environment. The red line
depicts the equivalent gusts determined by removing the aircraft’s dynamics from the
simulation results. Finally, the light blue line is the stochastic portion of the CFD gust
that acted on the center of the fuselage during simulation.
100
101
10-4
10-2
100
102
104
DDG-81 Hover Location: WOD Case: 25kts / 0deg
Freq (rad/sec)
w -
Velo
city C
om
ponent [(f
t/s)2
]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.39: w - Velocity PSD of the DDG-81 Initial Simulation Results
112
0 20 40 60 80 100 120-4
-3
-2
-1
0
1
2
3
4
5LHA Landing Spot 8: WOD Case: 30kts / 0deg
u -
Velo
city C
om
ponent [ft/s]
time [s]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.40: u - Velocity Component for LHA 30 kts / 0 deg Initial Simulation Results
0 20 40 60 80 100 120-4
-3
-2
-1
0
1
2
3LHA Landing Spot 8: WOD Case: 30kts / 0deg
v -
Velo
city C
om
ponent [ft/s]
time [s]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.41: v - Velocity Component for LHA 30 kts / 0 deg Initial Simulation Results
113
0 20 40 60 80 100 120-4
-3
-2
-1
0
1
2
3
time [s]
w -
Velo
city C
om
ponent [ft/s]
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.42: w - Velocity Component for LHA 30 kts / 0 deg Initial Simulation Results
0 20 40 60 80 100 120-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
time [s]
p -
Velo
city C
om
ponent [r
ad/s
]
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Tustin
Simulink
Equivalent Gust
Figure B.43: p - Roll Component for LHA 30 kts / 0 deg Initial Simulation Results
114
0 20 40 60 80 100 120-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
time [s]
q -
Velo
city C
om
ponent [r
ad/s
]
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Tustin
Simulink
Equivalent Gust
Figure B.44: q - Pitch Component for LHA 30 kts / 0 deg Initial Simulation Results
0 20 40 60 80 100 120-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
time [s]
r -
Velo
city C
om
ponent [r
ad/s
]
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Tustin
Simulink
Equivalent Gust
Figure B.45: r - Yaw Component for LHA 30 kts / 0 deg Initial Simulation Results
115
100
101
10-3
10-2
10-1
100
101
102
103
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Freq (rad/sec)
u -
Velo
city C
om
ponent [(f
t/s)2
]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.46: u - Velocity PSD of the LHA 30 kts / 0 deg Initial Simulation Results
100
101
10-4
10-2
100
102
104
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Freq (rad/sec)
v -
Velo
city C
om
ponent [(
ft/s
)2]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.47: v - Velocity PSD of the LHA 30 kts / 0 deg Initial Simulation Results
116
B.8 LHA 30 kts / 30 deg Initial Simulation Results
The analysis used the UH-60A hovering in the LHA airwake with a WOD
condition of 30 kts / 30 deg at the hover location described in Appendix A. The dark blue
line is the results from the Tustin discrete filter. The green line depicts the disturbance
from the continuous filter that was excited in the SIMULINK environment. The red line
depicts the equivalent gusts determined by removing the aircraft’s dynamics from the
simulation results. Finally, the light blue line is the stochastic portion of the CFD gust
that acted on the center of the fuselage during simulation.
100
101
10-3
10-2
10-1
100
101
102
103
LHA Landing Spot 8: WOD Case: 30kts / 0deg
Freq (rad/sec)
w -
Velo
city C
om
ponent [(f
t/s)2
]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.48: w - Velocity PSD of the LHA 30 kts / 0 deg Initial Simulation Results
117
0 20 40 60 80 100 120-25
-20
-15
-10
-5
0
5
10
15
20
25LHA Landing Spot 8: WOD Case: 30kts / 30deg
u -
Velo
city C
om
ponent [ft/s]
time [s]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.49: u - Velocity Component for LHA 30 kts / 30 deg Initial Simulation Results
0 20 40 60 80 100 120-30
-20
-10
0
10
20
30LHA Landing Spot 8: WOD Case: 30kts / 30deg
v -
Velo
city C
om
ponent [ft/s]
time [s]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.50: v - Velocity Component for LHA 30 kts / 30 deg Initial Simulation Results
118
0 20 40 60 80 100 120-30
-25
-20
-15
-10
-5
0
5
10
15
20
time [s]
w -
Velo
city C
om
ponent [ft/s]
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.51: w - Velocity Component for LHA 30 kts / 30 deg Initial Simulation Results
0 20 40 60 80 100 120-1.5
-1
-0.5
0
0.5
1
1.5
time [s]
p -
Velo
city C
om
ponent [r
ad/s
]
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Tustin
Simulink
Equivalent Gust
Figure B.52: p - Roll Component for LHA 30 kts / 30 deg Initial Simulation Results
119
0 20 40 60 80 100 120-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
time [s]
q -
Velo
city C
om
ponent [r
ad/s
]
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Tustin
Simulink
Equivalent Gust
Figure B.53: q - Pitch Component for LHA 30 kts / 30 deg Initial Simulation Results
0 20 40 60 80 100 120-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
time [s]
r -
Velo
city C
om
ponent [r
ad/s
]
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Tustin
Simulink
Equivalent Gust
Figure B.54: r - Yaw Component for LHA 30 kts / 30 deg Initial Simulation Results
120
100
101
10-2
100
102
104
106
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Freq (rad/sec)
u -
Velo
city C
om
ponent [(f
t/s)2
]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.55: u - Velocity PSD of the LHA 30 kts / 30 deg Initial Simulation Results
100
101
10-2
100
102
104
106
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Freq (rad/sec)
v -
Velo
city C
om
ponent [(
ft/s
)2]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.56: v - Velocity PSD of the LHA 30 kts / 30 deg Initial Simulation Results
121
100
101
10-2
10-1
100
101
102
103
104
LHA Landing Spot 8: WOD Case: 30kts / 30deg
Freq (rad/sec)
w -
Velo
city C
om
ponent [(f
t/s)2
]
Tustin
Simulink
Equivalent Gust
CFD Gust
Figure B.57: w - Velocity PSD of the LHA 30 kts / 30 deg Initial Simulation Results
Appendix C
Online AR Filter Results
Appendix C outlines the results found using online filter fitting methods to
estimate the airwake properties.
C.1 Online Burg AR Model Fitting Results
This section outlines the results using an online Burg AR Model to fit gust filters
to the UH-60A hovering in the DDG-81 airwake at the standard hover location outlined
in Appendix A.
100
101
-10
0
10
20
30
40
50
60
ug (
dB
)
DDG81 25 kts 0 deg / AR Burg Method - 6th Order
Frequency (rad/sec)
Off-line
20 sec
40 sec
60 sec
80 sec
100 sec
120 sec
Figure C.1: Identified Longitudinal Gust Velocity Gust Filters
100
101
-20
-10
0
10
20
30
40
vg (
dB
)
DDG81 25 kts 0 deg / AR Burg Method - 6th Order
Frequency (rad/sec)
Off-line
20 sec
40 sec
60 sec
80 sec
100 sec
120 sec
Figure C.2: Identified Lateral Gust Velocity Gust Filters
100
101
-30
-20
-10
0
10
20
30
40
50
wg (
dB
)
DDG81 25 kts 0 deg / AR Burg Method - 6th Order
Frequency (rad/sec)
Off-line
20 sec
40 sec
60 sec
80 sec
100 sec
120 sec
Figure C.3: Identified Vertical Gust Velocity Gust Filters
124
100
101
-30
-25
-20
-15
-10
-5
0
5
10
15
20
pg (
dB
)
DDG81 25 kts 0 deg / AR Burg Method - 6th Order
Frequency (rad/sec)
Off-line
20 sec
40 sec
60 sec
80 sec
100 sec
120 sec
Figure C.4: Identified Roll Rate Gust Filters
100
101
-35
-30
-25
-20
-15
-10
-5
0
5
10
qg (
dB
)
DDG81 25 kts 0 deg / AR Burg Method - 6th Order
Frequency (rad/sec)
Off-line
20 sec
40 sec
60 sec
80 sec
100 sec
120 sec
Figure C.5: Identified Pitch Rate Gust Filters
125
100
101
-40
-30
-20
-10
0
10
20
r g (
dB
)
DDG81 25 kts 0 deg / AR Burg Method - 6th Order
Frequency (rad/sec)
Off-line
20 sec
40 sec
60 sec
80 sec
100 sec
120 sec
Figure C.6: Identified Yaw Rate Gust Filters
Appendix D
Stochastic Simulation Modeling Results
This appendix outlines the results of the simulation modeling results from the
stochastic airwake that was implemented in Genhel-PSU and Sikorsky’s Genhel.
D.1 Genhel-PSU Simulation Hovering Results
This section presents the results from performing a simulated hover at four spots
within the stochastic airwake model. The results are compared to results of the same
maneuver performed the DDG-81 CFD airwake that the stochastic filters were derived
from. The case names in each figure explain the X-Y-Z location in the airwake
coordinate system that the hover was performed at. For example, 454-31-33 relates to
454 feet behind the origin of the ship, 31 feet to the starboard side of the ship’s center
line, and 33 feet above the bottom of the ship. Each section presents the PSD of pilot
stick inputs to maintain the hover as well as PSD of the winds acting on the center of the
aircraft’s fuselage.
127
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 439-0-42
Freq (rad/sec)
Late
ral S
tick, X
a [(%
)2]
Stochastic
CFD
Figure D.1: Lateral Sick Input at 439-0-42
100
101
10-5
10-4
10-3
10-2
10-1
100
101
DDG-81 WOD Case: 25kts / 0deg at 439-0-42
Freq (rad/sec)
Longitudin
al S
tick, X
b [(%
)2]
Stochastic
CFD
Figure D.2: Longitudinal Sick Input at 439-0-42
128
100
101
10-5
10-4
10-3
10-2
10-1
100
101
DDG-81 WOD Case: 25kts / 0deg at 439-0-42
Freq (rad/sec)
Colle
ctive S
tick, X
c [(
%)2
]
Stochastic
CFD
Figure D.3: Collective Stick Input at 439-0-42
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 439-0-42
Freq (rad/sec)
Pedals
, X
p [(
%)2
]
Stochastic
CFD
Figure D.4: Pedal Input at 439-0-42
129
100
101
10-4
10-2
100
102
104
DDG-81 WOD Case: 25kts / 0deg at 439-0-42
Freq (rad/sec)
U -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure D.5: Longitudinal Velocity acting on Fuselage at 439-0-42
100
101
10-2
10-1
100
101
102
103
104
DDG-81 WOD Case: 25kts / 0deg at 439-0-42
Freq (rad/sec)
V -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure D.6: Lateral Velocity acting on Fuselage at 439-0-42
130
100
101
10-2
10-1
100
101
102
103
DDG-81 WOD Case: 25kts / 0deg at 439-0-42
Freq (rad/sec)
W -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure D.7: Vertical Velocity acting on Fuselage at 439-0-42
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 454-0-52
Freq (rad/sec)
Late
ral S
tick, X
a [(%
)2]
Stochastic
CFD
Figure D.8: Lateral Sick Input at 454-0-52
131
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 454-0-52
Freq (rad/sec)
Longitudin
al S
tick, X
b [(%
)2]
Stochastic
CFD
Figure D.9: Longitudinal Sick Input at 454-0-52
100
101
10-5
10-4
10-3
10-2
10-1
100
101
DDG-81 WOD Case: 25kts / 0deg at 454-0-52
Freq (rad/sec)
Colle
ctive S
tick, X
c [(
%)2
]
Stochastic
CFD
Figure D.10: Collective Stick Input at 454-0-52
132
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 454-0-52
Freq (rad/sec)
Pedals
, X
p [(
%)2
]
Stochastic
CFD
Figure D.11: Pedal Input at 454-0-52
100
101
10-4
10-2
100
102
104
DDG-81 WOD Case: 25kts / 0deg at 454-0-52
Freq (rad/sec)
U -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure D.12: Longitudinal Velocity acting on Fuselage at 454-0-52
133
100
101
10-2
10-1
100
101
102
103
104
DDG-81 WOD Case: 25kts / 0deg at 454-0-52
Freq (rad/sec)
V -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure D.13: Lateral Velocity acting on Fuselage at 454-0-52
100
101
10-4
10-2
100
102
104
DDG-81 WOD Case: 25kts / 0deg at 454-0-52
Freq (rad/sec)
W -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure D.14: Vertical Velocity acting on Fuselage at 454-0-52
134
100
101
10-8
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 454-31-33
Freq (rad/sec)
Late
ral S
tick, X
a [(%
)2]
Stochastic
CFD
Figure D.15: Lateral Sick Input at 545-31-33
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 454-31-33
Freq (rad/sec)
Longitudin
al S
tick, X
b [(%
)2]
Stochastic
CFD
Figure D.16: Longitudinal Sick Input at 545-31-33
135
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 454-31-33
Freq (rad/sec)
Colle
ctive S
tick, X
c [(
%)2
]
Stochastic
CFD
Figure D.17: Collective Stick Input at 545-31-33
100
101
10-8
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 454-31-33
Freq (rad/sec)
Pedals
, X
p [(
%)2
]
Stochastic
CFD
Figure D.18: Pedal Input at 545-31-33
136
100
101
10-3
10-2
10-1
100
101
102
103
DDG-81 WOD Case: 25kts / 0deg at 454-31-33
Freq (rad/sec)
U -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure D.19: Longitudinal Velocity acting on Fuselage at 545-31-33
100
101
10-3
10-2
10-1
100
101
102
103
DDG-81 WOD Case: 25kts / 0deg at 454-31-34
Freq (rad/sec)
V -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure D.20: Lateral Velocity acting on Fuselage at 545-31-33
137
100
101
10-3
10-2
10-1
100
101
102
103
DDG-81 WOD Case: 25kts / 0deg at 454-31-33
Freq (rad/sec)
W -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure D.21: Vertical Velocity acting on Fuselage at 545-31-33
100
101
10-8
10-6
10-4
10-2
100
DDG-81 WOD Case: 25kts / 0deg at 544-37-58
Freq (rad/sec)
Late
ral S
tick, X
a [(%
)2]
Stochastic
CFD
Figure D.22: Lateral Sick Input at 454-37-58
138
100
101
10-8
10-6
10-4
10-2
100
DDG-81 WOD Case: 25kts / 0deg at 544-37-58
Freq (rad/sec)
Longitudin
al S
tick, X
b [(%
)2]
Stochastic
CFD
Figure D.23: Longitudinal Sick Input at 454-37-58
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 544-37-58
Freq (rad/sec)
Colle
ctive S
tick, X
c [(
%)2
]
Stochastic
CFD
Figure D.24: Collective Stick Input at 454-37-58
139
100
101
10-8
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg at 544-37-58
Freq (rad/sec)
Pedals
, X
p [(
%)2
]
Stochastic
CFD
Figure D.25: Pedal Input at 454-37-58
100
101
10-4
10-2
100
102
104
DDG-81 WOD Case: 25kts / 0deg at 544-37-58
Freq (rad/sec)
U -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure D.26: Longitudinal Velocity acting on Fuselage at 454-37-58
140
100
101
10-3
10-2
10-1
100
101
102
103
DDG-81 WOD Case: 25kts / 0deg at 544-37-58
Freq (rad/sec)
V -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure D.27: Lateral Velocity acting on Fuselage at 454-37-58
100
101
10-4
10-2
100
102
104
DDG-81 WOD Case: 25kts / 0deg at 544-37-58
Freq (rad/sec)
W -
Velo
city a
t F
usela
ge [(f
t/sec)2
]
Stochastic
CFD
Figure D.28: Vertical Velocity acting on Fuselage at 454-37-58
141
D.2 Genhel-PSU Simulation Trajectory Maneuver Results
This section presents the results from performing a simulation in which the
helicopter was prescribed a desired trajectory. The pilot model, described in Section 2.1,
emulated a pilot flying the desired flight path. This section presents the trajectory path
and PSDs of stick inputs for two separate flight paths. The first hover started aft and to
the starboard side of the desired landing spot. The aircraft then flew forward and ended in
a final hover near the desired landing spot. The second trajectory path was similar, but
started on the port side of the ship and at a different height. The flight path ended near the
desired landing spot. The aircraft positions are offset in the x-axis by 225 ft in the x-axis
and 3.4 feet in the z-axis due to the coordinate differences between the airwake origin and
Genhel-PSU ship origin.
142
-1000
100-600-400-200
-50
0
50
100
150
North-South Position (xN) [ft]
East-West Position (yE) [ft]
Altitud
e (
h)
[ft]
Flight Path
Trajectory Path
Figure D.29: Three Dimensional View of Trajectory Path 1
143
0 20 40 60 80 100 120-1000
-500
0
XN [ft]
0 20 40 60 80 100 120-100
0
100
YE [ft]
0 20 40 60 80 100 1200
50
100
h [ft]
Time [s]
Figure D.30: Aircraft Position Time History of Trajectory Path 1
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg
Freq (rad/sec)
Late
ral S
tick, X
a [(%
)2]
Stochastic
CFD
No Airwake
Figure D.31: Lateral Stick Input for Trajectory Path 1
144
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg
Freq (rad/sec)
Longitudin
al S
tick, X
b [(%
)2]
Stochastic
CFD
No Airwake
Figure D.32: Longitudinal Stick Input for Trajectory Path 1
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg
Freq (rad/sec)
Colle
ctive S
tick, X
c [(
%)2
]
Stochastic
CFD
No Airwake
Figure D.33: Collective Stick Input for Trajectory Path 1
145
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg
Freq (rad/sec)
Pedals
, X
p [(
%)2
]
Stochastic
CFD
No Airwake
Figure D.34: Pedal input for Trajectory Path 1
-100-50
050
100
-600
-400
-200
-50
0
50
100
150
(YE) [ft]
(XN) [ft]
Altitud
e (
h)
[ft]
Flight Path
Trajectory Path
Figure D.35: Three Dimensional View of Trajectory Path 2
146
0 20 40 60 80 100 120-1000
-500
0X
N [ft]
0 20 40 60 80 100 120-50
0
50
YE [ft]
0 20 40 60 80 100 1200
50
100
h [ft]
Time [s]
Figure D.36: Aircraft Position Time History of Trajectory Path 2
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg
Freq (rad/sec)
Late
ral S
tick, X
a [(%
)2]
Stochastic
CFD
No Airwake
Figure D.37: Lateral Stick Input for Trajectory Path 2
147
100
101
10-4
10-3
10-2
10-1
100
101
102
DDG-81 WOD Case: 25kts / 0deg
Freq (rad/sec)
Longitudin
al S
tick, X
b [(%
)2]
Stochastic
CFD
No Airwake
Figure D.38: Longitudinal Stick Input for Trajectory Path 2
100
101
10-4
10-2
100
102
104
DDG-81 WOD Case: 25kts / 0deg
Freq (rad/sec)
Colle
ctive S
tick, X
c [(
%)2
]
Stochastic
CFD
No Airwake
Figure D.39: Collective Stick Input for Trajectory Path 2
148
D.3 Sikorsky GENHEL Piloted Simulation Results
This section presents the PSDs of piloted inputs for each of the six pilots for
piloted simulations performed with Sikorsky GENHEL model. The stochastic airwake
model is compared to the CFD airwake model for the DDG-81 with a WOD condition of
25 knots / 0 deg.
100
101
10-6
10-4
10-2
100
102
DDG-81 WOD Case: 25kts / 0deg
Freq (rad/sec)
Pedals
, X
p [(
%)2
]
Stochastic
CFD
No Airwake
Figure D.40: Pedal input for Trajectory Path 2
149
100
101
10-8
10-6
10-4
10-2
100
Frequency (rad/s)
PS
D o
f Late
ral S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.41: Lateral Stick Input for Pilot 1
100
101
10-8
10-6
10-4
10-2
100
102
Frequency (rad/s)
PS
D o
f Long. S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.42: Longitudinal Stick Input for Pilot 1
150
100
101
10-6
10-4
10-2
100
102
104
Frequency (rad/s)
PS
D o
f C
oll.
Stick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.43: Collective Stick Input for Pilot 1
100
101
10-8
10-6
10-4
10-2
100
102
Frequency (rad/s)
PS
D o
f P
edals
, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.44: Pedal Input for Pilot 1
151
100
101
10-8
10-6
10-4
10-2
100
Frequency (rad/s)
PS
D o
f Late
ral S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.45: Lateral Stick Input for Pilot 2
100
101
10-6
10-5
10-4
10-3
10-2
10-1
100
Frequency (rad/s)
PS
D o
f Long. S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.46: Longitudinal Stick Input for Pilot 2
152
100
101
10-3
10-2
10-1
100
101
102
103
Frequency (rad/s)
PS
D o
f C
oll.
Stick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.47: Collective Stick Input for Pilot 2
100
101
10-5
10-4
10-3
10-2
10-1
100
101
Frequency (rad/s)
PS
D o
f P
edals
, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.48: Pedal Input for Pilot 2
153
100
101
10-6
10-5
10-4
10-3
10-2
10-1
100
Frequency (rad/s)
PS
D o
f Late
ral S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.49: Lateral Stick Input for Pilot 3
100
101
10-6
10-4
10-2
100
102
Frequency (rad/s)
PS
D o
f Long. S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.50: Longitudinal Stick Input for Pilot 3
154
100
101
10-6
10-4
10-2
100
102
104
Frequency (rad/s)
PS
D o
f C
oll.
Stick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.51: Collective Stick Input for Pilot 3
100
101
10-6
10-4
10-2
100
102
Frequency (rad/s)
PS
D o
f P
edals
, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.52: Pedal Input for Pilot 3
155
100
101
10-8
10-6
10-4
10-2
100
Frequency (rad/s)
PS
D o
f Late
ral S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.53: Lateral Stick Input for Pilot 4
100
101
10-6
10-4
10-2
100
102
Frequency (rad/s)
PS
D o
f Long. S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.54: Longitudinal Stick Input for Pilot 4
156
100
101
10-4
10-2
100
102
104
Frequency (rad/s)
PS
D o
f C
oll.
Stick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.55: Collective Stick Input for Pilot 4
100
101
10-6
10-4
10-2
100
102
Frequency (rad/s)
PS
D o
f P
edals
, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.56: Pedal Input for Pilot 4
157
100
101
10-5
10-4
10-3
10-2
10-1
100
Frequency (rad/s)
PS
D o
f Late
ral S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.57: Lateral Stick Input for Pilot 5
100
101
10-6
10-4
10-2
100
102
Frequency (rad/s)
PS
D o
f Long. S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.58: Longitudinal Stick Input for Pilot 5
158
100
101
10-4
10-2
100
102
104
Frequency (rad/s)
PS
D o
f C
oll.
Stick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.59: Collective Stick Input for Pilot 5
100
101
10-8
10-6
10-4
10-2
100
102
Frequency (rad/s)
PS
D o
f P
edals
, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.60: Pedal Input for Pilot 5
159
100
101
10-6
10-4
10-2
100
102
Frequency (rad/s)
PS
D o
f Late
ral S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.61: Lateral Stick Input for Pilot 6
100
101
10-5
10-4
10-3
10-2
10-1
100
101
Frequency (rad/s)
PS
D o
f Long. S
tick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.62: Longitudinal Stick Input for Pilot 6
160
100
101
10-4
10-2
100
102
104
Frequency (rad/s)
PS
D o
f C
oll.
Stick, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.63: Collective Stick Input for Pilot 6
100
101
10-6
10-4
10-2
100
102
Frequency (rad/s)
PS
D o
f P
edals
, [(
%)2
]
No Airwake
CFD Airwake
Stochastic Airwake
Figure D.64: Pedal Input for Pilot 6