Delft University of Technology

Novel Methods in Helicopter Performance Flight Testing

Arush, I.

DOI
10.4233/uuid:3ffd0639-a889-4882-8537-2f81a6671e8e
Publication date
2023
Document Version
Final published version
Citation (APA)
Arush, I. (2023). Novel Methods in Helicopter Performance Flight Testing. [Dissertation (TU Delft), Delft
University of Technology]. https://doi.org/10.4233/uuid:3ffd0639-a889-4882-8537-2f81a6671e8e

Important note
To cite this publication, please use the final published version (if applicable).
Please check the document version above.

Copyright
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent
of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Takedown policy
Please contact us and provide details if you believe this document breaches copyrights.
We will remove access to the work immediately and investigate your claim.

This work is downloaded from Delft University of Technology.

For technical reasons the number of authors shown on this cover page is limited to a maximum of 10.
 N OVEL M ETHODS IN H ELICOPTER
P ERFORMANCE F LIGHT -T ESTING

Dissertation

for the purpose of obtaining the degree of doctor

at Delft University of Technology
by the authority of the Rector Magnificus prof.dr.ir. T.H.J.J. van der Hagen;
Chair of the Board for Doctorates
to be defended publicly on
Wednesday 8 November 2023 at
12:30 o’clock

by

Ilan ARUSH

Master of Science in Industrial and Management Engineering,

Ben-Gurion University, Israel
born in Rehovot, Israel
This dissertation has been approved by the promotors.

Composition of the doctoral committee:

Rector Magnificus, chairperson
Prof.dr.ir. M. Mulder Delft University of Technology, promotor
Dr. M.D. Pavel Delft University of Technology, promotor

Independent members:
Prof.dr.ir. G. Jongbloed Delft University of Technology
Prof.dr. I. Yavrucuk München University of Technology, Germany
Prof.dr. M. Gennaretti Roma Tre University, Italy
Dr. L. Ingham Sikorsky, a Lockheed Martin company, USA
Prof.dr. A. Gangoli Rao Delft University of Technology, reserve member

The work described in this thesis has been carried out at the Control and Simulation
section at Delft University of Technology and at the National Test Pilot School
located in Mojave, California.

Keywords: Multivariable Polynomial, Singular Value

Decomposition, Corrected Variables, Non-
Dimensional Variables, Helicopter Performance,
Flight Testing, Optimization under Constraints
Cover design: Ilan Arush. Photos courtesy of the National Test
Pilot School, Mojave California
Printed by: Ipskamp Printing, the Netherlands

Copyright © 2023 by I. Arush

ISBN: 978-94-6473-231-3

An electronic version of this dissertation is available at

http://repository .tudelft.nl/.
 To my beloved family
Shoshi, Ofek and Maya
CONTENTS

1 Introduction 1
1.1 Background and relevance .......................................................................... 1
1.2 Helicopter Performance .............................................................................. 3
1.2.1 Available Power ................................................................................. 4
1.2.2 Power Required.................................................................................. 7
1.2.2.1 Hover Performance ............................................................. 8
1.2.2.2 Level Flight Performance .................................................... 9
1.3 Conventional Methods for Performance Flight Testing ......................11
1.3.1 Available Power Flight-Test Method ............................................12
1.3.2 Hover Performance Flight Testing ...............................................13
1.3.3 Level Flight Performance Flight Testing .....................................14
1.4 Problem Statement .....................................................................................15
1.4.1 Available Power ...............................................................................16
1.4.2 Power Required for OGE Hover..................................................17
1.4.3 Power Required for Level Flight ...................................................18
1.5 Research Goals and Objectives ................................................................20
1.6 Research Limitations and Scope ..............................................................21
1.7 Methods Accuracy Comparison ...............................................................23
1.8 Thesis Outline .............................................................................................24
1.9 Thesis Publications .....................................................................................27

2 Helicopter Performance Theory & Conventional Testing Methods 29

2.1 Hover Performance....................................................................................29
2.2 Level Flight Performance ..........................................................................44
2.2.1 The Induced Power in Level Flight ..............................................47
2.2.2 The Profile Power in Level Flight .................................................52
2.2.3 The Parasite Power in Level Flight ...............................................57
2.3 Conventional Methods for Performance Flight Testing ......................61
2.3.1 Available Power Flight-Test Method ............................................62
2.3.1.1 Phase I – The Engine ‘Rules of Operation’ ...................63

v
 2.3.1.2 Phase II – Maximum Available Power ........................... 65
2.3.2 Hover Performance Flight Testing............................................... 67
2.3.2.1 Non-Dimensional Hover Performance .......................... 68
2.3.2.2 Un-Referring to Conditions of Choice ........................... 72
2.3.2.3 Extremum Hover Performance ....................................... 73
2.3.3 Level Flight Performance Flight Testing ..................................... 75
2.3.3.1 Non-Dimensional Level-Flight Performance ................ 76
2.3.3.2 Constant Weight over Sigma (W/σ) Method ................ 77
2.3.3.3 Constant Weight over Delta (W/δ) Method ................. 79
2.3.3.4 Un-Referring to Conditions of Choice ........................... 80

3 A Multivariable Approach in Gas-Turbine Engine Testing 83

3.1 Chapter Overview ...................................................................................... 83
3.2 Introduction ................................................................................................ 84
3.3 The Conventional Single-Variable Method............................................ 85
3.3.1 Phase I – Engine ‘Rules of Operation’ ........................................ 85
3.3.2 Phase II – Maximum Available Power ........................................ 88
3.4 The MPOC Method for Engine Available Power Determination ..... 91
3.4.1 Phase I – Multivariable Empirical Models for the Rules of
Operation ......................................................................................... 92
3.4.2 Phase II – Fitting suggested models with experimental data.... 95
3.4.3 Phase III – selecting the right model for the task ...................... 97
3.4.4 Phase IV – max. GT engine output power estimation ............ 100
3.5 Maximum Power Estimation Comparison .......................................... 108
3.6 Summary and Conclusions ..................................................................... 109

4 A Singular Value Approach in Helicopter Flight Testing Analysis 111

4.1 Chapter Overview .................................................................................... 111
4.2 Introduction .............................................................................................. 112
4.3 Gas-Turbine Engine Performance Flight Testing .............................. 116
4.3.1 Principles of MPOC Method ...................................................... 116
4.3.2 Hypothesis testing and P-values ................................................. 117
4.3.3 Prediction goodness comparison between candidate models 119

vi
 4.4 Singular Values Approach for Model Screening ................................. 123
4.4.1 The SVD Theorem ....................................................................... 123
4.4.2 SVD implementation for model screening ............................... 124
4.4.2.1 The LSV – Models to PDs correspondences.............. 128
4.4.2.2 The RSV – Engines to PDs correspondences ............ 129
4.4.3 Selection of the best multivariable polynomial model ............ 131
4.5 Comparison to conventional methods and applications ................... 134
4.6 Conclusions and Summary ..................................................................... 137

5 Hover Performance Testing based on Dimensionality Reduction 139

5.1 Chapter Overview ................................................................................... 139
5.2 Introduction ............................................................................................. 140
5.3 Conventional Method for Hover Performance Testing .................... 143
5.4 Corrected-Variable Screening using Dimensionality Reduction ...... 149
5.4.1 Phase One - Original list of CVs for hover performance ...... 150
5.4.2 Phase Two - Screening for essential CVs using DR ................ 154
5.4.3 Phase Three - Deriving a practical empirical model ................ 161
5.5 The CVSDR Model Prediction Accuracy (OGE Hover) .................. 161
5.6 A Comparison Between the Conventional and CVSDR Methods .. 163
5.7 Conclusions and Summary ..................................................................... 165

6 Level Flight Performance Flight Testing 167

6.1 Chapter Overview ................................................................................... 167
6.2 Introduction ............................................................................................. 168
6.3 Level-Flight Performance Testing – The Conventional Way ........... 172
6.3.1 Constant Weight over Sigma (W/σ) Method ........................... 172
6.3.2 Constant Weight over Delta (W/δ) Method ............................ 174
6.3.3 Example Application - Constant (W/σ) Method..................... 175
6.4 The CVSDR Method for Level-Flight Performance Testing ........... 186
6.4.1 Phase One – Original list of CVs for level flight perf. ............ 187
6.4.2 Phase Two – Screening for essential CVs. ................................ 192
6.4.3 Phase Three – Deriving a practical empirical model ............... 199

vii
 6.5 Practical Guidance for the CVSDR Method in Level-Flight ............ 201
6.6 The CVSDR Model Prediction Accuracy (Level-Flight) ................... 204
6.6.1 Prediction Accuracy within the same coefficient-of-weight ... 204
6.6.2 Prediction Accuracy within a different coefficient-of-weight 209
6.7 Conventional and CVSDR Methods Comparison .............................. 211
6.8 Conclusions and Summary ..................................................................... 215

7 Conclusions and Recommendations 217

7.1 Novel Vs. Conventional Flight Test Methods –Main Differences .. 217
7.2 Conclusions............................................................................................... 219
7.2.1 Flight Testing for Power Available ............................................. 220
7.2.2 Flight Testing for Power Required in OGE Hover ................. 222
7.2.3 Flight Testing for Power Required in Level Flight .................. 225
7.3 Recommendations ................................................................................... 229
7.4 Closing Remarks ...................................................................................... 231

A. Gas-Turbine Engine Dimensional Analysis 233

B. High Speed Approximation, 10K Ft., Standard Day 241

C. Research Helicopters Description 243

The Bell Jet-Ranger Helicopter ...................................................................... 243
The MBB BO-105 Helicopter........................................................................ 249

References 254

Acknowledgements 269

Curriculum VitÆ 273

List of Publications 275

viii
List of Figures

Figure 1.1. Gas-Turbine engine aging process. ........................................................ 7

Figure 1.2. Power curve of a conventional helicopter in level flight. .................10
Figure 1.3. The Bell-Jet Ranger helicopter used for the research. .......................22
Figure 1.4. The MBB BO-105 helicopter used for the research..........................22
Figure 1.5. Thesis outline illustration.......................................................................27
Figure 2.1. The main-rotor blade in a hover flight. ...............................................31
Figure 2.2. The ideal and a -18° linear blade twist in an OGE hover.................36
Figure 2.3. Induced power of a -18° linear-twisted blade in an OGE hover. ...37
Figure 2.4. Main rotor power components in an OGE hover. ...........................39
Figure 2.5. Maximal disk-loading values for various helicopters. ........................41
Figure 2.6. The power-loading (PL) and the disk-loading (DL) relationship. ...43
Figure 2.7. The theoretical ground effect on the induced power in a hover. ....44
Figure 2.8. The main-rotor tip path plane (TPP). ..................................................46
Figure 2.9. The power curve of a helicopter in level-flight. .................................46
Figure 2.10. The theoretical induced velocity in level flight. ................................49
Figure 2.11. The error induced by using the high-speed approximation. ..........51
Figure 2.12. High-speed (HS) approximation validation chart (SSL) .................51
Figure 2.13. An example main-rotor disk in forward flight. ................................54
Figure 2.14. Max. airspeed of a conventional helicopter in forward flight. .........57
Figure 2.15. Parasitic drag breakdown for an example helicopter. .....................60
Figure 2.16. Example of 3rd order empirical model of gas-turbine engine.........64
Figure 2.17. The iterative procedure for engine available power disclosure. .....66
Figure 2.18. Example of an installed engine available power chart. ...................67
Figure 2.19. Non-dimensional OGE hover performance data............................72
Figure 2.20. Explicit presentation of OGE hover performance. ........................73
Figure 2.21. Example helicopter OGE hover ceiling determination. .................75
Figure 2.22. Non-dimensional level flight performance. ......................................78
Figure 2.23. Level-flight performance of an example helicopter. .......................81
Figure 3.1. Nom-dimensional single variable engine performance. ....................87

ix
 Figure 3.2. Estimated maximum continuous power of the example engine. .... 89
Figure 3.3. The MTU250-C20 engine power estimation errors using single-
variable models. .......................................................................................................... 90
Figure 3.4. Mean and SD of the single-variable estimation errors...................... 91
Figure 3.5. Estimation errors for the 10 proposed multivariable models.......... 97
Figure 3.6. Various multivariable empirical models performance ...................... 99
Figure 3.7. The engine internal rule of operation................................................ 102
Figure 3.8. A simultaneous presentation of all engine variables. ...................... 107
Figure 3.9. A simultaneous presentation of all engine variables. ...................... 108
Figure 3.10. MPOC and single-variable methods comparison.......................... 109
Figure 4.1. Corrected output power prediction errors ....................................... 118
Figure 4.2. Output power prediction performance............................................. 120
Figure 4.3. Test-statistics of models number 47 to 130 ..................................... 120
Figure 4.4. Top ten performing models. .............................................................. 121
Figure 4.5. Top ten performing models for the EC-145 engine. ...................... 122
Figure 4.6. The conceptual interpretation of SVD of matrix Z. ....................... 127
Figure 4.7. The relative strength of the seven Principle Dimensions (PDs). .. 128
Figure 4.8. Models to PDs correspondences (LSV). .......................................... 129
Figure 4.9. Engines to PDs correspondences (RSV). ......................................... 131
Figure 4.10. The Combined Normalized Scores (CNSs) for all 512 engine
models........................................................................................................................ 134
Figure 4.11. The mean errors of engines output power estimations. ............... 135
Figure 5.1. Non-dimensional OGE hover performance.................................... 145
Figure 5.2. Non-dimensional OGE hover performance (Sorties 1-3). ............ 147
Figure 5.3. Power prediction errors for Sortie 4 (base model). ......................... 148
Figure 5.4. The conceptual interpretation of SVD of Z’ in OGE hover
performance. ............................................................................................................. 156
Figure 5.5. The Singular Values (SVs) of Matrix Z’. ........................................... 157
Figure 5.6. Dimensions to CVs correspondence................................................. 158
Figure 5.7. Steps required for dimensionality reduction. ................................... 160
Figure 5.8. Power prediction errors for Sortie 4 (CVSDR model). .................. 162
Figure 5.9. The conventional and CVSDR methods prediction comparison. 163

x
Figure 6.1. Level flight performance (ND) of a BO-105 helicopter. ............... 176
Figure 6.2. Power prediction errors of the BO-105 (single-sortie app.). ......... 179
Figure 6.3. Mean of absolute power prediction errors (single-sortie app.). ...... 179
Figure 6.4. Prediction errors quantiles to theoretical normal quantiles ............... 181
Figure 6.5. Power prediction errors to advance-ratio correlation (single-sortie
approach). ................................................................................................................. 182
Figure 6.6. Power prediction errors of the BO-105 (cluster of sorties
approach). ................................................................................................................. 183
Figure 6.7. Mean of absolute prediction errors (single & cluster of sorties
comparison). ............................................................................................................. 184
Figure 6.8. Graphical presentation of all 36 CVs for level-flight perf.. ........... 192
Figure 6.9. The conceptual interpretation of SVD of Z’ in level-flight
performance. ............................................................................................................ 195
Figure 6.10. The normalized singular values of the level-flight performance. 196
Figure 6.11. Correspondence between CVs and level-flight dimensions. ....... 198
Figure 6.12. CVSDR level flight performance testing- Sorties planning
sequence. ................................................................................................................... 201
Figure 6.13. Conventional and CVSDR power prediction errors. ................... 206
Figure 6.14. Mean of power prediction errors - conventional and CVSDR
methods..................................................................................................................... 207
Figure 6.15. Prediction errors to advance ratio correlation. .............................. 209
Figure 6.16. Power prediction errors for Sortie 5 (CVSDR method). ............. 210
Figure 7.1. Conclusions to RQ’s Mapping ........................................................... 220
Figure C.1. The Allison T63-A-720 gas turbine engine ..................................... 245
Figure C.2. The Jet-Ranger flight-controls .......................................................... 246
Figure C.3. The horizontal stabilizer .................................................................... 247
Figure C.4. The Jet-Ranger flight instruments fed by the Pitot system. .......... 248
Figure C.5. The main-rotor assembly of the BO-105 helicopter ..................... 250
Figure C.6. The rear end of the BO-105 fuselage ............................................... 251
Figure C.8. The BO-105 Instrument Panel ......................................................... 253

xi
List of Tables

Table 3.1. Third order polynomials for GTE performance modeling ...............93
Table 3.2. Empirical model predictors ....................................................................93
Table 4.1. Gas-turbine engines used for the analysis ......................................... 115
Table 4.2. List of MPOC engine predictors ........................................................ 116
Table 4.3. List of 10 top-performing models for the BO-105 helicopter. ...... 121
Table 4.4. List of 10 top-performing models for helicopter GT engines ........... 133
Table 5.1. Summary of OGE hover conditions .................................................. 144
Table 5.2. Variables and dimensions involved in hover performance. ............ 150
Table 5.3. Corrected Variables to represent the OGE hover performance. ...... 154
Table 6.1. Summary of flight-test conditions for Sorties 1 to 4 ....................... 176
Table 6.2. Variables and dimensions involved in level-flight performance. ... 188
Table 6.3. Corrected-Variables for level-flight performance ............................ 191
Table 6.4. A step-by-step guidance for CVSDR level-flight perf. testing............ 203
Table 6.5. Summary of flight-test conditions for Sortie 5. ................................ 210
Table 7.1. A step-by-step guidance for CVSDR OGE hover testing .............. 224
Table 7.2. A step-by-step guidance for CVSDR level-flight testing................. 228
Table A.1 – GT engine - summary of variables and dimensions involved ..... 234
Table C.1 – The Bell Jet Ranger performance specifications ........................... 248
Table C.2 – The MBB BO-105 performance specifications. ............................ 253

xiii
Nomenclature

A Matrix containing numerical regressors

Ad , Adisk Main-rotor disk area
C Averaged chord length (main-rotor blades)
C d0 Zero-lift drag coefficient (main rotor blades)
CP 
P Coefficient of power (non-dimensional)
 a Ad  R 
3

CW 
W Coefficient of weight (non-dimensional)
 a Ad  R 
2

CNg 
Ng Corrected engine compressor speed

CSHP 
SHP Corrected engine output power (shaft horse power)
 
CTGT 
TGT Corrected engine temperature (turbine gas temperature)

CW f 
Wf Corrected engine fuel flow
 
D Main rotor blade aerodynamic drag force
Er Prediction error vector; difference between model (i) to actual
i
measured power
E R( j ) Mean of absolute power prediction errors for sortie (j)
M Mach number
Ng Gas turbine engine compressor speed
P Total power required for flight
pa Ambient air static pressure
po Standard sea-level air pressure (14.7 psi)
R Main-rotor radius
Rair Specific gas constant for air (=287 J/Kg·K)
SHP Engine output power (shaft horse power)
S xi Standard deviation in sampled variable Xi
Sref Aerodynamic drag reference area
T Thrust produced by the main rotor system
Ta Ambient air static temperature
To Standard sea-level static air temperature (288.15K)
TGT Engine temperature (turbine gas temperature)
TRQ Engine output shaft torque
VCW The main rotor chord wise velocity
Vih Induced velocity in a hover (average), main rotor disk
Vi(r) Induced velocity at blade station r
VT True airspeed
W Helicopter gross-weight
Wf Engine fuel-flow
Xcg Helicopter longitudinal center of gravity location
Z, (Z’) Corrected variables matrix (normalized)
a Speed of sound

xv
 ai,bi,ci Generic single variable polynomial coefficients
b Main-rotor number of blades
⃗
𝒃 Vector representing measured CSHP
fe Fuselage equivalent flat-plate area for drag
fi Engine multivariable regressors, i=1,2,3,…
gj, hk Inequality constraints, Equality constraints
ki Induced power correction factor
q Dynamic pressure
rx , y Linear correlation coefficient between two variables x,y
ti Test-statistics of model (i) prediction errors
𝛼𝑗𝑖 ,  i  ,  i  Generic multivariable polynomial coefficients
γ Heat capacity ratio for air (=1.4, non-dimensional)
𝜆𝑖 , 𝜂𝑖 Lagrange multipliers (equality constraints, inequality constraints)
  Pa P0 Static pressure ratio (non-dimensional)
𝜂𝑚 Main rotor mechanical efficiency
  Ta T0 Static temperature ratio (non-dimensional)
Θ Main rotor blade pitch angle

VT Advance ratio (non-dimensional)
R
i Mean value of prediction errors, model (i)
a Ambient air static density
o Standard sea level static air density (1.225 kg/m3)
,   Singular values matrix of Z (normalized)
  a 0 Static density ratio (non-dimensional)
 i  i 1,2,..., r  Singular values of a generic matrix of rank ‘r’
R 
bc Main-rotor solidity ratio (non-dimensional)
R
ψ Main rotor blade azimuth angle
i , i Generic non-dimensional (ND) variable
 i* ,  i* Generic corrected variable (ND for a specific helicopter type)
Ω,ω Main-rotor angular speed

Abbreviations

BET: Blade Element Theory

BEMT: Blade Element Momentum Theory
CFD: Computational Fluid Dynamics
CMIV: Constant Momentum Induced Velocity
CSHP: Corrected Shaft Horse Power
CTGT: Corrected Turbine Gas Temperature

xvi
CV: Corrected Variable
CVSDR: Corrected Variables Screening using Dimensionality Reduction
DL: Disk Loading
EFPA: Equivalent Flat Plate Area
FAA: Federal Aviation Administration
FM: Figure of Merit
FW: Fixed Wing
GTE: Gas Turbine Engine
HIGE: Hover In Ground Effect
HOGE: Hover Out of Ground Effect
HUMS: Health and Usage Monitoring System
ISA: International Standard Atmosphere
KKT: Karush Kuhn Tucker
LSV: Left Singular Vectors
MBB: Messerschmitt Bölkow Blohm
MPOC: Multivariable Polynomial Optimization under Constraints
MR: Main Rotor
NTPS: National Test Pilot School
PD: Principal Dimension
PDF: Probability Density Function
PL: Power Loading
PS: Problem Statement
RQ: Research Question
RSV: Right Singular Vectors
RW: Rotary Wing
SHP: Shaft Horse Power
SSL: Standard Sea Level
SVD: Singular Value Decomposition
TPP: Tip Path Plane
TR: Tail Rotor

xvii
Summary

Flight test engineering is an interdisciplinary science that gathers flight-test data

and develops methods with the objective of evaluating an aircraft or an airborne system
in its operational flight environment. The need for flight testing emanates as a
necessary effort that complements ground-based verification activities such as wind-
tunnel testing, simulators and computational modelling. Flight testing is a broad field
that involves many disciplines. Performance flight testing is one discipline that is
responsible of providing answers to questions like: How high can the aircraft fly? How
fast can it fly? How much power does the aircraft need in order to sustain specific
flight conditions of gross-weight, altitude and ambient air temperature? Or How long
can the aircraft remain airborne before it runs out of gas (or electric power)? A
profound data base for the performance of any type of aircraft is essential for their
safe and efficient operation.

This thesis focuses on performance flight-testing methods for conventionally-

configured helicopters, i.e., those that employ a single main rotor to generate lift and
thrust, and a single tail rotor to counter-act the torque effect of the main rotor. More
specifically, the scope of this research was limited to gas-turbine available power testing
and power required for out of ground effect (OGE) hover and power required for
level-flight (AKA cruise flight). The research was limited to the execution of up to ten
flight test sorties on two types of helicopters; the Bell Jet-Ranger and the MBB BO-
105 helicopters, both normally used for training at the National Test Pilot School
(NTPS) in Mojave, California.

The goal of this thesis is to develop new and improved flight-test methods to
rectify existing problems associated with the conventional methods. The conventional
method for the maximum available power of a gas-turbine relies on three independent,
single-variable polynomials that often yield poor prediction accuracy that sometimes
even defy basic engineering concepts. The conventional method for OGE hover
performance is overly simplified and neglects important blade non-linear effects. This
results in inaccurate empirical models for hover performance representation. The

xix
conventional flight-test method for level-flight performance incorporates several
drawbacks which not only make the execution of flight-test sorties inefficient and time
consuming, but also compromise the level of accuracy achieved. This conventional
level-flight method fails to specifically address non-linear effects such as blade-tip
compressibility and drag-divergence that often results in inaccurate predictions,
especially at high altitude and low air temperature conditions.

The research intended to develop new flight-test methods for the available power
of a gas-turbine engine and for the power required for hover and level-flight. Both new
methods are based on multivariable polynomial approach. The research was initiated
with the development of a new method for the maximum available power of a gas-
turbine engine. A novel method, referred to as the ‘Multivariable Polynomial
Optimization under Constraints’ (MPOC), was developed. This method seeks for a
third order multivariable polynomial to describe the engine output power as a function
of the other three variables of the engine (compressor speed, temperature and fuel-
flow). The maximum available engine power is realized by solving an optimization
problem of maximization under constraints. For this optimization, the Karush-Khun-
Tucker (KTT) method was used successfully. For the exemplary BO-105, the standard
deviation of the output power estimation error was reduced from 13 hp (conventional
method) to only 4.3 hp by using the proposed method. Expanding the flight-test data
base to include seven different engines reveals that the multivariable polynomials
approach of the proposed method performed much better with all seven engines, as
compared to the conventional single-variable approach. The maximum average
prediction error was only 0.2% as compared to a maximum average prediction error
of 1.15%, yielded by the conventional method.

The research effort conducted for the OGE hover performance was concluded
successfully with the development of the novel “Corrected Variables Screening using
Dimensionality Reduction” (CVSDR) method for hover performance. This novel
method combines fundamental dimensional analysis to generate a list of candidate
corrected-variables (CVs) to represent the hover performance problem, then screens
for the most essential ones by means of dimensionality reduction, implemented by

xx
singular-value-decomposition (SVD). This phase of the research was executed with
four sorties on the Bell Jet-Ranger helicopter and produced a total of five conclusions.
The most significant conclusion was that power predictions of the CVSDR method
were 1.9 times more accurate than the conventional method. At the 95% confidence
level, the CVSDR method deviated by an average of only 0.9 hp (0.3% of the maximum
continuous power of the example helicopter) from the actual power required to hover,
whereas power predictions from the conventional method deviated by an average of
1.7 hp.

The final phase of the research concentrated on developing a new flight-test

method for the level-flight regime. This effort spanned over five distinct sorties using
the BO-105 helicopter. Similar concepts used for the hover performance testing were
expanded and adapted for level-flight performance flight testing. The CVSDR method
for level flight performance can be regarded (abstractly) as an expansion of the CVSDR
method for OGE hover into a higher dimensional space. This phase of the research
was aimed at addressing five research questions and yielded ten conclusions. The top
three conclusions were that (1) the power predictions accuracy achieved using the
CVSDR method for level-flight was nearly 21% better (on average and at the 95%
confidence level), as compared to the prediction accuracy yielded from the
conventional method. (2) the CVSDR method made planning and execution of flight-
test sorties more efficient and time conserving. It is estimated to reduce flight-time for
data gathering by at-least 60%, and (3) the CVSDR method is not restricted by the
high-speed approximation, hence is also appropriate for the low-airspeed regime, and
can potentially bridge the empirical modelling gap between the hover and level-flight
regimes.

The novel flight-test methods developed within this research (the MPOC for the
available power of a gas-turbine engine and the CVSDR for OGE hover and level-
flight performance) are recommended to be used by the helicopter flight-testing
community, as they were shown to increase accuracy and promote execution
efficiency.

xxi
 This thesis produced six recommendations concerning possible future expansion
of the work already done during the current research. These include an expansion of
the CVSDR method into more areas of performance testing like vertical and forward
flight climb, partial power and unpowered descent, etc. Another continued research
recommendation relates to the applicability and efficiency of the CVSDR method to
relevant vertical-lift aircraft that combine both RW and FW characteristics. It is also
recommended that continued research look into the potential and feasibility of
employing the CVSDR method for empirical modelling used by Health and Usage
Monitoring Systems (HUMS) installed in helicopters.

xxii
 If you are in trouble, an airplane can fly over and drop flowers,
but a helicopter can land and save your life.
Igor Ivanovich Sikorsky

1 I NTRODUCTION

1.1 B ACKGROUND AND RELEVA NCE

F light test engineering is an interdisciplinary science that gathers flight-test data

and develops methods with the objective of evaluating an aircraft or a system in
its operational flight environment. The need for flight-testing means that the system
or the vehicle under testing requires accurate and efficient assessment of its
characteristics and performance while operating in its flight environment, rather than
just relying on the results of ground-based verification methods such as wind tunnels,
simulators, and software models [1]. There are many disciplines involved in flight
testing based on the nature of the questions in search. Such ones include, for example,
performance assessment, structural integrity testing, stability, and handling-qualities
evaluation, etc. Performance flight-testing is an expensive activity that requires efficient
and accurate methods for determining the aircraft performance in its certified flight
envelope and under a wide range of atmospheric conditions. Such methods involve
careful considerations regarding testing techniques and flight-test data analysis.

Accurate performance prediction of any aircraft is essential for a safe and

efficient operation of the air-vehicle. Knowing, in advance and with high level of
confidence, the answers to questions like “how high can the aircraft fly under specific
atmospheric conditions and gross weights? How fast can the aircraft fly? How long
can the aircraft remain airborne before it runs out of gas?” etc., are essential for
ensuring safety of flight and mission compliance. Although aircraft manufacturers
provide early-stage aircraft performance predictions, those are normally based on

1
1 | INTRODUCTION

practical engineering simplifications and assumptions. Performance validation through

an expensive and lengthy phase of flight-testing is inevitable.

Conducting a performance flight-test campaign is not limited to new aircraft

programs. Performance flight-testing programs are also required during the life cycle
of an aircraft. It is common-practice for aircraft operators to design and implement
modifications that alter the aerial-vehicle baseline performance. These modifications
can be limited in scope or even a full-scale upgrade programmes, initiated by either
economic, political, or operational reasons. The helicopter is no exception in this
regard. This type of ‘low and slow’ aircraft requires efficient and accurate performance
flight-testing methods, either to be used by the manufacturers, or by the common
operator (civilian or military) for post productions modifications and upgrades.
Moreover, the limited flight envelope of the helicopter, as compared to the fixed-wing
airplane, gives its operators the confidence and motivation to implement post-
production structural modifications that warrant limited-scope performance flight-test
campaign [2-6].

The performance charts and tables published by the helicopter manufacturers are
based on a certain available power level. As explained in Subsection 1.2.1 hereinafter,
the maximum available power out of the engine(s) deteriorates as the engine(s) matures
and accumulates an increasing number of working hours. For this reason, the
published performance of the helicopter is based on the minimum allowed level of the
available power, i.e., just when it is time for the engine(s) to be overhauled. It is
common for borderline missions to evaluate the feasibility of a specific helicopter to
execute the specific challenging mission. For this, the operator needs to execute ad-
hoc performance flight testing using the specific helicopter in order to conclude about
mission performance compliance. Accurate and efficient performance flight testing
methods are of high relevance for helicopter operators who wish to know the precise
performance of their particular helicopter.

2
 1.2 | H E L I C O P T E R P E R F O R M A N C E

1.2 H ELICOPTER P ERFORMANCE

One might wonder what does ‘helicopter performance’ exactly mean? According
to Cambridge dictionary, performance is defined as: “how well a person, machine, etc.
does a piece of work or an activity”. The Collins dictionary defines someone’s or
something’s performance as: “how successful they are or how well they do something”.
According to Meriam-Webster dictionary: “PERFORM implies action that follows
established patterns or procedures or fulfils agreed-upon requirements and often
connotes special skill”. The previous section has already alluded that ‘helicopter
performance’ has something to do with answering questions like how high can the
helicopter fly? How heavy can it hover? How long can it stay airborne? etc. The FAA
[7] defines aircraft performance as: “a term which is used to describe the ability of an
aircraft to accomplish certain things that make it useful for certain purposes”. It
continues and provides examples like the ability to carry heavy loads or to fly at high
altitudes. Definitely, not a sharp and elegant definition for aircraft performance. Prouty
[8] also struggles with this performance definition and provides the following
explanation instead; helicopter performance analysis is made to answer the questions:
How high? How fast? How far? How long? The results of the analysis may be used in
design trade-off studies, in a pilot’s handbook, in a set of military standard aircraft
characteristics charts, or in a sales brochure. Another explanation for helicopter
performance is provided by Gessow & Myers (1967) [9]: “The precise estimation of
helicopter performance depends on an accurate determination of the thrust produced
and the power required by the rotor in those conditions” (p. 66). Leishman (2006) [10]
explains the term ‘helicopter performance’ in the introduction to the performance
chapter as: “the estimation of the installed engine power required for a given flight
condition, determination of maximum level flight speed, evaluation of the ceiling (in
and out of ground effect), or the estimation of the endurance or range of the
helicopter”.

The previous definitions and explanations for helicopter performance draw a clear
distinction between two parts of this term. One is the amount of power available for
use by the helicopter. The other is the amount of power required to sustain any

3
1 | INTRODUCTION

specific flight condition. These two parts constitute the term helicopter performance
for flight-testing. The power available is provided by the power-plant installed in the
helicopter. This power-plant can be based on a single-engine or on a multi-engine
configuration. The power required is the amount of power needed to maintain the
helicopter under a specific flight and ambient conditions. The power available and
power required are independent of each other and can be visualized as the two hands
of a scale, the performance scale. As long as the amount of power generated by the
power plant is equal to or larger than the power required for the specific flight
conditions, the performance of the mission is feasible.

1.2.1 Available Power

Commercial and military helicopters are powered by mainly two types of engines:
reciprocating (piston) engines and gas-turbine (GT) engines. In recent years, few
programs were conducted to demonstrate and study the feasibility of use of electric
engines in helicopters. The Firefly program introduced by Sikorsky innovations in 2010
is a good example to these types of technology demonstration programs [11]. Although
some progress was made with the idea of electric propulsion of helicopters, there is
still a way to go before electric propulsion turns into a common way to power
helicopters for all their types of missions. The use of gas-turbine engines is far more
popular than the use of reciprocating engine in helicopters. The superior ratio of power
to weight of the gas-turbine engine, makes it a better choice when it comes to medium
to large types of helicopters. For small size and light helicopters, the piston engine
might be considered. According to Moon and Yakovlev [12] in 2018 the gas-turbine
engine helicopters accounted for 69.1% of all in-operation helicopters worldwide. This
unrivalled popularity of the gas-turbine engine, as the preferred propulsion system for
helicopters, motivates development of efficient flight-testing methods that facilitate
accurate prediction of installed gas-turbine output power under a wide range of
atmospheric conditions.

4
 1.2 | H E L I C O P T E R P E R F O R M A N C E

Although all engine manufacturers measure their engine performance (‘bench

testing’) and provide charts that describe the engine output power under various
ambient conditions (‘engine deck’), there is still a necessity for flight-testing methods
to measure the actual installed gas-turbine engines output power. The following are
two main reasons to support this necessity and provide motivation to develop new
methods for gas-turbine engine performance flight testing:

1) The performance of a gas-turbine engine once installed in any type of a

helicopter, is different from its performance measured by the engine manufacturer on
a test-bench. Each helicopter type imposes specific levels of degradation in engine
output power as compared to the same exact engine, uninstalled original performance.
This is referred-to as the engine installation loss. According to Prouty [8], there are
various contributors to this degradation in the output power of an installed engine, as
compared to its performance outside of a helicopter. One of them is the engine inlet
structure element. The purpose of the inlet on any type of aircraft is to slow-down the
air flow prior it enters the engine. This deceleration process in the inlet involves loss
in total pressure due to friction and increase in static temperature, due to exhaust re-
ingestion [13,14] and installation of various heat-exchange devices. The alteration of
the thermodynamic properties of the air enters the engine are a cause for up to 5% of
power installation loss. In addition to power loss associated with inlets, many
helicopters are fitted with ‘particle-separator’ systems designed to protect the gas-
turbine engine by filtering the air before it gets into the engine. Taslim and Spring
(2010) [15] show that ‘particle-separator’ systems reduce the available power by 3% to
10%. Another system installed on military helicopters and is responsible to high power
loss (3-15%) is the infrared suppressor designed to protect the helicopter against IR
missiles by lowering the IR signature generated by the engine(s) exhaust [16,17]. Finally
for installation loss, is the power drained from the engine(s) via compressor bleed, for
the benefit of particular helicopter on-board systems operation. This type of power
loss can reach up to 20%. The engine performance as provided by the engine
manufacturer, is not sufficient for the task of total helicopter performance
determination. Explicit flight-testing of an installed engine output power is mandatory

5
1 | INTRODUCTION

for the total performance determination of any specific type of helicopter and even for
the determination of engine installation-loss themselves.

2) The gas-turbine engine performance as published by the engine manufacturer

represents a new gas-turbine engine. This is referred-to as a specification (‘spec’) engine
or a ‘commercial off-the-shelf’ (COTS) engine. Any gas-turbine engine is ensured to
deliver, as a minimum, the ‘spec’ engine performance. It is common for gas-turbine
engine manufacturers to provide new engines with even better performance than the
‘spec’ engine. As the engine accumulates flight hours and matures, its performance
degrades. This is the natural aging process of the gas-turbine engine. Once the engine
performance reaches a well-defined, minimum level of performance it is taken-off
from the helicopter and sent for overhaul. This minimum level of performance
delivered by the engine is frequently defined as the ‘reject-line’ of the engine. This
aging process of the engine is illustrated in Fig. 1.1. In this figure the available power
which is the engine output power is represented as a function of the various engine
parameters. The helicopter operator must know at any given phase of the gas-turbine
life cycle the actual performance the engine possesses, and more importantly the
margin of power it maintains above the minimum acceptable power (the reject-line).
For this reason, the helicopter operator needs an explicit flight-testing method for the
evaluation of the installed gas-turbine engine output power at any given phase of the
life cycle of the engine; either newly installed in the helicopter or just prior for it to be
retired and sent for overhaul.

6
 1.2 | H E L I C O P T E R P E R F O R M A N C E

Figure 1.1. Gas-Turbine engine aging process. The engine performance drops once
installed in a helicopter. As the engine accumulates flight-hours its performance decreases until it
meets the reject line and is taken-off the helicopter to be overhauled.

1.2.2 Power Required

As previously stated, the power required is the amount of power needed to

maintain the aircraft under specific flight and ambient conditions. The power required
is independent of the available power, although the two are often evaluated
simultaneously in flight-test campaigns. It is common practice within the flight-testing
community to break down the power required envelope into the following disciplines
[18-21]:

1) Hover performance, in and out of ground effect (HIGE/HOGE, respectively);

2) Vertical climb and decent performance;

3) Level flight performance; and

4) Forward flight climb and descent performance.

7
1 | INTRODUCTION

As stated in the subsequent Section 1.5 (Research Goals and Objectives), the current
research is limited to only two performance disciplines out of the helicopter power
required. These are the hover out of ground effect (HOGE) and the level flight, also
known as ‘cruise flight’. For this reason, only these two-helicopter power required
subjects are discussed hereinafter.

1.2.2.1 Hover Performance

Igor Sikorsky, the legendary helicopter developer, was once asked by an

anonymous scientist, a friend of his, when will the helicopter go faster than an airplane?
In a documented interview Igor Sikorsky replied that it will never go faster than an
airplane, but the helicopter will be able to do ‘number of jobs no other airplane will be
able to do’, referring mainly to the remarkable ability of the helicopter to stabilize in a
long-term hover. Indeed, the most distinguishing characteristic of a helicopter as stated
by Leishman (2006) [22] and by Gessow and Myers (1967) [23] is its ability to steadily
hover at any phase of its mission, given it has a sufficient power margin. Knowing the
power required to hover throughout all mission phases is crucial for any helicopter
flight crew.

For a conventional helicopter, i.e., one which employs a single main-rotor and
a single anti-torque tail-rotor, the entire lift force in the hover is generated by the single
main rotor. The set of main-rotor blades, referred to as rotary-wings, generate the lift
required to hold the helicopter airborne. The main rotor is the helicopter major power
consumer in a hover. The actual percentage of power it consumes changes in between
types of helicopters and for a given type of helicopter it varies based on the gross
weight, external configuration, and atmospheric conditions, but the ‘golden-rule’ for
this power consumption percentage is about 85% [19,20,24]. The remaining ~15% of
the hovering power is dissipated by the tail-rotor (5-10%), various accessory drives and
transmission loss. Typical values of helicopter transmission loss at nominal rotor speed
can be learnt from Lewicki and Coy (1987) [25] and Coy et al. (1988) [26]. The
mechanical efficiency of a Black Hawk helicopter transmission at full power (2,828

8
 1.2 | H E L I C O P T E R P E R F O R M A N C E

hp.) was measured between 97.3% and 97.5% (depends on the lubricant oil inlet
temperature). For the lower power rated transmission of the OH-58C helicopter
(317hp.) measured transmission loss were between 1.2% and 1.7%. This relative
portion of the total power consumed by the main rotor is known as the mechanical
efficiency of the helicopter, denoted as (ηm). Since the main rotor is responsible for
about 85% of the total power required in a hover, much attention is given by helicopter
manufacturers for its blades design. Moreover, since the main rotor is the most
significant power consumer in a hover, it dictates the conventional flight-test method
for hover performance, as initially presented in Subsection 1.3.2 and thoroughly
discussed in Chapter 2 of the thesis.

1.2.2.2 Level Flight Performance

The helicopter does not exhibit superior capabilities over other types of aircraft
when it comes to level-flight (‘cruise flight’). Nevertheless, a typical helicopter spends
most of its flying time in the level-flight regime. The relative time while cruising varies
based upon the type and the specific mission the helicopter was designed for.
Porterfield and Alexander (1970) [44] analysed data from various types of helicopters
and proclaimed that, on average, the helicopter spends 71% of its flight-time in level-
flight. The FAA (2008) [45] provides different estimates for two exemplary gas-turbine
helicopters. The first example is a utility business type which is estimated to spend
61% of its flight time while cruising, and the second example is for a transport
helicopter which is estimated to spend 73% of its flight time in level-flight. Regardless
of where this value for relative time spent in level-flight truly resides, the helicopter
spends most of its airborne time while cruising. This observation makes the evaluation
of level-flight performance utterly important in any new or modified helicopter
performance flight-test campaign.

The power consumed by a conventional helicopter in level-flight is composed

out of three major components as illustrated in Fig. 1.2. These power components are
the induced power, the profile power and the parasite (or parasitic) power. The first

9
1 | INTRODUCTION

two power terms are familiar from the hover domain, although their relative
magnitudes significantly change from the hover, based upon the airspeed of the
helicopter. The induced power, which is required for the creation of thrust, drops
rapidly with the increase of airspeed. From constituting about 85% of the power
required for hover, the induced power can drop to about only 10% of the total power
required for level flight at a maximum allowed cruising airspeed of a conventional
helicopter. The profile power, required to overcome the viscous effects between the
blades and the air, increases with airspeed and even becomes the dominant power
component, over a certain airspeed section. The parasite power component is due to
fuselage drag and it increases rapidly with airspeed (cubic relationship). The parasite
power is typically the dominant power component for a cruising helicopter at a high
airspeed. The general behaviour of the power required for level flight has the shape of
a ‘bathtub’, i.e., it decreases with airspeed increase until it gets to a minimum power
level (referred-to as the ‘bucket’), then the power level increases with airspeed, as it is
dominated by the parasite and profile power components.

Figure 1.2. Power curve of a conventional helicopter in level flight. The power required
in level flight comprises of three main components (induced, profile and parasite). The total
power in level flight decreases with airspeed increase until it gets to a minimum (‘bucket’).
Past the bucket airspeed, the power required for level flight increases with airspeed increase.

A more comprehensive discussion about the power required for a conventional

helicopter in level-flight is presented in Chapter 2, Helicopter Performance Theory
and Conventional Testing Methods.

10
 1.3 | C O N V E N T I O N A L M E T H O D S FOR PERFORMANCE FLIGHT T ESTING

Following this brief introduction to the relevant aspects of helicopter

performance, a short introduction to the applicable conventional performance flight
testing is presented.

1.3 C ONVENTIONAL M ETHODS FOR

P ERFORMANCE F LIGHT T ESTING

The fundamental question troubling the helicopter performance flight-testers

has always been how should the performance of the helicopter be tested? This broad
question can be further narrowed down to the following set of specific questions; at
which particular weights should the performance be measured? Under which
combinations of atmospheric conditions? Is it possible to standardize the
measurements obtained? Can the performance measurements in a specific flight be
used to predict the performance under different flight conditions? These types of
questions, which are essentials in performance flight-testing, can be addressed by using
tools of dimensional analysis and the Buckingham PI theorem [66] which is commonly
regarded as the fundamental theorem of dimensional analysis [67]. Evans [67] also
claims that the Buckingham PI theorem is not widely known and that its full generality
has not been exploited.

According to the Buckingham PI theorem, any physically meaningful problem

with numerous dimensional parameters involved, can be reduced to a lesser number
of significant non-dimensional parameters, based on the dimensions involved. The
Buckingham PI theorem is used in many walks of science and engineering [68-73].
Zohuri [74] concludes its Chapter 1 by stating that many engineering problems are too
complex to find a mathematically closed form of solution for them. In such cases, a
type of analysis, which involves the dimensions of the quantities entering the problem
may be useful.

The Buckingham PI theorem is no stranger to aerodynamics problems. In fact,

this theorem can be used to justify the projection of conclusions from wind-tunnel

11
1 | INTRODUCTION

tests, performed using reduced size models, onto the actual true-size aircraft. The
Buckingham PI theorem can also be used to justify the importance of the Reynolds
number in flow dynamics study, or the importance affect the Mach number has on
flight performance. Using dimensional analysis in flight-testing allows to collect data
under specific atmospheric and flight conditions and to project the data to particular
conditions of choice (interpolation and extrapolations). As a ground rule, the flight
testers try to refrain, as much as possible, from carrying-out extreme extrapolations to
their measured data. Breaking new grounds for the operators is the motto of the flight-
testers. Interpolation of flight-test data, on the other hand, is always blessed. The
competency of dimensional analysis to reduce the number of affecting variables and
to support interpolation and extrapolation of data, has made it a popular tool in
performance flight testing. By applying dimensional analysis tools, the flight-tester can
answer all performance questions raised above and at the same time, significantly
reduce the number of flight-test sorties required to quantify the performance of an
aircraft throughout its flight-envelope.

1.3.1 Available Power Flight-Test Method

The current method widely used within the flight-test community for determining
the available power any gas-turbine helicopter possess is based on the single-variable
analysis [18-21]. According to this method, all four engine performance variables
(output power, compressor speed, turbine gas-temperature and fuel-flow),
accompanied with their corresponding atmospheric conditions, are recorded during
steady engine operation conditions. For this, the helicopter is flown throughout its
certified flight-envelope and under diverse atmospheric conditions.

Once a substantial database is gathered, it is analysed with the goal of evaluating

the maximum power the engine is able to produce under various atmospheric
conditions. This procedure is carried-out in two phases, the first (Phase I) is to generate
a convenient mathematical model describing the dependency of the engine output
power with all other engine performance variables (typically a third order polynomial).

12
 1.3 | C O N V E N T I O N A L M E T H O D S FOR PERFORMANCE FLIGHT T ESTING

Phase I can be regarded as uncovering and defining (mathematically) the specific

engine ‘rules of operation’. The second phase (Phase II) uses the ‘rules of operation’
to derive the maximum output power of the engine under a wide range of atmospheric
conditions, and for the various engine power ratings i.e., the maximum continuous
power, the maximum 5-minute take-off rating power, etc.

This conventional flight test method for the available power of a gas-turbine
engine (GTE) is further discussed and demonstrated in Chapters 2 and 3 of this thesis.
The deficiencies associated with this flight-test method originated the first question of
this research (RQ1), as presented in Subsection 1.4.1. A comprehensive discussion
about the deficiencies of this conventional method is presented in Chapter 3 of the
thesis.

1.3.2 Hover Performance Flight Testing

The objective of hover performance flight-testing is to provide a detailed map of

the actual power required to sustain the specific type of helicopter at a hover (either in
or out of ground effect, IGE/OGE) for all certified gross-weights, external
configurations, main-rotor angular speeds and the surrounding atmospheric
conditions of air temperature, pressure, and density. This performance ‘map’ is
traditionally presented in a format of a graph, or a set of synchronized graphs and
plots. For this objective, the hover performance flight-tester task is to measure the
actual power required for hover throughout the flight envelope. Since it is impractical
to hover the helicopter in each combination of gross-weight, main-rotor angular speed,
atmospheric conditions of pressure altitude and temperature, the flight tester applies
means of dimension analysis, as previously discussed in the introduction of this Section
(1.3). Applying means of dimensional analysis allows the flight-tester to reduce the
number of planned flight test sorties to an achievable and practical number, and at the
same time to provide a detailed performance map that covers the entire certified flight
envelope of the aircraft.

13
1 | INTRODUCTION

The conventional, widely used, flight-test method for hover performance is

derived from the non-dimensional form of the mathematical relationship that
describes the power required for the main-rotor of a hovering helicopter. Since the
main-rotor is the principal power consumer in a conventional helicopter hover (about
85%), it justifies enforcement of the main-rotor power model onto the helicopter as
a whole. The flight-test team is required to regress a mathematical model which relates
between the power required for hover and the variables of gross-weights, main-rotor
angular speeds and atmospheric ambient conditions (pressure, temperature and
density). This empirical hovering model retrieved from flight-testing is then used,
analytically, to create the detailed performance map mentioned above. This is the
estimated power required to hover at any certified gross-weight and main-rotor speed,
and under any combination of atmospheric conditions.

The conventional flight-testing method for hover performance is further

discussed and demonstrated in Chapters 2 and 5 of this thesis. The deficiencies
associated with this flight-test method originated the second question of this research
(RQ2), as presented in Subsection 1.4.2. A comprehensive discussion about the
deficiencies of this conventional hover performance flight-test method is presented in
Chapter 5 of this thesis.

1.3.3 Level Flight Performance Flight

Testing

The objective of level-flight performance flight testing is to provide a detailed map

of the actual power required to maintain the specific type of helicopter at a level flight
conditions, for all certified gross-weights, external configurations, main-rotor angular
speed range and the surrounding atmospheric conditions of air temperature, pressure,
and density. This performance ‘map’ is traditionally presented in a format of a graph,
or a set of synchronized graphs and plots. Moreover, cross-referencing the power
required in level flight to the fuel-consumption data base, as evaluated during the
available power flight-testing phase (described in Subsections 1.2.1 and 1.3.1 above),

14
 1.4 | P R O B L E M S T A T E M E N T

enables to define the helicopter ‘best-effort’ airspeeds, such as airspeed for maximum
range and for maximum endurance. Similarly to the available-power and hover
performance flight-testing, the conventional method for level-flight performance flight
testing makes use of dimensional analysis concepts in order to reduce significantly the
number of planned flight test sorties.

The conventional flight-test method for level-flight performance of a

conventional helicopter is thoroughly discussed in the literature [8, 10, 18, 49, 76] and
demonstrated in numerous flight test reports [6,77,78]. At its core, this flight-test
method seeks for various discrete empirical relationships (for several non-dimensional
gross-weights referred-to as coefficient of weight) between the non-dimensional forms
of power (coefficient of power) and airspeed (advance ratio). For this, the flight-tester
is required to perform many ‘speed-runs’ while maintaining the coefficient-of-weight
at a constant value, a requirement that imposes execution difficulties, and is responsible
to one of the method deficiencies. The acquired set of empirical models is then used,
analytically, to predict the level flight performance of the tested helicopter at any
certified gross-weight and main-rotor speed, and under any combination of
atmospheric conditions.

The conventional flight-testing method for level-flight performance is further

discussed and demonstrated in Chapters 2 and 6 of this thesis. The deficiencies
associated with this flight-test method originated five research questions (RQ3, RQ4,
RQ5, RQ6 and RQ7), as presented in Subsection 1.4.3. A comprehensive discussion
about these deficiencies of the conventional method for level-flight performance
flight-testing is presented in Chapter 6 of this thesis

1.4 P ROBLEM S TATEMENT

The conventional flight-testing method for the evaluation of conventional

helicopter power available and power required for OGE hover and level-flight were
briefly presented in the preceding Section 1.3. A comprehensive discussion and

15
1 | INTRODUCTION

thorough demonstration of these methods, accompanied with authentic flight-test data

to highlight deficiencies, is presented in the subsequent Chapters 2, 3, 5 and 6 of this
thesis. This section lists seven deficiencies associated with the conventional
performance flight-testing method. These seven deficiencies constitute the problem
statement (PS’s) of this thesis, and derive the corresponding seven research questions
(RQ’s) this thesis attempts to address.

1.4.1 Available Power

The conventional method which is briefly described in Subsection 1.3.1 and

thoroughly demonstrated in Chapter 2 (Subsection 2.3.1) relies on the intrinsic
assumption of complete independency between all three limiting parameters of engine
output power (compressor speed, turbine gas temperature and fuel-flow). This
conventional method uses three independent, single-variable, empirical polynomials
for the purpose of predicting the maximum available power of the installed engine,
under a wide range of atmospheric conditions and engine power ratings (Eq.(2.34),
(2.35) and (2.36)). The available power predictions are frequently inaccurate and can
even contradict basic engineering rules. An example for this inaccuracy and
fundamental rules contradiction is provided in Chapter 3, Subsection 3.3.2. The poor
prediction performance of the conventional method can be mainly attributed to over
simplification of the gas-turbine engine output power, as a linear combination of
single-variable functions instead of the more realistic engineering problem it is, a
multivariable type of a problem. Power prediction based on the simplistic single-
variable approach fails to reveal the complex internal relationship between the power
limiting parameters of the engine, i.e., the compressor speed, the gas-turbine
temperature and the engine fuel-flow.

A designated BO-105 helicopter example presented and analysed in Chapter 3

demonstrates the deficiency of this conventional single-variable method. This BO-105
flight test sortie yielded a physically impossible behaviour that predicts a temperature-
limited engine to produce more power under a higher ambient-air temperature.

16
 1.4 | P R O B L E M S T A T E M E N T

Moreover, the power prediction errors using the three single-variable polynomials
(Eq.(2.34), (2.35) and (2.36)) are substantial and should be reduced (standard deviation
of 13hp, which is about 4% of the engine maximum continuous power).

PS1: The current flight-test method for the available power of gas turbine
helicopters is based on a simplistic single-variable approach (instead of a
multivariable approach) that often results in unacceptable power-prediction
errors and physically unrealistic available power modelling.

RQ1: Can a novel flight-test method be developed for the available power of a
gas-turbine helicopter, which demonstrates enhanced power prediction
accuracy as compared to the conventional method?

1.4.2 Power Required for OGE Hover

The conventional flight-test method for OGE hover performance is based on the
combined blade-element momentum theory (BEMT). This method is briefly described
in Subsection 1.3.2 and thoroughly demonstrated in Chapter 2 (Subsection 2.3.2), with
the supporting theory presented in Section 2.1. This flight-testing method seeks to find
an overly simplified empirical model to relate between two non-dimensional variables.
These are the coefficient-of-power (CP) and coefficient-of-weight (CW) as expressed by
Eq.(2.41). The current method fails to explicitly address significant non-linear effects
such as blades’ compressibility issues and power increase due-to drag-divergence.
These non-linear effects are more common for helicopters operating at high-
altitude/low air temperature conditions. Any non-linear effects measured during hover
performance flight-test sorties, are being averaged into one simplistic empirical model,
instead of being handled specifically and exclusively for their effects.

PS2: The conventional flight-test method for OGE hover performance is overly
simplified and does not account for rotor-blades non-linear effects. This
conventional flight-test method often yields empirical models that fail to

17
1 | INTRODUCTION

accurately and consistently predict the total power required to hover, under a
wide range of helicopter gross-weights and atmospheric conditions.

RQ2: Can a novel flight-test method for OGE hover performance of a

conventional helicopter, which demonstrates enhanced prediction accuracy as
compared to the conventional OGE hover method, be developed?

1.4.3 Power Required for Level Flight

The conventional flight-test method for helicopter level-flight performance is

based on a simplified equation that describes the power required to sustain a helicopter
in level-flight (Eq.(2.44)). This flight-test method is briefly described in Subsection
1.3.3 and thoroughly demonstrated in Chapter 2 (Subsection 2.3.3), with the
supporting theory presented in Section 2.2. Although widely used and being taught in
test-pilot schools around the world, this method incorporates several drawbacks which
not only make the execution of flight-test sorties inefficient and time consuming, but
also compromises the level of accuracy achieved. The following is a list of the five
deficiencies associated with the current level-flight performance flight-test method.
This list constitutes five problem statements (PS’s) that derive five corresponding
research questions (RQ’s):

PS3: The conventional flight-test method for level-flight performance fails to

specifically address non-linear effects such as blade-tip compressibility and
drag-divergence. This often results in inaccurate predictions for the power
required for level-flight, especially at high altitude and low air temperature
conditions.

RQ3: Can a novel flight-test method be developed for level-flight performance

of a conventional helicopter, which accounts for non-linear effects and
demonstrates enhanced prediction accuracy as compared to the current level-
flight method?

18
 1.4 | P R O B L E M S T A T E M E N T

PS4: The conventional flight-test method for level-flight performance entails

the executions of ‘speed-runs’ at constant coefficients-of-weights (Cw). This
makes the method inefficient, cumbersome, and time-consuming. Moreover,
the resulting empirical model is prone to elevated levels of inaccuracy since it
is merely a set of single power curves for constant coefficients-of-weight, rather
than a unified empirical model that accounts for the entire range of coefficients-
of-weights.

RQ4: Can a novel flight-test method for level-flight performance, which is more
convenient, efficient and time-saving than the current one, and produces a
unified empirical model for a range of coefficient-of-weights, be developed?

PS5: The conventional flight-test method for level-flight performance is based

on Glauert’s high-speed approximation, hence making it irrelevant for the low-
airspeed regime. The current flight-test method ignores the airspeed regime
from the hover to about 40 kts. (depending on the particular helicopter type and
configuration).

RQ5: Can a novel flight-test method for helicopter level-flight performance, that
also includes the low-airspeed regime, be developed?

PS6: The conventional flight-test method for level-flight performance of a

conventional helicopter incorporates no analytical means to account for the
helicopter centre-of-gravity location.

RQ6: Can a novel flight-test method for helicopter level-flight performance,

which includes analytical means to account for the centre-of-gravity location,
be developed?

PS7: The conventional flight-test method for level-flight performance requires

the flight-test crew to precisely adjust the main-rotor speed during the test. This
requisite makes the method unsuitable for types of helicopter that do not allow
for a trivial main-rotor speed manipulation by the flight-test crew.

19
1 | INTRODUCTION

RQ7: Can a novel flight-test method for helicopter level-flight performance,

which does not require adjustment of the main-rotor speed, be developed?

1.5 R ESEARCH G OALS AND O BJECTIVES

The general objective of this research is to develop new and improved flight
test methods for the available and required power of a conventional helicopter. This
general objective is further reduced to form the following set of particular and concise
objectives, as imposed by the limitations and the scope of the research.

(1) Develop an enhanced flight-test method to evaluate the available power of a gas-
turbine engine installed in a conventional helicopter. The proposed available-power
method shall present an improved prediction accuracy, as compared to the current
flight-test method. The proposed method shall rectify the identified deficiencies of the
current method (PS1), as specified by RQ1 in Subsection 1.4.1 above. The proposed
method shall demonstrate an improved prediction accuracy, as compared to the
current flight-test method.

(2) Develop an original flight-test method to evaluate the power required for OGE
hover of a conventional helicopter. The proposed OGE hover performance method
shall address and rectify the identified deficiencies of the current method (PS2), as
specified by RQ2 in Subsection 1.4.2 above. The proposed method shall demonstrate
an improved prediction accuracy, as compared to the current flight-test method.

(3) Develop an original flight-test method for the level-flight performance of a

conventional helicopter. The proposed level-flight performance method shall address
and rectify the identified deficiencies of the current method (PS3, PS4, PS5, PS6 and
PS7), as specified by RQ3, RQ4, RQ5, RQ6 and RQ7 in Subsection 1.4.3 above. The
proposed method shall demonstrate an improved prediction accuracy, as compared to
the current flight-test method.

20
 1.6 | R E S E A R C H L I M I T A T I O N S AND SCOPE

1.6 R ESEARCH L IMITATIONS AND S COPE

The research is limited to gas-turbine engine powered conventional type of

helicopters, i.e., those which employ a single main-rotor to generate lift and thrust and
a single tail-rotor to counter act the torque effect of the main-rotor. Other types of
helicopters that employ a different arrangement of rotors is not covered by this
research. The research required planning and execution of flight-test sorties to enable
comparison between the conventional and the proposed flight-test methods.
Dedicated flight-test sorties were limited to only two types of helicopters, the Bell Jet-
Ranger (Fig. 1.3) and the MBB (Messerschmitt-Bölkow-Blohm) BO-105 (Fig. 1.4),
normally used for training at the National Test Pilot School (NTPS) in Mojave,
California (https:\\www.ntps.edu). The comparison drawn between the proposed and
the current flight-test methods is mainly based on flight-test data gathered from these
two helicopters. A more detailed description of these two specially instrumented
helicopters used for the research is presented in Appendix C. - Research Helicopters
Description.

The scope of the research was limited to the execution of ten flight-test sorties,
one single sortie for the available-power, four sorties for the hover performance and
five sorties for level-flight performance flight-testing. The research sorties were
launched from one single geographic location (Mojave) and were conducted under the
prevailing atmospheric conditions. These research constraints restricted the varying
range of essential parameters affecting helicopter performance, such as ambient air
temperature, pressure altitude, centre of gravity location, etc.

The performance flight-test methods that were developed within this research
are valid and applicable for any type of a conventional gas-turbine powered helicopter,
configured with a single main rotor and a single tail rotor.

21
1 | INTRODUCTION

Figure 1.3. The Bell-Jet Ranger helicopter used for the research. Photo courtesy
of the National Test Pilot School.

Figure 1.4. The MBB BO-105 helicopter used for the research. Photo courtesy
of the National Test Pilot School.

22
 1.7 | M E T H O D S A C C U R A C Y C O M P A R I S O N

1.7 M ETHODS A CCURACY C OMPARISON

The main goal and objective of this thesis, as defined in Section 1.5 above, is
to develop new, more accurate performance flight-test methods, as compared to the
current conventional flight-test method. This section presents the procedure by which
the prediction accuracy achieved by each method, is evaluated and compared.

For the available power method (Chapter 3), the same flight-test data of 34
stabilized engine points are used to establish the empirical models for the engine
output power. First, using the conventional single-variable approach then by using the
proposed multivariable approach. Power estimation errors result from each method
are then compared for a trivial comparison. The more challenging task of the available
power testing is the estimation of engine maximum available power under various day
conditions. The maximum available power (continuous rating) of the helicopter is
estimated for various atmospheric conditions using each method. The physical
legitimacy of the two estimated maximum continuous power, under a wide range of
atmospheric conditions is then compared.

The prediction accuracy attained by each method for the required power
(Chapters 5, 6) is based on establishing an empirical model by using only a part of the
flight-test data. The empirical model is then used to predict the helicopter required
power under the conditions of the remaining flight-test data base, the data not used
for the empirical model. Power estimation errors arise from each method are
compared (literally), and by using hypothesis testing are projected from the particular
measured case to the general case. For the OGE hover performance method (Chapter
5) an empirical model based on the first three sorties is used to predict the hover
performance of Sortie 4. For the level-flight performance method (Chapter 6) a more
elaborated comparison scheme, in two tiers, is executed. The first tier is using flight-
test data from each sortie to predict the helicopter performance under conditions of
the other three sorties (referred to as the single-sortie approach). The second layer of
comparison is accomplished by predicting the helicopter performance in each one of
the four sorties by using an empirical model based on data taken from the other three

23
1 | INTRODUCTION

sorties. This second-tier comparison scheme is referred to as the cluster of sorties

approach.

1.8 T HESIS O UTLINE

The thesis is structured into seven chapters in order to answer the seven
research questions (RQ1 to RQ7 as specified in Section 1.4 above). This process of
addressing the seven RQ’s is illustrated in Fig. 1.5 and clarified hereinafter. In this
Chapter 1 (Introduction) the reader is briefly introduced to helicopter performance
theory and the relevant conventional flight-testing methods and their embedded
deficiencies. This preliminary description of the relevant helicopter performance and
the associated flight-testing methods is merely sufficient to allow the reader
understanding the major goals and objectives of this research.

Chapter 2 (Helicopter Performance Theory and Conventional Testing

Methods) provides a thorough discussion augmented with examples of the relevant
helicopter performance and the associated conventional flight-test methods. This
Chapter 2 serves as the literature review for the thesis and is crucial for a full
understanding of the succeeding chapters that propose an original and improved flight-
test method to address embedded deficiencies with the conventional flight-test
methods. The substantial portion of this research effort is presented in Chapters 3
through 6. Chapter 7 concludes the thesis by providing summary remarks, reiterate the
fulfilment of all research goals and objectives and offering recommendations for
further research. The following is a short description of each chapter (3 through 7) and
its role in the big scheme of the thesis:

Chapter 3 - A Multivariable Approach in Gas-Turbine Engine Flight-Testing.

This chapter presents an improved flight-test method for the evaluation of the
available power of installed gas-turbine engines in helicopters. The method called
‘Multivariable Polynomial Optimization under Constraints’ (MPOC) is demonstrated
using flight-test data from a BO-105 helicopter and exhibits a better prediction

24
 1.8 | T H E S I S O U T L I N E

capability, as compared to the conventional flight-test method for the available power.
Chapter 2 addresses research goal (1) of Section 1.5 and answers RQ1, as defined in
Subsection 1.4.1.

Chapter 4 - A Singular Values Approach in Helicopter Flight-Testing Analysis.

This chapter intends to further address research goal (1) of Section 1.5 by elaborating
on the process of choosing between empirical models. More specifically, it deals with
empirical multivariable polynomials that best represents the available power of a gas-
turbine engine. This chapter is intended to complement the MPOC method presented
and demonstrated in Chapter 3, and to provide the flight-test team a systematic and
repeatable methodology to choose in-between various empirical models for the task
of flight-test data representation. The proposed approach is based on the singular-
value-decomposition (SVD) concept and is demonstrated by using gas-turbine engine
flight-test data gathered from a variety of seven helicopters. One of the several
conclusions of this chapter is that although the SVD approach is demonstrated for
gas-turbine engine performance, it can and should be used for any type of helicopter
performance flight-testing analysis where empirical models are evaluated. The method
developed and presented in this chapter for gas-turbine engine performance,
undertakes an essential role in the development of new flight-test methods for the
hover and level-flight performance, as presented in Chapter 5 and Chapter 6
accordingly.

Chapter 5 – Hover Performance Flight-Testing Using Dimensionality

Reduction Approach. This chapter presents an improved flight-test method for the
evaluation of the power required to hover OGE. The method is based on concepts of
dimensional-analysis and dimensionality-reduction to construct an effective and
accurate empirical model for the OGE hover performance. The proposed method
called Corrected Variables Screening using Dimensionality Reduction (CVSDR) is
demonstrated using flight-test data from an instrumented Bell Jet-Ranger helicopter.
The CVSDR method displays a better prediction accuracy, as compared to the
conventional flight test method. Chapter 4 addresses research goal (2) of Section 1.5
and answers RQ2 as defined in Subsection 1.4.2.

25
1 | INTRODUCTION

Chapter 6 - Level Flight Performance Flight-Testing Using Dimensionality

Reduction Approach. This chapter presents an improved flight-test method for the
evaluation of the power required for level-flight. The method is based on similar
concepts used for the improved CVSDR hover flight-test method (Chapter 5) but
takes it further to higher dimensional space. The proposed CVSDR flight-test method
for level flight performance is demonstrated using flight-test data from an
instrumented MBB BO-105 helicopter. The proposed method is proved more accurate
and more efficient as compared to the conventional flight-test method for level-flight.
Chapter 6 addresses research goal (3) of Section 1.5 and answers RQ3, RQ4, RQ5,
RQ6 and RQ7 as defined in Subsection 1.4.3.

Chapter 7 – Conclusions and Recommendations. The last chapter of the thesis

opens with a conceptual discussion about the difference between the two flight testing
approaches, that of the conventional methods and that of the proposed methods.
Next, a concise list of 22 main conclusions drawn from the research is presented in
the context of the research goals and objectives. The chapter ends with few
recommendations about how this research can be further expanded.

26
 1.9 | T H E S I S P U B L I C A T I O N S

Figure 1.5. Thesis outline illustration. This flow-chart illustrates the structure of the thesis,
how the various chapters are interrelated and serve for answering all research questions (RQ’s).

1.9 T HESIS P UBLICATIONS

The thesis is based on papers that have been (or are to be) published in peer-
reviewed scientific journals and conference proceedings. The following is a list of the
relevant publications:

Chapter 3 is based on the following papers:

(1) Arush I, and Pavel M.D. “Helicopter Gas Turbine Engine Performance Analysis:
A Multivariable Approach.” Proceedings of the Institution of Mechanical Engineers,

27
1 | INTRODUCTION

Part G: Journal of Aerospace Engineering, Vol. 223, No. 3, March 2019.

https://doi.org/10.1177/0954410017741329

(2) Arush, I., & Pavel, M.D. (2018). “Flight testing and analysis of gas turbine engine
performance: A multivariable approach.” In C. Hermans (Ed.), Proceedings of the
44th European Rotorcraft Forum: Delft, The Netherlands, September 2018.

Chapter 4 is based on the following paper:

(3) Arush I, Pavel M.D; and Mulder M. “A Singular Values Approach in Helicopter
Gas Turbine Engines Flight Testing Analysis.” Proceedings of the Institution of
Mechanical Engineers, Part G: Journal of Aerospace Engineering 234, no. 12 (2020):
1851–65. https://doi.org/10.1177/0954410020920060.

Chapter 5 is based on the following paper:

(4) Arush I., Pavel M.D., and Mulder M., “A Dimensionality Reduction Approach in
Helicopter Hover Performance Flight Testing.” Journal of the American Helicopter
Society 67, no. 3 (2022): 129–41. https://doi.org/10.4050/JAHS.67.032010

Chapter 6 is based on the following paper:

(5) Arush I, Pavel M.D., and Mulder M. “A Dimensionality Reduction Approach in

Helicopter Level Flight Performance Flight Testing.” Journal of the Royal
Aeronautical Society, First View 13 July 2023. https://doi.org/10.1017/aer.2023.57

28
 The more you know, the more you know you don’t know.
Aristotle

2 H ELICOPTER P ERFORMANCE
T HEORY & C ONVENTIONAL
T ESTING M ETHODS

T his chapter elaborates on the relevant two disciplines of the helicopter power
required for hover and level-flight. The performance theory is demonstrated
using arbitrary helicopter types, making the abstract concepts more practical for the
reader to grasp. Once a sound foundation for the hover and the level-flight
performance has been laid down, the conventional flight-testing methods are
presented and illustrated by using authentic flight-test data.

2.1 H OVER P ERFORMANCE

The power consumed by the main-rotor (PM/R) of a hovering conventional

helicopter is comprised out of two main terms as presented in Eq.(2.1). The
relationship given by Eq. (2.1) involves the helicopter gross-weight (W), the ambient
air density (ρa), the main-rotor disk area (Ad), the main-rotor blade zero-lift drag
coefficient (Cd0), the main-rotor radius (R), the main-rotor angular velocity (Ω) and a
non-dimensional term (σR), defined as the ‘solidity-ratio’ of the main-rotor disk. As
implied by its name, the solidity ratio describes how solid is the rotor disk, i.e., the
relative disk area that is occupied by the blades. The first term of Eq.(2.1) is referred-

29
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

to as the induced-power (or the ideal-power) and was derived from the momentum
concept. This term which represents the amount of power required to generate lift (or
thrust) is comprised from the product of thrust (T), equals the helicopter weight (W)

in a hover, and the uniform induced velocity (Vih) across the disk (Eq.(2.2)). The

expression for the constant induced velocity across the disk is based on linear-
momentum conservation concept, first presented by Rankine in use of marine
propellers in 1865 [27], later refined by Froude in 1878 [28] and generalized by Glauert
in 1935 for aeronautical applications [29]. A detailed derivation of the uniform induced
velocity through the hovering main-rotor disk (Eq. (2.2)) was presented by Prouty [33].

W3 1
 Cd0  R  a Ad  R 
3
PM / R 
2  a Ad 8 (2.1)
profile
induced

T W
vih   (2.2)
2  a Adisk 2  a  R 2 

The second term in Eq. (2.1) is referred-to as the profile power and it
represents the amount of power required to overcome the viscous effects between the
main-rotor blades and the surrounding air. This second term is based on principles of
blade element theory (BET). One should recognise that Eq. (2.1) relies on few
simplifications such as a uniform induced-velocity distribution across the main-rotor
disk, a constant zero-lift drag coefficient (Cd0) and a constant chord length ( c ) along
the main-rotor blades.

Figure 2.1 (a) illustrates a hovering rotor-blade with a constant chord and
length of R, rotating at a constant angular velocity (Ω). An arbitrary blade element
distanced ‘r’ from the centre-of-rotation is subjected to a tangential velocity equals
(Ωr), and an induced velocity (Vih) which is assumed, for this preliminary approach,
constant across the hovering disk.

30
 2.1 | H O V E R P E R F O R M A N C E

Figure 2.1. The main-rotor blade in a hover flight. The blade is subjected to a tangential
velocity due to its rotation. The aerodynamic forces acting on each blade-element are integrated
and used to estimate the induced and profile power required by the rotor disk at a hover.

31
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

In hover, the aerodynamic forces acting on this blade element are illustrated in
Fig. 2.1 (b), (c) and are calculated as per the fundamental lift and drag terms for
incompressible flow (Eq.(2.3)).

The power required to overcome the profile-drag is defined as the profile

power and can be estimated by the mechanical work done by this profile-drag force
per unit of time. This becomes a simple multiplication between the elementary profile-
drag force (dD) and the tangential velocity as expressed in Eq. (2.4). Integrating along
the span of a single blade (from zero to the blade length, R) yields the profile power
of a single blade. The profile power for the whole rotor (Eq. (2.5)) is attained by
multiplying the single-blade profile power by the number of blades (b) resulting in the
profile power term presented in Eq.(2.1). Note that Eq.(2.5) is expressed by using the
solidity ratio of the disk (σR).

1 1
dL  a  r   cl  cdr   dD   a  r   cdo  cdr 
2 2
(2.3)
2 2

1
dPprofile  dD   r   a  r   cdo  cdr 
3
(2.4)
2

R
1 bc
Pprofile  b  dPprofile dr  Cd0  R a Adisk  R   R 
3
(2.5)
0
8 R

Referring back to the induced power term presented in Eq.(2.1), the reader
should appreciate it is attained by using a similar approach, as used for the profile
power. The elementary induced power is calculated by taking the product of the lift
(or thrust) and the induced velocity (Vih). Since this induced velocity is assumed
constant (for this simplistic approach where the actuator disk is used), integration
along the blade and accounting for the number of blades (b) becomes utterly simple.
The thrust generated by the disk which equals the helicopter gross-weight in a hover,
is multiplied by the constant induced velocity (Eq.(2.2)). This results in the exact
induced power term, as appears in Eq.(2.1).

32
 2.1 | H O V E R P E R F O R M A N C E

The induced power term neglects elementary engineering information regarding

the number of blades (b) and their shape. The uniform distribution of the induced-
velocity across the disk (Vih) represents the minimum possible induced power, or the
‘best-case-scenario’, as stated by Prandtl’s lifting-line theory [30]. According to this
1921 era theory, the necessary and sufficient condition for minimum induced drag (and
power) is that the downwash produced by the longitudinal vortices be constant along
the entire lifting line, i.e., a constant profile of induced velocity across the rotor disk
minimizes the induced power. This best-case scenario can be regarded analogous to
the elliptical lift distribution of a fixed-wing aircraft.

A more realistic analysis of the hovering disk, that considers the number of blades
(b) and their shape, was first presented by Gessow in 1948 [31]. This analysis, known
as the combined blade element momentum theory (BEMT), treats the actuator disk as
a combination of infinitesimally thin annuli (rings) of constant induced velocity profile.
The practicality of this approach is that the induced velocity through the disk is only
considered constant with respect to the blade station (r), regardless of the azimuth
angle (ψ). This approach is more accurate compared to the previous analyses [27-29]
that assume a uniform induced velocity throughout the disk area.

The quadratic equation in induced velocity (Eq.(2.6)) arises from equating the
elementary thrust generated by an infinitesimally thin ring of the disk, using the two
analytical approaches of conservation of linear momentum and blade-element theory
(BET), illustrated in Fig. 2.1(c). The BET assumes a symmetrical blade cross-section
and estimates the angle-of-attack (α) of each blade cross-section as the difference
between the pitch angle (Θ) and the induced angle (Φ). This quadratic equation is then
solved for a constant-chord blade, providing an explicit expression for the induced
velocity through the hovering disk (Eq.(2.7)), for any blade station (r), as measured
from the blade centre of rotation.

bccl bcr2cl
vi2  vi   0  r  [0,R] (2.6)
8 8

33
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

bccl  32 
vi (r )   1  1  r   r  [0,R] (2.7)
16  cbcl 


It follows from Eq.(2.7) that maintaining a constant value for the product of
the blade station and the pitch angle (r∙Θ) along a constant-chord blade, ensures a
uniform induced velocity distribution across the hovering disk. This, according to
Prandtl’s lifting-line theory, guarantees the minimal possible induced power in a hover
[30]. One can easily identify the widely known ‘ideal’ blade twist profile (Eq.(2.8)) is
merely this requirement for keeping the product of blade station (r) and blade pitch
angle (Θ) constant along the span of the blade. Rotor blades with constant-chord and
a variable pitch angle that follow the ‘ideal’ blade twist profile (Eq.(2.8)) yield a constant
induced-velocity across the hovering disk which can be calculated precisely by using
Eq.(2.2).

R
( r )  tip  r  R (2.8)
r

As stated by Prouty [32], it is impractical to manufacture rotor blades with the

‘ideal’ geometric twist profile due to the pitch angle tends to infinity close to the root.
Instead, most conventional blades are made with a variable pitch angle that follows a
linear change (angle reduction of Θ1) from the root to the tip. Although, not the ‘ideal-
twist’ this linear geometric twist defined in Eq. (2.9) offers a practical method to
substantially reduce the induced power of the helicopter at a hover. The other pitch
angle (Θ0) in Eq. (2.9) represents the collective pitch angle, commanded by the pilot.
Prouty [32] states that in-general, it is fair to say that high values of main-rotor blade
twist (Θ1) produce good hover performance. This also reflects on the primary mission
the helicopter was designed for. When designing blades, variations on the conventional
linear twist distribution should be considered only for special reasons [32].

r
  0  1 1  0, r  R (2.9)
R

34
 2.1 | H O V E R P E R F O R M A N C E

An arbitrary example helicopter is selected for illustration purposes. This

hovering helicopter weighs 18,000 lbs., has a 4 bladed main-rotor disk with an area of
about 2,262 ft2 and a blade tip speed of 725 fps. The blades have a constant chord
length of 1.73 ft. and a linear pitch reduction (geometric twist) of 18° (Θ1=−18°). The
coefficient-of-lift to AOA (α) line slope is approximated as 2π rad-1. The collective
angle can be controlled by the pilot in a range between 9.5° to 25.9°. This constitutes
the ‘collective-stick’ range provided to the pilot. Figure 2.2 presents the calculated
parameters required to support the hover using the two types of blades twist, the first
using the theoretical ideal-twist and the other by using the actual -18° linear twist.
Numerical solution suggests that with a linear-twist of -18° the pilot must command a
collective pitch angle (Θ0) of about 22.6° to generate the precise amount of thrust
needed for the OGE hover. This collective pitch angle command of the example
helicopter is achieved by using about 79% of the total collective-control throw.

As shown in the left graph of Fig. 2.2, there are two crossings in pitch angle
between the two types of geometric pitch schedules. The first is at a blade station of
about 0.39 and the other at about 0.86. The theoretical pitch angle at the tip of the
blade is 6.1° for the ideal-twist profile and 4.6° for the linear-twist. The induced
velocity through the disk at the various blade stations is also presented (top-right graph
of Fig. 2.2). While the ideal blade twist produces a constant induced velocity of about
41 fps. across the disk, the -18° linear twist profile yields a variable induced velocity,
with a maximal value of about 45.4 fps. at the 0.63 blade station. The blade thrust-
distribution in each case (ideal and linear) is calculated using Eq.(2.10), and presented
in the bottom-right graph of Fig. 2.2. For simplicity, this estimation neglects blade tip
loss. While the ideal blade twist yields a triangular thrust distribution, the -18° linear-
twist generates a non-linear thrust distribution that peaks at a blade station of 0.8 with
a maximal calculated value of 296.6 lbs./ft.

35
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

Figure 2.2. The ideal and a -18° linear blade twist in an OGE hover. The two blades pitch
schedules are presented in the left. On the top-right a comparison between the induced velocities
is calculated. The bottom-right graph provides a comparison between the cross-sectional thrust
generated by each type of geometric twist.

dT 1  vi 
 a c  r  cl  
2
 (2.10)
dr blade 2  r 

Next, the induced power required for each case of geometric twist is calculated
and then compared. The induced power of each blade cross-section is calculated by
taking the product of the thrust and the induced velocity for each blade element.
Integration along the blade produces the induced power required for a single blade.
The induced power for the entire main rotor is attained by accounting for the number
of blades (b). This procedure is presented as Eq.(2.11). Notice that for the ideal-twist
case, this integral is immediately reduced to a simple multiplication of the (constant)
induced velocity by the total thrust produced by the rotor.

Figure 2.3 presents the estimated induced power required to support the
specific OGE hover example, using the -18° linear-twisted blades rotor system. The
bottom and top right graphs of Fig. 2.3 present, separately, the two factors of the
induced power. These are the induced velocity distribution and the thrust distribution

36
 2.1 | H O V E R P E R F O R M A N C E

as a function of the blade station. The left graph of Fig. 2.3 presents the induced power
distribution which peaks at the 75% blade station with a value of 23.6 hp/ft.
Integrating the power distribution along the blade provides an induced power of
342 hp per blade, and a total induced power of 1,368hp for the 4 bladed main rotor.
In the case of ideally twisted blades, the required induced power is only 1,339 hp. As
expected, a lower value compared to the linear-twist case, but not by much. For the
particular hover case illustrated, the -18° linear-twisted blades require an induced-
power increase of about 2.2% (29 hp), as compared to the ideal-twisted blades.

R
Pi  b  vi dT (2.11)
0

Figure 2.3. Induced power of a -18° linear-twisted blade in an OGE hover. The two factors
of the induced power are presented on the right, the thrust distribution along the blade (top
graph) and the induced velocity distribution (bottom graph). The left graph presents the induced
power distribution along the span of the blade. The area under the curve represents the induced
power required by the blade, under the specific OGE hover conditions.

The main rotor induced power in a hover is considerably larger than the profile
power required by the blades (more than double). The ratio between the induced and
the total power required for the main rotor at a hover is known as the ‘Figure of Merit’
(FM) of the rotor-system, as presented in Eq. (2.12)[33-35]. The FM, a value bounded

37
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

between zero and one, is used by helicopter designers to articulate about the
aerodynamic efficiency of the hovering rotor. The higher the value is, the more
efficient the rotor is for the hovering conditions. Typical maximum FM values for
actual rotor systems at a hover range from 0.75 to 0.8 [36-38].

The relative quantities of induced and profile powers in hover are not constant
and vary with altitude. As the hovering altitude increases, the air density decreases
resulting in opposite tendencies of the induced and the profile power portions; the
induced increases while the profile decreases. Figure 2.4 presents a typical breakdown
of the power required to OGE hover by a 22 ft. in radius and 6.5% solidity-ratio (σR)
main rotor of an arbitrary 8,500 lbs. helicopter. This estimated power is presented for
a standard day condition (ISA), from sea-level to 10,000 ft. of pressure altitude. Mind
this power is for the main rotor only. The total power required to hover (for the entire
helicopter) can be estimated by accounting for the mechanical efficiency (ηm) of the
specific helicopter at a hover. Typical values for conventional helicopter mechanical
efficiency at an OGE hover are around 0.85, as stated by Richards [24]. Figure 2.3 also
presents the theoretical FM and the induced to profile powers ratio for the entire
altitude range between sea-level and 10,000 ft. of pressure altitude (standard day
conditions).

Pinduced 1
FM   (2.12)
Pinduced  Pprofile 1   Pprofile Pinduced 

38
 2.1 | H O V E R P E R F O R M A N C E

Figure 2.4. Main rotor power components in an OGE hover. Also presented are the
theoretical figure-of-merit (FM) and the ratio of the induced to profile powers.

From the data presented in Fig. 2.4, under standard day conditions and for
pressure altitude of 5,000 ft., the required induced power is about 571 hp, the required
profile power is about 191 hp., total power for the single main rotor is 762 hp and the
total power required to hover under the assumption of 85% mechanical efficiency is
about 896 hp. The estimated ratio between the induced and the profile powers is 2.99.
The main rotor FM under the conditions above is estimated to be 74.9%. The practical
interpretation is that 74.9% of the power consumed by the main rotor is directed
towards a beneficial purpose of creating lift. The rest 25.1% is just a waste of power
since it served to overcome the viscous effects between the blades and the air.

It is worth emphasizing that data presented in Fig 2.4 are a theoretical

estimation based on few assumptions and approximations. The precise FM is
dependent on the actual induced and profile powers of the main rotor. The induced
power used in Fig. 2.4 is based on an ideal main rotor, characterised by constant
induced velocity across the disk (Eq.(2.2)) which yields the minimum possible induced
power (that is the best-case scenario from an operator standpoint). This estimation
serves as a minimum bound for the induced power. Higher values of induced power
(profile power remains constant) would increase the actual FM of the rotor system.

39
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

Moreover, the estimated profile power for Fig. 2.4 is based on a constant zero-lift drag
coefficient (Cd0) of 1% for all blades. This means that all cross-sections of the main-
rotor blades have a constant 0.01 zero-lift drag coefficient. In reality, this value is not
constant and increases (much) with both angle of attack and Mach number of the blade
cross-sections. An increase of the profile power component (while keeping the
induced power constant) would result in FM reduction.

Another ratio with high significance to helicopter hover performance is the

disk loading (DL). This dimensional parameter defined in Eq. (2.13) has units of
pressure and indicates how loaded the rotor disk is. The higher the DL value is the
higher the induced velocity through the rotor disk is. Since, the thrust (T) produced by
the main-rotor equals the gross-weight of the helicopter in a hover it can be
represented as the ratio of the helicopter gross-weight (W) to the disk area (Ad). It can
be shown using the DL definition and the induced velocity through an ideal rotor disk
(Eq.(2.2)) that the DL serves as the theoretical dynamic pressure of the airflow under
the disk (the ‘down-wash’ or ‘wake’ of the rotor). Actual measurements of non-
uniform distributed induced velocity rotor in a hover show the local dynamic pressure
may be significantly higher (more than double) than the DL [39]. Nevertheless, an
important conclusion that can be drawn from this is that the dynamic pressure
underneath a hovering helicopter is related to the DL (helicopter gross-weight and
main-rotor disk area), regardless the atmospheric surrounding parameters. Figure 2.5
presents maximum weight DL values of twelve different types of helicopters under
standard sea level conditions. The maximum DL values range between 2.8 psf. (small
helicopters like the Robinson 22) to about 14.3 psf. for large transport helicopters like
the MI-26.

40
 2.1 | H O V E R P E R F O R M A N C E

Figure 2.5. Maximal disk-loading values for various helicopters. The maximal disk loading
(DL) values of helicopters varies from as low as 2.8 psf. for a Robinson-22 helicopter to high
values of over 14 psf. for large and heavy helicopters like the CH-53E and the MI-26.

T W
DL   (2.13)
Adisk  R 2

The power loading (PL) of a hovering disk is defined as the ratio between the
thrust produced by the rotor and the power it consumes (Eq.(2.14)). By consolidating
information from the three equations, Eq.(2.2), (2.12) and (2.13), the power loading
(PL) of the hovering main-rotor disk is expressed in terms of disk loading (DL), figure
of merit (FM) and ambient air density (Eq.(2.15)). The PL, yet another significant ratio
for hover performance, can be regarded as the power efficiency for hovering. This
dimensional ratio provides information about how many thrust units can the hovering
rotor produce for a unit of power demanded. More generally, it can be seen from
Eq. (2.15) that for a given FM, the power loading and the disk loading are inversely
proportional to each other. This means that a highly loaded disk cannot be a power
efficient hovering device. The inversely proportional relationship between the PL and
the DL is demonstrated in Fig. 2.6 for an imaginary rotor disk under standard sea-level
conditions, for three distinct FM values (0.7, 0.75, 0.8) for a range of DL between 3 to
15 psf. The following is a practical interpretation of Fig 2.6; given an arbitrary main
rotor with a FM of 0.75 and a disk area of 2,260 ft2. Using this rotor system to stabilize

41
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

a 15,820 lbs. helicopter in a hover (OGE) requires an amount of power of 1,472 hp.
to be provided to the main rotor only. This procedure involved calculating the relevant
DL value of 7 psf. (15,820 divided by 2,260) and using the 0.75 FM curve of Fig 2.6
to read the corresponding PL value of 10.75 lbs. to hp. The main rotor generates thrust
that equals the gross weight at a hover, i.e., a value of 15,820 lbs., therefore the power
required for the main rotor is about 1,472 hp. (15,820 divided by 10.75). Mind this
amount of power is required for the main rotor only. Projecting from the main rotor
onto the entire helicopter, requires information about the mechanical efficiency (ηm)
of the helicopter in a hover (OGE). By assuming a mechanical efficiency of 85% an
OGE hover power estimation of 1,732 hp. is yielded. A smaller size rotor-system with
a similar aerodynamic efficiency (0.75 FM) would require more power to sustain an
OGE hover flight. For example, decreasing the disk area by 20% would increase the
power to hover by about 11.8%, all other variables are kept constant.

T
PL  (2.14)
PM / R

FM 2  a
PL  (2.15)
DL

42
 2.1 | H O V E R P E R F O R M A N C E

Figure 2.6. The power-loading (PL) and the disk-loading (DL) relationship. The
inversely-proportional relationship between the PL and the DL of a rotor system is presented
for three distinct values of figure-of-merit (FM) and under standard sea level (SSL) conditions.

It is common practice and knowledge that hovering a helicopter in close

proximity to the ground is more efficient in terms of power required to sustain the
flight. The ground imposes an external constraint on the induced velocity through the
rotor system, which reduces the induced power component of the total power required
to hover for the same amount of thrust generated. Cheeseman and Bennett [40]
modelled the effect of the ground on the flow through the disk. The ground effect is
simplified for the hovering flight and presented as Eq.(2.16). A graphical illustration
of Eq. (2.16) is provided in Fig. 2.7. It follows from Eq. (2.16) that hovering at an
example height above the ground of 40% of the rotor diameter (distance between the
rotor plane and the ground), the induced power required to sustain the hover reduces
by 9.8% compared to the out of ground effect case. Moreover, this power reduction
estimation, Eq. (2.16)can be used to estimate the practical hover height for which the
helicopter gets out of ground effect. This IGE/OGE transition height is widely
considered above 1.2 times the rotor diameter, for which the reduction in the induced

43
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

power falls below 1%. This 1% reduction in power is practically undetectable to the
flight crew. This practical and useful relationship given by Eq. (2.16) has been validated
numerous times in more recent research work and performance flight testing
campaigns [41-43].

dvi Adisk
 (2.16)
viOGE 16  ZG 2

Figure 2.7. The theoretical ground effect on the induced power in a hover. This chart
demonstrates the theoretical reduction in the induced power of the rotor system (as compared
to the out-of-ground effect case), while hovering in close proximity to the ground. Data
presented are based on Eq.(2.16).

2.2 L EVEL F LIGHT P ERFORMANCE

The power consumed by a conventional helicopter in level-flight (Plvl) is

composed out of three major segments, as presented in Eq.(2.17). This equation can

44
 2.2 | L E V E L F L I G H T P E R F O R M A N C E

be regarded as an expansion of the power equation for the hover (Eq.(2.1)), once a
forward motion of the helicopter is initiated. The forward velocity of the helicopter
(VT) alters the two power terms already familiar from the hover flight (Induced and
Profile) and introduces a new power-term, the parasite power. The induced power
terms in Eq.(2.17) is an approximation that is only valid above a certain (vague and
unclear) airspeed. It is obvious this induced power term cannot be used for the hover
and low airspeed regime, since it will tend to infinity. The parasite power arises from
the aerodynamic drag of the helicopter fuselage. The expression ‘helicopter-fuselage’
in the context of the parasite power means any helicopter structural element excluding
the main and the tail rotors.

Equation (2.17) introduces a new non-dimensional variable (μ) that represents

the non-dimensional velocity of the helicopter, also known as the ‘advance-ratio’. This
non-dimensional variable is defined in Eq.(2.18) as the ratio between the helicopter
true airspeed (VT) projected onto the main-rotor tip path plane (TPP), and the main-
rotor blade tip tangential velocity at a hover (ΩR). The main-rotor TPP, also known
as the ‘plane of no flapping’, and the TPP angle of attack (αTPP) are illustrated in
Fig. 2.8. The TPP is that plane spanned by the tips of the blades and the TPP angle of
attack is defined as the angle between the helicopter velocity vector and the main-rotor
tip path plane. By assuming small αTPP and applying the small angle approximation, the
TPP projected true airspeed can be used as the true airspeed itself and the advance-
ratio (μ) converts into a more practical form as expressed by Eq.(2.19).

Plvl 
W2 1
 1
 Cd0  R  a Adisk  R  1  k p  2   a f eVT 3
2  a Adisk VT 8
3
2 (2.17)
induced profile parasite

VT cos(TPP )
 (2.18)
R

VT
  for cos(TPP )  1 (2.19)
R

45
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

Figure 2.8. The main-rotor tip path plane (TPP). The main rotor TPP is spanned by the tips

of the blades. The TPP angle of attack ( αTPP) is assumed small enough so that the projection of
the helicopter true airspeed (VT) onto the TPP can be approximated as the true airspeed itself.

Figure 2.9. The power curve of a helicopter in level-flight. The power required to sustain a
conventional helicopter in level-flight is comprised out of three main components, the induced
power (reduced with airspeed increase), the profile power (increases moderately with airspeed
increase) and the parasitic power which increases rapidly with airspeed increase. The power curve
demonstrates a local minimum point, known as the ‘bucket’, which corresponds to the airspeed
for minimum required power for level-flight.

46
 2.2 | L E V E L F L I G H T P E R F O R M A N C E

Figure 2.9 presents the general behaviour of all power terms and the total
power required against the airspeed of an example helicopter in level flight. Mind this
plot is not based on the approximated induced power term of Eq.(2.17) since this high-
speed approximation is irrelevant for the low-airspeed regime. The induced power for
this plot is based on the induced velocity form of Eq. (2.20) multiplied by the example
helicopter gross-weight.

2.2.1 The Induced Power in Level Flight

The first term in Eq. (2.17) is the induced power. This term is based on a simple
multiplication of the thrust generated by the rotor-disk, which equals the gross-weight
in level-flight, by the induced velocity through the rotor disk. The induced velocity
through the cruising disk, as appears in this term, is slightly more obscured than the
thrust produced and is based on Glauert’s ‘high-speed’ approximation. Glauert [46]
treated the main-rotor as an actuator-disk and by applying concepts of conservation
of linear momentum, accompanied with the lenient assumption of a uniformed-profile
induced velocity across the disk, the expression for the induced velocity in level-flight
was developed (Eq.(2.20)). Notice that Eq. (2.20) is also applicable for the hover case,
where the helicopter true airspeed (VT) is zero. For this case the induced velocity
reduces to the induced velocity at a hover (Eq.(2.2)).

This development of the constant induced velocity across the disk in level flight
is repeated by Leishman and Prouty [34,47] and its outcome (Eq.(2.20)) is known as
the constant momentum induced velocity (CMIV). Prouty [47] comments that a
more realistic view of the induced velocity through the disk in level flight should be
based on a complex vorticity pattern, consisting of trailing, shed, and bound vortex
elements associated with the lift and the change of the lift on each blade element. This
type of rigorous approach for the induced velocity across the rotor-disk in cruising
flight is presented by Vil’dgrube et al. [48]. That being said, Prouty [47] clarifies this
complexity and rigorous approach is of great importance when studying blade loads
and vibrations problems, but for performance calculations the use of the constant

47
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

momentum induced velocity represents the average of the complex velocity field and
provides reasonable accurate results.

2 2
VT2  VT2   W 
vi _ lvl (CMIV )        (2.20)
2  2   2  a Adisk 

The constant momentum induced velocity (Eq.(2.20)) can be further simplified

for cases where the induced velocity through the disk is negligible compared to the
true airspeed the helicopter is flying at (VT). The same momentum analysis approach
is applied, but this time the true airspeed the helicopter flies at is used as the velocity
the actuator disk is subjected to, neglecting any alteration caused by the induced
velocity. This yields Eq. (2.21) known as the ‘Glauert’s high-speed approximation’
which can be used for cases where the true airspeed the helicopter flies at are ‘much’
higher than the induced velocity through the cruising rotor. Mind that the first term of
Eq. (2.17) is precisely the product of the lift force the rotor disk is generating (equals
the gross-weight) and the high-speed approximation for the induced velocity in level-
flight (Eq.(2.21)).

W vih2
vi _ lvl ( HS )   (2.21)
2 a AdiskVT VT

One should be asking for a more precise definition for the true airspeed from
which the ‘high-speed’ approximation is valid. The previous loose guidance of when
the helicopter true airspeed is ‘much’ higher than the induced velocity through the
rotor disk is vague and impractical. Leishman [34] provides a definite criterion for the
practical validity of the high-speed approximation. This criterion is set for the advance
ratio (μ) to be larger than 0.1. A brief observation of the two methods for calculating
the induced velocity (Eq.(2.20),(2.21)) reveals that by using the advance-ratio value as
a criterion for the high-speed approximation validity, the user will be faced with
inconsistent approximation errors. The advance-ratio alone cannot be used as a
validity-criterion for the high-speed approximation. A particular helicopter can be

48
 2.2 | L E V E L F L I G H T P E R F O R M A N C E

flown under a wide range of disk loading (DL) values that alters the relative proportion
of the induced velocity calculated in both ways, the CMIV and the high-speed
approximation. Figure 2.10 presents a comparison between the two methods for
calculating the induced velocity through the main-rotor disk for six example
helicopters at their maximum certified disk loadings and under standard sea-level
conditions. Leishman’s advance-ratio criterion (μ equals 0.1) is shown on all plots.
Figure 2.10 shows that applying the same threshold of μ equals 0.1 (translated to a
narrow range of true airspeeds between 39 to 43 kts. for the six example helicopters)
resulted in erratic differences between the two calculated induced velocities. This is
attributed to the different disk loadings of the example helicopters.

Figure 2.10. The theoretical induced velocity in level flight. Data are presented for six
example helicopters. Data are calculated using two methods. One according to the constant
momentum induced velocity (CMIV) approach, and the other using the high-speed (HS)
approximation.

The estimation-error using Glauert’s high-speed approximation as compared

to the CMIV approach is calculated per Eq.(2.22) and presented for the six example
helicopters in Fig. 2.11. Note the estimation-errors presented in Fig. 2.11 were
calculated for standard sea-level conditions. While using μ equals 0.1 as a criterion
yielded acceptable estimation errors for few helicopters (below 2.3%), this criterion

49
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

was found unacceptable for the BO-105, UH-60 and the MI-26 helicopters. The
resulted unacceptable estimation errors for the three helicopters were 3.9%, 6.4% and
12.1%, respectively.

An improved tool to assess the validity (or the level of inaccuracy introduced)
of the high-speed approximation under various helicopter conditions was developed
and is presented in Fig. 2.12. The estimation error associated with the approximated
induced velocity, as calculated by Eq.(2.22), is presented against the helicopter true-
airspeed for various discrete disk-loading values, ranging from 2 to 15 psf., and for
standard sea-level conditions. Figure 2.12 also includes an illustrated example to assess
the minimum airspeed for which the high-speed approximation is valid, given a 5 psf.
disk-loading and an allowed estimation error of up to 2%. Using Fig. 2.12 the minimum
true airspeed is 42.5 kts. It is clear from Fig. 2.12 that for a given estimation error, the
minimum airspeed for the validity of the high-speed approximation is proportional to
the helicopter disk-loading. While a true airspeed of 33 kts. is valid for an acceptable
2% estimation error for a low 3 psf. disk-loaded helicopter, the minimum valid airspeed
for a 14 psf. disk-loaded helicopter is 71 kts., for the same error budget of 2%.
Establishing a legitimacy criterion for the high-speed approximation which is solely
based on the advance ratio, as given by Lishman [34], is simply unacceptable. A similar
type of graph to Fig. 2.12 but for 10K ft. of pressure altitude (standard day conditions)
is provided in Appendix B of this dissertation.

 vi _ lvl ( HS )  vi _ lvl (CMIV ) 

ErrHS  100   (2.22)
 vi _ lvl (CMIV )
 

50
 2.2 | L E V E L F L I G H T P E R F O R M A N C E

Figure 2.11. The error induced by using the high-speed approximation. This chart
demonstrates the percentage of error caused by using Glauert’s high-speed (HS) approximation.
The error is calculated with respect to the induced velocity based on the constant momentum
induced velocity (CMIV) method. Data are presented for six example helicopters and under
standard sea-level (SSL) conditions.

Figure 2.12. High-speed (HS) approximation validation chart. This fan-type chart can be
used as a tool for assessing the acceptable minimum true-airspeed for the high-speed
approximation, given the disk-loading and the required estimation error. This chart is applicable
for standard sea-level (SSL) conditions.

51
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

2.2.2 The Profile Power in Level Flight

The second term in Eq. (2.17) is the profile power. This term is merely an
extension of the profile power component from the hover (see Eq.(2.1)) but with a
correction-term that makes it relevant to the level flight regime. This correction term
is based on an empirical coefficient (kP) multiplied by the advance-ratio squared (μ2) to
represent the proportional rise in profile power with the level-flight airspeed. Since this
kP is an empirical coefficient, it also accounts for the profile power of the tail-rotor.
Typical values for kP vary between 4.65 [8,10,18-20] to a value of 4.7 presented by
Stepniewski [49].

Taking the main-rotor from the hover into a forward flight disrupts the
symmetric velocity-field surrounded the rotor blades. In the hover, each one of the
main-rotor blades senses the same chord-wise velocity profile, regardless of the
azimuth angle of the blade (ψ). Once the rotating main-rotor is flooded with the
forward flight airspeed, the chord-wise velocity profile of each blade varies based on
the azimuth angle of the blade. The chord-wise velocity (VCW) each cross-section of
the blade senses is expressed in Eq. (2.23) as a function of the true airspeed of the
helicopter (VT), the azimuth angle (ψ) and the station of the blade (r).

VCW (r , )  r  VT sin    r  [0, R] (2.23)

The main-rotor disk in forward flight can be divided into two halves: the
advancing and the retreating sides. For a rotor system rotating in a counter clockwise
orientation (as viewed from above), the advancing side (azimuth angles between zero
and 180°) is on the right. Rotor blades on the advancing side are exposed to a relative
velocity profile which is higher than the tangential velocity, results from the angular
motion of the rotor. The advancing side is subjected to higher relative velocities as
compared to the retreating side of the disk. The advancing blades are exposed to the
tangential velocity resulting from the angular motion of the blade, added with the
chord-wise component of the helicopter airspeed. For the retreating blade it is the
opposite, the chord-wise component of the helicopter airspeed is subtracted from the

52
 2.2 | L E V E L F L I G H T P E R F O R M A N C E

tangential velocity of the blade. This asymmetry is the cause for the formation of three
main areas on the main-rotor disk in forward flight. (1) A circular-section area on the
advancing side, which is subjected to high Mach number and compressibility issues.
This area stretches from the tip of the disk and grows inboard, as the helicopter
increases its airspeed. (2) A reverse-flow area in which all blade elements face an air
flow from the trailing edge to the leading edge. This reverse flow area, located on the
retreating side of the disk, is a perfect circle with a diameter of exactly the product of
the advance ratio (μ) and the main-rotor radius (R). The centre of the reverse flow area
is always located at an azimuth angle of ψ=270°, regardless of the helicopter airspeed.
(3) The blade-stall area on the retreating side of the disk. This area stretches from the
edge of the disk inbound and grows with helicopter airspeed increase.

Figure 2.13 presents the theoretical compressibility and reverse flow areas of
an example main-rotor disk in level flight. This example uses the formerly 22 ft. in
diameter example helicopter for three distinct values of advance-ratio (0.1, 0.3, and
0.5) and two types of atmospheres; standard sea-level and standard-day 10K ft.
pressure altitude. All calculations for Fig. 2.13 are based on the example helicopter
standard angular speed (Ω) of 324 RPM (33.9 radians per second.). The Mach number
each blade cross-section is subjected to (MBE) is calculated from Eq.(2.23) divided by
the applicable speed of sound (Eq.(2.24)). The same information is used to annotate
the constant Mach lines in Fig. 2.13. For this, the blade station (r) is explicitly expressed
as a function of the desired Mach number, the applicable speed of sound, the advance-
ratio and the azimuth angle (Eq.(2.25)).

a   air RairTa (2.24)

M BE a
r   R sin   (2.25)


53
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

Figure 2.13. An example main-rotor disk in forward flight. An example main-rotor disk is
subjected to two atmospheric conditions and three distinct advance ratios. The reverse-flow area
grows proportionally to the advance ratio. The advancing side of the disk faces higher Mach
numbers with increase in altitude (decrease in speed of sound) and with increase in advance ratio.

The profile power in level-flight (Eq.(2.17)) assumes constant profile drag

coefficient (Cd0). This assumption is valid for an airflow-regime free of compressibility
effects, i.e., characterised by Mach numbers below the drag divergence Mach number.
Figure 2.13 presents the estimated maximal Mach numbers exist on an example main
rotor, under fairly moderate atmospheric conditions. Flying on a standard day, 10K ft.
pressure altitude at advance-ratios above 0.3 clearly places a substantial circular section
of the main-rotor disk above 0.8M and inside the drag-divergence region of many
aerofoils [47,50-53]. This drag-divergence region is characterized by an abrupt increase
in the profile drag coefficient due to the formation of shockwaves, responsible for
significant drag increase. Flying the helicopter under more extreme conditions of
higher altitudes and lower ambient temperatures, would increase the disk area
subjected to compressibility issues, resulting in a steep increase in the profile power.
As stated by Prouty [47] “if any blade element exceeds the drag divergence Mach

54
 2.2 | L E V E L F L I G H T P E R F O R M A N C E

number for its aerofoil, the power required will be higher than that calculated from the
closed-form equations” (p. 177).

The two particular zones on the rotor disk, the circular-section infested with
compressibility-effects on the advancing side, and the blade-stall area on the retreating
side are the reason for the practical (low) airspeed limitation of the conventional
helicopter. As the airspeed increases on the advancing blade, certain blade elements on
the outbound part of the blade might experience critical Mach values associated with
formation of shockwaves on their upper and lower surfaces. This sudden change in
the flow regime around those blade element causes a jump in the aerodynamic centre
of the relevant cross-sections and a pitch-down moment. Mind this chain of events,
formation, and deformation of shock-waves, repeats periodically on a rotating rotor
disk. The phenomenon briefly described here is commonly known from fixed-wing
airplanes as the ‘Mack Tuck’. A flight-test campaign on the Sikorsky NH-3A (modified
Sea-King, SH-3A) compound helicopter was aimed to investigate flight characteristics
of a helicopter in the high airspeed regime. The flight-test team encountered an
unstable blade-twist singularity due-to the Mach Tuck phenomenon. This severe
dynamic problem described by Paul [54] was the cause for elevated blade-loads above
a tip Mach number of 0.92. This value is commonly considered as the practical
limitation for the maximum tip Mach number a main rotor-disk can endure in forward
flight. Recalling the maximal Mach number a generic blade section senses is the tip at
the 90° azimuth angle, this practical criterion is expressed as Eq.(2.26).

VT  R
M tip (max)   0.92 (2.26)
 air RairTa

The retreating side of the rotor-disk is subjected to an increasing zone of blade

stall. Increasing the helicopter airspeed results in lower relative speeds on the outboard
sections of the retreating blades. This combined with the flap-down motion of the
retreating blade (ψ between 180° to 270°) makes it more susceptible for stalling. There
are many types of stall characteristics based on the shape of the aerofoil (thickness,
thickness to chord ratio, symmetry, etc.) and the prevailing non-dimensional variables

55
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

of the flow (Reynolds and Mach numbers). A thorough discussion of blade stall is
beyond the scope of this dissertation and can be found in Ref [55, 56]. The outboard
elements of the retreating blade are subjected to a special type of stall, referred to as
the dynamic stall. It was found that when the AOA on an aerofoil is increased rapidly,
it can postpone the stall AOA and temporarily generate a higher coefficient of lift, as
compared to the static or the quasi-static case. This phenomenon is thoroughly
explained and demonstrated in Ref [57-61]. Mind the dynamic aspect of the rotor
blades stall can arise from (1) the flapping motion of the blade (defined as ‘plunge’
motion in the context of dynamic stall. (2) the commanded pitch oscillations. (3) the
flap-wise bending structural mode and (4) the torsion structural mode of the blade.

McCroskey et al. [57] provide stall results for seven different helicopter aerofoils
oscillating in pitch at 0.3M. All aerofoil sections demonstrated a similar increased
maximal coefficient of lift. No significant correlation was found between the amount
of increase (between 0.7 and 0.8) and the shape of the aerofoil. Prouty [32] provides a
quick and easy guideline for preventing excessive retreating blade stall: “It is also
generally accepted that for conventional helicopters at maximum speed, the tip speed
ratio limit should not exceed 0.5 to avoid retreating blade stall” (p. 646). This practical
criterion is expressed in Eq.(2.27).

VT
  0.5 (2.27)
R

The two conflicting criteria from Eq. ((2.26),(2.27)) are presented on the VT, ΩR
plane in Fig. 2.14. As the airspeed increases the requirement for restraining the
advancing blade compressibility issues results in a decrease of the rotor angular speed,
whereas the criteria for preventing excessive retreating blade stall requires an increase
in the main-rotor angular speed. The two contradictory requirements meet to define
an equilibrium point that represents the maximal-practical airspeed of a conventional
helicopter in forward flight and the corresponding main-rotor tip speed. This
maximum true airspeed is about 203 kts. (Standard sea-level conditions) and about

56
 2.2 | L E V E L F L I G H T P E R F O R M A N C E

196 kts. (Standard-day, 10K ft. pressure altitude). The corresponding main-rotor tip
speeds are 685 and 661 fps.

Figure 2.14. Maximum airspeed of a conventional helicopter in forward flight. The two
contradicting constraints for restraining compressibility issues on the advancing blade and for
preventing excessive retreating blade stall define the practical maximal airspeed for a
conventional helicopter in forward flight.

Notice that the two equations can be combined to form an explicit form
(Eq.(2.28)) to represent the maximum practical airspeed of a conventional helicopter.
This insight relates back to Igor Sikorski’s famous quote about the helicopter to never
be able to fly faster than the airplane, as already mentioned in the introduction to hover
performance in Chapter 1, Subsection 1.2.2.1.

M tip (max)
VTMAX   air RairTa (2.28)
3

2.2.3 The Parasite Power in Level Flight

The third term in Eq. (2.17) is the parasite power. This power term is required for
the support of the aerodynamic drag generated by the helicopter fuselage. This
‘fuselage’ term holds for any structural element of the conventional helicopter,

57
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

excluding the main and tail rotor blades. The main and tail rotor blades are already
accounted for in the power equation (Eq.(2.17)) as the profile power. The parasite
power term is based on the same fundamental mechanical concept previously used for
the induced and profile power components. The mechanical work done by a force on
an object is defined by the scalar-multiplication of the two vectors, the force and the
object’s displacement. The rate of work (power) is then expressed by the evaluation of
the work done per unit of time. This ratio of displacement per unit of time is precisely
the true airspeed of the helicopter, and when multiplied by the fuselage aerodynamic
drag force the parasite power term is attained.

The parasite power term, as appears in Eq.(2.17), is slightly manipulated and is

explained hereinafter. Since the conventional helicopter flies at relatively low altitudes
and airspeeds characterized by Mach numbers below 0.3, the flow-field surrounding
its fuselage can be treated as incompressible flow, as explained by Dommasch et al.
[62]. The fuselage aerodynamic drag-force (Df) is calculated as per Eq. (2.29) which is

the product of the dynamic pressure (q), the drag reference area (Sref) and the drag-

coefficient based on the drag reference area (CDf). For an incompressible flow the
dynamic pressure is calculated as per Eq.(2.30).

D f  qS ref CDf (2.29)

1
q aVT2 (2.30)
2

For reasons of convenience and ease of common-base comparison, the product

of coefficient-of-drag (CDf) and drag reference area (Sref) is expressed by a different
product, which has the same numerical value. This new product consists of a
convenient coefficient-of-drag (equals 1), associated with a new reference drag area,
defined as the ‘equivalent flat plate area’ (EFPA) and depicted by fe (Eq.(2.31)). By

using these new terms, the fuselage aerodynamic drag-force (Df) is conveniently
expressed as Eq.(2.32). The power required to provide against this aerodynamic drag

58
 2.2 | L E V E L F L I G H T P E R F O R M A N C E

(Df) is merely the product of the aerodynamic drag force and the true airspeed of the
helicopter, as expressed by (Eq.(2.33)).

Sref CDf  f e 1  f e (2.31)

1
Df   aVT2 f e (2.32)
2

1
Pparasite  D f VT  aVT3 f e (2.33)
2

Helicopter airframes are less aerodynamic than their fixed-wing counterparts, this
source of drag can be very significant, as shown by Leishman [10]. According to
Leishman [50] the drag of a helicopter fuselage may be up to one order of magnitude
higher than that of a fixed-wing aircraft of the same gross weight. As presented by
Prouty [47] and Rosenstein et al. [63], typical values for helicopter EFPA range from
as low as 5 ft2, for small and clean designs, up to 60 ft2 for large flying cranes. Leishman
[10] presents data of a variety of helicopter designs as a plot on the EFPA, gross-weight
plane. This plot suggests helicopter designs fall into two major categories of ‘clean’ and
‘utility’ helicopters (excluding the large flying cranes) and within each category the
EFPA value is approximately proportional to the square root of the helicopter gross-
weight. This empirical observation is referred to as the ‘square-cube’ law.

According to Sheehy [64] the main-rotor hub and landing gear are the two major
contributors to the parasite drag of the helicopter. While, the drag arises from the
landing-gear can be significantly reduced by fairings, or even entirely eliminated by
retracting the lading-gear into the fuselage, the rotor hub drag contribution cannot be
easily reduced and accounts for approximately 20-30% of the total parasite drag.
Moreover, fully articulated rotor hub designs can contribute to about 60% of the total
parasite drag of a helicopter designed for high-speed operations. Prouty [8] provides
parasite drag breakdown estimates for an undisclosed example helicopter (Fig. 2.15),
accompanied with the following personal note: “I have never known of an airplane, or

59
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

a helicopter drag estimator who was pleasantly surprised by flight-test results showing
an overestimation of the drag” (p. 304). For illustration purposes, this estimated EFPA
of 19.3 ft2 converts using Eq. (2.33) to a parasite power of 677 hp, while flying at a
true airspeed of 150 kts under standard sea-level (SSL) conditions.

Figure 2.15. Parasitic drag breakdown for an example helicopter. The two main
contributors to the parasitic drag of a conventional helicopter are the main-rotor hub with an
EFPA of 7 ft2 (36.3% of total parasite drag) and the fuselage with an EFPA of 5.8 ft2 (30.1%).
The landing gear system is the third largest contributor, with a combined main and nose landing
gear EFPA of 2 ft2 (10.4%).

Maintaining a constant external configuration does not guarantee a permanent

parasite drag. The fuselage EFPA varies as the fuselage angle of incidence changes.
This variation in the fuselage angle of incidence may be attributed to changes in
airspeed or even to migration of the centre-of-gravity (CG). Any change in the fuselage
angle of incidence, not only varies the EFPA of the fuselage but also causes for EFPA
changes in all components attached to the fuselage, i.e., nacelles, landing gear, stub-
wings, and stabilizer surfaces. Notice that changing the angle of incidence on
aerodynamic surfaces results in not only skin-friction change but also induced drag
change. Flight tests conducted on the prototype OH-58A helicopter with a flat-plate
canopy showed [65] that while flying at the cruise airspeed of this helicopter (102 kts.),
an aft migration of the longitudinal CG, from the mid-point to the maximum aft

60
 2.3 | C O N V E N T I O N A L M E T H O D S FOR PERFORMANCE FLIGHT T ESTING

position resulted in 2.2 ft2 (22%) increase in the EFPA. A forward migration of the
CG, from the mid-point to the maximum forward position, resulted in a more subtle
EFPA increase of 0.7 ft2 (7%). The 2.2 ft2 ERPA increase is equivalent to a power
increase of about 24.3 hp (7.7% of available power), under the relevant flight-
conditions.

Level flight performance flight testing conducted on the 18K lbs. UH-60A
helicopter, equipped with the External Stores Support System (ESSS), demonstrated a
strong relationship between the EFPA and the longitudinal CG location of the
helicopter [6]. A 15 inches longitudinal CG migration from an aft fuselage station
(FS 358”) to a forward FS 343” resulted in an EFPA increase of 9.6 ft2 while flying
straight and level at airspeeds between 40 to 140 kts. This 9.6 ft2 increase in EFPA is
equivalent to a power increase of about 274 hp (9.6% of available power) while flying
at 140 kts under SSL conditions.

2.3 C ONVENTIONAL M ETHODS FOR

P ERFORMANCE F LIGHT T ESTING

As already mentioned in the introduction to this thesis (Section 1.3), the

fundamental question troubling the helicopter performance flight-testers has always
been how should the performance of the helicopter be tested? More specifically, at
which particular gross-weights? Under which combinations of atmospheric
conditions? Can measurements be standardized? Is it possible to use data from a
specific flight to predict performance under different and arbitrary flight conditions?
These types of questions which are essentials in performance flight-testing can be
addressed by using tools of dimensional analysis and the Buckingham PI theorem [66]
which is commonly regarded as the fundamental theorem of dimensional analysis, as
stated by Evans [67].

The Buckingham PI theorem states that any physically meaning problem with
numerous dimensional parameters involved, can be reduced to a lesser number of

61
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

significant non-dimensional parameters, based on the dimensions involved. The

competency of dimensional analysis to reduce the number of affecting variables and
to support interpolation and extrapolation of data, has made it a popular tool in
performance flight testing. By applying dimensional analysis tools, the flight-tester can
answer all performance questions raised above and at the same time, significantly
reduce the number of flight-test sorties required to quantify the performance of an
aircraft throughout its flight-envelope.

2.3.1 Available Power Flight-Test Method

The current method widely used within the flight-test community for determining
the available power any gas-turbine helicopter possess is based on the single-variable
analysis, as presented in few flight testing method text books [18-21] and practicably
demonstrated in Belte and Stratton [78] and Benson et al. [86]. This conventional flight
test method is further demonstrated, with its major deficiencies emphasized, in
Chapter 3 of this thesis. For this, authentic flight test data from a MBB (Messerschmitt-
BÖlkow-Blohm) BO-105 helicopter are used. According to this flight-test method, all
four engine performance variables (output power, compressor speed, temperature, and
fuel-flow), accompanied with their corresponding atmospheric conditions, and are
recorded during steady engine operation conditions. For this, the helicopter is flown
throughout its certified flight envelope and under diverse atmospheric conditions. As
demonstrated by Jackson [75], it is essential for multi-engine helicopters to handle data
from each engine separately. Each engine is a separate entity with potentially a different
level of maximum available power that might be operated under slightly different
atmospheric conditions, even if installed on the same helicopter.

Once a substantial database is gathered, it is analysed with the goal of evaluating

the maximum power the engine is able to produce under various atmospheric
conditions. This procedure is carried-out in two phases. The first (Phase I) is to
generate a convenient mathematical model to describe the dependency of the engine
output power with all other engine performance variables (compressor speed,

62
 2.3 | C O N V E N T I O N A L M E T H O D S FOR PERFORMANCE FLIGHT T ESTING

temperature and fuel-flow). Phase I can be regarded as uncovering (mathematically)

the specific engine ‘rules of operation’. The second phase (Phase II) concentrates on
deriving the maximum output power of the specific installed engine under a wide range
of atmospheric conditions i.e., performance under hot day or cold day conditions, and
for the various engine power ratings i.e., the maximum continuous power, the
maximum 5-minute take-off rating power etc.

2.3.1.1 Phase I – The Engine ‘Rules of Operation’

The essence of this phase is to derive simple empirical models to represent the
dependency between the engine output power and each one of the performance
variables of the engine i.e., the engine temperature (TGT), compressor speed (Ng) and
fuel-flow (Wf). As already discussed above, dimensional analysis plays a major role in
helicopter performance flight-testing. Using the Buckingham PI theorem, the engine
performance problem can be simplified to include a set of only four corrected
variables. These corrected variables are essentially non-dimensional magnitudes which
bear units and for this reason they are addressed by Knowles [73] as the GT engine
corrected variables, rather than non-dimensional. Specifically, these corrected variables
are the corrected output power of the engine (CSHP), the corrected compressor speed
(CNg), the corrected engine temperature (CTGT) and the corrected fuel-flow into the
engine (CWf). The definitions of these non-dimensional variables are presented in the
nomenclature and the rigorous procedure to derive these corrected variables is
presented as Appendix A in this thesis.

By applying common methods of linear regression, a set of third order single-

variable polynomials is retrieved to relate between the corrected output power and
each one of the other corrected variables of the GT engine, as given by Eq.(2.34),
Eq.(2.35) and Eq.(2.36). Third order polynomials are employed for the reason they are
the lowest order that enable modelling an inflection point, a fundamental behaviour of
the gas-turbine engine. These polynomials, based on actual flight-test data, serve as
empirical models to represent the ‘rule of operation’ of the specific GT engine, as

63
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

installed in the particular type of helicopter and at the specific phase of its life cycle.
An example of actual experimental polynomials is presented in Fig 2.16.

3
CSHP   ai  CNg 
n
(2.34)
n 0

3
CSHP   bi  CTGT 
n
(2.35)
n 0

CSHP   ci  CW f 
3
n
(2.36)
n 0

Figure 2.16. Example of 3rd order empirical model of gas-turbine engine. The presented
three 3rd order polynomials are derived from actual flight-test data and serve as an empirical
model for the installed gas-turbine engine (GTE).

As previously mentioned, the requirement for these polynomials to be of the

third order is to ensure the empirical model captures the inflection point of the engine
performance. This inflection point represents a fundamental property of any gas-
turbine engine for which the rate of change in output power with respect to the engine
variables (compressor speed, temperature, or fuel-flow) changes its sign. For the same
amount of engine variable increase, the resulted increase in engine output-power is to

64
 2.3 | C O N V E N T I O N A L M E T H O D S FOR PERFORMANCE FLIGHT T ESTING

be reduced beyond the inflection point, as compared to the output-power increase

below this inflection point.

2.3.1.2 Phase II – Maximum Available Power

The second phase of the available power flight-test method uses the empirical
models retrieved from Phase I to define the maximum available power the installed
engine can generate, for any power rating and under any atmospheric conditions as
selected by the flight tester. This phase involves an iterative process, as illustrated by
Fig. 2.17 and explained hereinafter. The maximum engine power can be limited by
either one of the following parameters: the engine temperature (TGT), the engine
compressor speed (Ng), the engine fuel-flow (Wf) or the maximum output shaft torque
(TRQ). These limitations are well known to the helicopter operator and their values
are typically different for the various power ratings of the engine. The iterative process
of evaluating the available power commences by defining the relevant power rating. Is
it the continuous power rating with no time limitation? Or is it for the take-off which
is limited to only 5 minutes of operation.

Once the power rating is decided, the type of day must be defined. A standard
day (ISA) is defined as one with a temperature of 15°C at sea level which decreases by
1.98°C for every 1,000 ft. of climb. Other day conditions are based on the ISA day and
are symbolised as the temperature difference from the standard day conditions. For
example, ISA+10 represents a day for which the ambient air temperature at sea level
is 5°C. The elapsed rate of the temperature with altitude is assumed as 1.98°C per 1,000
ft., although it is seldom the reality. Once a type of day is decided upon, for each
pressure altitude selected the relative temperature (θ) and relative pressure (δ) are
calculated. For each pressure altitude the three-engine limiting corrected variables can
be calculated and plugged into the three flight-test based polynomials retrieved in
Phase I. Each one of these polynomials yields a corrected output power which can
easily be turned into actual output power and output shaft torque.

65
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

Figure 2.17. The iterative procedure for engine available power disclosure. The maximum
available power of an installed gas-turbine engine (GTE) is disclosed by executing an iterative
procedure based on the empirical model of the GTE and its operating limitations.

The maximum output power of the engine for the selected pressure altitude would be
the minimum of the three values yielded from the empirical polynomials and the
transmission limitation (the output shaft torque limitation). Once the maximum
available power for the selected pressure altitude is defined, another iteration for the
next pressure altitude is carried out. This procedure is repeated for other types of day
and all relevant power ratings of the engine. Figure 2.18 presents an example outcome
of this procedure, a plot that specifies the maximum available power an installed GT
engine is capable of delivering, for a range of pressure altitudes. Note that Fig. 2.18 is
not related to the specific set of polynomials presented in Fig. 2.16. Figure 2.18
specifies the maximum available power of the engine, for a continuous power rating
and for two distinct types of day. It shows that for standard day conditions (ISA) this
specific installed GT engine is limited by the transmission of the helicopter (569 hp.),
from sea level up to a pressure altitude of 3,500 ft. Above 3,500 ft. the engine becomes
compressor-speed (Ng) limited. Under hot-day conditions (ISA+25°C), the maximum
continuous power is limited by the compressor speed throughout the altitude range

66
 2.3 | C O N V E N T I O N A L M E T H O D S FOR PERFORMANCE FLIGHT T ESTING

presented, from sea-level up to 10,000 ft. At sea-level the installed engine is capable of
continuously producing 486 hp. This continuous maximum power reduces to only 403
hp. at a pressure altitude of 10,000 ft.

Figure 2.18. Example of an installed engine available power chart. The data represent the
available power for continuous operation rating of a GTE, under standard day (ISA) and hot-
day (ISA+25°C) conditions.

2.3.2 Hover Performance Flight Testing

As already mentioned in the introduction (Subsection 1.3.2), the objective of

hover performance flight testing is to provide a detailed map of the actual power
required to sustain the specific type of helicopter at a hover (either in or out of ground
effect, IGE/OGE) for all certified gross-weights, external configurations, main-rotor
angular speeds and the surrounding atmospheric conditions of air temperature,
pressure, and density. This performance ‘map’ is traditionally presented in a format of
a graph, or a set of synchronized graphs and plots. For this objective, the hover
performance flight-tester task is to measure the actual power required for hover
throughout the flight envelope. Since it is impractical to hover the helicopter in each

67
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

combination of gross-weight, main-rotor angular speed, atmospheric conditions of

pressure altitude and temperature, the flight tester applies means of dimension analysis
as previously discussed in the introduction of this section (2.3). By applying means of
dimensional analysis, the flight-tester can both reduce the number of planned flight
test sorties to an achievable and practical number, and at the same time to provide a
detailed performance map that covers the entire certified flight envelope of the aircraft.

2.3.2.1 Non-Dimensional Hover Performance

The conventional, widely used, flight-test method for hover performance is

based on Eq. (2.1) which describes the power required for the main rotor at an OGE
hover. The main-rotor system is the principal power consumer in a conventional
helicopter hover. Its relative consumption varies between different types of
helicopters, and for various flight conditions of a specific type of helicopter, but
nevertheless it consumes an immense power portion surrounding 85% [19,20,24]. The
remaining ~15% of the hovering power is dissipated by the tail-rotor, all sorts of
accessory drives and systems and transmission loss. For this reason, the hover flight
test method of the complete helicopter is based on the power of the main-rotor and
the applicable Eq. (2.1).

Equation (2.1) is normalized by dividing both of its sides by the term

(  a Adisk  R  ) which is a product of the ambient air density, main-rotor disk area
3

and the blade tip tangential speed cubed. This division yields Eq. (2.37) which simply
represents the non-dimensional version of Eq. (2.1). The reader should recall that Eq.
(2.1) describes the ideal case, or the ‘best-case-scenario’, of minimum possible induced
power in a hover. Leishman [34] compensates for the non-ideal case by implementing
an empirical correction factor (ki) as presented in Eq.(2.38). This (ki) is called the
induced power correction factor and its typical average value is about 1.15. A power
correction value of 1.15 indicates 15% increase in the actual main rotor induced power,
as compared to the ideal case of constant induced velocity across the hovering disk.

68
 2.3 | C O N V E N T I O N A L M E T H O D S FOR PERFORMANCE FLIGHT T ESTING

Mind that Eq. (2.39) merely defines the main-rotor solidity ratio (σR), and the two
non-dimensional variables known as the ‘coefficient-of-power’ (CP), and the
‘coefficient-of-weight’ (CW). It is common for applicable textbooks, papers and flight-
test reports to interchangeably use either the coefficient-of-weight or the coefficient-
of-thrust (CT) since the thrust equals the weight at a hover, hence CT equals the
coefficient-of-weight (CW).

2 1
CPM / R   CW   Cd0 R
1.5
(2.37)
2 8

2 1
CPM / R  ki  CW   Cd0 R  ki  1.15
1.5
(2.38)
2 8

PM / R W bc
CPM / R   CW   R  (2.39)
a Adisk  R  a Adisk  R  R
3 2

The conventional flight-test method for hover performance is based on

imposing the main-rotor power onto the helicopter as a whole. This approach is
presented as Eq.(2.40), where CP is the coefficient-of-power based on the total power
in hover. The task of the flight-tester is to relate between all measured coefficient-of-
power (CP) and coefficients of weight (CW) while hovering under a wide range of gross-
weights (W), main-rotor angular speeds (Ω) and ambient air atmospheric conditions.
This task can be regarded, mathematically, as Eq. (2.41) for which the flight-tester is
required to define the two constants (α1, α2) for a particular type of helicopter and/or
helicopter external configuration.

1  2 1 
CP   CW   Cd0 R 
1.5
 ki (2.40)
m  2 8 

69
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

CP  1  CW    2
1.5


 2 k 1 1 
 1  : 2   Cd 0  R   (2.41)

 2 m m  8 

The flight-test team is required to plan and execute numerous hover test-points
in order to cover the entire flight envelope of the test article. This includes all certified
gross-weights, from minimum to the maximum certified, the entire ambient air
temperatures and pressure-altitudes the helicopter is expected to fly at, and throughout
the governed range of the main-rotor angular speed (Ω).

There are two fundamental techniques to execute the precise hover sorties for
data gathering. The first is the free-flight hover and the second is the tethered hover.
The first technique requires the flight-test crew to stabilize the helicopter at a hover
and record the essential data to regress the CP to CW relationship, as presented in
Eq.(2.41). Variation of gross weights is achieved by physical ballast added/removed
from the helicopter. Altering the atmospheric conditions is done either by changing
testing sites or by hovering in formation to another aerial vehicle, equipped with a low
airspeed system capable of establishing a true hover flight.

The second technique of tethered hover is more complicated and requires

additional preparation effort. For this technique the helicopter is attached via a tether
and an instrumented load-cell device to the ground. The tension in the tether is
continuously measured by the load-cell, recorded by the flight-test instrumentation
package and presented real time to the flight-test crew. This manner, the thrust
generated by the rotor system counter both the physical weight of the helicopter added
with the tether tension.

The main advantage of the tethered hover technique over the free-flight
technique is that reconfigure the helicopter for different gross-weight is done
immediately, just by raising the collective-stick for more thrust. It does not require the
flight-test crew to land and to add or remove ballast for the next gross-weight data

70
 2.3 | C O N V E N T I O N A L M E T H O D S FOR PERFORMANCE FLIGHT T ESTING

planned. Regardless of the hover flight-test technique, whether it is the free-flight or

the tethered hover technique, it is utterly important to establish a true and accurate
aerodynamically hovering flight. This means that the relative motion between the
helicopter and the ambient air is limited to 3 kts. A relative motion that exceeds 3 kts.
is outside of the allowed tolerance for this test, since it significantly reduces the induced
power, hence deceiving the test results.

A closer look at Eq. (2.41) can provide practical limitations on the empirical
values for α1 and α2. These limitations allow the flight-tester to perform a basic ‘sanity-
check’ to validate the empirical non-dimensional hover performance equation yielded
from the test. As discussed before, typical values for the induced power correction
factor (ki) and the mechanical efficiency (ηm) at hover are 1.15 and 0.85, respectively.
This dictates an expected nominal α1 value of about 0.957. Furthermore, since it is
physically impossible for the value of (ki) to plunge below 1, an established α1 value
below 0.83 should trigger a detailed investigation about either the data analysis process,
or the validity of the flight test sorties execution. A possible reason for a lower-than-
expected α1 value can be attributed to hover-data gathering under high relative winds
(above the 3 kts. limitation). This will cause for a lower induced power component,
hence a lower-than-expected α1 value. The expected value for α2 is more trivial and can
be easily interpreted from the solidity ratio, zero-lift drag coefficient and the nominal
mechanical efficiency at a hover.

Figure 2.19 presents a genuine relationship between the measured coefficient-

of-power (Cp) and the coefficient-of-weight (Cw) raised to the 1.5 power, as measured
during a limited scope hover performance performed for this research. A linear
regression based on minimum squares is performed to retrieve the two coefficients (α1,
α2). The specific non-dimensional OGE hover performance of the evaluated helicopter
is presented in Eq.(2.42). This simple equation is assumed to encapsulate the entire
OGE hover performance of the evaluated helicopter. This authentic Bell Jet Ranger
(OH-58C) flight-test data are further discussed and analysed in Chapter 5 within the
context of deficiencies associated with this current flight-test method.

71
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

CP  1.184  Cw   3.839 105

1.5
(2.42)

Figure 2.19. Non-dimensional OGE hover performance data. The data represents a limited-
scope hover performance flight testing campaign of the Bell Jet-Ranger helicopter that includes
76 stabilized OGE hover points.

2.3.2.2 Un-Referring to Conditions of Choice

For the common helicopter operator this non-dimensional OGE hover

performance as presented by Eq. (2.42) does not tell much. It cannot be used explicitly
for flight planning purposes. This implicit (or convoluted) OGE hover performance
information needs to be simplified and be presented in an accessible manner to the
common operator. This simplification procedure, called ‘un-referring the data to
specific conditions of choice’, is discussed and demonstrated hereinafter.

The process starts with reinstituting the explicit definitions of the coefficient-
of-power and coefficient-of-weight into the established non-dimensional hover
performance (Eq.(2.42)). This back-substitution yields an explicit multivariable
function that relates between the power required to hover and the following three
independent variables of gross-weight, ambient air density and main-rotor angular

72
 2.3 | C O N V E N T I O N A L M E T H O D S FOR PERFORMANCE FLIGHT T ESTING

speed (Eq.(2.43)). This allows the flight-tester to generate the performance ‘map’
mentioned in the introduction of Subsection 2.3.2 above. The power required to
sustain a hover flight can be predicted for any arbitrary combination of gross-weight,
ambient air density (atmospheric conditions) and main-rotor angular speed. An
example OGE hover performance chart based on Eq. (2.43) is presented as Fig. 2.20.

1.5
P  W 
 1.184    3.839 105 
 a Adisk  R 
3
  A  R  
3
 a disk 
(2.43)

P  1.184
W3
 a Adisk

 3.839 105  a Adisk  R 
3


Figure 2.20. Explicit presentation of OGE hover performance. This graph is based on data
extracted from the non-dimensional OGE hover performance of the Bell Jet-Ranger helicopter
(Eq.(2.43)), and the procedure defined as ‘un-referring data to conditions of choice’.

2.3.2.3 Extremum Hover Performance

The previously presented information about the power required to hover a

particular helicopter can be combined with the information regarding the available
power of the helicopter (Subsection 2.3.1 above) to form what is known as the
extremum hover performance of the helicopter. The extremum hover performance

73
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

relates to the hover ceiling of the helicopter, i.e., the maximum altitude the helicopter
can hover at. Another extremum hover performance aspect is the maximum gross-
weight for which the helicopter can establish a hover flight, under various atmospheric
conditions. This extremum hover performance evaluation can be accomplished once
both the power available, and the power required to hover flight-test campaigns are
concluded. Figure 2.21 demonstrates this procedure for a hot-day condition
(ISA+20°C). By overlaying the available power on the required power, the extremum
hover performance is exposed. On the one hand, the available power of the installed
gas-turbine engine reduces with altitude increase (unless limited by the transmission).
On the other hand, the power required to hover increases with altitude increase. Both
contradicting tendencies reach an equilibrium point which defines the maximum hover
altitude (hover ceiling), for a specific gross-weight, specific type of day and a particular
engine power rating (continuous, take-off rating, etc.). For clarifying the data presented
in Fig. 2.21, the continuous rating OGE hover ceilings of this example helicopter,
while operated at standard main-rotor speed (354 RPM) and under ISA+20°C
conditions are 9,380, 6,560 and 3,730 ft., for gross-weights of 2,700, 2,900 and
3,100 lbs., accordingly.

The procedure to conclude about the maximum gross-weight the helicopter

can hover at for various atmospheric conditions and power ratings, is similar, although
not explicitly demonstrated here. For this, Eq.(2.43) should be rearranged to solve for
the gross-weight, while all the other variables are treated as known values (P, the
available power, the desired atmospheric conditions expressed by the air density, and
the main-rotor angular speed).

74
 2.3 | C O N V E N T I O N A L M E T H O D S FOR PERFORMANCE FLIGHT T ESTING

Figure 2.21. Example helicopter OGE hover ceiling determination. The graph shows the
hover ceiling of the Bell Jet-Ranger helicopter for three distinct gross-weights, under hot-day
conditions (ISA+20°C).

2.3.3 Level Flight Performance Flight

Testing

As previously mentioned in the introduction of this thesis (Subsection 1.3.3), the

objective of level-flight performance flight testing is to provide a detailed map of the
actual power required to maintain the specific type of helicopter at a level flight
conditions, for all certified gross-weights, external configurations, main-rotor angular
speed range and the surrounding atmospheric conditions of air temperature, pressure,
and density. This performance ‘map’ is traditionally presented in a format of a graph,
or a set of synchronized graphs and plots. Moreover, cross-referencing the power
required for level flight with the fuel-consumption data base, as evaluated during the
available power flight-testing phase (described in Subsections 1.2.1 and 2.3.1 above),
enables to define the helicopter ‘best-effort’ airspeeds, such as airspeeds for maximum
range and for maximum endurance. It is impractical for the flight-tester to measure
the actual power required for level-flight throughout the flight envelope of the
helicopter, and for all possible combinations of configurations and atmospheric
conditions. Application of dimensional analysis concepts and means, allows the fight-

75
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

tester to both, reduce the number of planned flight test sorties to an achievable and
practical number, and to provide a detailed performance map that covers the entire
flight envelope of the helicopter.

2.3.3.1 Non-Dimensional Level-Flight Performance

The conventional flight-test method for determining the level-flight

performance of a conventional helicopter is thoroughly discussed in the literature [8,
10, 18, 49, 76] and demonstrated in numerous flight-test reports [6, 77, 88]. This
method originates from Eq. (2.17) which describes the power required to sustain the
helicopter in level flight. Equation (2.17) is converted into a non-dimensional form
(Eq.(2.44)) by dividing both sides of the equation by the mathematical-term

(  a Adisk  R  ). The non-dimensional (ND) power required for level-flight equation

3

(Eq.(2.44)) relates between only three variables, the coefficient-of-power (Cp), the
advance-ratio (μ) and the coefficient-of-weight (Cw). All other terms in Eq. (2.44) are
constants for a specific helicopter and configuration.

CP 
CW2 1
2 8
 
 Cd0  R 1  k p  2 
1 fe
2 Adisk
3 (2.44)

The conventional flight-test method for level-flight performance seeks to

simplify this rather already simple three-variable relationship. The three-variable
relationship described in Eq. (2.44) is further reduced into sets of two-variable
mathematical relations that describes the association between the coefficient-of-power
(Cp) and the advance-ratio (μ), for discrete values of coefficient-of- weight (Cw). The
conventional flight-test method seeks to find this exact empirical relationship (Cp to
μ) for the helicopter entire coefficient-of-weight spectrum. For this, the flight-test
crew executes numerous ‘speed-runs’ while maintaining a constant coefficient-of-
weight. The method of assuring a constant coefficient-of-weight during the speed run
of the helicopter is what defines the flight-test method and can be achieved in two

76
 2.3 | C O N V E N T I O N A L M E T H O D S FOR PERFORMANCE FLIGHT T ESTING

ways: (1) the constant weight over sigma (W/σ) method; and (2) the constant weight
over delta (W/δ) method.

2.3.3.2 Constant Weight over Sigma (W/σ) Method

This flight-test method is the foremost popular and recognized method

conducted for the evaluation of the level-flight performance on the conventional
helicopter. In this method the flight-test crew maintains the coefficient-of-weight at a
certain value by keeping the main-rotor speed constant and maintaining a constant
ratio of weight (W) to the air relative density (σ). As presented in Eq.(2.45), the air
relative density is defined as the ratio between the ambient air density (𝜌a) and the
standard sea level air density (𝜌o). Maintaining a constant ratio of weight to relative
density (W/σ) is achieved by a gradual adjustment of the cruise altitude for the speed
runs as the helicopter burns fuel and becomes lighter. This constant W/σ method is
demonstrated mathematically in Eq.(2.45). The required altitude change in-between
test points of the speed-runs is calculated in real time by the test-crew. Typically, the
flight-test campaign for a specific helicopter configuration requires the execution of
five sorties, each conducted at a different coefficient-of-weight value. The various
coefficient-of-weights shall cover the entire certified envelope of the helicopter. Each
speed run consists of at least eight different airspeeds, beginning at some ‘arbitrary’ low
airspeed to the maximum level flight airspeed defined either by maximum available
power (Vh), or by the manufacturer’s definition for the ‘never-exceed’ airspeed (VNE).

W W W 1 1 a
CW        (2.45)
 a Adisk  R 
2
o Adisk  R 
2
  2
o Ad R 2
o
held  fixed

Figure 2.22 presents a genuine non-dimensional (ND) relationship between

the measured coefficient-of-power (Cp) and the advance-ratio (μ) for a single
coefficient-of-weight (CW) of 5.79×10-3. Data are gathered for twelve distinct advance-
ratios during a dedicated constant W/σ level flight performance flight-test sortie

77
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

performed for this research. The presented range of advance-ratio (0.098 to 0.28)
translates for the specific helicopter type and test conditions into a true airspeed range
in between 42 to 118 kts. A linear regression based on minimum squares is performed
to retrieve the four coefficients required to define the particular 3rd order polynomial
(Eq.(2.46)). This polynomial represents the non-dimensional level-flight performance
of the BO-105 helicopter for the specific coefficient-of-weight (CW = 5.79×10-3) and
for the tested external configuration. This authentic MBB BO-105 helicopter flight-
test data is further discussed and analysed in Chapter 6 within the context of
deficiencies associated with this current flight test method.

CP  0.0119 3  0.0218 2  0.0057   0.0007  CW  5.79 103 (2.46)

Figure 2.22. Non-dimensional level flight performance. The graph shows the relationship
between the coefficient-of-power (Cp) and the advance ratio (μ) for a CW value of 5.79×10-3 as
measured on a BO-105 helicopter. Flight test sortie was based on the constant W/σ method.

78
 2.3 | C O N V E N T I O N A L M E T H O D S FOR PERFORMANCE FLIGHT T ESTING

2.3.3.3 Constant Weight over Delta (W/δ) Method

The second and less common approach of maintaining a constant coefficient-

of-weight during the speed run is called the ‘weight over delta’ method. This method
is demonstrated mathematically in Eq.(2.47). Note that the air relative pressure ratio
(δ) is defined as the ambient air static pressure (Pa) over the standard sea level air
pressure (P0). By using the equation of state (Eq.(2.48)), the ambient air density is
expressed using the ambient static-temperature (Ta) ambient pressure (Pa) and the
specific gas constant of the air (Rair). It is evident from Eq. (2.47) that by holding a
constant ratio of weight over the relative pressure (W/δ) and a constant ratio of static
ambient pressure over the angular rotor speed squared (Ta/Ω2), the flight-test crew
assures a constant coefficient-of-weight during the various speed runs. The only
advantage this method has over the W/σ method is that by maintaining a constant
ratio the flight-test crew can control a particular blade tip Mach number. For flight
conditions where compressibility is an issue, the test-crew can avoid gathering flight-
test data contaminated with compressibility effects. Nevertheless, level-flight
performance is typically required for flight conditions that include compressibility
effects. This constant W/δ flight-test method requires even more flight-test sorties
than the amount required for the W/σ method. This increased number of sorties is
mostly attributed to the complexity and cumbersome associated with the continuous
adjustments of the main-rotor angular speed.

W W W Ta Rair
CW     2
 a Adisk  R 
2
  Po    Po Adisk R 2
 Adisk  R 
2
 held  fixed
 Rair Ta  (2.47)
P  Po
  a  a 
Po Rair Ta

Pa   a Rair Ta (2.48)

79
2 | HELICOPTE R PE RFORMA NCE T HEORY & CONVENTIONAL T ESTING ME THODS

2.3.3.4 Un-Referring to Conditions of Choice

As stated for the hover performance is Subsection 2.3.2.2 above, this non-
dimensional level-flight performance (Eq.(2.46)) will not be appreciated by the
common helicopter operator, since it cannot be used explicitly for the task of flight
planning. However, this implicit information can be simplified or ‘un-referred’ to
specific conditions of choice. According to this conventional method, the coefficient-
of-power (Cp) and the advance-ratio (μ) are interrelated in level-flight as per Eq.(2.46)
as long as the coefficient-of-weight of the helicopter equals 5.79×10-3. For a normal
operations main-rotor speed of 423 RPM (blade tip speed of 715 fps.), this specific
coefficient-of-weight value can be converted into a range of gross-weight and ambient
density combinations (Eq.(2.49)). Exhausting the one degree of freedom by choosing
an arbitrary gross-weight of 4,850lbs (within the tested range) defines the applicable
ambient density of 0.002 slug/ft3. This ambient air density corresponds to an altitude
of 5,744 ft. under standard day conditions. The coefficient-of-power can also be
reduced into a dimensional value as expressed by Eq.(2.50). Next, by using the
empirical level-flight model retrieved from the flight-test campaign (Eq.(2.46)), the
actual power required to sustain a 4,850lbs. BO-105 helicopter in level-flight and under
the relevant conditions (5,744 ft. standard day, 423 RPM main-rotor speed) is known.
This is expressed as Eq.(2.51) and illustrated in Fig. 2.23. This tedious procedure is
repeated for the entire flight envelope covered by the flight-testing sorties.

W R  715 fps
CW   5.79  103  2
 a Adisk  R  Adisk 818 ft
2

(2.49)
W [lbs.]  2.422  10  a [ slug / ft ]
6 3

80
 2.3 | C O N V E N T I O N A L M E T H O D S FOR PERFORMANCE FLIGHT T ESTING

Figure 2.23. Level-flight performance of an example helicopter. The graph is based on

extracted data from the ND level-flight performance (Eq.(2.46)) and the procedure discussed in
Subsection 2.3.3.4, ‘Un-Referring to Conditions of Choice’.

P
CP   P[hp]  1.089  106 CP (2.50)
 a Adisk  R 
3 R  715 fps
Adisk 818 ft 2
 a  0.002 slug / ft 3

 
P[hp]  1.089 106 0.0119  3  0.0218 2  0.0057   0.0007    [0.098, 0.28] (2.51)

Once a theoretical foundation for the relevant sections of helicopter

performance and the associated conventional flight test methods have laid down, the
substantial portions of the research are presented in the following chapters. Chapter 2
is intendent to provide the crucial theoretical background to allow the reader for a full
understanding of the succeeding chapters of the thesis that discuss deficiencies
embedded within the conventional flight test method and propose enhanced methods
instead.

81
 Nothing takes place in the world whose meaning is not that of
some maximum or minimum.
Leonard Euler

3 A M ULTIVARIABLE A PPROACH IN

G AS -T URBINE E NGINE T ESTING

3.1 C HAPTER O VERVIEW

H elicopter performance relies heavily on the available output power of the

engine(s) installed. A simplistic single-variable approach is often used within the
flight-testing community to reduce flight-test data in order to predict the available gas-
turbine engine power under various atmospheric conditions. This conventional
approach which often results in unrealistic power predictions was previously debated
in Subsection 2.3.1. This chapter presents a novel method for analysing flight-test data
of a helicopter gas turbine engine. The so-called “Multivariable Polynomial
Optimization under Constraints” (MPOC) method is capable of providing an
improved estimation of the engine maximum available power. The MPOC method
relies on optimization of a multivariable polynomial model subjected to equalities and
inequalities constraints. The Karush-Khun-Tucker (KKT) optimization method is
used with the engine operating limitations serving as inequalities constraints.

This Chapter 3 was published as a journal paper (i) and as a conference paper (ii):

i. Arush, I., and Pavel, M.D., “Helicopter Gas Turbine Engine Performance Analysis: A
Multivariable Approach”, Proceedings of the Institute of Mechanical Engineers, Part G:
Journal of Aerospace Engineering, Vol. 223, No. 3, March 2019.
ii. Arush, I., & Pavel, M.D., “Flight testing and analysis of gas turbine engine performance:
A multivariable approach.” In C. Hermans (Ed.), Proceedings of the 44th European
Rotorcraft Forum: Delft, The Netherlands, September 2018.

83
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

The MPOC method is applied to a set of flight-test data of a Rolls Royce/Allison

MTU250-C20 gas turbine, installed on a MBB BO-105M helicopter. It is shown that
the MPOC method can predict the engine output power under a wider range of
atmospheric conditions and that the standard deviation of the output power estimation
error is reduced from 13hp in the single-variable method to only 4.3hp using the
MPOC method (over 300% improvement).

3.2 I NTRODUCTION

Flight testing is an expensive activity that requires efficient methods for

determining correctly the helicopter performance. Such methods involve
considerations regarding testing techniques and data reduction of the raw flight-test
data. This chapter relates to the flight-test methodology performed for defining the
maximum available power of a helicopter gas-turbine engine. Unlike the conventional
single-variable method, the novel method presented in this chapter is based on
multivariable polynomials defined for the engine parameters, i.e., shaft output power,
compressor speed, temperature and fuel-flow. It is shown that this multivariable
approach results in more realistic and accurate modelling of the gas-turbine engine
output power.

This chapter is structured as follows: following this short introduction, the

conventional single-variable method is applied in Section 3.3 to a set of authentic flight
test data (34 stabilized test points) of a Rolls Royce/Allison MTU250-C20 gas turbine
engine, installed as the left engine on a MBB BO-105 helicopter used for training at
the National Test Pilot School in Mojave, California. The two phases of the
conventional methodology, as previously presented in Chapter 2 (Subsection 2.3.1),
are closely executed to determine the maximum available power of this particular
MTU250-C20 gas-turbine engine. In Section 3.4 the novel MPOC methodology is
presented and demonstrated by using the same flight test data used for the
conventional single-variable method. Final conclusions and recommendation are
provided in Section 3.5.

84
 3.3 | T H E C O N V E N T I O N A L S I N G L E - V A R I A B L E M E T H O D

3.3 T HE C ONVENTIONAL S INGLE -V ARIABLE

M ETHOD

The useful performance of any helicopter depends on the amount by which

the power available exceeds the power required [79]. The conventional single-variable
method widely used by the flight-test community for determining the maximum
output power of the helicopter engine is based on gathering stabilized engine(s)
parameters (such as temperature, compressor speed, fuel-flow and shaft output power)
accompanied by their corresponding atmospheric conditions prevailed during the test
[18-21]. These flight-test data are collected while flying the helicopter throughout its
certified envelope and collecting engine parameters to their approved operating
limitations. Once a substantial data base is gathered it can be analysed with the final
goal of deriving the maximum shaft output power that the turbine engine can deliver
under various combinations of atmospheric conditions. One should comprehend that
the limiting factor for the maximum output power could change under different
atmospheric conditions. For example, under hot day conditions the engine maximum
output power could be limited by the engine temperature, while under relatively cold
day conditions the engine compressor speed could limit the maximum output power
the engine can deliver. The flight-test methodology must provide the answer to the
following two questions: what is the maximum output power, and what is the related
limiting factor. The limiting factor can be either one (or a combination) of the engine
temperature, the engine compressor speed or the fuel flow to the engine. Another
common power-limiting factor is the maximum transmission torque. Although this
limiting factor is not an engine limitation ‘per-se’, it has a fundamental effect on
maximum output power of the engine.

3.3.1 Phase I – Engine ‘Rules of Operation’

The first step in analysing the specific BO105 gas turbine engine data is to ‘correct’
or ‘non-dimensionalize’ the raw flight-test data of 34 stabilized test points. The four

85
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

engine parameters: shaft output power, compressor speed, temperature and fuel-flow
are corrected using the corresponding atmospheric ambient conditions and are
converted into, CSHP, CNg, CTGT, and CWf respectively. As previously mentioned
in Chapter 2, the definitions of these non-dimensional variables are presented in the
nomenclature and the rigorous procedure to derive these corrected variables is
provided in Appendix A.

By applying common methods of linear regression, the following set of third

order single-variable polynomials is retrieved to relate between the corrected output
power and each one of the other corrected variables of the specific gas-turbine engine,
as given by Eq.(3.1), Eq.(3.2), and Eq.(3.3). Note that third order polynomials are
employed for the reason they are the lowest order that enable modelling an inflection
point, a fundamental behaviour of the gas-turbine engine.

CSHP  f1  CNg   a1  CNg 3  a2  CNg 2  a3  CNg   a4 

 
  (3.1)
 a a a a  0.009947 2.95 273.47 8153.2 
 1 2 3 4  

CSHP  f 2  CTGT   a1  CTGT 3  a2  CTGT 2  a3  CTGT   a4 

  (3.2)
 
 
  a1 a2 a3 a4    3.328 10 0.068 43.87 9256.7  
5

CSHP  f  CW   a  CW 3  a  CW 2  a  CW   a 
 3 f 1 f 2 f 3 f 4

  (3.3)
 
 
  a1 a2 a3 a4    9.37 10 0.002 2.56 234.32 
6


These polynomials, based on actual flight-test data, serve as empirical models

to represent the ‘rule of operation’ of the specific MTU250-C20 gas turbine engine,
installed as the left engine on the specific BO-105 helicopter, and at the specific phase
of its life cycle. Each polynomial is treated like the ‘finger print’ of the specific installed
engine in the particular helicopter, representing the mathematical relationship between
the corrected output power and the separate corrected engine variable (temperature,

86
 3.3 | T H E C O N V E N T I O N A L S I N G L E - V A R I A B L E M E T H O D

compressor speed and fuel-flow). Figure 3.1 presents the 34 stabilized test points of
the specific MTU250-C20 gas turbine engine, accompanied by the three best-fit third
order polynomials specified in Eq.(3.1), Eq.(3.2) , and Eq.(3.3).

As previously mentioned, the requirement for these polynomials to be of the

third order is to ensure the empirical model captures an inflection point of the engine
performance. This inflection point represents a fundamental property of any gas-
turbine engine for which the rate of change in output power with respect to a particular
engine variables (compressor speed, temperature, or fuel-flow) changes its sign. For
the same amount of engine variable increase, the resulted increase in engine output-
power is to be reduced beyond the inflection point, as compared to the output-power
increase below this inflection point.

Figure 3.1. Nom-dimensional single variable engine performance. Data represents 34

stabilized engine operation points of a RR/Allison MTU250-C20 engine installed as a left engine
on a BO105 helicopter. The corrected engine output power (CSHP) is separately presented
against each of the other corrected variables: corrected engine temperature (CTGT), corrected
compressor speed (CNg), and corrected fuel-flow (CWf, presented in pounds per hour units).

87
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

3.3.2 Phase II – Maximum Available Power

The second phase of the single-variable method uses the empirical models
retrieved from Phase I (Eq.(3.1), Eq.(3.2), and Eq.(3.3)) to define the maximum
available power the installed engine can deliver, for any desired power rating and under
any atmospheric conditions as selected by the flight tester. This phase involves an
iterative procedure, as previously explained in Chapter 2 (Subsection 2.3.1.2) and
illustrated by Fig. 2.17. The data presented in Fig. 3.2 were derived by following the
relevant iterative procedure with the specific polynomials (Eq.(3.1), Eq.(3.2), and
Eq.(3.3)). Figure 3.2 shows the synthesised data for up to 12,000 ft. of pressure-altitude
and for five distinct day conditions; a standard day (ISA), 10°C and 20°C hotter than
standard, and 5°C and 10°C colder than standard day conditions. Figure 3.2 presents
the estimated maximum continuous output power of the engine based on a set of 34
stabilized engine flight-test data points.

The continuous power rating of this type of engine was set at engine
temperature of 738°C and compressor speed of 105%. For the fuel-flow a fictitious
limitation (@ 450 pounds per hour) was used. Note that for this specific type of engine
and under the atmospheric conditions used for Fig. 3.2, the engine fuel-flow is known
to be a non-limiting factor. The maximum continuous power limitation associated with
the transmission torque was set at 344 hp. It can be easily seen from Fig. 3.2 that for
ISA, ISA-5 and ISA-10 day conditions the helicopter maximum power is limited by
the transmission, from sea-level up to 790 ft., 2800 ft. and 3800 ft. above sea-level
respectively. For higher pressure-altitudes the limiting factor swaps from the
transmission to the engine temperature. As for ISA+10°C and ISA+20°C day
conditions, the analysis suggests the engine output power is expected to be
temperature limited immediately above sea-level.

88
 3.3 | T H E C O N V E N T I O N A L S I N G L E - V A R I A B L E M E T H O D

Figure 3.2. Estimated maximum continuous power of the example engine. Note the
specific MTU250-C20 engine installed as the left engine on the tested BO105 helicopter is
transmission limited for continuous operation under ISA, ISA-5°C and ISA-10°C conditions.

The major disadvantage of this single-variable analysis method lies in the intrinsic
assumption of independency between the rules of operation in all three engine limiting
factors. This drawback manifests itself by the unrealistic relative behaviour of the three
lines of ISA, ISA-5°C and ISA-10°C crossing each other above pressure-altitude of
8,000 ft. as seen in Fig. 3.2. It is physically impossible for a temperature limited engine
to deliver more power whilst the ambient temperature is higher.

The absolute errors between the actual measured engine output power and the
corresponding predicted values using the reduced polynomials (Eq.(3.1), Eq.(3.2), and
Eq.(3.3)) are calculated as per Eq.(3.4),(3.5),(3.6), and are presented in Fig. 3.3.

Er
CNg
 
 CSHPi  f1  CNgi  , i  1,,34 (3.4)

Er
CTGT
 
 CSHPi  f 2  CTGTi  , i  1,,34 (3.5)

Er
Wf
 
 CSHPi  f3 W f i  , i  1,,34 (3.6)

89
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

Figure 3.3. The MTU250-C20 engine power estimation errors using single-variable
models. Note the relative large estimation errors of up to 30 hp using the engine temperature
variable.

These errors were found to be normally distributed about a practically zero mean.
Figure 3.4 shows the error standard deviation for each prediction channel plotted
against its relevant error mean. This figure also includes a horizontal bar to represent
the 95% confidence level interval range for the mean of the error. This bar shows
where the mean of the error can be found for the 95% confidence level. Inspecting
this figure one can immediately see that the output power prediction, based on engine
temperature (Eq.(3.2)) presents the worst performance; the relevant standard deviation
of this error is 13 hp. and under 95% confidence level the mean of the estimation
could be found anywhere along a range of ±4.6 hp. For the specific engine/helicopter
combination tested, a standard deviation of 13 hp is considered a substantial error
value for power predictions.

90
 3.4 | T H E MP O C M E T H O D FOR ENGI NE AVAILABLE POW ER DETERMINATION

Figure 3.4. Mean and standard deviation of the single-variable estimation errors.
The engine temperature based estimation presented the worst performance with an error
standard deviation of 13 hp.

Concluding, the conventional single-value method used for determining the

maximum output power of the helicopter gas-turbine engine can result in large errors
and unrealistic prediction trends. Next section presents a novel, more accurate method
for available power determination of a helicopter gas-turbine engine.

3.4 T HE MPOC M ETHOD FOR E NGINE

A VAILABLE P OWER D ETERMINATION

This section presents a novel flight-test method, referred to as ‘Multivariable

Polynomial Optimization under Constraints’ (MPOC), for the task of helicopter gas-
turbine engine available power determination. This method requires no change to the
way engine performance flight-test sorties are carried out, only to the flight-test data
analysis. Using the elegant method of projection onto subspaces a list of mathematical
candidate models is derived to best represent the relationship between the engine
output power and the engine other variables. The maximum output power of the

91
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

engine is assessed as an optimization problem under constraints. The nature of this

optimization (maximization) problem has both equalities and inequalities constraints.
For this, the Karush-Khun-Tucker (KKT) optimization method which deals with both
type of constraints is utilized. The MPOC method presented in this section hereinafter
is exemplified with the same MTU250-C20 gas turbine engine flight-test data, used for
the conventional single-variable method demonstration in Section 3.3.

The MPOC method is implemented in four phases; phase I is to establish a list

of candidate multivariable models to describe the gas-turbine engine rules of operation,
Phase II is to fit the candidate models with experimental flight-test data, Phase III
concentrates on choosing the right empirical model to represent the gas-turbine engine
performance, and in Phase IV the chosen empirical model is used for estimating the
maximum available power under a wide range of atmospheric conditions.

3.4.1 Phase I – Multivariable Empirical

Models for the Rules of Operation

A convenient mathematical relationship needs to be found for representing the

flight-test data. Polynomials serve great role in flight-testing due to their simplicity
which makes them suitable candidates for best-fit type models. Various math model
search algorithm were developed in the literature of specialty for optimizing regression
models of multivariate experimental data obtained in aviation. For examples see
Ulbrich [80, 81] and Zhao and Xue [82]. The MPOC method seeks for a multivariable
polynomial limited to the third order as in the conventional single-variable method.
The first step of the MPOC method is finding candidate multivariable polynomial
models to relate between the corrected shaft output power (CSHP), the corrected
compressor speed (CNg), corrected engine temperature (CTGT) and corrected fuel
flow to the engine (CWf). For simplification and based on common practice, six basic
two-variable polynomials of the third order are defined using the three independent
engine variables. This results in six different combinations as presented in
Table 3.1. Each mathematical term presented in Table 3.1 yields six lower order terms

92
 3.4 | T H E MP O C M E T H O D FOR ENGI NE AVAILABLE POW ER DETERMINATION

resulting in a long list of 42 regressors (predictors). However, many of the lower order
terms are merely duplicates and can be dismissed. Filtering out repeating terms gives
an updated list of regressors as presented in Table 3.2. This table corresponds to a list
of 18 candidate regressors to work with for a best fit mathematical expression under
the generic expression as given by Eq.(3.7).

Table 3.1. Third order polynomials for GTE performance modeling. This list of third-
order polynomials and their lower-order terms yields the empirical model regressors.

# Mathematical term List of lower-order terms

1 (CNg)3(CTGT) CNg, (CNg)2, (CNg)3, CTGT, (CTGT)(CNg), (CTGT)(CNg)2

2 (CNg)3(CWf) CNg, (CNg)2, (CNg)3, CWf, (CWf)(CNg), (CWf)(CNg)2

3 (CTGT)3(CNg) CTGT, (CTGT)2, (CTGT)3, CNg, (CNg)(CTGT), (CNg)(CTGT)2

4 (CTGT)3(CWf) CTGT, (CTGT)2, (CTGT)3, CWf, (CWf)(CTGT), (CWf)(CTGT)2

5 (CWf)3(CNg) CWf, (CWf)2, (CWf)3, CNg, (CNg)(CWf), (CNg)(CWf)2

6 (CWf)3(CTGT) CWf, (CWf)2, (CWf)3, CTGT, (CNg)(CTGT), (CNg)(CTGT)2

      n  18
n
CSHP  f CNg , CTGT , CW f   0    i fi CNg , CTGT , CW f (3.7)
i 1

Table 3.2. Empirical model predictors. An updates list of regressors for best fit hierarchical
math regression model.

Single Variable Regressors Double Variable Regressors

f1=(CNg)3 f10=(CNg)(CTGT)
f2=(CNg)2 f11=(CNg)(CWf)
f3=CNg f12=(CTGT)(CWf)
f4=(CTGT)3 f13=(CNg)2(CTGT)
f5=(CTGT)2 f14=(CNg)2(CWf)
f6=(CTGT) f15=(CTGT)2(CWf)
f7=(CWf)3 f16=(CNg)(CTGT)2
f8=(CWf)2 f17=(CNg)(CWf)2
f9=(CWf) f18=(CTGT)(CWf)2

93
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

With the 18 derived regressors one has an enormous amount of possible

models to check. The case can be thought as a combination of 1, 2, 3… 18 functions
from a set of 18 regressors i.e., 262,143 possibilities as per Eq.(3.8), for which N
represents the number of possibilities.

18  18  18  18! 18! 18!

N              262,143 (3.8)
1 2 18  1!17! 2!16! 0!18!

The number of possible combinations can be reduced by setting a base model

which is a linear combination of the elementary regressors f1 to f9 (Eq.(3.10)). The
polynomial as given by Eq.(3.10) is addressed in this chapter as model number 1. This
way, the problem has been reduced to finding a model which will be constructed from
Model 1 superimposed with any combination of the regressors f10 to f18. The number
of combinations is now reduces to 512 as per Eq.(3.9).

9 9 9 9! 9! 9!

N '  1           1     512 (3.9)
   
1 2  
9 1!8! 2!7! 0!9!

     
  31 CN g   41  CTGT    51  CTGT  
3 2 3 2
CSHPM 1  11 CN g   21 CN g
(3.10)
  61  CTGT    71 CW f      
3 2
 81 CW f   91 CW f   01

This still represents a substantial number of combinations but more

manageable. Within the limited scope of this chapter, a performance comparison
between ten different models from the 512 is presented. Model 1 presented as
Eq.(3.10) is merely being added with the nine regressors (f10 to f18 of Table 3.2), one at
a time. This process of providing candidate multivariable polynomials is presented
mathematically as Eq.(3.11). Equation (3.12) presents the suggested model number 4
(CSHPM4) as a particular case of the generic formula described by Eq.(3.11).

K 8
   
9
CSHPMK   iK fi CN g , CTGT , CW f    Kj f j CN g , CTGT , CW f
i 0 j 10 (3.11)
 f0 1, K  1,2,,10

94
 3.4 | T H E MP O C M E T H O D FOR ENGI NE AVAILABLE POW ER DETERMINATION

     
  34 CN g   44  CTGT    54  CTGT  
3 2 3 2
CSHPM 4  14 CN g   24 CN g

  64  CTGT    74 CW f         CTGT  
3 2
 84 CW f   94 CW f  10
4
CN g (3.12)
 11
4
CN g CW f   124 CTGT  CW f   04

3.4.2 Phase II – Fitting the suggested

models with experimental data

This subsection presents the method used to fit the ten proposed multivariable
models (Eq.(3.11), for M=1 to 10) with actual experimental flight-test data. The
method used is based on a linear Algebra concept known as projection onto subspaces
[83] and is demonstrated hereinafter for Model 1. The 34 flight-test data points of the
example MTU250-C20 gas turbine engine considered in this chapter are next
substituted in Eq.(3.10). This gives a linear system of 34 equations with ten unknowns
(the coefficients  n1 ). This system of equations is compactly represented as Eq.(3.13).

 A    b (3.13)

The matrix A is of size of (34x10) and contains the numerical regressors as

columns, α is a column vector (34x1) containing the unknown coefficients and 𝑏⃗ is a
column vector (34x1) representing the measured experimental corrected output power
of the engine (CSHP). Substituting the regressors of the proposed model number 1
into Eq.(3.13) gives Eq.(3.14).

 11 
 
  CNg  3
 CNg1 2
CNg1 CTGT1  3
CTGT1  2
CTGT1 CWf1  3
CWf1  2
CWf1 1   2 
1
 1     CSHP1 
  CNg 3  CNg 2 2  CTGT2 3 CTGT2 2 CWf 2 3 CWf 2 2 1   3   CSHP2  (3.14)
1
CNg 2 CTGT2 CWf 2
 2
  1  
  CNg 3  CNg3 2 CNg 3  CTGT3 3 CTGT3 2 CTGT3 CWf 3 3 CWf 3 2 CWf 3 1   4   CSHP3 
 3
  1  
 . . . . . . . . . .   5   . 
    
. . . . . . . . . . 6  1 .
     
 . . . . . . . . . 
.   1  . 
   7   CSHP 
  CNg33   CNg33 2 CTGT33 3 CTGT33 2  CWf33 3  CWf33 2
3
CNg 33 CTGT33 CWf33 1   8  
1 33 
     CSHP34 
  CNg 3  CNg34 2 CNg 34  CTGT34 3 CTGT34 2 CTGT34 CWf 34 3 CWf 34 2 CWf 34 1   91 
 34
1 
 0

95
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

This system of equations is over-determined and does not have an exact

solution. However, one can look for the “closest” solution for this system, i.e. the
“best-fit” solution. This best-fit solution is denoted as { ̂ }. The matrix constructed
from [ATA]-1AT is the projection matrix which when multiplied by the vector 𝑏⃗ yields
a solution in a subspace of matrix A (Eq.(3.15)). This solution serves as a best-fit or
the closest solution one can determine.

1
ˆ    AT A AT  b (3.15)

Following the above-described procedure one can immediately solve for the
10 coefficients of model number 1, see Eq.(3.16)

    A
1
1 T
A AT  CSHP (3.16)
i 

For the numerical set of flight-test data exemplified in this chapter, model
number 1 as given in Eq.(3.10) is presented as Eq.(3.17).

11 
 1   0.0105 
 2   2.8486 
 1  
 3   250.48 
 1  
 4   2.386  10 
5

 1   0.046874 
 
1
i
  
  5  

 (3.17)
 6  
1 30.406 
 1   8.556  105 
 7   
 1   0.043963 
 8  
 1   5.6956 
 9   945.18 
 01 

Similar procedure was repeated for all other nine candidate models.

96
 3.4 | T H E MP O C M E T H O D FOR ENGI NE AVAILABLE POW ER DETERMI NATION

3.4.3 Phase III – selecting the right model

for the task

Consider the prediction errors of models number 1 to 10 per an experimental

data point as presented in Fig. 3.5 and calculated according to Eq.(3.18).

Er
MK
 
 CSHPi   CSHPMK i  i  1,34, K  1,,10 (3.18)

For completeness reasons, Fig. 3.5 includes data obtained from the
conventional single-variable analysis method presented in Fig. 3.3. Looking at Fig. 3.5
one can see that, even before any statistical tool is used, each MPOC proposed
multivariable polynomial is performing better in predicting the engine output power
as compared to the conventional method. However, only one empirical model is
required. Since a projection from a limited sample of experimental flight-test data to
the entire population needs to be made, inferential statistics tools is utilized. In general,
an empirical model is best replicating the experimental data if both the mean and
variance of the estimation errors are zero. Obviously, this hypothetical perfect model
is not to be found, however the following two approaches look for the closest one.

Figure 3.5. Estimation errors for the 10 proposed multivariable models. This figure also
includes the estimation errors yielded by the single-variable method. The multivariable empirical
models performed far better in estimating the output power of the MTU250-C20 gas turbine
engine, as compared with the experimental data.

97
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

(1) The p-value approach. The p-value approach (‘p’ stands for probability)
is used to compare between the different ten proposed models. The idea behind the
p-value is thoroughly discussed by Guttman et al. [84]. This statistical test concept
involves stating two contradicting hypotheses and use the experimental data to either
support or to reject the first hypothesis (the Null-Hypothesis, H0). In our analysis H0
is set to claim that each of the multivariable models has an array of estimation errors
with a zero mean. The level of significance for this statistical analysis was set at 1%
(meaning 99% of confidence level). The p-values returned from normal distribution
tables represent the smallest significant level that lead to rejecting the Null-Hypothesis.
In general, low p-values cast a doubt on the validity of the Null-Hypothesis and once
submerge under the significance level of the test, the Null-Hypothesis must be rejected
and the Alternative-Hypothesis should be accepted instead. One may think about the
p-value as the probability that one would observe a more extreme statistic than actually
observed if the Null-Hypothesis were true. All models except for model number 10
strongly supported the Null Hypothesis for the 1% significance level set. All first 5
models returned similar p-values, ranging from 0.999 to 1 with model number 2 being
the only one to return a computed p-value of 1. The p-value approach resulted in the
elimination of model number 10 from the list.

(2) Mean-Variance Plane. A complementary approach to the p-value concept

was to compare the models performance on the mean-variance plane. Figure 3.6
presents the paired values of mean and standard-deviation (square root of the variance)
of the estimation prediction errors obtained for the first nine proposed models.

98
 3.4 | T H E MP O C M E T H O D FOR ENGI NE AVAILABLE POW ER DETERMINATION

Figure 3.6. Various multivariable empirical models performance. This figure presents the
nine multivariable empirical models performance on the mean-standard deviation plane. Model
number 10 was omitted from this figure due to an outstanding mean of estimation error of 4 hp.

Concluding from the two approaches and the relative performance of all ten
multivariable empirical models involved, model number 2 (Eq.(3.19),(3.20)) was
selected as the one to best represent the engine output power. Model number 2 is
further used in the subsequent Subsection 3.4.4 for the demonstration of the MPOC
method.

     
  32 CN g   42  CTGT    52  CTGT  
3 2 3 2
CSHPM 2  12 CN g   22 CN g
(3.19)

 62  CTGT    72 CW f       
CTGT  CN g   02
3 2
 82 CW f   92 CW f  10
2

 12   0.0165 
 2  
  2   3.837 
 2  
  3   380.69 
 2  5 
  4   3.36  10 
  2   0.075 
 5   
  i
2
   62    41.809  (3.20)
 2  5 
  7   8.35  10 
 2  
 8   0.043 
  2   5.577 
 9  
10 2  
0.1486 
 2  
  0   2242.4 

99
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

3.4.4 Phase IV – maximum gas-turbine

engine output power estimation

Once acquiring a multivariable polynomial to best describe the change in

corrected engine output power based on other engine corrected parameters
(compressor speed, temperature and fuel-flow), one can look for the maximum
available output power of the engine under various atmospheric conditions. The
engine output power is limited by reaching one (or more) of its parameters.
Determining the maximum output power is equivalent to a mathematical problem of
finding an extremum point (maximum output power) under constraints (the engine
variables: compressor speed, temperature and fuel-flow). Finding an extremum point
of a multivariable function under constraints is of a totally different nature from the
case of extremum of a single-variable function. The typical approach for the
multivariable case is to use the Lagrange multipliers, but this approach works with
equalities constraints only, whereas the problem we have in hand involves both
equalities and inequalities constraints.

One applicable method for optimization under both equalities and inequalities
constraints is the KKT (Karush-Kuhn-Tucker) thoroughly discussed by Singiresu [85].
According to this KKT approach, Eq.(3.21) provides the general Lagrange equations
required for satisfying extremum points of a multivariable function f(xi) subjected to
‘m’ number of inequalities constraints, g(xi), and ‘l’ number of equalities constraints
given by h(xi). As per Eq.(3.21) ηj represent the Lagrange multipliers associated with
the inequalities constraints and λk represent the Lagrange multipliers associated with the
equalities constraints.

f m  g j  l  hk 
   j       0  i  1, 2,3,..., n
xi j 1  xi  k 1  k xi 
x   x1 , x2 ,..., xn  (3.21)
g j ( x)  0   j  1, 2,..., m 
hk ( x)  0   k  1, 2,..., l 

100
 3.4 | T H E MP O C M E T H O D FOR ENGI NE AVAILABLE POW ER DETERMINATION

The function to be maximized is the empirical model number 2 (Eq.(3.19)

,(3.20)) subjected to several engine operational constraints. For this specific
optimization problem to be solvable, at least two equality constraints need to be
provided. Those are fulfilled with the engine internal rule of operation, as explained
hereinafter. Implementing similar approach as described in Subsection 3.4.3 above
with the p-value and comparative evaluation on the mean-standard deviation plane, a
best-fit surface was calculated to constitute the example MTU250-C20 gas turbine
engine multivariable internal rule of operation. This type of surface which describes the
relationship between the corrected engine temperature (CTGT) and both the corrected
compressor speed (CNg) and the corrected fuel-flow (CWf), complemented with the
experimental data points, is presented in Fig. 3.7.

The first equality constraint denoted as h1 and presented in its implicit form as
Eq.(3.22) relates between the corrected engine temperature and the corrected
compressor speed. The second equality constraint is denoted as h2 and represents
relationship between the corrected compressor speed and the corrected fuel-flow
(Eq.(3.23)). Note that h1 and h2 constraints are projections of the multivariable rule of
operation onto two planes; the CTGT-CNg plane and the CNg-CWf plane,
respectively

h1 : CTGT  a1  CNg   a2  CNg   a3  CNg   a4  0

3 2

(3.22)
 a1   0.0117 
a   
 2  2.9739 
    
 a3   258.49 
a4   7050 

     
3 2
h2 : CNg  b1 CW f  b2 CW f  b3 CW f  b4  0
(3.23)
 b1  6.492  106 
b   
 2   0.00433 
    
b3   1.0621 
b4   2.9888 
 

101
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

Figure 3.7. The engine internal rule of operation. This figure presents the relationship
between the engine corrected temperature and the engine corrected compressor speed and
corrected fuel-flow. The circles plotted are the example MTU250-C20 engine data points, which
few are obscured by the best-fit surface.

The inequalities constraint for the engine maximum output power are simply
the operational limitations imposed on the engine. For the exemplary MTU250-C20
gas turbine engine those are the continuous rating of the engine, denoted as g1 to g3
and are presented as equations (3.24) to (3.26).

105
g1 : CNg  0 (3.24)


738
g 2 : CTGT  0 (3.25)


450
g3 : CW f  0 (3.26)
 

The partial differential equations based on Eq.(3.21) and the KKT conditions
specified as equations (3.24) to (3.26) for a maximization problem result in equations
(3.27) to (3.29).

102
 3.4 | T H E MP O C M E T H O D FOR ENGI NE AVAILABLE POW ER DETERMINATION

  CSHPM 2    h1 
 1  1  2  0 (3.27)
  CNg    CNg 

  CSHPM 2 
 2  1  0 (3.28)
  CTGT 

  CSHPM 2    h2 
 3  2 0 (3.29)

 CW f  
 CW f 

Equations (3.27) to (3.29) can be rearranged compactly as presented in

Eq.(3.30).

   CSHP  
 M2  
  h1   
   CNg    1 0 0  1   1
     CNg    2 
   CSHPM 2      
  CTGT    0 1 0 1 0    3  (3.30)
       
  h2 
   CSHP    0 0 1 0    1 
 M2
  
 CW f    2 

  CW f
   



The system of partial differential equations (Eq.(3.30)) describes conditions for

candidate engine corrected variables representing maximization of the engine output
power. This set of equations does not have a unique solution but a solution with two
degrees of freedom for the three distinct cases it represents. The first case (Case I) is
when the compressor speed is at its maximum value, i.e., the engine output power is
limited by the compressor speed. The second case (Case II) is when the output power
is limited by the engine temperature and the last case (Case III) represents a fuel-flow
limited engine. Splitting Eq.(3.30) into the three individual cases and applying the KKT
conditions on the Lagrange multipliers associated with the inequalities constraints (η1,
η2, η3) eliminates the two degrees of freedom and makes each one of these cases to
have a unique solution. The three cases are demonstrated hereinafter:

103
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

1) Case I – compressor speed limited engine.

Application of the relevant KKT conditions for this case imposes the
following conditions on the Lagrange multipliers associated with the inequalities
constraints (Eq.(3.31)).

1  0,2  0,3  0 (3.31)

Combining Eq.(3.31) and Eq.(3.30) results in the following system of equations

(Eq.(3.32)):

   CSHP    1  0 
 M2  
  h1    
   CNg    1  1   CNg  105 
     CNg    1    
   CSHPM 2        
  CTGT    0 1 0  
  1   738  (3.32)
    
  h2     CTGT 
 
   CSHP    0 0    2 
 M2
  
 CW f    CW  450 

  CW f
   

 

f
  

2) Case II – temperature limited engine.

Application of the relevant KKT conditions for this case imposes the
following conditions on the Lagrange multipliers associated with the inequalities
constraints (Eq.(3.33)).

1  0,2  0,3  0 (3.33)

Substituting Eq.(3.33) into Eq.(3.30) results in the following system of

equations (Eq.(3.34)):

104
 3.4 | T H E MP O C M E T H O D FOR ENGI NE AVAILABLE POW ER DETERMINATION

   CSHP    2  0 
 M2  
  h1    
   CNg    0  1   CNg  105 
     CNg    2    
   CSHPM 2        
  CTGT    1 1 0  
  1   738  (3.34)
  CTGT 
     h2      
   CSHP    0 0    2 
 M2
  
 CW f    CW  450 

  CW f
   

 

f
  

3) Case III – fuel-flow limited engine.

Finally, the third case is when the maximum output power of the engine is
bounded by reaching the maximum fuel-flow the pump is capable of delivering to the
engine. Application of the KKT conditions for this case imposes the following
conditions on the Lagrange multipliers associated with the inequalities constraints
(Eq.(3.35)).

1  0,2  0,3  0 (3.35)

Combining Eq.(3.35) with Eq.(3.30) results in the following set of equations

(Eq.(3.36)):

   CSHP    3  0 
 M2  
  h1    
   CNg    0  1   CNg  105 
     CNg    3    
   CSHPM 2        
  CTGT    1 1 0  
  1   738  (3.36)
    
  h2     CTGT 
 
   CSHP    1 0    2 
 M2
  
 CW f   CW  450 

  CW f
   

 

f
  

For demonstration purposes, the specifics of Case II (temperature limited

engine) are used with the exemplary MTU250-C20 gas turbine engine flight-test data.
Similar methodology can be applied to find the maximum output power for the other

105
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

two cases, compressor speed limited engine (Case I) and fuel-flow limited performance
(Case III).

The set of equations specified in Eq.(3.34) has a solution if and only if (IFF)
the rank of the system matrix is the same as the rank of the auxiliary matrix. This
solution would be unique if both ranks equal three (the three unknowns of the problem
which are the Lagrange multipliers). This requirement for a unique solution can be
stated mathematically as in Eq.(3.37).

   CSHPM 2  
0   1
 h
  1 
  h1     CNg    CNg  
0  1 
   CNg    
     CSHPM 2  
rank  1 1 0   rank  1 1 0 3 (3.37)
     CTGT  
  h2   
0  
0   h2    CSHPM 2  

 
 CW f  


0 0  

 
 CW f  
 CW f  


Instead of pursuing for a pair of corrected compressor speed (CNg) and

corrected fuel-flow (CWf) under a limited corrected temperature (CTGTlimit) to satisfy
Eq.(3.34), one can simplify the process by using the following “back-door” approach:
for each and every combination of atmospheric conditions a pair of candidate
corrected compressor speed and corrected fuel-flow will be suggested via the engine
internal rule of operation (Eq.(3.22) and (3.23)). These candidate pairs complemented
with the engine temperature limit will then be evaluated for fulfilment of the KKT
conditions required for maximization of the engine output power. Since the
equations specified in Eq.(3.34) have a unique solution, they can be rearranged as in
Eq.(3.38). The three engine parameters (candidates for maximum output power) can
be used in Eq.(3.38) to solve for the Lagrange multipliers. The three candidate
simultaneous engine parameters are then proved valid, as ones that define a maximum
output power of a temperature limited engine, if and only if the solution of the system
specified as Eq.(3.38) is achieved while coinciding with the KKT conditions required
for the case.

106
 3.4 | T H E MP O C M E T H O D FOR ENGI NE AVAILABLE POW ER DETERMINATION

 1 1     CSHP    2  0 
 1   M2 
 
 2  
  h1 CNg   h1 
CNg  h2 CW f      CNg  
  
 CNg  105 
  
   1 1     CSHPM 2     (3.38)
 1     h 0 
    KKT :  738 
   1
 2 
CNg   h1 
CNg  h2 CW f      CTGT   CTGT 
 
    CSHP   
 1   M2
  CW  450 
0 0

  h2 CW f  
   CW f
   



f
  

This “back-door” procedure was executed by using the engine internal rules of
operation (Eq.(3.22) and (3.23)) for different type of day conditions (ISA, ISA+10°C,
ISA+20°C, ISA-5°C, and ISA-10°C). Figure 3.8 presents the maximum output power
of the exemplary MTU250-C20 gas turbine engine alongside with all the KKT
requirements as a function of pressure-altitude for an ISA day conditions. It is evident
that all of the KKT requirements are met.

Figure 3.8. A simultaneous presentation of all engine variables. This figure presents the
exemplary MTU250-C20 engine parameters between sea level to 12,000 ft. of pressure altitude
and under standard day conditions (ISA). The engine maximum continuous output power is
limited by its temperature (738°C). Note the fulfilment of all KKT requirements.

The estimated maximum continuous output power of the exemplary MTU250-

C20 gas turbine engine as a function of pressure-altitude for different day conditions
is presented in Fig. 3.9. The maximum continuous output power of the engine is either
transmission limited or temperature limited under all atmospheric conditions

107
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

presented in Fig. 3.9. Note the KKT requirements were omitted from Fig. 3.9 although
they were all met.

Figure 3.9. A simultaneous presentation of all engine variables. This figure presents the
exemplary MTU250-C20 engine parameters between sea level to 12,000 ft. of pressure altitude
and under standard day conditions (ISA). The engine maximum continuous output power is
limited by its temperature (738°C). Note the fulfilment of all KKT requirements.

3.5 M AXIMUM P OW ER E STIMATION

C OMPARISON

The estimated maximum engine output power was compared using both the
conventional single-variable and the MPOC methods. This comparison is presented
in Fig. 3.10. From this figure one can observe that both methods demonstrate similar
results for atmospheric conditions close to those prevailed during the actual flight-tests
(ISA+21°C); however, while the conventional single-variable method completely
collapses under standard (ISA) and colder day conditions, the MPOC method
predicted reasonable and logical estimations for ISA and colder day conditions. The
fundamentally wrong estimation provided by the single-variable method by which a
temperature-limited engine delivers more power under higher ambient temperatures,
is rectified by the MPOC method.

108
 3.6 | S U M M A R Y AND CONCLUSIONS

Figure 3.10. MPOC and single-variable methods comparison. This figure shows that while
the conventional single-variable method collapses under the estimation for engine maximum
continuous output power for standard and colder day conditions, the MPOC method provides
logical maximum output power estimations.

3.6 S UMMARY AND C ONCLUSIONS

The output power of a helicopter gas turbine engine is a multivariable problem

that can be non-dimensionalized as any other physically meaning problem. Over
simplification of the problem as linear combination of single-variable models does not
provide sufficient accuracy and frequently provides unrealistic estimations for
maximum output power under atmospheric conditions different than those prevailed
during the test. The novel method presented in this chapter referred to as the
Multivariable Polynomial Optimization under Constraints, or MPOC for short, is
based on multivariable polynomials. These polynomials demonstrate a substantial
better performance in estimating the output power of an exemplary MTU250-C20 gas-
turbine engine installed in a MBB BO105 helicopter. The P-value concept
complemented with a comparative performance on the mean-standard deviation plane
were used successfully as an inferential statistical tool for sorting between various
candidate multivariable models to represent the gas turbine engine output power.

109
3 | A MULTIVARIABLE APPROACH IN GAS-TURBINE ENGINE TESTING

The prediction of the maximum output power of the gas-turbine engine can
be regarded, mathematically, as an optimization problem of a multivariable function
subjected to both equalities and inequalities constraints. The equalities constraints are
based on the experimental data and the inequalities are provided by the engine
operating limitations. While the conventional single-variable method provides
unrealistic estimations for certain atmospheric conditions, the novel MPOC method
demonstrates adequate prediction performance for a wider range of atmospheric
conditions. Although the conventional single-variable method is simple to use it should
be utilized only as a first estimation and not as a formal analysis tool in the process of
estimating the maximum output power of a gas turbine engine. The approach
presented in this chapter is next expanded in Chapter 4 of this dissertations to include
flight-test data of other types of helicopters and engines. This also includes a
comparative analysis between a broader base of candidate multivariable polynomials
in order to better understand which type of regressors are performing better in
modelling the output power of a gas turbine engine.

110
 Algebra is generous. She often gives more than is asked of her.
Jean le Rond D’Alembert

4 A S INGULAR V ALUE A PPROACH

IN H ELICOPTER F LIGHT T ESTING
A NALYSIS

4.1 C HAPTER O VERVIEW

T he process of empirical models evaluation is at the core business of experimental

flight-testing data analysis. Accurate and convenient flight-testing of helicopter
engine(s) available power is crucial for predicting the total helicopter performance.
Common practice in estimation of in-flight helicopter gas turbine engine power consist
of a reduction of flight-test data into simplistic single-variable analysis approach. While
such an approach is convenient for practical use, it often results in unrealistic
predictions of the available engine(s) power. A novel approach for the gas-turbine
engine maximum available power problem, referred to as the Multivariable Polynomial
Optimization under Constraints (MPOC) method, was introduced in Chapter 3. This
chapter is intended to complement the MPOC method and answer the question of
which multivariable-polynomial can be generally used in representing helicopter gas-
turbine engine performance?

This Chapter 4 was published as the following journal paper: Arush, I., and Pavel, M. D., and
Mulder, M., “A Singular Value Approach in Helicopter Gas Turbine Engine Flight Testing
Analysis”, Proceedings of the Institute of Mechanical Engineers, Part G: Journal of Aerospace
Engineering, April 2020. https://doi.org/10.1177/0954410020920060.

111
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

In this sense, a variety of seven gas-turbine engines installed on different helicopters

are analysed, each one giving 512 possible polynomial models to be used for available-
power calculations. While conventional statistical methods of hypothesis-testing failed
in providing the answer to the question stated above of which the best general
empirical model for representing engine performance is, an alternative approach based
on the Singular-Value-Decomposition (SVD) theorem, was proven successful in
providing the answer. Moreover, this approach presented in this chapter yielded a
short list of ten simple and convenient multivariable-polynomials, best representing the
performance of all seven engines analysed as a group.

4.2 I NTRODUCTION

Flight test engineering is an interdisciplinary science that gathers data and

develops methods with the objective of evaluating an aircraft or a system in its
operational flight environment. This requisite for flight-testing means that the system
or the vehicle under testing requires accurate assessment of its characteristics while
operating in its flight environment rather than just relying on the results of ground-
based verification methods such as wind tunnels, simulators, and software models [1].
There are many disciplines involved in flight-testing based on the nature of the
questions in search. Such ones include, for example, performance assessment,
structural integrity testing, handling-qualities evaluation, etc. Regarding helicopter
performance assessment, the useful performance of any helicopter is directly derived
from the amount by which the engine power (the available-power) surpasses (or falls
below) the power required by the main and tail rotor systems, the drag of the fuselage
and all other consumers of power for the specific conditions [79].

The “off-the-shelf” engine available power as given by the manufacturer

changes once installed in a particular type of helicopter. It typically reduces due to inlet
loss. Moreover, the maximum output-power of the installed engine degrades as it
matures. Therefore, the actual available-power of the installed engine during a
particular phase of its life is of high practicality to helicopter users. This chapter relates

112
 4.2 | I N T R O D U C T I O N

to the methods used in flight-test engineering for measuring helicopter gas turbine
engine performance and estimating the maximum available output power under a wide
range of environmental conditions.

The conventional flight-test method widely used for determining the

maximum power of a helicopter gas turbine engine relies on empirical single-variable
polynomials. This method is thoroughly discussed and demonstrated in Chapter 2
(Subsection 2.3.1) and Chapter 3. The method requires the collection of stabilized
engine parameters while flying the helicopter throughout its operational envelope. The
four main raw engine variables measured in flight (compressor speed, temperature,
fuel-flow and the output power) are normalized (or ‘corrected’) using the surrounding
atmospheric conditions. By applying linear regression methods, three third-order
single-variable polynomials are defined, representing the empirical relation between
the corrected engine power (CSHP) and each one of the three engine corrected
variables: corrected compressor speed (CNg), given by Eq.(4.1), corrected temperature
(CTGT), given by Eq.(4.2) and corrected fuel-flow (CWf), given as Eq.(4.3).

  ai  CN g   n  3
n i
CSHP  f1 CN g  (4.1)
i 0

n
CSHP  f 2  CTGT   bi  CTGT   n  3
i
(4.2)
i 0

    n  3
n i
CSHP  f3 CW f  ci CW f (4.3)
i 0

The maximum available power of the installed engine is next estimated by using
these three empirical single-variable polynomials as demonstrated in Chapter 2
(Subsection 2.3.1.2). The three calculated values of the engine output power are first
compared with each other and then against the maximum transmission torque
(transmission limitation). This comparison is performed through an iterative process
executed for various atmospheric conditions. The maximum available power under
various atmospheric condition is then prescribed as the minimum value out of all four

113
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

values compared. The main advantage of the single-variable method lies in its
simplicity. The flight-tester does not need to be confused with which mathematical
model to choose, since the method is based on third-order single-variable polynomials.
However, this simplicity is also the method’s biggest disadvantage since (1) it requires
careful analysis of the data especially when the required flight conditions are outside
of the limitations of the helicopter; (2) it may not replicate performance limiting factors
that depend on actual flight conditions although matching non-dimensional values has
been targeted successfully; and (3) it frequently yields poor estimations of the
maximum engine output power, especially under atmospheric conditions outside of
the actual tested range. A comprehensive demonstration of the poor estimation using
the single-variable method is presented in Chapter 3 (Section 3.3).

The novel “Multivariable Polynomial Optimization under Constraints” (MPOC)

method to estimate the maximum output power of a helicopter gas turbine engine
more accurately and under a wider range of atmospheric conditions is presented in
Chapter 3. The main advantages to analysing data using MPOC models over the single-
variable method are: (1) it gives the ability to determine the relative influence of one
or more predictor variables to the criterion value; (2) it has the ability to identify
outliers, or anomalies; and (3) it gives a superior estimation precision. As demonstrated
in Chapter 3 for the exemplary MTU250-C20 gas-turbine engine installed in a MBB
BO105 helicopter, the MPOC provided a more accurate engine power estimation (in
excess of 300%) when compared to the single-variable method. However, the main
weakness of the MPOC method is that it struggles with a large number of possible
multivariable-polynomials (more exactly 512 polynomials) to choose from without
clear and decisive guidelines.

The primary objective of this chapter is to address this disadvantage of MPOC

method by developing a systematic and repeatable approach on which specific, pre-
defined multivariable-polynomial models shall be used. For this goal the MPOC
method is applied to a large set of flight-test data gathered from seven different types
of helicopters as presented in Table 4.1. Using the Singular-Value-Decomposition
(SVD) approach, the relative performance of 512 different potential models is

114
 4.2 | I N T R O D U C T I O N

compared towards the objective of identifying the best performing multivariable

polynomial model to be generally used by the MPOC method.

This chapter is structured as follows: right after the introduction, the MPOC
method, as applied to a set of flight-test data gathered from a MBB BO-105 helicopter,
is reviewed. The MPOC review in Subsection 4.3.1 also presents the procedure of
fitting the candidate multivariable polynomials with the flight-test data. Next in
Subsection 4.3.2, the conventional method of hypothesis-testing is used
(unsuccessfully) for the task of screening between all 512 candidate multivariable
models and choosing the best-performing empirical model, with respect to a group of
seven distinct engines (Table 4.1). This unsuccessful screening attempt is then rectified
in Section 4.4 which presents a novel method based on the Singular-Value-
Decomposition (SVD) theorem. This novel screening method was used successfully
with the seven gas-turbine engines in producing a short list of accurate and convenient
multivariable-polynomial models. This list is provided in Table 4.4. Section 4.5 draws
a short comparison between this chapter findings and other similar studies. A summary
and conclusions portion in Section 4.6 completes this chapter.

Table 4.1. Gas-turbine engines used for the analysis. The following table lists the
seven different gas-turbine engines used for the MPOC analysis.

Engine Engine Model Helicopter Installed Rated Pwr. Installation

No. [hp.] Config.
1 RR MTU250-C20B MBB BO-105M 420 Twin
2 Turbomeca Arriel 1E2 Eurocopter EC-145 740 Twin
3 Allison T63-A-700 Bell OH-58C 420 Single
4 Turbomeca Arriel 1C2 Aerospatiale SA365-N2 700 Twin
5 PW-207E MD-902 Explorer 710 Twin
6 Turbomeca Arriel 1M1 Eurocopter AS-565 780 Twin
7 GE T700-GE-701A Sikorsky UH-60A 1700 Twin

115
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

4.3 G AS -T URBINE E NGINE P ERFORMANCE

F LIGHT T ESTING

4.3.1 Principles of MPOC Method

Unlike the conventional single-variable polynomial method, the MPOC method

is seeking for a multivariable-polynomial model representing the engine power while
capturing the interrelation between the engine variables. The maximum engine power
can then be assessed as an optimization problem of a multivariable-function under
constraints. Such a multivariable approach applied to engine analysis results in a more
accurate and realistic available power prediction as it contains the intrinsic couplings
between all engine variables. Chapter 3 demonstrates that the empirical model for the
engine output power (Eq.(4.4)) should rely on a basic model, superimposed with any
possible combination of nine regressors (f1 to f9), as listed in Table 4.2.

CSHPM (i )  CSHPM (1)  f (CN g , CTGT , CW f )

(i 1)
i  1, 2,3,..., 512  Model number i (4.4)

Table 4.2. List of MPOC engine predictors. The following table lists nine engine regressors
to be superimposed on Model 1.

f1=(CNg)(CTGT) f4=(CNg)2(CTGT) f7=(CNg)(CTGT)2

f2=(CNg)(CWf) f5=(CNg)2(CWf) f8=(CNg)(CWf)2
f3=(CTGT)(CWf) f6=(CTGT)2(CWf) f9=(CTGT) (CWf)2

The basic model, referred to as Model 1 and denoted hereinafter as CSHPM(1), is a

third-order multivariable-polynomial in all engine variables given as Eq.(4.5). One
should acknowledge there are 511 different combinations of choosing from regressors
f1 to f9 of Table 4.2, as demonstrated by Eq.(4.6). Adding the basic model 1 with the
511 possible combinations sets the total number of candidate models to be as large as

116
 4.3 | G A S -T U R B I N E E N G I N E P E R F O R M A N C E F L I G H T T E S T I N G

512. Having a list of 512 candidate models is impractical as the flight-tester still needs
to undertake a tedious task of evaluating the performance of each candidate model
(Eq.(4.4)) against the actual flight-test data.

CSHPM (1)  11CN g3   12CN g2   31CN g   14CTGT 3   51CTGT 2   61CTGT 

(4.5)
  71CW f3  81CW f2  91CW f   01 Model number 1

9  9 9 9! 9! 9!
N              511 (4.6)
1  2  9  1! 8! 2! 7! 0! 9!

4.3.2 Hypothesis testing and P-values

All 512 proposed polynomial models can be fitted with actual experimental flight-
test data, yielding the specific coefficients for the best-fit solution. A practical method
to solve for the best-fit coefficient, based on linear concept known as projection onto
subspaces, is thoroughly described by Strang [83] and demonstrated in Chapter 3
(Subsection 3.4.2). The best-fit solution obtained for any candidate polynomial model
can be used to evaluate how precisely this model predicts the actual measured flight-
test data. The corrected engine power (CSHP) is estimated by substituting the measured
independent variables in the model, i.e., corrected engine compressor speed (CNg),
corrected engine temperature (CTGT) and corrected engine fuel-flow (CWf). The
prediction errors of the arbitrary chosen model 122 for each measured data point of
the exemplary MTU250-C20B gas turbine engine of Chapter 3 are then calculated
using Eq.(4.7) and presented graphically in Fig. 4.1. The prediction errors of model
122 are approximately normally distributed about a practically zero mean (actual mean
is -4x10-10 hp).

Er
122
 
 CSHPi   CSHP122 i  i  1,34 (4.7)

The conventional approach in flight-testing assessing prediction goodness is

based on hypothesis testing and the associated p-values assigned. This approach

117
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

follows from the Central Limit Theorem and is thoroughly discussed in literature
[84,87]. In a nutshell, one can set-up a hypothesis (‘the null hypothesis’) with regards
to the mean value of the prediction errors and by using the actual measured data, the
probability of falsely rejecting this hypothesis (making a ‘type-I’ error) is calculated.
This probability numeral is known as the p-value and once it falls under a predefined
value (the statistical significance level) it raises doubts about the statistical validity of
the null hypothesis.

Figure 4.1. Corrected output power prediction errors. This figure shows the MTU250-C20B
gas turbine engine corrected output power (CSHP) prediction errors using the arbitrary
polynomial Model 122.

Once again, the process is demonstrated by using the exemplary MTU250-C20B

engine data and the arbitrary chosen, Model 122. The hypothesis assigned claims all of
model 122 prediction errors have a mean of zero. The test-statistic of this two-sided
case is calculated as per Eq.(4.8) to be an extremely low value of -5.47x10-10.

t122 
E r
122
(4.8)
S n

In Eq.(4.8) the symbol ‘n’ represents the number of measured test-points and ‘S’
stands for the sample standard deviation with respect to the estimation errors of the
engine power. One should realize for this particular case, low test-statistics values

118
 4.3 | G A S -T U R B I N E E N G I N E P E R F O R M A N C E F L I G H T T E S T I N G

return large p-values and vice versa. This extremely low test-statistic value returns a
calculated p-value of 1 (the maximum available due to software rounding errors). There
is no statistical data to support rejection of the null hypothesis, meaning that model
number 122 predicts the MTU250-C20B engine performance with zero mean errors.
Theoretically, this makes Model 122 an excellent multivariable model for the available
power prediction.

4.3.3 Prediction goodness comparison

between candidate models

Assuming the arbitrary-chosen model number 122 is the “perfect” multivariable-

polynomial model to represent the output power of the BO-105 engine, how will all
other 511 candidate models perform? Repeating the previous analysis presented in
Subsection 4.3.2 for all other candidate models returned far too many calculated p-values
of 1. The immediate conclusion one can draw is that the p-value by itself is not an
effective screening tool. Since in our case the p-value and the absolute-value of the
test-statistic are inversely proportional to each other, it is reasonable to use the test-
statistic value itself as an indicator for prediction goodness. The screening process
should be based then on minimum values of the test-statistics in lieu of a maximum p-
values.

Figure 4.2 presents a wide perspective of the test-statistics, mean of prediction

errors and errors standard deviations for all 512 candidate models. Figure 4.3 presents
a closer look (“zooming”) at the test-statistic of a group of only 84 candidate models,
those involving the base model (Eq.(4.5)) superimposed with any combination of three
predictors out of the list of the nine (f1 to f9 in Table 2). Note that each one of those
84 models returned a perfect computed p-value of 1, including models number 62, 88,
107 and 112 which seem to stand out from the group. The conclusion arising from
Fig. 4.3 is that the conventional approach of screening models using the p-value is not
practical for the specific task of finding the best empirical model to represent gas-
turbine engine performance.

119
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

Figure 4.2. Output power prediction performance. This figure shows a wide perspective of
all 512 candidate models performance in predicting the exemplary MTU250-C20B engine output
power.

Figure 4.3. Test-statistics of models number 47-130. This figure shows the test-statistics of
84 candidate models (models number 47-130) involving the base Model 1 (Eq.(4.5))
superimposed with any combination of three regressors from Table 4.2.

The absolute-values of the test-statistics are then used instead of the p-values.
Figure 4.4 presents the test-statistics (absolute-value) of the top ten performing models
for the MTU250-C20B engine. Table 4.3 specifies these models in details. Examining
Table 4.3, no obvious pattern can be detected with respect to which regressors yield
the best prediction performance. Nevertheless, the number of regressors used in the

120
 4.3 | G A S -T U R B I N E E N G I N E P E R F O R M A N C E F L I G H T T E S T I N G

model has no immediate obvious effect on the prediction performance. Within the set
of ten top-performing models there are models which involve additional one, two,
three or four regressors to be superimposed over the basic model number 1. The trivial
question to be asked next is do these 10 top performing models also excel when applied
to different gas-turbine engines? Can findings from the MTU250-C20B engine
installed in the BO-105 helicopter be generalized to other types of helicopter gas-
turbine engines? These enquiries are addressed hereinafter.

Figure 4.4. Top ten performing models. This figure shows the test-statistics (absolute value)
of the top ten performing models for the MTU250-C20B engine installed in the BO-105
helicopter.

Table 4.3. List of 10 top-performing models for the BO-105 helicopter.

Auxiliary Regressors Model Number

Involved † 367 3 53 127 61 12 65 94 199 355
f1=(CNg)(CTGT) x x x x
f2=(CNg)(CWf) x x x x X
f3=(CTGT)(CWf) x x x
f4=(CNg)2(CTGT) x x x
f5=(CNg)2(CWf) x x
f6=(CTGT)2(CWf) x x x x
f7=(CNg)(CTGT)2 x x x
f8=(CNg)(CWf)2 x x x x
f9=(CTGT)(CWf)2 x x x x
† Regressors to be superimposed over the basic Model 1 (Eq.(4.5)).

121
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

Consider next a different type of gas-turbine engine (Engine 2 as per Table 4.1)
installed on a different type of helicopter and a new set of flight-test data. Performing
similar analysis reveals completely different findings from the BO-105 case. Figure 4.5
presents test-statistics of the 10 top-performing models for engine number 2. Further
analysis was undertaken to include flight-test data from five other types of gas-turbine
engines installed on different helicopters, as presented in Table 4.1. Results merely
confirmed the previously stated conclusion that the best performing model to describe
helicopter gas-turbine power, if it exists, cannot be found using a conventional
approach of screening between models using hypothesis testing, neither based on the
p-value nor on the test-statistics. Concluding this section, an alternative general
approach needs to be taken. The alternate approach for screening between empirical
models relates to the Singular-Value-Decomposition (SVD) theorem and is discussed
and demonstrated in the next section of this chapter.

Figure 4.5. Top ten performing models for the EC-145 engine. This figure shows the test-
statistics (absolute value) of the top ten performing models for the Turbomeca Arriel 1E2 engine
installed in the EC-145 helicopter.

122
 4.4 | S I N G U L A R V A L U E S A P P R O A C H FOR MODEL SCREENING

4.4 S INGULAR V ALUES A PPROACH FOR

M ODEL S CREENING

The singular values approach for screening between various engine output power
model candidates is derived from a mathematical theorem known as the Singular-
Value-Decomposition (SVD). This theorem which relates to the field of linear algebra
is briefly introduced in the following subsection, before it is applied for the task of
candidate models screening.

4.4.1 The SVD Theorem

The theory and mechanics of the SVD are thoroughly discussed in Strang [88]. In
a nutshell, this theorem states that any matrix from any size which holds real numbers
as entries can be decomposed as a product of 3 unique and special matrices as shown
in Eq.(4.9). One should view this decomposition as a way of finding convenient
orthogonal bases for both the column-space and the row-space of an arbitrary real
matrix.

 u1,1 u1,2  u1, n 

u    v1,1  vn,1 
u2, n   1
v2,1
  v vn,2 
 2,1 u2,2
2 v2,2 
Z  U V T          1,2
        
      


 r   v1,r v2, r vn,r 
u
 m,1 um,2  um,n  
(4.9)
1   2  ...   r  0

Consider a real matrix Z to be of size ‘m’ by ‘n’ (denoted (m,n)) and rank ‘r’. Matrix
Z can then be expressed as a product of the three unique matrices:

(1) Matrix U called the “left-singular-vectors” (LSV) is an orthonormal matrix of

size (m,r). The columns of this matrix are unity-norm vectors which are orthogonal to
each other. This set of vectors serves as a basis for the column-space of matrix Z.

123
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

(2) Matrix Σ is a diagonal matrix (size (r,r)) which holds the singular-values of Z
as entries along its diagonal. The singular-values are non-negative real numbers which
can be arranged along the diagonal in a descending order.

(3) Matrix V called the “right-singular-vectors” (RSV) is an orthonormal matrix

of size (n,r). The columns of this matrix (or the rows of the transposed matrix, VT) are
unity-norm vectors which are orthogonal to each other. The set of these vectors serves
as a basis for the row-space of matrix Z.

The SVD of a real matrix can alternatively be regarded as a linear combination of

‘r’ rank-one matrices (Eq.(4.10)). This complementary manner to look at the SVD is
referred-to as the spectral decomposition of the matrix Z. With this approach, any
real matrix Z of rank ‘r’ can be “approximated” as a lower ranked matrix (lower than
rank ‘r’). This reduction in the rank of a matrix is the essence of the dimensionality
reduction of matrix Z.

 u1,1   u1,2 
u  u 
 2,1   2,2 
Z   1    v1,1 v2,1  vn,1    2    v1,2 v2,2  vn,2  
   
     
u  u 
 m,1   m,2 
 u1, r 
u 
 2, r 
+...   r     v1, r v2, r  vn, r  (4.10)
 
  
u 
 m, r 
 1   2  ...   r  0

4.4.2 SVD implementation for model

screening

The SVD theorem can be implemented to identify latent dimensions or concepts

in the gas-turbine engine flight-test data. For this, matrix Z is defined with its elements

124
 4.4 | S I N G U L A R V A L U E S A P P R O A C H FOR MODEL SCREENING

to indicate measures of excellence (scores) for each specific multivariable-polynomial

model in predicting performance of each specific engine tested (see Eq.(4.11)).

  1,1  1,2  1,3  1,4  1,5  1,6  1,7 

 
  2,1  2,2  2,3  2,4  2,5  2,6  2,7 
  3,1  3,2  3,3  3,4  3,5  3,6  3,7 
 
. . . . . . . 
Z  
1
  i , j  (4.11)
. . . . . . . ti , j
 
 . . . . . . . 
  511,2  511,3  511,4  511,5  511,6  511,7 
 511,1 
  512,7 
 512,1  512,2  512,3  512,4  512,5  512,6

This matrix Z is of size (512, 7) with its rows representing all 512 candidate
multivariable-polynomial models and its columns representing the various
engines/helicopters tested. For example, engine number 1 is represented by the most
left column and engine number 7 by the most right column of matrix Z.

Next step is to assign scores as elements of matrix Z to quantify level of precision

each model predicts a specific engine. As explained before, these scores are based on
the absolute-values of the relevant test-statistics (Eq.(4.11)). Since prediction goodness
and test-statistics (absolute-value) are inversely proportional to each other, that is the
smaller the test-statistic absolute-value is the better the model represents the
experimental data, the reciprocals of all test-statistics (absolute-value) are used as
elements in matrix Z. The variable ti,j as appears in Eq.(4.11) represents the test-statistic
calculated for model number (i) using the flight-test data of engine number (j). Note
that matrix Z encapsulates the entire flight-test data base.

Equation (4.9) displays the SVD decomposition of matrix Z into its three unique
matrices as defined above. The idea of linearly-independent vectors to span a base in
space can be regarded as an exposure of hidden dimensions in the data. The
conceptual interpretation of the SVD of matrix Z is illustrated in Fig. 4.6 and further
explained hereinafter:

125
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

The rank of matrix Z represents the number of independent hidden Principal

Dimensions (PDs). The diagonal singular-values matrix (Σ) has all PDs represented by
elements along its main diagonal. These elements are arranged in a descending order
and indicate the relative ‘strength’ of appearance of each PD in the flight-test data.

The left singular-vectors (LSV) matrix has seven columns, each with 512 elements.
These seven columns are orthonormal vectors which represent the level of
correspondence between each one of the 512 models and an identified PD in the data.
As illustrated in Fig. 4.6 the first column vector indicates correspondence between
each one of the 512 models to the first (and the most significant) PD identified in the data.
The second column vector specifies level of correspondence between all 512 models
to the second most significant PD, and so on. Figure 4.6 explicitly notates one element
of the left-singular vector matrix (third row and sixth column) as an example to indicate
the level of correspondence between model number 3 and PD number 6.

The right singular vectors (RSV) matrix has seven rows (the rank of matrix Z)
with seven elements each (the seven engines in the flight-test data base). As illustrated
in Fig. 4.6, these rows of VT (or the columns of V) represent the level of
correspondence between each specific engine (denoted by the column number of VT)
and a Principal Dimension (denoted by the row number of VT). The first row vector
indicates relative levels of correspondence between all 7 engines and the first (and the
most significant) PD. The second row specifies the relative strength between all
engines and the second most significant PD, and so on. Figure 4.6 specifies one
element of the right-singular vector matrix (7th row and 2nd column) as an illustration
of the level of correspondence between engine number 2 and PD number 7 (the least
significant PD exposed in the data).

The relative strength of each PD which is indicated by the corresponding

singular-value is then normalized as per Eq.(4.12).

i
ˆ i  r
r rank ( Z ) (4.12)
 k
k 1

126
 4.4 | S I N G U L A R V A L U E S A P P R O A C H FOR MODEL SCREENING

Figure 4.6. The conceptual interpretation of SVD of matrix Z. This matrix decomposition
is used as a tool for screening between 512 distinct empirical models based on their relative
prediction performance using flight-test data from seven distinct gas-turbine engines.

Figure 4.7 presents the normalized seven PDs singular-values. One can observe
that the major PD detected in the data holds a relative strength of 36%, while the
following two PDs (PD2 and PD3) share an almost similar relative strength of 23%
and 22%, respectively. The combination of the first four PD’s encapsulates about 96%
of the PDs representation in the data.

127
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

Figure 4.7. The relative strength of the seven Principle Dimensions (PDs). This figure
presents the normalized strength of each identified PD as demonstrated by the normalized
Singular-Values (SV’s) of matrix Z (Eq.(4.12)).

4.4.2.1 The LSV – Models to PDs correspondences

The absolute-value of each element along a column vector of the LSV indicates the
level of correspondence between a specific model (row number of the vector) and the
relevant PD. Each element along the column vectors is normalized as per Eq.(4.13).

U (i, j )
Uˆ (i, j )  512
, j  1, 2,..., r  r  rank ( Z ) (4.13)
 U (i, j )
i 1

Figure 4.8 presents a collage of seven plots to indicate the normalized elements
along the seven columns of the LSV as level of correspondence between each one of
the 512 candidate models and the seven PDs. The first plot represent correspondences
between each candidate model and the first and most significant PD (PD1). It is
evident from this plot that Model 320 demonstrates the strongest correspondence to
PD1. The other plots on Fig. 4.8 are broadening the spectrum of models to PD’s
correspondence. Model 125 demonstrates the strong correspondence to PD2, model

128
 4.4 | S I N G U L A R V A L U E S A P P R O A C H FOR MODEL SCREENING

367 to PD3, model 4 to PD4, models 49 and 226 to PD5, model 7 to PD6 and model
282 to PD7.

Figure 4.8. Models to PDs correspondences (LSV). This figure presents the normalized
correspondences (Eq.(4.13)) between all 512 candidate models and the seven identified PDs.

4.4.2.2 The RSV – Engines to PDs correspondences

The absolute-value of each element along a row vector indicates the level of
correspondence between a specific engine (column number of the row-vector) and the
relevant PD. Each element along a row is normalized as per Eq.(4.14).

129
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

V (i, j )
Vˆ (i, j )  7
, i  1, 2,..., r  r  rank ( Z ) (4.14)
 V (i, j )
j 1

Figure 4.9 presents a collage of seven plots to indicate the normalized elements
along the seven row-vectors as level of correspondence between engines and PDs. It
follows from the first plot in Fig. 4.9 that PD2 is mainly driven by two engines; engine
number 1 and engine number 4. In a more general context, these two engines share a
substantial similarity with respect to performance models through the second most
significant PD (PD2). This demonstrates the capability of the SVD decomposition to
detect latent dimensions in the data, hence to expose hidden similarities between
different types of engines.

The other six plots in Fig. 4.9 continue to expose the similarity shared between
engines 1 and 4 through PD3. The most significant PD1 is mostly driven by engine
number 7, PD4 by engine number 2, PD5 by engine number 5 and PD7 by engine
number 6.

130
 4.4 | S I N G U L A R V A L U E S A P P R O A C H FOR MODEL SCREENING

Figure 4.9. Engines to PDs correspondences (RSV). This figure presents the normalized
correspondences (Eq.(4.14)) between all seven engines and the principal dimensions (PDs).

4.4.3 Selection of the best multivariable

polynomial model

Once the SVD theorem and its practical interpretation for flight-test data analysis
has been demonstrated, the fundamental question raised in this chapter can be
readdressed, namely, is it practicable to find a general approach to the MPOC method
for best prediction of the gas turbine engine available power? Can the flight-test data
recommend a short list of multivariable polynomial models best describing the

131
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

helicopter gas-turbine engine performance? As concluded in Subsection 4.3.3 above,

the conventional method of hypothesis-testing provided confusing and incoherent
results. For this a new matrix (W) is defined as per Eq.(4.15).

 ˆ 
 1 0 
W   Uˆ  ˆ   ˆ 
 



 (4.15)
0 ˆ r 


Matrix W is the product of the normalized LSV matrix and the normalized
singular-values matrix. This matrix is of the same size of matrix Z, i.e. 512 rows and 7
columns. Each column of W represents the relative correspondence of the 512 models
to the relevant PD (the column number). Adding all column vectors of matrix W to
each other results in a single column vector {S} with 512 elements (Eq.(4.16)).

S   wi  W 

r
w1 w2  . . wr  (4.16)
i 1

Practically, each element of the column-vector {S} holds a normalized value for
the overall/combined performance of each model in predicting the output power of
the “generic” engine, a hypothetical engine that represents all engines tested. The
elements of the column-vector {S} can be regarded as the Combined Normalized
Scores (CNSs) of each one of the 512 models used in predicting the performance of a
gas-turbine engine in general. Figure 4.10 presents the CNS for all 512 candidate
multivariable polynomial models. Based on the highest CNS achieved, the best
empirical model describing the gas-turbine engine performance is model number 320.

This outcome can be expanded to provide a short list of the ten top-performing
multivariable polynomial for the seven engines tested (see Table 4.4). From Table 4.4
one can find the similarities between this list and the one formulated for Engine 1
(Fig. 4.4) and for Engine 2 (Fig. 4.5). Although engines number 1 and 2 ‘sent’ few of
their top 10 performing models as “representatives” to the final ten top-performing
models list, neither one nor the other shared the best final model proposed based on
their level of correspondence to the most significant PD1. The engine that

132
 4.4 | S I N G U L A R V A L U E S A P P R O A C H FOR MODEL SCREENING

demonstrates the maximum correspondence to PD1 was Engine 7 (Fig. 4.9) and its
top-performing model comes leading in the final list. As presented in Table 4.4, model
number 320 involves the basic ten predictors as given by Eq.(4.5), superimposed with
five other predictors: f1, f4, f6, f8 and f9. One should notice that adding more predictors
to the basic model, Model 1 (Eq.(4.5)), does not necessarily correlate with prediction
performance improvement. The final top-ten list actually includes two models which
are using only one extra predictor to the basic Model 1. These are models number 4
and number 3. Model number 4 (Eq.(4.17))) uses f1 as the auxiliary predictor and
Model 3 (Eq.(4.18)) uses f2. When analysis requires simple model to use, either one of
the two is suitable. Another point worth addressing is how well the basic model
performs in the bigger scheme of all seven engines? It appears that model number 1
attains the 173rd place, at the top of the second trimester of the pack of all 512 models.

     
  34 CN g   44  CTGT    54  CTGT  
3 2 3 2
CSHPM ( 4)  14 CN g   24 CN g
(4.17)
 CTGT   CW f  CW f  CW f  CTGT  CW f 
3 2
  64   74  84   94  10
4
  04

     
  33 CN g   43  CTGT    53  CTGT  
3 2 3 2
CSHPM (3)  13 CN g   23 CN g
(4.18)
  63  CTGT    73 CW f        
3 2
 83 CW f   93 CW f  10
3
CN g  CW f   03

Table 4.4. List of 10 top-performing models for helicopter gas turbine engines.

Auxiliary Model Number

Regressors
Involved* 320 367 125 4 3 157 305 237 53 103
f1=(CNg)(CTGT) x x x x
f2=(CNg)(CWf) x x x
f3=(CTGT)(CWf) x x x x x x
f4=(CNg)2(CTGT) x x
f5=(CNg)2(CWf) x x x x x x
f6=(CTGT)2(CWf) x x
f7=(CNg)(CTGT)2 x x
f8=(CNg)(CWf)2 x x x x x
f9=(CTGT)(CWf)2 x x x x x
* Regressors to be superimposed to the basic-model expressed as Model 1 (Eq.(4.5))

133
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

Figure 4.10. The Combined Normalized Scores (CNSs) for all 512 engine models. This
figure presents the CNS values for all 512 multivariable engine models based on data from the
seven engines. Model number 320 outperforms all other empirical models in predicting the
output power of the seven gas-turbine engines tested.

4.5 C OMPARISON TO CONVENT IONAL

METHODS AND APPLICAT IONS

The conventional method for estimating helicopter installed gas turbine engine
output power is based on single-variable analysis method. The innovative MPOC
method, based on multivariable polynomial models, was shown in Chapter 3 to
significantly improve prediction of maximum available power under a wider range of
atmospheric conditions. The main weakness of the MPOC method is that it struggles
with a large number of candidate multivariable polynomial models to choose from.
Table 4.4 addresses this shortcoming by providing a brief list of 10 best-performing
multivariable polynomial models to be used with the MPOC method. Figure 4.11
presents the mean of the prediction error of all these multivariable polynomial models
using the seven engines of Table 4.1.

The mean of the prediction errors presented in Fig 4.11 were calculated as per
Eq.(4.19). In this equation the variable CSHPi represents the measured engine power

134
 4.5 | C O M P A R I S O N TO CONVENTIONAL METHODS AND APPLICATIONS

for data point “i” and CSHPj is the engine output power as estimated by sequential
model number “j”. The parameter “n” represents the number of measured data points.



1 n  100 CSHPi  CSHPj
j   i   (4.19)
n i 1  CSHPi 

 

Figure 4.11. The mean errors of engines output power estimations. This figure presents
the mean of output power prediction errors for all seven engines tested using the top-ten
multivariable models listed in Table-4.4 and the conventional single-variable model.

Figure 4.11 also presents the prediction performance of the seven engines of
Table 4.1 using the conventional method based on the single-variable models (Eq.(4.1)
-(4.3)). One can notice that the multivariable polynomial models performed much
better in predicting all seven engine output-power. The maximum average prediction-
error using a multivariable model was measured to be only 0.2%. This relatively low
prediction error belongs to the two models 3 and 4 whilst predicting the output power

135
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

of Engine 4 (as per Table 4.1). Comparing the prediction performance of the
multivariable polynomial models to those achieved using single-variable models
disclose a clear advantage for the multivariable models. The single-variable models
returned much higher prediction errors for all seven engines tested. One can see in
Fig. 4.11 these prediction errors reached up to 1.15% (for Engine 3).

A different multivariable approach for helicopter engine performance

determination is presented by Gomez et al. [89]. Although prediction accuracy
achieved is not specifically discussed in Gomez [89], the method presented completely
ignores the engine temperature as a predictor for the engine performance model. The
engine temperature is essential for the determination of maximum available power,
since the engine output power is often limited by this variable reaching the maximum
allowed. Discarding the engine temperature from the performance model, makes this
approach useless for the MPOC method. Another fundamental difference between
the MPOC method and the one presented by Gomez [89] is the order of the
polynomials used to describe the engine output power. The MPOC method is based
on third-orders, while Gomez [89] uses second-orders. Limiting the engine
performance model to a second-order only, prevents an inflection point and therefor
fails from modelling an important physical characteristic of the engine that the rate of
power increase with engine temperature increase must reduce, while operating the
engine close to its limitations.

All multivariable models presented in Table 4.4 were found to estimate the output
power of the seven different engines tested with an average accuracy of no more than
0.2% for each model tested. The absolute prediction error for a single measured point
never exceeded 4.1% for all seven different engines tested. Similar analysis, based on
conventional single-variable models, returned best estimation errors of only 8%.
Putting the work presented in this chapter in the larger context of gas turbine engine
performance and comparing the prediction accuracy achieved using the multivariable
polynomial with prediction accuracy of commonly used research simulation tools, such
as Turbomatch [90] reveals similar or better results. Goulos et al. [91] uses Turbomatch
to predict helicopter gas-turbine performance for their work. Chapter 3 of Goulos et

136
 4.6 | C O N C L U S I O N S AND SUMMARY

al. [91] reports that the model has been matched at design point conditions with public
domain data in terms of specific fuel consumption (SFC) with an accuracy of 0.3%.

Heng et al. [92] presents a method of calculating gas-turbine engine output power
based on flow-field simulation and aerodynamics modelling. The engine output-power
estimation is based on the engine outlet temperature. Predictions for engine outlet
temperature were validated against five measured steady-state engine operating data
points (output power between 340 to 1,394 hp) using five similar-type but different
helicopter gas turbine engines tested on a bench. The reported temperature estimation
errors ranged between 2.4% and 4.1% for all 25 data points measured.

Simple and accurate mathematical models that represent the available output
power of the engine, such as presented in Table 4.4, can efficiently be used not only
for the immediate prediction of a specific helicopter performance, but also in relevant
adjacent research which requires a gas-turbine engine power model. Examples of such
are improvement of existing gas-turbine engine technologies, where current
performance is needed for comparison (see Zhang and Gummer [93]). Other examples
can be improvement and validation of helicopter performance where the engine output
power is needed as a module in the big scheme of total helicopter performance as
presented by Savelle and Garrard [94], and Yeo et al. [95].

4.6 C ONCLUSIONS AND S UMMARY

The process of empirical models evaluation is at the core business of experimental

flight-testing data analysis. A commonly used technique in experimental flight testing
to sort between candidate models is the one based on hypothesis-testing and the
associated P-values. The hypothesis-testing method is thoroughly demonstrated in
Subsection 4.3.2 above for the purpose of selecting the best empirical multivariable-
polynomial model to represent the BO-105 engine. After applying minor adjustments,
the hypothesis-testing method was implemented successfully in ranking all 512
candidate multivariable-polynomial models based on their relative performance.

137
4 | A SINGULAR VALUE APPROACH IN HELICOPTER FLIGHT TESTING ANALYSIS

However, the hypothesis-testing approach became completely ineffective once the

experimental data-base was expanded to include six more engines. The method failed
providing a clear answer to the question of which is the best-performing model when
the entire experimental data from all seven engines is analysed as a whole.

The Singular-Value-Decomposition (SVD) approach was used successfully where

the hypothesis-testing method failed and produced a list of top-performing
multivariable polynomial models to describe gas turbine engine performance in
general. The SVD approach was also used for exposing latent similarities between
different engines with respect to their performance models.

Analysis showed no correlation between the number of auxiliary predictors used

in the multivariable-polynomial model and the power prediction accuracy. The fourth
and fifth best-performing models out of the 512 evaluated incorporated only 11
predictors, making either one of them a great choice, if analysis simplicity is
paramount.

Although the SVD approach is demonstrated in this chapter using engine data, it
is not bounded only to gas-turbine engine testing and available power flight-testing. In
the following chapters of this dissertation, the SVD approach is implemented for other
disciplines of helicopter performance flight-testing, for which empirical models are
being evaluated.

138
 It does not matter how beautiful your theory is. If it does not
agree with experiments, it’s wrong.
Richard P. Feynman

5 H OVER P ERFORMANCE T ESTING

BASED ON D IMENSIONALITY
R EDUCTION

5.1 C HAPTER O VERVIEW

T he power required to hover a helicopter is fundamental to any new or modified

helicopter performance flight-testing effort. The conventional flight-test method
that was previously discussed in Chapters 1 and 2 (Subsections 1.3.2 and 2.3.2) is based
on relating two non-dimensional variables (coefficient of power and coefficient of
weight). This single-variable method is overly simplified and neglects compressibility
effects in the power required to hover under a wide range of gross weight and
atmospheric conditions. This chapter presents an alternative flight test method for
hover performance that addresses the deficiencies of the conventional method, as
stated in Subsection 1.4.2 (PS2). This novel hover performance testing method is
referred-to as the ‘Corrected-Variables Screening using Dimensionality Reduction’
(CVSDR). The method uses an original list of 15 corrected-variables derived from
fundamental dimensional analysis.

This Chapter 5 was published as the following journal paper: Arush I., Pavel M.D., and Mulder
M., “A Dimensionality Reduction Approach in Helicopter Hover Performance Flight
Testing”, Journal of the American Helicopter Society 67, No. 3 (2022): 129–41.
https://doi.org/10.4050/JAHS.67.032010.

139
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

This list is further reduced by means of dimensionality reduction to include only the
essential and effective hover performance predictors. The CVSDR method is
demonstrated and tested in this chapter using flight-test data of a Bell Jet-Ranger and
shows that at the 95% confidence level; the averaged prediction error is only 0.9hp
(0.3% of the helicopter maximum continuous power). Using the same set of flight-test
data, the conventional method yields a much larger average prediction error of 1.7 hp.
Although demonstrated in this chapter with a specific type of helicopter, the CVSDR
method is applicable for hover performance flight-testing of any type of a conventional
helicopter configured with a main rotor and a tail rotor.

5.2 I NTRODUCTION

The most distinguishing characteristic of a helicopter is its ability to steadily hover

at any phase of its mission given it has a sufficient power margin [22,23]. Knowing the
power required to hover is fundamental to any new or modified helicopter flight-test
effort. As already discussed in Chapter 2, the conventional flight-test method for hover
performance is based on the combined blade-element momentum (BEM) theory and
is overly simplified. This simplification often yields empirical models that fail to
accurately and consistently predict the total power required to hover under a wide
range of helicopter gross weight and atmospheric conditions. Bousman [96]
demonstrates this drawback by using Out-of-Ground Effect (OGE) hover
performance testing of five different flight test programs and reporting inconsistency
in OGE hover performance of up to 5% of which the source of the error could not
be explained.

As already mentioned in Chapter 1 (Subsection 1.4.2), a major disadvantage of the

conventional OGE hover flight-test method is that it does not address main rotor
blade compressibility effects as those are often assumed to have negligible effect on
the hover performance. An example of this frequently taken assumption can be found
in the study on uncertainty quantification in helicopter performance by Siva et al. [97].
The ability to account for compressibility effects, mostly related to blade tip Mach

140
 5.2 | I N T R O D U C T I O N

number and shape, is essential for accurate hover performance predictions. This
relation is well illustrated by computational fluid dynamic (CFD) simulations used to
predict hover performance of rotor systems. Jacobson and Smith [98] presents hover
performance comparison between predictions from a hybrid CFD methodology and
measured hover performance of a rotor with three different blade tip configurations
at three different tip Mach numbers (0.55, 0.6 and 0.65). They state that future work is
needed to understand why CFD models do not predict the same impact of the tip
shape as measured in the experiment. Moreover, one of Jacobson and Smith [98]
conclusions states that hover performance predictions from the hybrid methodology
CFD improve as tip Mach numbers reduce. This conclusion solidifies the significance
compressibility effects have on hover performance. Garcia and Barakos [99] provide
another example to show compressibility effects should not be neglected from hover
performance predictions. Their work, which focuses on accurate rotors hover
performance predictions using modern CFD methods with modest computer
resources, shows the significance the tip shape and Mach number have on hover
performance of a rotor system.

Measuring compressibility effects in flight-testing of a full-scale helicopter and not

just a rotor system requires the hover trials to be performed in high altitude and low
air temperatures. Whereas in the past, these kind of high altitude hover trials were
challenging since they required high-ground reference points, recent technological
developments show potential to make these trials more practical in the future.
Matayoshi et al. [100] present results from a flight-test evaluation of a helicopter
airborne LiDAR (Light Detection and Ranging) system. This system can measure
accurately three-axis true airspeed which is crucial for high altitude hover performance
trials. Boirun [101] attempts to rectify the disadvantage of the conventional hover
flight-test method by including compressibility effects into the empirical performance
model of the helicopter. However, his approach does not determine a definitive single
empirical model to include compressibility effects. Instead, various curves for different
values of main-rotor tip speeds are offered.

141
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

Obtaining an accurate single empirical model to predict the hover performance is

highly beneficial since it can also be used for real-time applications. A single empirical
hovering model can be used in conjunction with existing algorithms that predict the
gross weight of the helicopter for real time hovering performance. Abraham and
Costello [102] present such a practical algorithm to estimate the gross weight and
center of mass of a helicopter in flight and report the algorithm works well in hover.

Chapters 3 and 4 present an alternative and more accurate approach to helicopter

performance flight-testing, using multivariable polynomials as empirical models. This
approach was proven in Chapter 3 more accurate in the prediction of the available
power of a helicopter under a wide range of atmospheric conditions as compared to
the conventional single-variable flight-test method. Chapter 4 provides a systematic
method for screening between candidate multivariable predictors. This multivariable
polynomial approach is next applied in this chapter (Chapter 5) to the power required
to sustain a helicopter in an OGE hover, without taking any lenient assumptions such
as negligible compressibility effects.

This Chapter 5 is structured as follows: after a short introduction, the

conventional method for hover performance testing is briefly reminded and
demonstrated using flight-test data from a Bell Jet-Ranger helicopter. Flight-test data
from three distinct sorties, totalling 56 data points are used to obtain an empirical
single-variable model for the power required to OGE hover. The expected level of
accuracy is then evaluated while using this empirical model to predict the power
required to hover, under condition of the 20 hover points of Sortie 4. Unsatisfied with
the level of prediction accuracy, an alternative method referred to as the “Corrected-
Variables Screening using Dimensionality Reduction” (CVSDR), is presented and
discussed in Section 5.4. The CVSDR method is applied to the same Jet-Ranger flight-
test data and yields an alternative empirical multivariable model for power required to
OGE hover. The level of prediction accuracy expected from the CVSDR driven model
is discussed in Sections 5.5 and compared with the conventional method in Section
5.6. The later Section 5.6 also provides possible reasoning to explain the different

142
 5.3 | C O N V E N T I O N A L M E T H O D FOR HOVER PE RFORMANCE T ESTING

prediction accuracy between the two methods. Section 5.7 presents the conclusions
and summary and concludes this chapter.

5.3 C ONVENTIONAL M ETHOD FOR H OVER

P ERFORMANCE T ESTING

The conventional single-variable flight-test method for determining the power

a helicopter required for a hovering flight is thoroughly discussed in Chapter 2
(Subsection 2.3.2), and demonstrated in many helicopter hover performance papers
[77,103-105]. In a nutshell, this method seeks to uncover the linear relation between
the coefficient of power (𝐶𝑝 ) and the coefficient of weight raised to the 1.5 power
(𝐶𝑤 )1.5, i.e., to realize the two coefficients 𝛼1 and 𝛼2 of Eq.(5.1) for a particular type of
helicopter.

CP  1  CW 
1.5
 2

 P W 
 (5.1)
 P
C  , C  2
 a Ad  R   a Ad  R  
3 W



The flight-test team is required to plan and execute numerous hover test points
in order to cover the entire flight envelope of the helicopter under test. This includes
all certified gross-weights (W), the entire atmosphere the helicopter is expected to fly
at (which defines the ambient air density, 𝜌𝑎 ), and throughout the governed range of

the main-rotor angular speed (ω,Ω). As already discussed in Chapter 2, there are two
fundamental techniques to execute the precise hover sorties for data gathering. The
one is the free-flight hover and the other, which requires more preparation efforts and
coordination, is the tethered hover. The pros and cons of each technique are discussed
in Chapter 2 (Subsection 2.3.2.1).

Next, the conventional method is demonstrated using free-flight OGE

hovering flight-test data obtained during from four distinct sorties of a Bell Jet-Ranger

143
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

helicopter. The four sorties were conducted under different atmospheric and gross-
weight conditions, as summarized in Table 5.1.

Table 5.1. Summary of OGE hover conditions. This table presents the ambient conditions
and range of Bell Jet-Ranger parameters during the free-flight OGE hovers.

Sortie W [lbs.] CW [x10-3] Pressure Altitude Ta [°C] Mtip

[ft.]
1 2900 - 3000 3.298 – 4.032 2200 - 6600 11 to 18 0.59 – 0.62
2 2850 - 2960 2.986 – 4.046 3100 - 6100 10 to 15 0.59 – 0.61
3 2850 - 2980 3.161 – 3.739 700 - 6350 -2 to 3 0.61 – 0.64
4 2700 - 3060 3.043 – 4.062 425 - 6800 20 to 26 0.59 – 0.61

Figure 5.1 presents the total of 76 matching pairs of coefficient of power (𝐶𝑝 )
and coefficient of weight raised to the 1.5 power ((𝐶𝑤 )1.5) measured in all four sorties.
All 76 OGE hover points were obtained using the free-flight (un-tethered) flight-test
technique. Specialty Flight Test Instrumentation (FTI), which was calibrated for the
test, sampled relevant parameters at a rate of 10 cycles per seconds. The helicopter was
stabilized at each hover point for a duration of at least 20 sec., and sampled data was
averaged over this period of time post flight. The power required to hover was reduced
from the engine output torque and the free-turbine speed which were both sampled
by the FTI. The gross weight (W) of the helicopter was calculated by subtracting the
fuel used from the take-off all up weight. All hover points were conducted under the
restrict limitation for the relative wind to be less than 3 kts. For ground referenced
hover points the relative was measured using a ground based anemometer and for high
altitude hover points, an independent helicopter with an independent Low Airspeed
Indicator (LAI) was used as a hover reference for the tested Jet-Ranger.

144
 5.3 | C O N V E N T I O N A L M E T H O D FOR HOVER PE RFORMANCE T ESTING

Figure 5.1. Non-dimensional OGE hover performance. This figure shows the non-
dimensional OGE hover performance of a Bell Jet-Ranger. The figure is split into four
distinct sorties (Sortie 1-4), as per the conditions specified in Table 5.1.

The level of accuracy achieved using the conventional method was assessed by
taking the following approach: flight-test data from the first three sorties was used for
the derivation of an empirical OGE hover model, obtained from a linear regression.
Then, the accuracy and effectiveness of this empirical model was evaluated by
comparing its predictions with the actual flight-test data gathered in Sortie 4. The
reason for this specific partition of predicting the performance of Sortie 4 by using
data obtained from the first three sorties was to challenge the method to the fullest
extent possible. It is evident from Table 5.1 that Sortie 4 was executed under a wider
range of gross weights and pressure altitudes, not covered by the first three sorties. By
applying this specific partition, the empirical hovering model is challenged with an
extrapolation task.

Linear regression was executed in order to describe the relationship between

the coefficient of power (𝐶𝑝 ) and the coefficient of weight raised to the 1.5 power
((𝐶𝑤 )1.5). The 56 flight-test hover points of Sorties 1-3 were substituted in Eq.(5.1),
yielding a linear system of 56 equations with only two unknowns (𝛼1 and 𝛼2 ). This set
of equations is compactly represented as Eq.(5.2).

145
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

 A   b (5.2)

The matrix A is of size (56,2) and contains the numeral values of the coefficient
of weight (𝐶𝑤 ) raised to the 1.5 power as the first column, and a unity vector as the
second column. The column vector α is of a size (2, 1) and contains the coefficients
(α1 and α2). The column vector 𝑏⃗ is of size (56,1) and contains the numerical values of
the measured coefficient of power (𝐶𝑝 ) for all hover points. The explicit representation
of Eq.(5.2) is presented as Eq.(5.3).

 C 1.5
w 1   C P1 
 1.51   
 C w2 1   C P2 
   
     1    
   (5.3)
     2    
   
      
 1.5 1   
C w56  C P56 

The system of equations represented by Eq.(5.3) is over-determined and does

not have an exact solution. However, one can look for the ‘closest’ solution of this
system, i.e., the “best-fit” solution denoted as 𝛼̂ (see Strang [83]). The matrix
constructed from [ATA]-1AT is defined as the projection-matrix, and when multiplied
by the vector 𝑏⃗ provides the best-fit solution or the “closest” solution one can look
for (Eq.(5.4)). Although this specific example solves for only two coefficients (𝛼1 , 𝛼2 ),
this method is applicable for an over-determined system with any arbitrary number of
coefficients.

1
ˆ    AT A AT b (5.4)

For the exemplary Bell Jet-Ranger considered in this chapter, the regressed
empirical OGE hover model is presented as Eq.(5.5).

 1.175  CW   4.118 105  Base Model 

1.5
CP ( S13 )
(5.5)

146
 5.3 | C O N V E N T I O N A L M E T H O D FOR HOVER PE RFORMANCE T ESTING

Figure 5.2 presents all 56 data points from the first three sorties and the “best-
fit” solution (Eq.(5.5)).

Figure 5.2. Non-dimensional OGE hover performance (Sorties 1-3).

The errors between the measured and the predicted OGE hovering power for
Sortie 4 were calculated in accordance with Eq.(5.6) and are presented in Fig. 5.3.


 
  C pi  1.175 CWi

 4.118 105   ai Ad i R  , i  1, 2,..., 20
1.5 3
Er (5.6)
base   

The prediction errors ranged up to an absolute maximum value of 11.7 hp, a

mean of -3.7 hp and a variance of 18.1 hp2. For the type of helicopter tested, a power
deviation of more than 1.6 hp (absolute value) is already noticeable to the aircrew. The
averaged prediction error of -3.7 hp (over-estimate) with a variance of 18.1 hp2 is
therefore considered substantial. The conventional approach in flight-testing for
assessing how accurate a model predicts the actual performance is based on
hypothesis-testing. This approach which follows from the central-limit theorem is
thoroughly discussed in the literature [84,87]. In a nutshell, a hypothesis is set (the
‘null-hypothesis’) and by using the test-statistic (Eq.(5.7))) the validity of the null-
hypothesis is assessed against the alternative hypothesis.

147
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

E r base  0
tbase  , 0  1.6hp : n  20 (5.7)
Sbase n

For the specific case presented, the null-hypothesis assigned is that on-
average the power-to-hover predicted by the empirical model obtained (Eq.(5.6)) does
not differ from the true measured power by more than ±1.6 hp (deviation mismatch
noticeable to the Jet-Ranger aircrew). This null hypothesis is tested against the
alternative that on-average the power to hover from Eq. (5.6) shows an absolute
prediction error of more than 1.6 hp.

Figure 5.3. Power prediction errors for Sortie 4 (base model). This figure shows the power
to OGE hover prediction errors generated by the base model (Eq.(5.6)) for the conditions of
Sortie 4.

The relevant test-statistic for this hypothesis testing is calculated per Eq.(5.7).
The symbol ‘n’ represents the number of measured test points of Sortie 4 (n=20) and
‘S’ stands for the standard deviation of the prediction errors of the empirical hover
model (Eq.(5.6)) which are presented in Fig. 5.3. The calculated value for the test-
statistic (Eq.(5.7)) was found to be 2.18. Inferential statistical analysis based on the
sampled data from Sortie 4 shows the probability for making a type-I error by rejecting
the null-hypothesis to be only 4.2%. This low probability for a type-I error is below
the 5% significance level accustomed in helicopter performance flight-testing. The

148
5.4 | C O R R E C T E D -V A R I A B L E S C R E E N I N G USING DI MENSIONA LITY RE DUCTION

practical meaning of this test is that there is significant statistical evidence at the 95%
confidence level to reject the null hypothesis and adopt the alternative hypothesis
instead. It can be concluded that on-average and at the 95% confidence level, the
power required to hover predictions (Eq.(5.6)) deviates by more than 1.6 hp from the
actual measured power. Complementary statistical analysis shows that on-average and
at the 95% confidence level, the hover-power predictions based on Eq.(5.6) deviate by
up to 1.7 hp from the actual measured power. This noticeable prediction error of the
conventional hovering model is to be expected. One should doubt the linear relation
between the coefficient of power and the coefficient of weight raised to the 1.5 power
(Cp, (Cw)1.5). Merely by looking at Fig. 5.2 one should doubt if the relation is actually
linear and whether there are some other latent factors that affect the relation between
the data points.

Concluding this subsection, the conventional flight-test method for assessing

the OGE power required to hover can result in substantial estimation errors as
demonstrated for the prediction of Sortie 4. Statistical analysis at the 95% confidence
level shows that on-average the hover-power predictions based on Eq.(5.6) deviate by
up to 1.7 hp from the actual measured power. Next, in Sections 5.4 an alternative
analysis method with an improved prediction accuracy is proposed.

5.4 C ORRECTED -V ARIABLE S CREENING

USING D IMENSIONALITY R EDUCTION

An alternative analysis method for the power required to hover is proposed,

referred-to as the ‘Corrected-Variables Screening using Dimensionality Reduction’
(CVSDR). This method requires no variation to the way flight-test sorties are carried-
out, only the analysis method is modified. The method involves three phases. In phase
one, an original list of corrected-variables is generated for a multivariable analysis
approach. In phase-two, this list of corrected-variables is refined based on concepts of
dimensionality reduction. Phase three starts once the bare-essential list of corrected

149
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

variables is defined, and an empirical multivariable model is fitted to the flight-test

data. The entire derivation process is demonstrated hereafter using the same flight test
data from a Bell Jet-Ranger helicopter that was used in Section 5.3.

5.4.1 Phase One - Original list of corrected-

variables for hover performance

Phase one of this method starts by proposing the dimensional variables that affect
the physical problem of the amount of power needed for a helicopter in a hover. These
are the ambient static pressure, Pa, the ambient static temperature, Ta, the helicopter
gross-weight, W, the main-rotor disk area, Ad, the main rotor angular speed, ω, and
the main-rotor height above the ground, h. The power required to hover, P, can be
represented mathematically as Eq.(5.8) and Eq.(5.9) in an implicit form. The
dimensions involved are presented in Table 5.2, where ‘M’ represents mass, ‘L’
represents length and ‘T’ represents time.

P  f ( Pa , Ta ,W , Ad ,  , h) (5.8)

fˆ ( P, Pa , Ta ,W , Ad ,  , h)  0 (5.9)

Table 5.2. Variables and dimensions involved in hover performance. This table
presents all major variables affecting the OGE hover performance problem and
associated dimensions.

# Physical Variable Notation Dimension

1 Power Required to Hover P [M][L]2[T]-3
2 Ambient static Pressure Pa [M][L]-1[T]-2
3 Ambient static Temperature Ta [L]2[T]-2
4 Helicopter Gross-Weight W [M][L][T]-2
5 Main-Rotor Disk Area Ad [L]2
6 Main-Rotor Angular speed ω [T]-1
7 Main-Rotor Height Above Ground h [L]

150
5.4 | C O R R E C T E D -V A R I A B L E S C R E E N I N G USING DI MENSIONA LITY RE DUCTION

The physical problem of OGE hover performance has seven variables involved
with three dimensions (L,M,T). According to the Buckingham Pi-theorem [66] the
complexity of the problem can be reduced from the seven dimensional-variables
dependent to only four Non-Dimensional (ND) variables. These four ND variables
are next defined as products of the dimensional variables. The four ND variables are
denoted by 𝜋𝑖 . Since there are seven dimensional variables to construct four ND
variables, three dimensional variables are used as repeating variables in the ND
products (𝜋𝑖 ). There are 35 different options to choose three variables out of seven
for the case where the order does not matter (combinations). This sets a fairly tedious
task of screening between 35 different options, defining the best appropriate manner
to describe the ND helicopter hover performance. The derivation is demonstrated for
only one of the 35 options available. The following example involves setting the main-
rotor disk area, the ambient static pressure and the ambient static temperature as
repeating variables. The four ND products are defined in Eq.(5.10).

   A a  P b T c  P  
 1 d a a

   A   P  T  f W  
d e
 2 d a a 
  (5.10)
 3   Ad   Pa  Ta    
g h i

 

 4   Ad  j
 Pa  k
 Ta  m
 h  


Next, the dimensional analysis procedure requires to replace each of the

dimensional variables with its dimensions and to enforce each one of the four 𝜋𝑖
parameters to be non-dimensional. This process is demonstrated as per Eq.(5.11).

151
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

 b c
 2  2 
1    L2   M   L   ML   M b 1 L2 a b  2c  2T 2b  2c 3  M 0 L0T 0 
a

    LT 2   T 2   T 3  
   
 
 f
d  M   L2   ML 
e 
 2    L2      M e 1 2 d  e  2 f 1 2 e  2 f  2
L T  M 0 0 0
L T
    LT 2   T 2   T 2  
 
 i
 (5.11)
 g  M   L2   1 
h

 3    L   2   2     M L
    
2 h 2 g h 2 i
T 2 h 2 i 1
M LT
0 0 0

  LT   T   T  
 m 
 j  M   L2  
k

 4    L   2   2   L   M L
    
2 k 2 j k 2 m 1
T 2 k 2 m
M LT
0 0 0

  LT   T  

Each one of the 𝜋 products yields 3 equations with 3 unknowns, which are the
exponents. Solving for the exponents of 𝜋1 is demonstrated in Eq.(5.12). The same
process is repeated for each one of the other ND variables, 𝜋2 , 𝜋3 and 𝜋4 .

 M  : b  1  0 
 
  0 1 0  a   1  a   1 
           
 L  : 2a  b  2c  2  0    2 1 2  b    2  b    1  (5.12)
   0 2 2   c   3   c   1 / 2 
          
T  : 2b  2c  3  0 
 

Based on Eq.(5.12) the first ND variable, 𝜋1 , can be written as Eq.(5.13).

P
1  (5.13)
Ad  Pa Ta

The interest is in developing a method to gather hover performance for a specific

helicopter and not in drawing a comparison between different types of helicopters.
Therefore, the ND variable (Eq.(5.13)) can be further simplified. The main-rotor disk
area (Ad) is constant, and the static pressure (Pa) and temperature (Ta) of the ambient
air can be expressed as per their ratio to the standard sea level values (Eq.(5.14)).

Pa  P0   , Ta  T0   (5.14)

152
5.4 | C O R R E C T E D -V A R I A B L E S C R E E N I N G USING DI MENSIONA LITY RE DUCTION

This gives a simplified expression for 𝜋1 defined as 𝜋1∗ in Eq.(5.15).

P P P 1 P P
1      Const  
Ad Pa Ta Ad P0 T0 A P d 0 
T0   Ad P0 T0    
(5.15)
P
1* 
 

Since this term has dimensions and is not a pure ND, it is better defined as a
“corrected” variable (CV) to describe the hover performance of a specific helicopter.
It can be used to facilitate the forthcoming analysis.

Similar analysis performed for 𝜋2 , 𝜋3 and 𝜋4 yielded the corresponding three

corrected-variables (𝜋2∗ , 𝜋3∗ and 𝜋4∗ ). The hover performance of a specific helicopter
can now be simplified as presented as Eq.(5.16).

W  h
 2*  ,  3*  ,  4*  (5.16)
  Ad

One should be noted that 𝜋4∗ is a true ND variable which represents the ND
height of the main-rotor above the ground. This ND variable is beneficial only if the
hover performance deals with in-ground-effect (IGE). This thesis is limited to the out
of ground effect (OGE) only and does not address the ground effect on hover
performance.

Identical dimensional analysis was repeated to evaluate all other 34 possibilities of

choosing three dimensional variables out of the seven. Ten options were found to not
having a unique solution, and few other options returned repeated ND variables.
Overall, the analysis yielded 15 different corrected-variables which can be used for the
specific hover performance analysis. Table 5.3 summarizes all 15 corrected-variables
in an array form that indicates which of the three dimensional-variables (power, weight
and main-rotor angular speed) is used in the specific corrected-variable. Three of the
∗ ∗ ∗
corrected-variables (𝜋13 , 𝜋14 and 𝜋15 ) are based on all three dimensional-variables.

153
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

Table 5.3. Corrected Variables (CV) to represent the OGE hover performance.

Power Weight Main-Rotor & Power, Weight and

Based Based angular-speed Main-Rotor angular
based speed based
Power based P P P P
 1*   6*   4*  13
*

  W   W
P P 2  3
 5*  P4 10  P 
 8*  14
*
 2
*

W 5 
 W3
P  P 2
 9*  12
*
 P 
W 2  3 15
*

W
Weight based W W2
 2*  11
*

  
Main-Rotor 
 3* 
Angular speed 
based 
7 
*


5.4.2 Phase Two - Screening for essential

CVs using dimensionality reduction

Phase Two of the proposed CVSDR method is to refine the list of 15

corrected-variables (Table 5.3) generated from fundamental dimensional analysis and
to select only the essentials for the task of acquiring an empirical model to represent
the OGE hover performance of a helicopter. A power based corrected-variable needs
to be expressed as a function of other corrected-variables. It is immediately evident
∗ ∗ ∗
that the three corrected-variables (𝜋13 , 𝜋14 , 𝜋15 ) cannot serve as effective predictors
since each one of them simultaneously involves all three major variables of power,
weight and angular speed of the main rotor. Even prior to implementing
dimensionality reduction tools, the list of candidate corrected variables is reduced to
12 candidate predictors.

154
5.4 | C O R R E C T E D -V A R I A B L E S C R E E N I N G USING DI MENSIONA LITY RE DUCTION

The proposed dimensionality reduction approach for hover CVs screening is

based on the Singular Value Decomposition (SVD) theorem which is thoroughly
presented and discussed in Chapter 4 (Subsection 4.4.1). As a reminder, the SVD
theorem states that any generic real matrix can be uniquely decomposed into a set of
three matrices as given in Eq.(5.17). Consider a real matrix Z to be of size (m,n) and
rank ‘r’. This matrix Z can be expressed as a product of the following three unique
matrices: matrix U, an orthonormal matrix of size (m,r) called the “Left-Singular-
Vectors” (LSV); matrix Σ, a diagonal matrix which holds along its diagonal the
singular-values of Z; and matrix V, an orthonormal matrix of size (n,r) called the
“Right-Singular-Vectors” (RSV).

 u1,1 u1,2  u1, n 

u    v1,1  vn,1 
u2, n   1
v2,1
  v vn,2 
 2,1 u2,2
2 v2,2 
Z  U V   
T
      1,2
          (5.17)
      


 r   v1, r v2, r vn, r 
u
 m,1 um,2  um, n  
1   2  ...   r  0

The SVD theorem can be implemented for refining the corrected-variable

(CV) list and to identify those which stand-out from the group of 12 as the most
effective predictors for the OGE hover empirical model. A comparable approach is
performed in the process of gas-turbine empirical models screening presented in
Chapter 4. For this task of screening the most effective CVs for hover performance,
matrix Z is filled with numeral entries of the 12 corrected-variables as evaluated for
the first three flight-test sorties of the Bell Jet-Ranger helicopter. Matrix Z becomes of
size (56,12); 56 rows that each represents a distinct single hover point and 12 columns
that represent the 12 CVs (𝜋1∗ to 𝜋12
∗
). Next is to normalize the columns of Z to have
a mean of zero and a variance of 1. For this, each entry along the columns of Z is
normalized as per Eq.(5.18).

 i*   *
 i'  i
, i  1, 2,...,11,12 (5.18)
S *
i

155
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

The normalized Matrix Z (defined as Z’) is then decomposed into its unique
three matrices as per Eq.(5.17). As expected, the rank of Z’ is 12 representing the
dimensionality of the flight-test data. The OGE hover performance problem as
appears in matrix Z’ can be represented by using all 12 CVs (𝜋1∗ to 𝜋12
∗
). However, not
all corrected-variables have the same level of significance in representing the variance
in the flight-test data held by matrix Z’. The singular-values (𝜎𝑖 ) which are arranged
in a descending order along the main diagonal of matrix Σ are key to understanding
the level of importance each corrected-variable (‘i’) holds. The conceptual
interpretation of the SVD of Z’ for the specific problem of OGE hover performance
is illustrated in Fig. 5.4 and is further explained herein.

Figure 5.4. The conceptual interpretation of SVD of Z’ in OGE hover performance. This
figure presents how the abstract SVD of matrix Z’ (normalized predictors) should be interpreted
for the task of screening out the most effective predictors for OGE hover performance
representation.

156
5.4 | C O R R E C T E D -V A R I A B L E S C R E E N I N G USING DI MENSIONA LITY RE DUCTION

The 12 singular-values of the diagonal matrix Σ are normalized as per Eq.(5.19)

i
ˆ i  , i  1, 2,3,...,11,12 (5.19)

k 1 k
12

The normalized singular values are presented in Fig. 5.5 along with a
cumulative-sum plot of all normalized singular values. The main conclusion one can
draw from Fig. 5.5 is that the dimensionality of the general OGE hover problem can
be practically reduced from 12 (the general case) to only five for the specific OGE
hover analysed. The empirical model representing the general OGE hover
performance can be substantially simplified for the specific case analysed, to include
only five CVs, instead of the original 12. The cumulative sum plot presented in
Fig. 5.5 indicates that 98% of the variance in the flight-test data stored in matrix Z (or
Z’) can be captured by using only the first five most significant CVs. Also from
Fig. 5.5, it can be noticed that the most significant dimension of the problem is
responsible to 52% of variance in the flight-test data, the second dimension explains
19% of variance in the data, and the third, fourth and fifth can explain 13%, 8% and
6%, respectively.

Figure 5.5. The Singular Values (SVs) of Matrix Z’. This figure shows the relative
significance of the 12 SVs (or dimensions) involved in the specific OGE hover performance.

157
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

The following necessary question one might have is “which are the most
significant corrected-variables?” This question is answered by evaluating the absolute
values of the entries of the RSV matrix. As illustrated in Fig. 5.4, each row of the RSV
indicates the level of correspondence to a specific singular-value (or a dimension) of
the problem. For example, the first row of the RSV specifies the level of
correspondence each one of the 12 corrected-variables has to the first (and most
significant) singular-value. The second row of the RSV indicates the correspondence
between all 12 corrected-variables to the second most significant dimension of the
problem and so on. Since the dimensionality of the problem was reduced from 12 to
5, it is required to evaluate the first five rows of the RSV matrix in order to expose the
most significant CVs of the OGE hover problem. Figure 5.6 presents the significance
of each CV to each one of the five substantial dimensions of the OGE hover
performance problem by indicating the normalized values (as per Eq.(5.20)) of the
entries along the first five rows of the RSV matrix.

V (i, j )
Vˆ (i, j )  , i  1, 2,3, 4,5 (5.20)

12
j 1 V (i, j)

Figure 5.6. Dimensions to CVs correspondence. This figure shows the correspondence
between each one the 12 CVs to the detected dimensions of the OGE hover problem.

158
5.4 | C O R R E C T E D -V A R I A B L E S C R E E N I N G USING DI MENSIONA LITY RE DUCTION

The left singular vectors (LSV) matrix has no significant role in the type of
analysis addressed in this chapter since it only indicates level of correspondence
between each one of the 56 OGE hover points and the singular-values of Z. This type
of correspondence between particular hover test-points and the various dimensions of
the OGE hover performance is deemed irrelevant to the subject of this analysis.

The following five conclusions are drawn from Fig. 5.5 and Fig. 5.6:

(1) The first and most significant dimension of the OGE hover problem holds
∗
for 52% of variance in the data, and is best represented by 𝜋12 .

(2) The second most significant dimension of the OGE hover problem holds
∗
for 19% of variance in the data, and is best represented by 𝜋11 .

(3) The third dimension of the OGE hover problem holds for 13% of variance
in the data, and is best described by 𝜋2∗ .

(4) The fourth dimension of the problem holds for 8% of variance in the data,
∗
and is best represented by 𝜋12 .

(5) The least significant dimension in the truncated list of 5 dimensions holds
for only 6% of variance in the data and is best represented by 𝜋9∗ , followed by 𝜋7∗ .

Since only one power-based predictor is required for the empirical model in
∗
quest and the previous conclusions suggest two (𝜋12 and 𝜋9∗ ), it was decided to use the
∗
one that shows the highest correspondence with the first dimension which is 𝜋12 .
Furthermore, 𝜋9∗ is replaced with 𝜋7∗ as the predictor which best represents the fifth
dimension of the OGE hover problem.

Finally for Phase Two, a conceptual empirical model to represent the OGE
hover performance of the example helicopter can be stated as Eq.(5.21). This
conceptual relationship is next pursued with a first-order linear model as described in
Eq.(5.22).

159
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

P 2  W2 W  
12
*
 f1 (11
*
,  2* ,  7* )   f1  , ,
     (5.21)
 3   

P 2  W2  W    
 1     2    3    4 (5.22)

 3         

The numerous steps executed for dimensionality reduction in Phase Two are
summarized as a flowchart presented in Fig. 5.7.

Figure 5.7. Steps required for dimensionality reduction. This figure presents the seven steps
required for screening between the various CVs and choosing the most significant ones.

160
 5.5 | T H E C V S DR M O D E L P R E D I C T I O N A C C U R A C Y ( O G E H O V E R )

5.4.3 Phase Three - Deriving a practical

empirical model

The proposed model (Eq.(5.22)) is fitted with the 56 flight-test OGE hover
points from the first three sorties. This regression process is based on the ‘least-
squares’ method as previously explained in Section 5.3. The refined OGE hover
model, based on the CVSDR method and the flight-test data from the first three
sorties, is presented as Eq.(5.23). This empirical model is addressed hereinafter as
Model number 1 and denoted as M1.

 1   0.134 
 W      7.99 
P 2 2
W      2  
 1     ,  (Model 1) (5.23)
     2    3    4  3   926.5 
  3
       
  4  2  105 
 

5.5 T HE CVSDR M ODEL P REDICTION

A CCURACY (OGE H OVER )

The OGE hover model generated by the CVSDR method (Eq.(5.23)) is next
evaluated for its expected level of accuracy. For this, Model 1 (Eq.(5.23)) is used to
predict the power required to OGE hover under the conditions of Sortie 4. The errors
between the predicted power and the actual measured power were calculated in
accordance with Eq.(5.24), and are presented in Fig. 5.8.

  W2  W       i i3
ErM 1  Pi   1     2    3     4  , i  1, 2,..., 20 (5.24)
       i   i  i
2
 i

Prediction errors ranged up to a maximum absolute deviation of 8.5 hp. The

mean of the prediction errors for the 20 hover points of Sortie 4 was calculated to be
-2.3 hp with a variance of 9.7 hp2.

161
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

Figure 5.8. Power prediction errors for Sortie 4 (CVSDR model). This figure shows the
power to OGE hover prediction errors generated by the CVSDR model (Eq.(5.23)) for the
conditions of Sortie 4.

Parallel statistical analysis as discussed in Section 5.3 for the base-model was
performed in order to evaluate the level of accuracy to be expected from the CVSDR-
based OGE hover model (Model 1, Eq.(5.23)). The applicable test-statistic for the
relevant hypothesis testing was calculated per Eq.(5.25).

E rM 1  0
tM 1  , 0  1.6 hp , n  20 (5.25)
SM 1 n

The symbol ‘n’ represents the number of measured test points of sortie 4
(n=20) and ‘SM1’ stands for the standard deviation of the prediction errors of Model 1
(the standard deviation of the data presented in Fig. 5.8). Test-statistic was found to
be 1.06. Inferential statistical analysis based on the sampled data from Sortie 4 show a
significant probability of 30.1% for making a type-I error by rejecting the null-
hypothesis. This probability for a type-I error is well above the 5% significance level
accustomed in helicopter performance flight-testing. Practically, there is no significant
statistical evidence at the 95% confidence level to reject the null hypothesis therefore
it has to be accepted. Complimentary statistical analysis shows that at the 95%
confidence-level, Model 1 (Eq.(5.23)) predictions deviate on-average by up to 0.9 hp

162
 5.6 | A C O M P A R I S O N B E T W E E N THE C O N V E N T I O N A L A N D C V SD R M E T H O D S

from the actual measured power to hover. This value of 0.9 hp is well below the
deviation threshold of 1.6 hp noticeable to the Bell Jet-Ranger aircrew.

5.6 A C OMPARISON B ETW EEN THE

C ONVENTIONAL AND CVSDR M ETHODS

As previously noted in Section 5.3, the OGE hover flight-test data obtained
from a Bell Jet Ranger helicopter in a course of 4 different sorties were divided into
two groups. The first, which consisted of data from the first three sorties, was used to
develop an empirical model to represent the power for OGE hover. This model was
evaluated for accuracy while used to predict hover points of Sortie 4. Two different
models were used, the base-model which relies on the conventional hover flight-testing
method (the single-variable, Cp to (Cw)1.5 method), and another multivariable
empirical model derived from the proposed CVSDR method. Figure 5.9 presents a
comparison between the prediction errors of the two OGE hover models, the
conventional method (Eq.(5.5)) and the proposed multivariable Model 1 (Eq.(5.23)).

Figure 5.9. The conventional and CVSDR methods prediction comparison. This figure
compares the prediction accuracy achieved by both methods (conventional and CVSDR) for the
conditions of Sortie 4.

163
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

The conventional model predicts the hover points of Sortie 4 with an average
error of –3.7 hp and a variance of 18.1 hp2, whereas the proposed Model 1 yields better
predictions with an average error of -2.3 hp and a narrower variance of 9.7 hp2.
Hypothesis testing aimed at projecting from the particular case of Sortie 4 to the
general case, shows that at the accustomed 95% confidence level Model 1 predictions
deviate on-average by only 0.9 hp. Power predictions of the conventional model
deviate, on average, by a significant 1.7 hp, which is noticeable to the Jet-Ranger
helicopter aircrew. This power deviation of 1.7 hp can be translated to a gross-weight
difference of about 15 lbs. under the conditions tested (Sorties 1 - 4). The power to
hover prediction of the proposed CVSDR method was found to be substantially more
accurate than the conventional method as its deviation from the actual power was 1.9
times less than the conventional method.

One might question why is it that Model 1 predicts the power required to hover
more accurately than the conventional model? First and foremost, the CVSDR
method does not assume beforehand which predictors should be used in the empirical
model. Instead, the list of the potential 15 predictors is reduced to the most essential
and effective ones based on the specific flight-test data analyzed. This approach by
itself provides more flexibility which allows for more accurate modelling. Specifically
and as emphasized in the introduction to this chapter, compressibility effects have
substantial influence on hover performance of rotors as reported by current CFD
analysis. The conventional model neglects compressibility and drag-divergence effects,
whereas the multivariable Model 1 employs a predictor to represent the blade tip
∗ 𝑊𝜔 2
Mach-number (𝜋11 = ), therefore capable of representing compressibility and
𝛿𝜃

drag-divergence effects. The inherent assumption of the conventional single-variable

method for a constant zero-lift drag coefficient (Cd0) cannot be held valid for a wide
range of Mach numbers and for high values of main rotor disk-loading. Hovering at
low ambient temperatures (high Mach tip numbers) and at high gross-weights might
be responsible for some sections of the main rotor disk to be subjected to drag-
divergence conditions.

164
 5.7 | C O N C L U S I O N S AND SUMMARY

The two predictors (𝜋7∗ , 𝜋11

∗
) used in Model 1 can provide the extra degree of
freedom in modelling compressibility effects, which are absent in the conventional
model (the Cp to (Cw)1.5 method).

5.7 C ONCLUSIONS AND S UMMARY

The proposed CVSDR hover performance flight-testing method requires no

modification to the manner helicopter hover performance flight-test sorties are carried
out. The change is to the procedure of the data analysis. An original list of 15 corrected
variables (predictors) to represent the general hover performance of a helicopter was
formulated by means of dimensional analysis. This list is further reduced by applying
concepts of dimensionality reduction to include only the most essential and effective
CVs to represent the hover performance problem. For the particular Jet Ranger case
demonstrated in this chapter, this list of 15 CVs was reduced to only four essential
predictors. Those four CVs represented 98% of variance in the specific hover
performance data, and were applied in an empirical model to represent the OGE hover
performance of the Jet-Ranger.

The CVSDR method showed great potential as it was used successfully with OGE
hover flight-test data. The power predictions of the proposed CVSDR method were
compared to those of the conventional single-variable method, and were found to be
1.9 times more accurate. At the 95% confidence level, the CVSDR method deviated
by an average of only 0.9 hp from the actual power to hover, whereas the conventional
method deviated by an average of 1.7 hp.

Although demonstrated in this chapter using flight-test data of a Bell Jet-Ranger

helicopter, the CVSDR method is applicable and can be used for OGE hover flight-
testing of any other types of conventional helicopters, which employ a single main
rotor and a single tail rotor. The CVSDR method, at its core, is using dimensionality
reduction concepts to select the most the most effective and essentials predictors of
any physically meaningful problem. This competency of the CVSDR method can also

165
5 | HOVER PERFORMANCE T ESTING BASED ON DIME NSIONALITY REDUCTION

be applied to other types of helicopter performance testing which seek to relate ND

variables. After proved successful for hover performance testing, the following
Chapter 6 continues to develop the CVSDR method and expands it further into higher
dimensional space of level-flight performance flight-testing.

166
 Philosophy consists mostly of kicking up a lot of dust and then complaining
that you can’t see anything.
Gottfried Leibniz

6 L EVEL F LIGHT P ERFORMANCE F LIGHT

T ESTING

6.1 C HAPTER O VERVIEW

T he evaluation of the power required in level-flight is essential to any new or

modified helicopter performance flight-testing effort. The conventional flight-
test method is thoroughly discussed in Chapters 1 and 2 (Subsections 1.3.3 and 2.3.3).
This testing method is overly simplified as it is based on approximations of the induced
and profile power components. The method incorporates several drawbacks which,
not only make the execution of the flight-test sorties inefficient and time consuming,
but also compromise the level of accuracy one can expect from the empirical model
yielded. This chapter proposes an alternative flight test method for level-flight
performance of a conventional helicopter that addresses and rectifies all the identified
deficiencies of the conventional method, as stated in Subsection 1.4.3 (PS3-PS7). The
novel method, referred-to as the ‘Corrected-Variables Screening using Dimensionality
Reduction’ (CVSDR), is practically an expansion of the hover method discussed in
Chapter 5. The CVSDR flight-test method for level flight performance can be regarded
as an expansion of the hover CVSDR method into a higher dimensional space.

This Chapter 6 was published as the following journal paper: Arush I., Pavel M.D., and
Mulder M., “A Dimensionality Reduction Approach in Helicopter Level Flight Performance
Flight Testing”, Journal of the Royal Aeronautical Society, First View 13 July 2023.
https://doi.org/10.1017/aer2023.57

167
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

The CVSDR method for level-flight performance uses an original list of 36 corrected-
variables (CVs) derived from fundamental dimensional analysis principles. This list of
candidate predictors is reduced by means of dimensionality reduction to retain only
the most essential and effective predictors. The CVSDR method is demonstrated and
evaluated for prediction accuracy level in this chapter by using flight-test data of a
MBB BO-105 helicopter. It is shown that the CVSDR method predicts the power
required for level-flight about 21% more accurately than the conventional method,
while lowering the required flight time by an estimate of at least 60%. Unlike the
conventional method, the CVSDR is not bounded by the high-speed approximation
associated with the induced power estimation, therefore it is also relevant to the low
airspeed regime. This low-airspeed relevancy allows the CVSDR method to bridge
between the two important flight regimes of a helicopter, hover and level-flight. The
CVSDR method for level-flight performance is applicable to any type of conventional
helicopter.

6.2 I NTRODUCTION

The helicopter spends most of its flying-time in the level flight regime. The
relative time while cruising varies based upon the type and the specific mission the
helicopter was designed-for. Porterfield and Alexander [44] analysed data from various
types of helicopters and proclaimed that on average the helicopter spends 71% of its
flight-time in level-flight. The FAA [45] provides different estimates for two exemplary
turbine helicopters. The first example is a utility business type helicopter which
estimated to spend 61% of its flight time while cruising and the second example
presented is for a transport helicopter which is estimated to spend 73% of its flight
time in level-flight. Regardless of where this value for relative time spent in level-flight
truly resides, it is fair to say the helicopter spends most of its flight time while cruising.

The helicopter performance flight test team may be tasked to execute a level
flight performance test campaign for various reasons; it might be for a limited-scope
validation of existing performance charts for certification purposes; or it might be for

168
 6.2 | I N T R O D U C T I O N

the task of updating performance charts due to external configuration modification; or

it even be required for a full-scope level flight performance campaign, for which a
complete set of charts and/or tables is required to specify the level flight performance
of a brand new helicopter type. Whatever the reason is, the performance flight test
team has a need for an efficient and accurate method to evaluate the helicopter
performance in level-flight.

The conventional flight-test method for helicopter level flight performance is

based on a simplification of the equation for the power required to sustain a helicopter
in level flight, as already discussed in Chapter 2 (Subsection 2.3.3). This method is
further demonstrated in this chapter using flight-test data of a MBB BO-105
helicopter. Although widely used, common practice shows that this flight-testing
method is inefficient, time-consuming and includes few drawbacks which seriously
compromises the accuracy of the empirical power model it yields. The following is a
compilation of the main disadvantages of the conventional flight-testing method, as
listed in Chapter 1 (Subsection 1.4.3) as the problem statements (PS3 to PS7).

First, the conventional method reduces a multi-dimensional physical problem

into a three non-dimensional variable one. The three non-dimensional variables are
the coefficient of power, Cp, the advance ratio, μ, and the coefficient of weight, Cw.
The conventional method provides no comprehensive tools for addressing the effect
of rotor blades compressibility on the power required for level flight. Boirun [101]
addresses the compressibility effect in his work but his approach does not determine
a decisive unified empirical model to include compressibility effects. Instead, various
curves for different values of main-rotor tip Mach numbers are presented in the format
of a ‘carpet-plot’. Obtaining an accurate and unified empirical model to predict the
level flight performance is highly desirable since it can also be easily used and
implemented for real-time applications.

Second, the current method requires executions of various airspeed runs at

constant coefficients of weight (Cw). This requirement makes the method inefficient,
cumbersome and time consuming. Moreover, the resulting empirical model is prone

169
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

to elevated levels of inaccuracy since it is merely a set of single power curves for
constant Cw, rather than a unified empirical model which accounts for the entire range
of coefficients of weight.

Third, the conventional method takes the high-speed approximation for which
the induced velocity of the air through the main-rotor disk is assumed negligible
compared to the airspeed the helicopter flies at. By adopting this approximation, the
conventional method becomes irrelevant for the low airspeed regime.

Fourth, the current method has no analytical means to account for the
helicopter center-of-gravity location although numerous flight-test campaigns show
substantial dependency between the helicopter longitudinal center-of-gravity and the
power required for level-flight. For example, Buckanin et al. [6] present an increase of
about 10 square-feet in the equivalent flat-plate drag area of a Blackhawk helicopter
resulting from a 15 inches forward migration of the center-of-gravity in level flight.

Finally, the conventional method requires the flight test crew to precisely
control the main rotor-speed. This requirement makes the current flight-test method
unsuitable for helicopters which their main-rotor speed control system cannot be easily
overridden by the pilot.

Chapters 3-5 presented an alternative and more accurate approach to

helicopter performance flight-testing, using multivariable polynomials as empirical
models. Chapter 3 discussed this approach for gas-turbine power testing, and
demonstrated an increased prediction accuracy (in excess of 300%). This multivariable
approach was also used successfully in the prediction of hover performance in Chapter
5; taking this multivariable approach reduced the average prediction error by about
47% as compared to the conventional hover flight test method. The systematic
procedure to screen between candidate predictors, which is at the core of the CVSDR
method, is discussed in Chapter 4.

The goal of this Chapter 6 is to expand this multivariable polynomial approach

for the greater benefit of improving the level-flight test method. The CVSDR method

170
 6.2 | I N T R O D U C T I O N

is stretched to accommodate a more complicated helicopter performance problem

than the hover performance discussed in Chapter 5. Abstractly, this Chapter 6 can be
regarded as a rigorous expansion of hover CVSDR method into a higher dimensional
space of level-flight performance. The proposed CVSDR level-flight performance
method is aimed at addressing all identified drawbacks of the conventional method, as
specified in this Introduction and specifically listed as the five problem statements PS3-
PS7 in Chapter 1 (Subsection 1.4.3).

This chapter is structured as follows: after the short introduction, the

conventional level flight performance testing is discussed and demonstrated in
Subsection 6.3 by using flight-test data from a MBB BO-105 helicopter. The flight-test
data obtained from four distinct sorties totalling 44 data points are used to generate
four empirical models to represent the helicopter required power in level-flight. Each
empirical model is then used to predict the power required for the other three sorties.
This procedure is implemented to assess the accuracy level one can expect by using
the conventional flight-test method. Next, in Section 6.4, an alternative method
referred to as the ‘Corrected-Variables Screening using Dimensionality Reduction’
(CVSDR) is proposed. This method is demonstrated by using the same flight-test data
used with the conventional method. Section 6.5 provides a summary of the CVSDR
method and a practical step-by-step guidance to facilitate the execution of this method
by future flight-test crew. Next in Section 6.6, the expected prediction accuracy of the
CVSDR method is assessed. This expected accuracy evaluation is executed in few
different combinations of sorties to be used as data-base for model building, and
sorties to be predicted by those empirical models. Section 6.7 provides a
comprehensive comparison between the conventional and the proposed CVSDR
methods. Section 6.8 concludes and summarizes the chapter.

171
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

6.3 L EVEL -F LIGHT P ERFORMANCE T ESTING -

T HE C ONVENTIONAL W AY

The conventional flight test method for level flight performance of a helicopter is
thoroughly discussed in Chapter 2 (Subsection 2.3.3) and only briefly reminded in this
chapter. As per Eq.(6.1), the conventional method attempts to define sets of empirical
relationships between the coefficient-of-power (Cp) and the advance-ratio (µ) for
various discrete values of coefficient-of-weight (Cw). The different Cw values need to
span the entire operational envelope of the helicopter.

CP 
CW2 1
 
 Cd0  R 1  k p  2 
2 8
1 fe 3
2 Adisk


(6.1)
 P W VT 
CP  , C  ,  = 
 a Ad  R   a Ad  R  R 
3 W 2
 

For this, the flight-test crew needs to execute numerous ‘speed-runs’ while
maintaining a constant coefficient-of-weight. The technique by-which the coefficient
of weight is held constant throughout the speed-runs, defines the specific flight-test
method. Ensuring a constant coefficient-of-weight during the speed run can be
attained in two ways: (1) the constant “weight over sigma (W/σ)” method; and (2) the
constant “weight over delta (W/δ)” method.

6.3.1 Constant Weight over Sigma (W/σ)

Method

As previously discussed in Chapter 2, the constant “weight over sigma” is the

foremost popular method for level-flight performance of a conventional helicopter.
Following this method, the flight-test crew maintains the coefficient-of-weight at a
certain value by keeping the main-rotor angular speed (ω,Ω) constant and maintaining

172
 6.3 | L E V E L -F L I G H T P E R F O R M A N C E T E S T I N G - T H E C O N V E N T I O N A L W A Y

a constant ratio of weight (W) to the air relative density (σ). As presented in Eq.(6.2),
the air relative density is defined as the ratio between the ambient air density (𝜌a) and
the standard sea level air density (𝜌o). Maintaining a constant ratio of weight to relative
density (W/σ) is achieved by a gradual adjustment of the cruise altitude for the speed
runs as the helicopter burns fuel and becomes lighter. The required altitude change in-
between test points of a specific speed-run is calculated in real time by the test-crew.
It is common to encounter few iterations before the accurate altitude is reached. The
extent of altitude climb between consecutive data points relates directly to the fuel
consumption of the helicopter and the efficiency of the flight test crew to stabilize the
helicopter in the desired conditions. This altitude climb is typically between a few tens
to a few hundreds of feet. Once the new altitude is reached, the pilot needs to stabilize
the helicopter at the new airspeed and to validate (or to readjust) the main-rotor
angular speed remains constant. Note that the pilot is required to ‘stay-on-conditions’
for the entire duration necessary for the engine(s) to reach thermal equilibrium,
followed by the data gathering period of time. Typically, the flight-test campaign for a
specific helicopter configuration requires the execution of five sorties, each conducted
at a different coefficient-of-weight value. The various coefficient-of-weights shall
cover the entire certified envelope of the helicopter. Each speed run consists of at least
eight different airspeeds, beginning at some ‘arbitrary’ low airspeed to the maximum
level flight airspeed defined either by maximum available power (Vh), or by the
manufacturer’s definition for the ‘never-exceed’ airspeed (VNE).

W W W 1 1
CW     
 a Adisk  R 
2
o Adisk  R 
2
  2
o Ad R 2
held  fixed const .
(6.2)
  a 
  
 o 

173
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

6.3.2 Constant Weight over Delta (W/δ)

Method

The second and less common approach of maintaining a constant coefficient-

of-weight during the speed run is called the ‘weight over delta’ method. This method
is demonstrated mathematically in Eq.(6.3). Note that the air relative pressure ratio (δ)
is defined as the ambient air static pressure (Pa) over the standard sea level air pressure
(P0). By using the equation of state (Eq.(6.4)), the ambient air density is expressed using
the ambient static-temperature (Ta) ambient pressure (Pa) and the specific gas constant
of the air (𝑅𝑎𝑖𝑟 ). It is evident from Eq.(6.3) that by holding a constant ratio of weight
over the relative pressure (W/δ) and a constant ratio of static ambient pressure over
the angular rotor speed squared (Ta/Ω2), the flight-test crew assures a constant
coefficient-of-weight during the various speed runs. As previously discussed in
Chapter 2 (Subsection 2.3.3.3), the only advantage this method has over the W/σ
method is that it allows the flight-test crew to gain some limited control over
compressibility effect of the main rotor advancing blades.

W W W Ta Rair
CW     2
 a Adisk  R 
2
  Po    Po Adisk R 2
 Adisk  R 
2
 held  fixed
 Rair Ta  (6.3)
P  Po
  a  a 
Po Rair Ta

Pa   a Rair Ta (6.4)

This constant W/δ flight-test method requires even more flight-test sorties than
the amount required for the W/σ method. This increased number of sorties is mostly
attributed to the complexity and cumbersome associated with the continuous
adjustments of the main-rotor angular speed.

The following procedure illustrates how cumbersome and time consuming the
conventional flight-test method is. For a small size and light helicopter, such as the

174
 6.3 | L E V E L -F L I G H T P E R F O R M A N C E T E S T I N G - T H E C O N V E N T I O N A L W A Y

BO-105, it takes about five minutes to obtain one data point. One should appreciate
that out of those five minutes, only about two are essential for engine(s) thermal
equilibrium attaining and data gathering. There is about 60% of time wasted due to the
inefficiency of the conventional flight test technique. The requirement of at least eight
data points (different airspeeds) for each constant Cw and evaluating five different
values of coefficient-of-weight translates into at least 3 hours and 20 minutes of flight.
This duration should be regarded as an optimistic estimation based on small sized
helicopters. Executing level flight performance flight-test campaign on a large and
heavy helicopter might even double this time duration. Proposing an alternative flight
test method that eliminates the requirement for flying at constant coefficient of weight
has the potential for saving at least 2 hours of flight time for the same amount of
required data points (60%).

According to the conventional flight-test technique, as long as the helicopter flies

straight and level at a constant coefficient-of-weight (Cw), its level flight performance
can be uniquely represented by a single curve of coefficient-of-power (Cp) to
advance-ratio (μ). One should question how extensively can the single variables that
constitute the coefficient of weight be varied, while keeping the coefficient of weight
constant before an effect on the coefficient of power is noticeable? In other words,
how realistic is the assumption on which the conventional level-flight test technique is
built upon?

6.3.3 Example Application - Constant

Weight over Sigma (W/σ) Method

The conventional method (constant weight over sigma approach) is demonstrated

within the context of the deficiencies associated with this method, using flight-test data
obtained from a MBB BO-105 helicopter. The power required to sustain level flight at
various airspeeds was recorded during four distinct sorties. All four sorties, totalling
44 stabilized level flight points, were conducted at a targeted coefficient-of-weight
(5.79x10-3) with a tight tolerance between -0.3% to +0.7%. The main-rotor angular

175
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

speed was kept constant at 100% (equivalent to 423 RPM for the specific helicopter)
throughout the four sorties, as required by the conventional method. All physical
values for gross-weight and atmospheric conditions are summarized in Table 6.1.
Figure 6.1 presents all 44 data points of Sorties 1-4 as matching pairs of coefficient-of-
power (Cp) and advance ratio (μ) accompanied with third order polynomial best-fit
curves.

Table 6.1. Summary of flight-test conditions for Sorties 1-4.

Sortie Gross Average Pressure Ambient Cw* Average

# weight* Long. C.G. Altitude* Air Temp.* [x10-3] Cw
[lbs.] [in.] [ft.] [°C] [x10-3]
1 4890 - 5012 123.8 4000 - 4670 12 to 14 5.78 – 5.80 5.79
2 4760 - 4865 123.9 5040 - 5400 12 5.77 – 5.83 5.80
3 4270 - 4380 123.5 7770 - 8520 9 to 10 5.78 – 5.80 5.79
4 3890 - 3960 124.4 11210 - 11820 0 to 2 5.78 – 5.80 5.79
* values represent the range of change during the sortie

Figure 6.1. Level flight performance (ND) of a BO-105 helicopter. This figure presents the
coefficient of power (Cp) against the advance ratio (μ) measured under conditions of four
distinct sorties listed in Table 6.1. Data points are accompanied with third order best-fit curves.

The first concern to be discussed is with the uniqueness of the coefficient-of-

power (Cp) to advance ratio (μ) curve for the four sorties executed. As previously

176
 6.3 | L E V E L -F L I G H T P E R F O R M A N C E T E S T I N G - T H E C O N V E N T I O N A L W A Y

noted, all four sorties were conducted at the same coefficient of weight and hence
should all generate a unique coefficient-of-power (Cp) to advance-ratio (μ) curve. One
can immediately doubt it, just from observing Fig. 6.1. It is quite evident that not all
44 flight-test data points belong to the same (Cp) to (μ) curve. As listed in Table 6.1
the coefficient-of-weight (Cw) was held constant within a tight tolerance range of 1%.
The expected variance in the coefficient-of-power (ΔCp) due to the variance in Cw
(ΔCw) can be estimated by a sensitivity analysis to Eq.(6.1). This derivation is presented
explicitly as Eq. (5).

CP C
CP   CW   W  CW  , CW  5.79 103 (6.5)
CW 

This analysis show that the actual 1% variance in Cw should only be responsible
to a ΔCp of 0.02%, under a high advance ratio of 0.3. For a low advance ratio of 0.1
the expected variance in Cp should reach up to only 0.06%. The actual variance in Cp
during the four sorties reached 11% in low advance ratios of about 0.1, and 9% for
high advance ratios of about 0.3. This variance in (Cp) cannot be entirely explained by
the 1% variance in (Cw), therefore casting severe doubts on the soundness of this
conventional flight test method.

The level of accuracy achieved using the conventional flight-test method was
assessed in two ways. The first and the foremost trivial assessment was to use each
single sortie for the prediction of power required in each one of the other three sorties,
then comparing the prediction to the actual power measured. This simplistic approach
is addressed hereinafter as the single sortie approach. The second approach for
accuracy assessment was to base the power prediction of each sortie on a conglomerate
of flight-test data from the other three sorties. This approach is referred-to hereinafter
as the cluster of sorties approach.

1) The single-sortie approach: linear regressions were performed to retrieve

four distinct third order polynomials to describe the non-dimensional level-flight
performance of the BO-105 helicopter for the particular tested coefficient of-weight
(Eq.(6.6)).

177
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

CP ( j )  a3j  3  a2j  2  a1j   a0j , j  1, 2,3, 4 (6.6)

Each one of those four third order polynomials (CP(1), CP(2), CP(3), and CP(4)) was
used to predict the power required for level flight under the conditions of the other
three sorties. For example, the third order polynomial based on Sortie 1 was used to
predict power required for level flight under the conditions of sorties 2, 3 and 4. The
third order polynomial retrieved from Sortie 2 was used to estimate the power required
for level flight under the conditions of Sorties 1, 3 and 4 and so on. Power estimations
were compared to the actual measured values and prediction errors for each data point
were calculated as per Eq.(6.7).

 
Er( j )i  CP ( j )i  a3j i3  a2j i2  a1j i  a0j   A
i d  i R 3 , i  1, 2,... (6.7)

Figure 6.2 presents a summary of all prediction errors retrieved for all four sorties.
These errors are presented in horse-power units and as a function of the corresponding
advance-ratio (μ). It is worth noting that positive prediction errors mean under
estimation of the power required and a negative value represents an over estimation of
power. From an operator stand-point, underestimation is the worst-case scenario since
the helicopter demands for more power than predicted and planned for. This extra
power needed might not be available from the engine or the engines, jeopardizing a
successful execution of the mission. On the other hand, overestimation of the power
required can only contribute to inefficient planning and execution of the mission.

The prediction errors presented in Fig. 6.2 reveal a dissatisfying accuracy

performance of the conventional method. For example, power prediction errors for
Sortie 1 ranged between -20 hp (overestimate) to +18 hp (underestimate) using flight-
test data from Sortie 2. Using flight-test data from Sortie 4 to predict power levels of
Sortie 1 resulted in enormous overestimation errors that ranged between -50 hp to -2
hp. The means of the absolute prediction errors for each sortie were calculated as per
Eq.(6.8) and are presented in Fig. 6.3.

178
6.3 | L E V E L -F L I G H T P E R F O R M A N C E T E S T I N G - T H E C O N V E N T I O N A L W A Y

1 n
E R( j )   Er ( j )i (6.8)
n i 1

Figure 6.2. Power prediction errors of the BO-105 (single-sortie approach). This figure
presents the level-flight power prediction errors yielded by the conventional power models,
based on the single-sortie approach.

Figure 6.3. Mean of absolute power prediction errors (single-sortie approach). This figure
presents the mean of the absolute errors yielded for each sortie by the conventional empirical
models and based on the single-sortie approach.

179
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

The average power prediction errors range from 7.2 hp to 28 hp and are
considered by the author unacceptable for the task of level flight power prediction. It
is worth noting that for the specific type of helicopter tested, any power deviation
above (or below) 4 hp from the expected value is clearly evident to the aircrew. The
BO-105 helicopter (like many other types of helicopters) is not equipped with an
instrument that explicitly presents the engines output power in hp units; however, it is
equipped with a torque-meter gauge (‘steam-gauge’ style), installed on the instrument
panel, that indicates both engines output shaft torques. The smallest detectable
resolution of this gauge translates into a 4 hp quantity.

One might debate whether these samples of prediction errors presented in

Fig. 6.2 were drawn from a normally distributed population. For this a Quantile-
Quantile (QQ) plot is presented in Fig. 6.4. This plot compares the test data, the
prediction errors samples in the case presented, to a theoretical sample drawn from a
normally distributed population. A sample of data that comes from a normally
distributed population would manifest itself as a straight line on the QQ plot. It is clear
from Fig. 6.4 that all sampled prediction errors for sorties 1 through 4 do not come
from a normally distributed population. Taking Sortie 1 as an example, the inflection
of the curves might indicate that the largest (and smallest) estimate errors are not as
extreme as would be expected in a normal distributed population. The QQ plots for
Sortie 4 show a different behavior than those of Sortie 1. The curves inflect in a way
that might indicate heavier tails of the Probability Density Function (PDF) as
compared to a PDF of a normally distributed population. This means more extreme
prediction errors are expected from both sides, under-estimation and over-estimation,
compared to a normal distributed population.

The correlation between the power prediction level and the advance ratio was
studied. For this, the correlation coefficient (r) between the prediction error and the
advance ratio was calculated for each combination of sortie predicted and sortie used
to base the empirical prediction model on. The correlation coefficient was calculated
as per Eq.(6.9), where (n) represents the number of data points (sample size) and (S)
stands for the standard deviation of the sample.

180
 6.3 | L E V E L -F L I G H T P E R F O R M A N C E T E S T I N G - T H E C O N V E N T I O N A L W A Y

1 n
  Er   i  n  Er  
n  1 i 1
rEr ,   (6.9)
S Er S 

Figure 6.4. Prediction errors quantiles to theoretical normal quantiles (“QQ plot”). This
figure shows the sampled prediction errors don’t come from a normally distributed population.

Figure 6.5 presents these correlation coefficients for all four Sorties. Sorties 1,2
and 3 had twelve data points and Sortie 4 had only eight. The sample size affects the
correlation coefficient value to be considered significant. At the accustomed 95%
confidence level and for a sample size of twelve, a correlation coefficient of 0.58
(absolute value) and above indicates significant correlation between the two variables.
For a smaller sample size of eight (Sortie 4), significant correlation between two
variables (95% confidence level) is indicated by a correlation coefficient of 0.71 and
above. Figure 6.5 clearly indicates a significant correlation between the power
prediction errors and the advance ratio. The correlation value peaks when Sorties 1
and 2 are used to predict the power levels of sorties 3 and 4 (and vice versa). The
conclusion taken from this correlation analysis is there might be one (or few) latent
dimensions which is (are) missed by the conventional flight-test method. The empirical
prediction models based on the conventional method fail to equally estimate power
levels regardless of the advance ratio.

181
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

Figure 6.5. Power prediction errors to advance-ratio correlation (single-sortie approach).

This figure shows a significant correlation between the prediction errors using the conventional
method and the advance-ratio.

2) The cluster of sorties approach: similarly to the single-sortie approach, four

linear regressions were performed to retrieve four distinct third order polynomials to
describe the non-dimensional level-flight performance of the BO-105 helicopter for
the particular tested coefficient of-weight (Eq.(6.6)). The difference from the single-
sortie approach is that data used for the regression was based on a conglomerate of
three distinct sorties. Each one of these third order polynomials was used to predict
the power required for level flight under the conditions of the fourth Sortie, the one
not used for the linear regression. For example, data measured in Sorties 1, 2 and 3
was used to regress a third order polynomial which was used to predict the power
required of Sortie 4. Power estimation from each third order polynomial were
compared with the actual measured values and the estimation errors were calculated
as per Eq.(6.7). Figure 6.6 presents a summary of all prediction errors retrieved for all
four sorties. The prediction errors are presented in horse-power units and as a function
of the corresponding advance-ratio.

182
 6.3 | L E V E L -F L I G H T P E R F O R M A N C E T E S T I N G - T H E C O N V E N T I O N A L W A Y

Figure 6.6. Power prediction errors of the BO-105 (cluster of sorties approach). This figure
presents the level-flight power prediction errors yielded by the conventional power models,
based on the cluster of sorties approach.

Subscribing to the cluster of sorties approach slightly improves the prediction

performance. The power prediction errors of Sortie 1 ranged from -23 hp
(overestimate) to 13 hp (underestimate). Using flight test data measured in Sorties 1,3
and 4 to predict power levels of Sortie 2 yielded prediction errors between -14 hp to
19 hp. The power predictions errors for Sortie 3 ranged between -27 hp to 5.5 hp and
power predictions for Sortie 4 were all underestimating the true measured power by
up to 37 hp. The four means of the absolute prediction errors for each sortie were
calculated as per Eq.(6.8) and are presented in Fig. 6.7, alongside the absolute
prediction errors yielded from the single-sortie approach (Fig. 6.3). The averaged
absolute power prediction errors ranged between 8.8 hp to 22.9 hp (mean of 13.4 hp
with a standard deviation of 6.5 hp).

183
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

Figure 6.7. Mean of absolute prediction errors (single & cluster of sorties comparison).
This figure presents a comparison between the power prediction errors yielded by the two
approaches, the single-sortie and cluster of sorties.

Inferring from the specific averaged prediction errors presented in Fig. 6.7 to the
general case is based on hypothesis-testing. The null-hypothesis assigned is that on-
average the power required for level-flight, as predicted by the conventional flight-test
method (using the cluster-of-sorties approach) and the empirical model obtained
(Eq.(6.6)) does not differ from the true measured power by more than ±4 hp. This
null hypothesis is tested against the alternative that on-average the power required for
level-flight as estimated by the conventional method differ from the actual power by
more than 4 hp (absolute value). The motivation for setting 4 hp as the threshold for
the null-hypothesis is derived from the reasoning that for the BO-105 helicopter any
power deviation above (or below) 4 hp is noticeable to the aircrew. As previously
explained in this Subsection, the amount of power produced by the engines is
(implicitly) presented to the aircrew by the engines torques meter gauge. The smallest
detectable resolution of this gauge translates into a 4 hp quantity.

The relevant test-statistic for this hypothesis-testing is calculated per Eq.(6.10) for
which the symbol ‘n’ represents the number of sorties and ‘S’ stands for the standard

184
 6.3 | L E V E L -F L I G H T P E R F O R M A N C E T E S T I N G - T H E C O N V E N T I O N A L W A Y

deviation of the averaged power prediction errors that were calculated per Eq.(6.8) and
presented in Fig. 6.7.

t
E r  0 ,   4hp
0 (6.10)
S n

The calculated value for the test-statistic was 2.89. Inferential statistical analysis
shows the probability for making a Type-I error by rejecting the null-hypothesis to be
small (3%). This small probability for a Type-I error fall below the 5% significance
level accustomed in helicopter performance flight-testing. The practicality of this test
is that there is sufficient statistical evidence to reject the null hypothesis and to adopt
the alternative hypothesis instead. There is practically no statistical evidence to support
the null-hypothesis assigned. Complementary statistical analysis shows that on-average
and at the 95% confidence level, the level-flight power predictions based on the current
method and Eq. (6) deviate by ±5.8 hp from the actual measured power.

The poor power-prediction performance of the current flight-test method is to be

expected. As discussed above, the current level-flight performance method assumes
that for a constant coefficient-of-weight the coefficient-of-power is solely dependent
on the advance ratio, regardless of any compressibility effects that might be present.
Based on data and analysis presented above, this is clearly not a sound assumption to
make. Another potential contributor to the unsatisfactory power prediction might be
related to the change of the longitudinal center-of-gravity. As mentioned in the
introduction to this chapter, a longitudinal migration in the center-of-gravity should
have an effect on the total drag area of the fuselage, hence affecting the power required
for level flight.

The next section of the chapter presents an alternative flight test method for level-
flight performance with an improved prediction accuracy, as compared to the
conventional method. This method is based on the SVD concept, first introduced in
Section 4.4 for empirical model screening in available power testing. This SVD
approach was then reused in Chapter 5, for the novel CVSDR hover performance

185
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

flight-testing. The following section continues this course of research with the
presentation of the CVSDR method for level flight performance, which can be
regarded abstractly as a rigorous expansion of hover CVSDR method into a higher
dimensional space.

6.4 T HE CVSDR M ETHOD FOR L EVEL -

F LIGHT P ERFORMANCE T ESTING

The CVSDR method for level-flight performance aims to rectify all identified
drawbacks of the existing method, while providing better prediction accuracy as
compared to the conventional method. The CVSDR method is implemented in three
phases. Employment of this method by flight-testers requires recitation of only the last
two phases since the first phase is generic to all conventional helicopters. Phase one
deals with the generation of an original list of CVs for a multivariable analysis. In Phase
two this list of corrected variables is refined based on concepts of dimensionality
reduction through SVD. Phase three of the proposed method focuses on finding an
empirical multivariable model using the bare-essential CVs (‘predictors’) identified in
Phase 2. This list of CVs serves as an orthogonal base for the specific helicopter level-
flight performance. The complete CVSDR method is demonstrated using the same
BO-105 helicopter flight-test data, already presented in Section 6.3. Using the same
flight-test data allows for a genuine comparison of the prediction accuracy achieved
from each one of the two methods, the conventional and the CVSDR.

A practical and convenient summary of the method is presented in the next

section of the chapter (Section 6.5). This summary is intended to serve as a guide for
the flight-tester who wishes to evaluate the power required for level flight of a
conventional helicopter using the CVSDR method. This Summary provides brief
directions with regards to level-flight data base establishment and analysis.

186
 6.4 | T H E C V S DR M E T H O D FOR L E V E L - F L I G H T P E R F O R M A N C E T E S T I NG

The helicopter spends most of its flying-time in the level flight regime. The relative
time while cruising varies based upon the type and the specific mission the helicopter
was designed-for. Porterfield and Alexander [44] analysed data from various

6.4.1 Phase One – Original list of corrected

variables for level flight performance

The physical problem of the power required to sustain a helicopter in level-

flight (out-of-ground effect) was re-evaluated using tools of dimensional analysis [66,
67, and 73]. The procedure starts by proposing variables that are expected to affect the
power required in level-flight. These are the ambient static pressure (Pa), the ambient
static temperature (Ta), the helicopter gross-weight (W), the true airspeed the
helicopter flies at (VT), the main-rotor disk area (𝐴𝑑 ), the main rotor angular speed
(𝜔), and the longitudinal location of the center-of-gravity, Xcg. The power required to
hover, P, can be represented mathematically as Eq.(6.11) and Eq.(6.12) in implicit
form.

P  f ( Pa , Ta ,W ,VT , Ad ,  , xcg ) (6.11)

fˆ ( P, Pa , Ta , W , VT , Ad ,  , xcg )  0 (6.12)

The dimensions involved are presented in Table 6.2. ‘M’ represents mass, ‘L’
represents length and ‘T’ represents time.

187
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

Table 6.2. Variables and dimensions involved in level-flight performance. This

table presents all major variables affecting the level-flight performance problem and
associated dimensions.

# Physical Variable Notation Dimension

1 Power Required for Level-Flight P [M][L]2[T]-3
2 Ambient static Pressure Pa [M][L]-1[T]-2
3 Ambient static Temperature Ta [L]2[T]-2
4 Helicopter Gross-Weight W [M][L][T]-2
5 True Airspeed VT [L][T]-1
6 Main-Rotor Disk Area Ad [L]2
7 Main-Rotor Angular speed ω [T]-1
8 Longitudinal Center-of-Gravity Xcg [L]

The physical problem of power required for level flight involves eight variables
with three dimensions (L,M,T). According to the Buckingham Pi-Theorem [66] the
complexity of the problem can be reduced from eight dimensional variables to only
five Non-Dimensional (ND) variables. Following the methodology presented by
Buckingham [20], these 5 ND variables (denoted by ψ) are formed as products of the
dimensional variables. Since there are eight dimensional variables to construct five ND
variables, three dimensional variables were used as repeating variables in the ND
products (ψ). There are 56 different options to choose three variables out of eight for
the case where the order does not matter (combinations). This requires a fairly tedious
task of screening between 56 different options in order to identify the best way of
describing the non-dimensional level-flight performance. The following is a
demonstration of only one combination out of the 56 options available. In this
particular demonstration, the three repeating variables are the ambient static
temperature (Ta), the helicopter gross-weight (W) and the main rotor disk area (Ad).
The five ND products (ψ) are defined in Eq.(6.13). According to Buckingham [66], the
repeating variables should be raised to some arbitrary powers, those are denoted as
a1,b1,c1,…,c5 in Eq.(6.13). As demonstrated hereinafter, these arbitrary powers are
identified as those numeric values that make the ψ products non-dimensional.

188
 6.4 | T H E C V S DR M E T H O D FOR L E V E L - F L I G H T P E R F O R M A N C E T E S T I NG

  T a1 W b1  A c1  P  

 i a d

  T a2 W b2  A c2  P  
 j a d a 
 
 k  Ta  W   Ad    
a3 b3 c3
(6.13)
 
 m  Ta  4 W  4  Ad  4 VT  
a b c

 
 n  Ta  5 W  5  Ad  5 xcg 

a b c
 


Next, the procedure requires to replace each one of the dimensional-variables

with their corresponding dimensions and to enforce each one of the five ψ products
to be non-dimensional. This process is demonstrated as per Eq.(6.14). Each one of the
ψ products yields three equations with three unknowns, which are the exponents.
Solving for the exponents of ψi is demonstrated in Eq.(6.15). The same process is then
repeated for each one of the other ND variables, ψj , ψk , ψm and ψn.

  21
a b1
c  2 
 i    L   ML   L2  1  ML   M b1 1 L2 a1  b1  2c1  2 T 2 a1  2b1 3  M 0 L0T 0 
   T 2   T 2     T 3  
 
 a
 L2  2  ML  b2 2 c2  M  
  j      L   M b2 1 L2 a2  b2  2c2 1 T 2 a2  2b2  2  M 0 L0T 0 
  2   
  LT  
  T   T 
2 2

 
 a3
 L2   ML  b3 2 c3  1  
 k    2   2   L     M 3 L 3 3 3 T 3 3  M L T
b 2 a  b  2c 2 a  2b 1 0 0 0
 (6.14)
  T   T  T  
 a4 
  L2   ML  4 2 c4  L 
b

 m    2   2   L     M L
     
b4 2 a4 b 4 2 c4 1
T 4 4 M LT
2 a 2 b 1 0 0 0

  T   T   
T 
 a5 
  L   ML 
2 b5

 n    2   2   L   L   M L
c5
 
2 b 2 a  b  2 c  1
T 2 a  2 b
M LT
0 0 0

5 5 5 5 5 5

  T   T  

 M  : b1  1  0 
 
   0 1 0  a1   1 a1   1/ 2 
           
 L  : 2a1  b1  2c1  2  0   2 1 2   b1   2   b1    1  (6.15)
   2 2 0   c1   3  c   0 
   1  
T  : 2a1  2b1  3  0 
 

189
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

Based on Eq.(6.15), the first ND variable (ψi) can be written as Eq.(6.16).

P
i  (6.16)
W Ta

This ND variable (Eq.(6.16)) can be further simplified once the ambient static
temperature is represented using its relative value (Eq.(6.17)). This gives a simplified
expression for ψi (Eq.(6.18)) denoted as ψi*. Since this term indeed carries dimensions
and is not a pure ND, it is better defined as a ‘corrected’ variable (CV).

Pa  P0   , Ta  T0   (6.17)

P P 1 P P
i      Const  
W Ta W T0 T0 W  W 
(6.18)
P
 i* 
W 

A similar analysis was conducted to reveal the other four ND variables (ψj, ψk,
ψm and ψn). These ND variables were further simplified to represent non-dimensional
variables of a particular helicopter, hence referred-to as corrected-variables (CVs). The
corresponding CV’s are denoted with an asterisk and presented as Eq.(6.19).

W  V X cg
 *j  ,  k*  ,  m*  T ,  n*  (6.19)
   R

The procedure demonstrated above was repeated for all other 55 possibilities
of choosing three variables out of eight. From all 56 options evaluated, 20 did not yield
a unique solution and a few other returned repeating ND variables. Overall, the
analysis yielded 36 distinct CVs which can be used for the helicopter level-flight
performance. Table 6.3 summarizes all 36 CVs in an array form to indicate which of
the five dimensional-variables (power, weight, true airspeed, main-rotor angular speed
and\or longitudinal center-of-gravity location) are used in the specific CV. This list of
CVs is also presented graphically in Fig. 6.8, where one can clearly observe the number

190
 6.4 | T H E C V S DR M E T H O D FOR L E V E L - F L I G H T P E R F O R M A N C E T E S T I NG

of dimensional-variables involved in each CV. There are 6 CVs which are based on
only one dimensional-variable (1-D), 16 CVs that include two dimensional-variables
(2-D), 13 CVs which employ three dimensional-variables (3-D) and only one CV (ψ36*)
which involves four dimensional-variables (4-D). Note this ψ36* was omitted from
Table 6.3 for reasons of formatting efficiency and is presented as Eq.(6.20)

Table 6.3. Corrected-Variables (CVs) for level-flight performance.

Power M/R Weight Airspeed C.G. based 3-D

based angular- based based variables
speed based
P P P P P
 4*   5*   11
*
  22
*
  6* 
Power P  W   VT 2
X cg  3 W
 1* 
based   P 2 P P 
 7*   12
*
  23
*

P  9* 
 3 W 2
X cg    W3
P P
 20
*
  25
*

 WVT
P P 
 21
*
  26
*

  
2 3
 W3
2 VT 
1 3  P 2   27
*

 24
*
    W
   
VT  W
 28
*


 W  2  *  VT X cg
 3*   8*   16
*
 P
M/R  
13
    29
*

angular-  VT3
 14
*
 2
 X cg 
speed  17
*
 P 2
based   30
*

 VT3
W 2
 2*  WX cg P
Weight   18
*
  31
*

  X cg
based W
 19
*
  32
*

P
 X cg
2
 X cg
3

VT  33
VT
 10  
* *
Airspeed   X cg
based P
 34
*

X cg
2
VT X cg 
 15
*

C.G. R PVT
 35
*

based 2
X cg 

P
 36
*
 (6.20)
W  X cg

191
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

1-D CVs

Figure 6.8. Graphical presentation of all 36 CVs for level-flight performance. This figure
presents the classification of all determined level-flight performance CVs with their traced
dimensionality.

6.4.2 Phase Two – Screening for essential

CVs.

Phase Two of the CVSDR method, as already presented for the hover
performance in Subsection 5.4.2, focuses on narrowing the list of all candidate CVs
(Table 6.3 above) to select only those most essential and effective CVs for the specific
helicopter level-flight performance data that is being analysed. A power-based
corrected variable needs to be expressed as a function of few other CV’s. For this, the
flight tester might be asking the following two questions:

(1) How many CVs are required for a sufficient description of the level-flight
performance?

(2) Which CVs should be used?

192
 6.4 | T H E C V S DR M E T H O D FOR L E V E L - F L I G H T P E R F O R M A N C E T E S T I NG

These questions are addressed in this phase of the CVSDR method. The
procedure of CVs selection, both the quantity and types of CVs, is based on principals
of dimensionality-reduction and the correlated mathematical procedure known as the
Singular Value Decomposition (SVD). This phase of the method is demonstrated
using the same MBB BO-105 level-flight test data presented and analysed in
Subsection 6.3.3 above.

The SVD theorem is thoroughly discussed in Chapter 4 (Subsection 4.4.1),

however its fundamentals are briefly reminded to the reader hereinafter. The SVD
theorem states that any generic real matrix can be uniquely decomposed into a set of
three matrices as given in Eq.(6.21). Consider a real matrix Z to be of size (m,n) and
rank ‘r’. This matrix Z can be expressed as a product of the following three unique
matrices: matrix U, an orthonormal matrix of size (m,r) called the “Left-Singular-
Vectors” (LSV); matrix Σ, a diagonal matrix which holds along its diagonal the
singular-values of Z; and matrix V, an orthonormal matrix of size (n,r) called the
“Right-Singular-Vectors” (RSV). From an algebraic point of view, this decomposition
is viewed as a convenient way to reveal an orthogonal bases for the column and row
spaces of matrix Z (given by matrix U and VT accordingly).

 u1,1 u1,2  u1, n 

u    v1,1  vn,1 
u2, n   1
v2,1
  v vn,2 
 2,1 u2,2
2 v2,2 
Z  U V   
T
      1,2
         (6.21)
      


 r   v1, r v2, r vn, r 
u
 m,1 um,2  um, n  
1   2  ...   r  0

Alternatively, the SVD decomposition can be regarded as a ‘spectral’

decomposition of any arbitrary real matrix Z. A generic real matrix Z of rank ‘r’ can
be expressed as a linear combination of ‘r’ rank-one matrices (Eq.(6.22)).

193
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

 u1,1   u1,2 
u  u 
 2,1   2,2 
Z   1  v1,1 v2,1  vn,1    2    v1,2
  v2,2  vn,2   ...
   
     
u  u 
 m,1   m,2 
 u1, n  (6.22)
u 
 2, n 
...   r     v1, r v2, r  vn, r 
 
  
u 
 m, n 

The practicality of this approach is that any real matrix Z can be approximated
as a lower ranked matrix by using only parts of its rows and columns basis. The
‘closeness’ between the original matrix and the approximated one can be assessed by
comparing the norm of the two. There is more than one way to measure the
‘magnitude’ of a matrix (various norms). The preferable norm for the proposed
CVSDR method is the Frobenius-norm [106]. This norm is defined as the square root
of the sum of all squares of the elements of the matrix. This norm can be expressed,
with few simple algebraic passages, as the square root of the sum of all singular-values
squares (Eq.(6.23)).

 zi, j 
m n 2
Z F
  12   22  ...   r2 (6.23)
i 1 j 1

The ability to approximate any arbitrary real matrix of rank ‘r’ by an increasing
sum of rank-one matrices is the essence of the dimensionality reduction concept.
Reducing the long list of 36 corrected variables (Table 6.3 and Eq.(6.20)) to a short
and practical list of effective CVs for the level-flight performance is precisely based on
this concept of dimensionality reduction.

The procedure starts with filling matrix Z with numeral entries of all 36 CV’s
as measured for the BO-105 level-flight sorties and already presented in Subsection
6.3.3. For this demonstration 36 stabilized level-flight points measured in Sorties 1, 2
and 3 are used. The columns of the matrix represent the various CV’s (ψ1* to ψ36*) and

194
 6.4 | T H E C V S DR M E T H O D FOR L E V E L - F L I G H T P E R F O R M A N C E T E S T I NG

the 36 rows represent the different test-points measured. Next is to normalize all
columns of Z to have a mean of zero and a variance equals 1. This is done by
normalizing each entry along the columns of Z by using Eq.(6.24).

 i*  i*
 i'  , i  1, 2,...,35,36 (6.24)
S *
i

Once matrix Z contains normalized columns it is defined as Z’ and can be

partitioned into the three unique matrices expressed by Eq.(6.21). The level-flight
performance as appears in matrix Z’ is represented by all 36 CVs (ψ1* to ψ36*).
However, not all CV’s possess the same significance in representing the variance
captured in the flight-test data. The singular-values (𝜎𝑖 ) which appear along the main
diagonal of matrix 𝛴 in a descending order are key to understanding the level of
importance each CV (‘i’) holds. The conceptual interpretation of the SVD of Z’ for
the specific problem of level-flight performance is illustrated in Fig. 6.9.

Figure 6.9. The conceptual interpretation of SVD of Z’ in level-flight performance. This

figure presents how the abstract SVD of matrix Z’ (normalized predictors) should be interpreted.

195
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

The 36 singular-values of matrix Σ are normalized as per Eq.(6.25) and are

presented in Fig. 6.10 alongside a cumulative-sum plot of all normalized singular-
values.

i
ˆ i  , i  1, 2,...,35,36 (6.25)
 k 1 k
36

Figure 6.10. The normalized singular values of the level-flight performance. This figure
represents the relative magnitude of all 36 dimensions involved in the level-flight performance
of the BO-105 helicopter.

One should deduce from Fig. 6.10 that the dimensionality of the level-flight
problem can be significantly reduced from a 36-dimension problem to only a
7-dimension one. In linear-algebraic terms, it can be stated that the level-flight
performance can be sufficiently described by a basis of only seven orthogonal CVs.
The cumulative sum plot presented in Fig. 6.10 shows that 96.7% of the total variance
in the flight-test data, as stored in matrix Z, can be presented by using the seven most
significant CVs. Also indicated by Fig. 6.10, the most significant dimension of the
specific level-flight performance problem analysed holds 35% of the variance in the
data. Comparing the Frobenius norm of matrix Z’ and its 7th order approximation

196
 6.4 | T H E C V S DR M E T H O D FOR L E V E L - F L I G H T P E R F O R M A N C E T E S T I NG

(the combination of the first seven rank-1 matrices) reveals a practically similar norm
of the two; 34.986 for Z’ and 34.983 for its 7th order approximation.

The identity of the seven most important CVs is solely indicated by the right-
singular-vector (RSV) matrix. As illustrated in Fig. 6.9, each row of the RSV indicates
the level of correspondence to a specific singular-value, or a dimension, of the
problem. For example, the first row of the RSV specifies the level of correspondence
each one of the 36 CVs has with to first (and most significant) singular-value. The
second row of the RSV indicates the correspondence between all 36 CVs to the second
most significant dimension of the problem, and so on. Since the dimensionality of the
problem is reduced from 36 to seven, it is required to evaluate only the first seven rows
of the RSV matrix. For this, the elements along the first seven rows of the RSV matrix
are normalized as per Eq.(6.26) and presented in Fig. 6.11. The significance of each
CV towards the seven substantial dimensions of the level-flight performance is then
concluded.

V (i, j )
Vˆ (i, j )  , i  1, 2,..., 6, 7 (6.26)

36
j 1 V (i, j)

The left singular vectors (LSV) matrix has no significant role in the type of
analysis addressed in this paper since it only indicates the “level of correspondence”
between each one of the level-flight test points and the singular-values of Z. This type
of correspondence between particular test-points and the various dimensions of the
level-flight performance was deemed irrelevant to the topic analysed.

197
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

Figure 6.11. Correspondence between CVs and level-flight dimensions. This figure
represents the relative correspondence between each CV and each one of the 36 dimensions of
the specific BO-105 level flight performance (rows of the RSV).

The following conclusions can be drawn from Fig. 6.10 and 6.11:

(1) The first and most significant dimension of the level-flight performance
analysed holds for 35% of variance in the data and is best represented by ψ1*. This CV
represents variance in power.

(2) The second most significant dimension of the level-flight performance

analysed holds for 21.7% of variance in the data and is best represented by ψ2*. This
CV represents the variance in gross weight of the helicopter.

(3) The third dimension of the level-flight performance analysed holds for
16.1% of variance in the data and is best described by ψ14*.

(4) The fourth dimension of the problem holds for 14.3% of variance in the
data and is best represented by ψ3*.

198
 6.4 | T H E C V S DR M E T H O D FOR L E V E L - F L I G H T P E R F O R M A N C E T E S T I NG

(5) The fifth dimension of the problem holds for 5.1% of variance in the data
and is best represented by ψ30*. This ψ30* involves power and since the first dimension
already yielded a power-based CV for the role of an independent CV for the physical
problem in-hand this CV was renounced. Next in-line (non-power related) to best
represent the fifth dimension were the two CVs ψ10* and ψ13* which could not be
differentiated with respect to their representation of the fifth dimension.

(6) The sixth dimension of the problem holds for 2.3% of variance in the data
and is best represented by ψ15*.

(7) the least significant dimension in the truncated list of seven dimensions
holds for only 2% of variance in the data and is best represented by the same CV
selected to represent the third dimension, which is ψ14*.

Finally for Phase Two, a conceptual empirical-model to represent the level-

flight performance of the MBB BO-105 helicopter, as resulted from the CVSDR
method, can be stated as Eq.(6.27). This relation involves six independent corrected
variables (CVs) and one power-based dependent CV.

P W  VT VT X cg 
 1*  f ( 2* , 14
*
, 3* , 10
*
, 13
*
, 15
*
)  f  ,  2  , , , ,  (6.27)
      R 

6.4.3 Phase Three – Deriving a practical

empirical model

Once the most influential CVs of the level-flight performance problem are
exposed, a practical empirical polynomial in the six independent CVs is pursued. The
physical nature of the problem (Eq.(6.1)) suggests a third order as the highest degree
to represent the power in level-flight. This puts a cap on the order of the empirical
polynomials to be explored. As a guideline for simplicity the prospective polynomial
needs to refrain from employing any cross-products of CVs as regressors. Numerous
configurations involving the six independent CVs were evaluated for their power

199
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

estimation accuracy using the 36 stabilized data points from the first three sorties
specified in Table 6.1. The particular polynomial presented as Eq.(28, 29) was selected
due to its best performance in representing the power measurements in the first three
sorties, i.e., yielding the least values for the mean and the variance of the estimation
errors. This empirical model is addressed hereinafter as Model 123 (denoted M123)
since it is based on flight-test data from sorties 1, 2 and 3.

2
  
P
 
W
  1M 123 

 W 
 
   2M 123      3M 123     4M 123 
  
2
  ...
  
2 3
V   VT   VT   VT 
...   5M 123  T    6M 123     7 M 123     8M 123    ... (6.28)
          
V 
2
V 
3
 X cg 
...   9M 123  T    10M 123  T    11M 123     12M 123
     R 

  1M 123 
   6.679 
  2M 123   4 
  5.85  10 
  3M 123   19.475 
   
 4M 123   1019.915 
5   
 M 123   418.52 
  6M 123   2.989 
   M123 (6.29)
  7 M 123   8.31 10 
3

  19336.47 
 8M 123   
9   6189.634 
 M 123   
 10M 123   742.462 
   116.372 
 11M 123   
   102479 

 12M 123 

200
 6.5 | P R A C T I C A L G U I D A N C E FOR THE C V SD R M E T H O D I N L E V E L - F L I G H T

6.5 P RACTICAL G UIDANCE FOR THE CVSDR

M ETHOD IN L EVEL -F LIGHT

A performance flight-test campaign starts with a careful planning of the

required sorties. The power required for level flight using CVSDR is no exception to
this rule. The flight tester should plan for a set of level-flight ‘speed runs’ to cover the
applicable and required flight envelope. With the aim of establishing a sound data base
to be analysed, the flight tester should gather level-flight performance that covers the
entire range of airspeed (VT), gross-weight (W), center of gravity (Xcg), main-rotor
angular speed (ω), and ambient air properties of pressure and temperature. Figure 6.12
provides a methodical approach for sorties planning and execution while using the
CVSDR method. The flight test campaign should be executed in three configuration-
based stages. Each stage includes a set of various ‘speed-runs’ (denoted as the numbers
1 to 9 in Fig. 6.12) conducted at various conditions of gross weight, altitude and main-
rotor angular speed. Every single ‘speed-run’ should be conducted from the lowest
practicable airspeed (hover if possible) to the highest attainable level flight airspeed,
with about eight different intermediate airspeeds.

Figure 6.12. CVSDR level flight performance testing- Sorties planning sequence. This
figure represents three configuration based stages for planning and execution of level-flight
performance testing, based on the CVSDR method.

201
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

At each stabilized airspeed point, the flight-tester should gather all data needed to
compute the CVs presented in Table 6.3 and Eq.(6.20). The flight-test campaign starts
with a middle center of gravity (c.g.) configuration (the left chart in Fig. 6.12), followed
by an aft c.g. configuration (the middle chart in Fig. 6.12) and end with a forward c.g.
configuration. The first sets of speed-runs should be conducted at high altitude and
high gross weight, this would extend the range of many weight-based CVs. For
helicopters that allow the crew to adjust the main-rotor speed under standard
procedures, few sets of speed runs shall be repeated three times for three distinct values
of main rotor speed that span the governed range (see example denoted as 1a, 1b and
1c in Fig. 6.12). Note that by following the directions of Fig. 6.12 closely, the flight-
test team are expected to acquire a data base of 17 distinct speed runs, totalling about
136 stabilized level flight data points. This would constitute a sound data base to be
analysed. Succeeding the establishment of this data base, the flight test data analysis
should be conducted by following the sequential eight steps of Table 6.4. This table is
intended to provide a practical, step-by-step guidance, to realize the three phases of
the CVSDR method as discussed in Subsections 6.4.1, 6.4.2 and 6.4.3 above.

202
 6.5 | P R A C T I C A L G U I D A N C E FOR THE C V SD R M E T H O D I N L E V E L - F L I G H T

Table 6.4. A step-by-step guidance for CVSDR level-flight performance testing.

Step Task Description & Instruction

Phase One – Establish an applicable list of CVs to represent the level-flight performance.
This phase is described in Subsection 6.4.1.

1 Compute all 36 CVs (Table 6.3) for each stabilized level-flight data point
measured. There should be 136 stabilized data points, If all sorties of Fig. 6.12
are exactly executed.
2 Arrange the computed CVs in a matrix form (this is matrix Z). The rows of Z
should represent the different data points and columns of Z should represent the
various CVs. If all sorties of Fig. 6.12 were closely executed, matrix Z should be
of size 136x36.

Phase Two – Screening for the most effective CVs using dimensionality reduction. This
phase is described in Subsection 6.4.2.

3 Normalize all columns of matrix Z as per Eq.(6.24) to have a zero mean and a
variance equals 1.
4 Decompose the normalized matrix Z into its three unique matrices (U,Σ and V)
using a Singular Value Decomposition (SVD) algorithm. Matrix U is also referred
to as the Left Singular Vectors (LSV), matrix Σ is called the singular values and
matrix V is called the Right Singular Vectors (RSV).
5 Normalize all singular values (entries along the main diagonal of matrix Σ) as per
Eq.(6.25). The normalized values represent the relative strength of the various
dimensions exist in the data. Determine the number of significant dimensions
involved in the specific level-flight performance data, based on the cumulative
sum of the normalized singular values (as presented in Fig.6.10).
6 Normalize the rows of matrix VT (RSV) as per Eq.(6.26). This normalization calls
for the absolute value of each element along the rows of RSV to be divided by
the sum of all elements absolute values along the corresponding row of RSV.
7 Identify the most significant CVs of the specific level-flight performance
analysed. The level of correspondence between each CV and an abstract
dimension of the level-flight problem is illustrated in Fig. 6.9. Note that only the
first significant rows of the normalized RSV should be evaluated. The number of
significant rows of RSV equals the number of significant dimensions retrieved in
sequential step 5 above. Example for this step is presented in Fig.6.11.

Phase Three – Forming a practical empirical model (Subsection 6.4.3)

8 Use the most significant CVs identified in sequential step 7 to form a practical
polynomial that uses the relevant CVs as regressors in this empirical model.

203
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

6.6 T HE CVSDR M ODEL P REDICTION

A CCURACY (L EVEL -F LIGHT )

The prediction accuracy achieved using the CVSDR method is evaluated

hereinafter in a build-up manner. First, it is evaluated against the conventional flight-
test method by using the flight test data from sorties 1 through 4, all conducted at the
same targeted coefficient of weight. Next, the CVSDR method is challenged to predict
level flight performance of a new sortie (Sortie 5), which was conducted under arbitrary
and varying coefficient-of-weights (Cw). This evaluation is performed only for the
purpose of challenging the CVSDR-based empirical model, and to experiment up to
what extent it can predict (extrapolate) the level flight performance of the same
helicopter but under arbitrary conditions. Note the empirical models retrieved using
the conventional method in Section 6.3 (Eq. (6.6)) are irrelevant for the prediction of
Sortie 5. These empirical models are representing the level flight performance of the
helicopter for a single and specific coefficient of weight (Cw), the one targeted in
Sorties 1 through 4. For Sortie 5, the comparison between the conventional and
CVSDR methods is trivial since the conventional method immediately fails.

6.6.1 Prediction Accuracy within the same

coefficient-of-weight

The latter two phases of the CVSDR method, Phases 2 and 3 as presented in
Subsection 6.4.2 and 6.4.3, are repeated by utilizing the three other combinations
available from the flight-test data of Sorties 1 through 4. An empirical model based on
flight-test data from Sorties 1, 3 and 4 (denoted M134) is used to predict power
required under the conditions of Sortie 2. This empirical model which employs nine
distinct regressors and a constant is presented in Eq.(6.30) without the numeral
coefficient. The same approach was repeated for the derivation of M234 and M124
(empirical models based on sorties 2,3,4 and 1,2,4 accordingly) for power levels
predictions of sorties 1 and 3 respectively. The two models, M234 and M124, employ

204
 6.6 | T H E C V S DR M O D E L P R E D I C T I O N A C C U R A C Y ( L E V E L - F L I G H T )

(each) eight regressors and a constant and are presented in Eq.(6.30). Mind that the
four empirical models (M123,M134,M234 and M124) share many of the same
regressors but are not exact. This is expected since they are based on slightly different
flight test data bases.

2
  
P
 
W
 1 

 W 

  2     3    4 
  
2
 
  
 VT   VT 
2
 VT 
3
 VT   X cg 
 5   6   7     8     9     10 M134 (6.30)

           X cg   R 

M234 :  3  0  M124 :  4  0

Power prediction errors were calculated for all four sorties in the same manner
demonstrated by Eq.(6.31), specifically for Sortie 4. Figure 6.13 presents these
calculated prediction errors against their corresponding advance ratios. For
comparison purposes, Fig. 6.13 includes the prediction errors obtained from the
conventional flight-test method (cluster of sorties approach).

 2
   

W 
  i
W 
  i

  1M 123     2M 123     3M 123  2    4M 123  
i
  i
  

 2 3

  VT   VT   VT   VT  
Er  Pi   i i   5M 123     6M 123     7 M 123     8M 123    
   i   i   i   i 
  (6.31)
   VT 
2
 VT 
3
 X cg  
  10M 123     11M 123     12M 123
 9M 123       
 i i  R i 

i  1, 2,...,8

205
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

Figure 6.13. Conventional and CVSDR power prediction errors. This figure provides a
comparison between the power prediction errors yielded from the CVSDR method and the
conventional method (cluster of sorties approach) for all four sorties.

The superiority of the CVSDR over the conventional method is immediately

evident from Fig. 6.13. Power prediction of Sortie 1 using M234 resulted in prediction
errors that ranged from -12.6 hp (overestimate) to 9.9 hp (underestimate). The
prediction errors mean was -1.7 hp with a standard deviation of 7 hp. This compares
to prediction errors ranging from -23.1 hp to 12.8 hp (averaged at -7.6 hp with a wide
standard deviation of 13 hp) achieved by using the conventional method. Comparing
the two methods for the other three sorties reinforces the prediction accuracy
advantage of the CVSDR method: for Sortie 2, the CVSDR prediction errors averaged
at 0.2 hp with a standard deviation of 7.3 hp as compared to a mean of -6.3 hp with a
standard deviation of 9 hp, yielded by the conventional method. For Sortie 3, the
respective comparisons were prediction error means of 1.4 hp and -6.4 hp in favor of
the CVSDR and standard deviations of 5.8 hp and 10.8 hp in favor of the CVSDR.
For Sortie 4, the CVSDR method achieved a prediction error mean of only -1.5 hp,
compared to an underestimation average of 22.9 hp. The CVSDR prediction errors
for Sortie 4 were also less scattered as demonstrated by the two standard deviations
(8.1 hp compared to 11.5 hp).

206
 6.6 | T H E C V S DR M O D E L P R E D I C T I O N A C C U R A C Y ( L E V E L - F L I G H T )

Figure 6.14 presents an alternative view of the data displayed in Fig. 6.13. The
means of the absolute prediction errors for each sortie were calculated as per Eq.(6.8)
and are presented alongside the corresponding values retrieved from the conventional
method (cluster of sorties approach). Once more, the CVSDR method performed
better for this comparison. The means of absolute errors for Sorties 1 through 4 were
6.3 hp, 5.2 hp, 5.1 hp and 7 hp accordingly. These means compare to 12.5 hp, 9.5 hp,
8.7 hp and 22.9 hp resulted from the conventional method.

Figure 6.14. Mean of power prediction errors - conventional and CVSDR methods. This
figure provides a comparison between the power prediction errors yielded by the CVSDR and
the conventional method (cluster of sorties approach) for all four sorties.

Inferring from the particular case of the four sorties to the general case is realized
by using the hypothesis testing, as demonstrated in Subsection 6.3.3 for the
conventional method. The null hypothesis assigned is that on-average the power
required for level-flight as predicted by the CVSDR method does not differ from the
true measured power by more than ±4 hp (the smallest deviation noticeable to the
BO-105 aircrew). This null hypothesis is tested against the alternative that on-average
the CVSDR estimated power for level-flight differ by more than 4 hp (absolute value)
from the actual power. The relevant test-statistic for this hypothesis-testing is
calculated per Eq.(6.10). The symbol ‘n’ represents the number of sorties and ‘S’ stands
for the standard deviation of the averaged power prediction errors, calculated per

207
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

Eq.(6.8) and presented in Fig. 6.14. The test-statistic was fairly large (4.11), mainly due
to the relative low standard deviation. Inferential statistical analysis shows the
probability for making a Type-I error by rejecting the null-hypothesis is very small
(1.3%) hence does not support the null-hypothesis. On average and at the accustomed
95% confidence level, the CVSDR power predictions deviate from the actual measured
power by ±4.8 hp. Although above the 4 hp threshold noticeable to the BO-105
aircrew, this average prediction error is about 17% lower than the ±5.8 hp achieved
using the conventional method.

The correlation coefficient between the prediction errors and the advance ratio
was calculated for all four sorties per Eq.(6.9). Figure 6.15 presents these coefficients
accompanied with those obtained from the conventional method, cluster of sorties
approach. It is evident the CVSDR prediction errors are not significantly correlation
to the advance ratio. As already explained in Subsection 6.3.3, any correlation
coefficient above 0.58 (absolute value) for sorties 1 through 3, and above 0.71 for
Sortie 4 indicates a statistically significant correlation.

It can be concluded that based on flight-test data from all four sorties, the power
prediction accuracy obtained from the CVSDR method is not related to the advance
ratio. Similar accuracy level is expected from the CVSDR method regardless of the
corresponding advance-ratio.

208
 6.6 | T H E C V S DR M O D E L P R E D I C T I O N A C C U R A C Y ( L E V E L - F L I G H T )

Figure 6.15. Prediction errors to advance ratio correlation. This figure presents the
correlation between power prediction errors to the advance ratio using both methods, CVSDR
and the conventional (cluster of sorties approach).

6.6.2 Prediction Accuracy within a different

coefficient-of-weight

One might inquire whether the adequate performance of the CVSDR method is
made possible only due-to the fact the power estimations were made for the same
coefficient-of-weight. For this, another Sortie (number 5) was conducted under
different values of coefficient-of-weight as specified in Table 6.5. Sortie 5 was executed
without the cumbersome restriction imposed by the conventional method for
maintaining a constant coefficient-of-weight and a constant main-rotor speed while
gathering the power required to sustain level flight at various airspeeds. The
coefficient-of-weight varied between 4.8x10-3 to 4.95x10-3 and was significantly
different from the value maintained constant during the first four sorties
(5.79x10-3).

209
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

Table 6.5. Summary of flight-test conditions for Sortie 5.

Gross Long. C.G.* Pressure Ambient Cw* Main Rotor

weight* [In.] Altitude* Temp. [x10-3] speed*
[Lbs.] [ft.] [°C] [RPM]
3920 - 4080 125.6 – 125.8 5980 - 6050 8 4.81 – 4.95 421 - 425
* values represent the range of change during the sortie

The four empirical models originated from the CVSDR method were used to
predict the power levels of ten stabilized data points of Sortie 5. These empirical
models are M123 defined in Eq. (6.28) and Eq.(6.29); M234, M134 and M124 specified
in Eq.(6.30). Prediction errors were calculated by subtracting the predicted power from
the measured value, this way a positive error represents an underestimation of the
actual measured power. All power estimation errors for Sortie 5 are presented in
Fig. 6.16 against the appropriate advance ratio. This figure also includes a presentation
of the average estimation error of the four models for each data point.

Figure 6.16. Power prediction errors for Sortie 5 (CVSDR method). This figure presents the
power estimation errors for Sortie 5 using four distinct CVSDR empirical models.

As expected, all four empirical models provided adequate prediction levels, even
for different and varying values of coefficient of weights. Prediction errors ranged

210
 6.7 | C O N V E N T I O N A L AND C V S DR M E T H O D S C O M P A R I S O N

from -11.1 hp to 10.1 hp for M134, -12.4 hp to 9.1 hp for M234, -9 hp to 12 hp for
M124 and from -10.6 hp to 8.7 hp for M123. The prediction-error means were all close
to zero (-0.9 hp, -1.3 hp, -0.7 hp and -0.4 hp for M134, M234, M124 and M123
accordingly) with relatively narrow standard deviations of 7.3 hp, 8.6 hp, 6.8 hp and
9.8 hp respectively. Hypothesis testing at the 95% confidence level shows no
statistically significant difference between the prediction performances of all four
empirical models. Moreover, no statistical significance was found between the
performance of each empirical model when acted on Sorties 1 to 4 (constant Cw) or
when acted on Sortie 5. That means one can expect adequate prediction performance
when using the CVSDR method for extrapolating to a different coefficient of weight.

The correlation-coefficients (r) between the power-prediction errors of each

empirical model and the advance-ratio were calculated per Eq.(6.9). The values were
significantly low; 0.17 for M134, 0.25 for M234, 0.42 for M124 and 0.41 for M123.
For the specific number of data points in Sortie 5 (10 data points) and the accustomed
95% confidence level, only a value of 0.632 and above indicates a significant
correlation between the two variables. It can be concluded that based on flight-test
data of Sortie 5 no significant correlation was found between the power prediction
errors using all four empirical models (M134, M234, M124 and M123) and the
advance-ratio.

6.7 C ONVENTIONAL AND CVSDR M ETHODS

C OMPARISON

The conventional flight-test method for level-flight performance is based on a

simplification of the physical problem and comprises several drawbacks which affect
the accuracy and efficiency of the method. This section draws a comparison between
the conventional and the CVSDR methods by dwelling on each one of the
conventional method’s drawbacks specified in the introduction to this chapter (Section
6.2).

211
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

First and foremost, the prediction accuracy to be expected from each method is
different. Figure 6.13 shows a comprehensive comparison between the prediction
errors attained from each method for all four sorties, totalling 44 flight-test data points.
Figure 6.14 compares between the two methods by presenting the mean of the
absolute prediction errors for each sortie. The superiority of the CVSDR method over
the conventional method is clear. The conventional method generates average absolute
prediction errors of 12.5 hp, 9.5 hp, 8.7 hp and 22.9 hp, as compared to 6.3 hp, 5.2 hp,
5.1 hp and 7 hp (respectively) yielded by the CVSDR method. Statistical analysis shows
that on-average (at the 95% confidence level) the CVSDR power predictions deviate
by up to 4.8 hp (absolute value) from the actual measured power. The corresponding
deviation obtained from the conventional method is 5.8 hp, an increase of nearly 21%.

The prediction errors generated from the conventional method were significantly
correlated to the advance-ratio, whereas the CVSDR method demonstrated prediction
accuracy with no correlation to the advance ratio. This correlation between the
prediction error and the advance ratio might suggests there is a latent phenomenon
related to the advance ratio which is missed by the conventional method and the
empirical model it yields.

The conventional method is aimed at constant coefficient-of-weight data. As

such, the empirical models retrieved from the first four sorties were useless for the
predictions of Sortie 5. The CVSDR method is more versatile in this manner and was
successfully used for the predictions of Sortie 5. Besides the versatility aspect, the
constant coefficient-of-weight restriction makes the execution of the conventional
flight-test method cumbersome and more time consuming, as compared to the
CVSDR method. As discussed in Section 6.3, the conventional method requires the
flight-test crew to continuously calculate (in real-time) and adjust the cruise altitude for
maintaining a constant coefficient of weight. For a small sized and light helicopter, this
inflates the flight time required for each data point from about 2 minutes to about 5
minutes. For large and heavy helicopters this inflation rate is even expected to increase
more. For example, on a flight-test campaign that requires five different coefficients
of weight, each including eight different airspeeds, the CVSDR method is expected to

212
 6.7 | C O N V E N T I O N A L AND C V S DR M E T H O D S C O M P A R I S O N

save about 2 hours of flight time. This is about 60% reduction in the flight-test duration
required by the conventional method. Moreover, losing the requirement for a
continuous adjustment of the cruise altitude based on the helicopter weight can free
up valuable crew resources and promote flight safety.

There are two approaches of maintaining a constant coefficient-of-weights (Cw)

during a speed runs. The first is to keep a constant ratio of weight over relative density
(W/σ), and a constant main-rotor angular speed. This approach is discussed
Subsection 6.3.1 and thoroughly demonstrated using BO105 data in Subsection 6.3.3.
The second approach for maintaining a constant coefficient-of-weight was not
demonstrated in the paper but is discussed in Subsection 6.3.2. This second approach
requires the flight-tester to maintain a constant ratio of static ambient temperature (Ta)
over the main-rotor angular speed squared (Ω2) during the speed-runs. These
requirements dictate a continuous involvement of the flight-test crew with the main-
rotor speed. For the first approach of constant main-rotor speed the crew needs to
continuously apply fine-tuning, either to compensate for a non-perfect control system
(M/R speed governor) functioning, or even to override an inherent scheduling profile
dictated by the govern control laws. When executing the second approach the flight
tester involvement with main-rotor speed adjustments is even more challenging since
they need to maintain a constant value of Ta/Ω2. Besides the fact this main-rotor speed
continuous manipulation during the test imposes inconvenience on the crew, there are
types of helicopters (the MD-902 Explorer as an example) which do not allow the crew
to adjust the main rotor speed under standard procedures. For these types of
helicopters, a precise execution of the conventional level flight performance testing
method is questionable, and undesirable scatter in the data is almost inevitable.

The CVSDR method does not force the test crew to follow any kind of main-
rotor speed profile, or to keep it fixed. Any variation in the main rotor speed regardless
of its initiation source (automatically by the control system or manually by the flight-
test crew) can be used as a valid flight-test data point. That said, the flight-tester should
be reminded that flight-test data should be collected throughout the flight envelope of

213
6 | LEVEL FLIGHT PE RFORMANCE F LIGHT T ESTING

the aircraft. For this reason, performance data should be collected for the entire range
of main-rotor angular speed under normal operations (as presented in Fig. 6.12).

Another drawback inherent to the conventional method and efficiently addressed

by the proposed CVSDR method is the influence of the center-of-gravity on the power
required for level-flight. As mentioned in the chapter introduction (Section 6.2),
migration of the center-of-gravity can affect the helicopter attitude, hence alter the
drag frontal area of the helicopter. Through this mechanism the power required to
sustain level flight is affected as well. Unlike the conventional method which neglects
this influence, the CVSDR method identified a corrected variable (ψ15*) which conveys
the effect of center-of-gravity migration into the empirical power model. For the
specific type of helicopter tested and the limited scope of tests, this center-of-gravity
was identified as the 6th concept in the data (𝜎6 ), responsible for 2.3% of variance in
the data (as presented in Fig. 6.11). Note that the specific data analyzed covers a limited
center-of-gravity travel range (between longitudinal stations 123.5 and 124.4 inch as
per Table 6.1), which represents only 6.4% of the allowed longitudinal center-of-
gravity of the BO105 helicopter. Expanding the flight-test data base to include level
flight performance data measured under a larger center-of-gravity travel range might
have resulted in a larger significance of the relevant corrected variable (ψ15*).

The conventional method is bounded by the high-speed approximation, meaning

it is relevant only for airspeeds in which the induced velocity through the main-rotor
disk is negligible as compared to the airspeed the helicopter flies at. This makes the
conventional method irrelevant for modeling and estimating power required in the
low-airspeed regime. The CVSDR method is not bounded by this high-speed
approximation and is indeed relevant for the low-airspeed regime. As seen in Fig. 6.16,
the CVSDR method was also applied to the low-airspeed regime and provided
adequate power estimations in this regime. Three power estimations were made for
the advance-ratios of 0.05, 0.07 and 0.08 representing true-airspeeds of 19, 30 and
35 kts. respectively. Those estimations were at a similar accuracy level as achieved for
the high-speed regime. Nevertheless, statistical analysis for Sortie 5 and the CVSDR

214
 6.8 | C O N C L U S I O N S AND SUMMARY

method showed no significant correlation between the power prediction errors and
the advance-ratio.

6.8 C ONCLUSIONS AND S UMMARY

The conventional flight-test method to evaluate helicopter performance in level-

flight includes many drawbacks which seriously compromise its accuracy and its
execution efficiency. The proposed CVSDR method aims at addressing those
downsides of the conventional flight-test method. The CVSDR method showed great
potential as it was used successfully with level-flight test data obtained from a MBB
BO-105 helicopter. The power prediction accuracy achieved using the CVSDR
method was nearly 21% better than the level of accuracy yielded from the conventional
flight-test method. Moreover, the CVSDR method does not require the test crew to
follow a strict and binding flight scheduling, as mandated by the conventional method.
This potentially makes the CVSDR more efficient and time conserving. The CVSDR
is estimated to reduce flight-time for data points gathering by at least 60%.

The CVSDR method is not restricted by the high-speed approximation and is

therefore relevant to the low airspeed regime, as opposed to the conventional flight-
test method. This low-airspeed regime relevancy can potentially bridge the empirical-
modelling gap between the two most important flight regimes of the helicopter, the
hover, and the level-flight.

Although demonstrated using flight-test data from a MBB BO-105 helicopter, the
CVSDR method is applicable for any other type of conventional helicopter in level-
flight.

215
 No matter how thin you slice it, there will always be two sides.
Baruch Spinoza

7 C ONCLUSIONS AND

R ECOMMENDATIONS

T he goal of this dissertation is to develop new and improved flight-testing methods

for the available power, the OGE hover and the level-flight performance of a
conventional, GT engine(s) powered helicopter. The identified drawbacks of the
current flight-test methods are spelled out in the problem statement of this thesis
(Subsection 1.4). These shortcomings can be sorted under two core categories of
inefficiency and inaccuracy. The two most important properties the performance
flight-tester is seeking for are those of efficiency and accuracy. Performance flight-
testing is all about acquiring accurate empirical models for the performance of the
aircraft by applying efficient methods that minimize time and effort.

7.1 N OVEL V S . C ONVENTIONAL F LIGHT

T EST M ETHODS –M AIN D IFFERENCES

The novel flight-test methods presented in this thesis are derived from the
same source as the conventional flight-test methods. This is the fundamental approach
of dimensional analysis. The performance flight-testing problem is addressed more
efficiently by reducing the number of the participating variables. The practicality for
the performance flight-tester is that the test matrix and the associated number of
planned sorties is immensely reduced. The first substantial difference between the two

217
7 | CONCLUSIONS AND RECOMMENDATIONS

performance flight-testing approaches (novel and the conventional) relates to the role
dimensional analysis plays within the method. While the conventional flight-test
methods are using the Buckingham PI Theorem [66] only as a mean for justification,
the novel approach uses it as a genuine tool that provides non-dimensional (ND)
variables. These ND variables are used as predictors (regressors) in the empirical
models explored.

The conventional dated methods originate from overly simplified physical

models that are ‘reversed-engineered’ to yield ND variables. While the conventional
methods are simple to comprehend, and even provide satisfactory interpolation
predictions for conservative helicopter and atmospheric conditions (‘center of the
envelope’), these methods easily fail, when challenged with performance prediction of
less benign conditions, where non-linear effects become a factor. The novel flight-test
methods employ an original list of ND variables to be used as predictors. This list is
reduced using tools of dimensionality reduction (SVD) to retain only the most essential
and effective ND variables for the specific performance problem. This method is
referred-to as the ‘Corrected-Variables Screening using Dimensionality Reduction’, or
shortly the CVSDR method. Unlike the conventional method, the proposed CVSDR
is not bounded by any predetermined simplifications and approximations. This
method adjusts and adapts based on the actual flight-test data analysed. This flexibility
of the CVSDR method allows to accommodate for the various sensitivities each type
of helicopter demonstrates and also for any kind of non-linear effects, as they become
even more significant while the helicopter operates under extreme conditions and close
to the boundaries of its envelope. The proposed CVSDR method accommodates for
the exceptionality of each helicopter and for uncommon flight-conditions. This stands
opposite to the conventional flight test methods which are bounded by conservative
approximations and simplifications.

Another fundamental difference between the two approaches is with respect

to multi-variability modelling. The conventional approach consistently avoids dealing
with multivariable models and prefers the use of several single-variable mathematical
relations over employment of one multivariable relationship. The conventional flight-

218
 7.2 | C O N C L U S I O N S

testing method for the available power of a gas-turbine (GT) engine is centred on single-
variable polynomials as discussed in Chapter 1 (Subsection 1.3.1) and demonstrated,
using actual flight-test data, in Chapter 2. The conventional approach simplifies a
multivariable-type problem by assuming the output power of a GT engine can be
treated as a linear combination of single-variable models. This notion of refraining
from employing multivariable models also guides the conventional flight-test methods
for the required power. Taking the level-flight performance as an example (Subsection
1.3.3), the three-variable empirical model in coefficient-of-power (Cp), advance-ratio
(μ) and coefficient-of-weight (CW) is traded in the name of simplicity for series of two-
variable models (Cp to μ), given at various discrete values of coefficient-of- weight
(CW).

7.2 C ONCLUSIONS

The main conclusions of this dissertation are summarized and presented

hereinafter with respect to the goals and objectives of the research (Section 1.5). All
conclusions are numbered for ease of tracking and referencing. This list of conclusions
relates to the specific seven Research Questions (RQ’s) specified in the problem
statement of the dissertation (Section 1.4). Figure 7.1 presents a ‘checkers-board’ type
plot that maps the correspondence between the conclusions and the seven RQ’s of
this thesis. A dark element in the rectilinear plot of Fig. 7.1 indicates a relationship
between the specific conclusion and a particular RQ. Note there are specific
conclusions which are related to more than one RQ and vice versa; there are few RQ
which are supported by more than one conclusion. Conclusion 22 is not related directly
to any RQ’s but refers to the conventional level-flight performance method.

219
7 | CONCLUSIONS AND RECOMMENDATIONS

Figure 7.1. Conclusions to RQ’s Mapping. This checkers-board type plot presents
the correlation between the 7 RQ’s and the 22 conclusions of the thesis. A dark element
represents a correspondence between a specific conclusion and a specific RQ.

7.2.1 Flight Testing for Power Available

This subsection relates to the first Research Question (RQ1) which is Can a novel
flight-test method be developed for the available power of a gas-turbine
helicopter, which demonstrates enhanced power prediction accuracy as
compared to the conventional method? For answering RQ1 a novel method,
referred to as the ‘Multivariable Polynomial Optimization under Constraints’ (MPOC),
was developed for the available power of a gas-turbine helicopter. The method, which
is presented in Chapter 3, seeks for a third order multivariable polynomial to describe
the corrected output power of a GT engine (CSHP) as a function of the three engine
corrected variables, the corrected compressor speed (CNg), the corrected temperature
(CTGT) and the corrected fuel-flow (CWf). The MPOC method is further developed
and tuned in Chapter 4, where a systematic and repeatable methodology to choose in-
between various empirical models is discussed. The following seven conclusions,
Conclusion (1) to Conclusion (7), are related to RQ1 and were drawn while developing
the MPOC method:

220
 7.2 | C O N C L U S I O N S

(1) The output power of a helicopter GT engine is a multivariable problem that

can be adequately described by a third-order multivariable polynomial in corrected
compressor speed (CNg), corrected temperature (CTGT) and corrected fuel-flow
(CWf).

(2) The prediction of the output power of a GT engine by using a multivariable

polynomial in corrected compressor speed (CNg), corrected temperature (CTGT) and
corrected fuel-flow (CWf) is more accurate as compared to single-variable polynomials.
For the example BO-105 flight-test data used in Chapter 3, the standard deviation of
the output power estimation error is reduced from 13hp using the current single-
variable method, to only 4.3hp by using a multivariable polynomial. Expanding the
flight-test data base to seven different engines in Chapter 4 reveals that the
multivariable polynomials performed much better with all seven engines, as compared
to the single-variable approach. The maximum average prediction error using a
multivariable polynomial model was only 0.2% as compared to a maximum average
prediction error of 1.15%, using the single-variable approach.

(3) The maximum available power of a GT engine under various atmospheric

conditions can be accomplished by finding the extremum (maximum) of a
multivariable polynomial that represents the output power of the engine, subjected to
both equalities and inequalities constraints. The novel method called Multivariable
Polynomial Optimization under Constraints (MPOC) seeks for optimizing a
multivariable polynomial representing the engine output power, while satisfying
equalities and inequalities constraints. The equalities constraints are the engine
empirical internal rules of operation and the inequalities constraints are the engine
operating limitations, i.e., the compressor speed limitation, the maximum allowed
engine temperature, and the maximum fuel-flow imposed by the engine fuel pump.

(4) The Karush-Khun-Tucker (KTT) optimization method was used

successfully with the MPOC method for the task of evaluating the maximum available
power of a GT engine, under various atmospheric conditions. While the current flight-
test method yielded unrealistic predictions for certain atmospheric conditions, the

221
7 | CONCLUSIONS AND RECOMMENDATIONS

proposed MPOC method demonstrated acceptable predictions for the maximum

available power of the engine for a wider range of atmospheric conditions.

(5) The process of empirical models evaluation and screening is at the core
business of experimental flight-test data analysis. The flight-test team needs to select
the most effective and accurate empirical model to represent the aircraft performance,
as reflected by the measured data. The conventional statistical method known as the
hypothesis-testing failed to differentiate between the many candidate multivariable
polynomials based on their performance in representing the output power of a GT
engine.

(6) A singular-value-decomposition (SVD) based procedure was used

successfully to distinguish between many candidate multivariable polynomials based
on their excellence level in representing the output power of a GT engine.
Furthermore, this SVD procedure is capable of exposing latent similarities between
different GT engines with respect to their output power models.

(7) No significant correlation was found between the number of predictors

used in the multivariable empirical model and the prediction accuracy of the GT engine
output power. Two specific multivariable polynomials that employ only 11 predictors
(out of 19 available) were identified as the fourth and the fifth best-performing models
(out of 512 candidates) for the output power of an example group of seven GT
engines.

7.2.2 Flight Testing for Power Required in

OGE Hover

This subsection relates to RQ2 which is Can a novel flight-test method for
OGE hover performance of a conventional helicopter, which demonstrates
enhanced prediction accuracy as compared to the conventional OGE hover
method be developed? For answering RQ2 a novel method, referred to as the

222
 7.2 | C O N C L U S I O N S

‘Corrected Variables Screening using Dimensionality Reduction’ (CVSDR), was

developed for the power required of a conventional helicopter in OGE hover. The
method is presented in Chapter 5 and its explicit steps are summarized in Table 7.1
presented hereinafter. The following five conclusions, Conclusion (8) to Conclusion
(12), are related to RQ2 and were drawn while developing the CVSDR method for
OGE hover performance:

(8) The power required for OGE hover of a conventional helicopter can be
adequately described by a multivariable first order polynomial in corrected variables
(predictors) retrieved from a rigorous dimensional analysis.

(9) The identity and number of corrected variables required for the OGE hover
multivariable empirical model (‘conceptual empirical model’) is obtained by the
CVSDR method. For the example Bell Jet-Ranger helicopter and the specific OGE
hover flight test data base presented in Chapter 5, the CVSDR method propose a list
of four corrected variables that represent 98% of the variance in the flight-test data.

(10) The power predictions of the CVSDR method were 1.9 times more
accurate than the conventional method, when used with OGE hover flight test data of
the example Bell Jet-Ranger helicopter. At the 95% confidence level, the CVSDR
method deviated by an average of only 0.9hp (0.3% of the maximum continuous power
of the example helicopter) from the actual power required to hover, whereas power
predictions from the conventional method deviated by an average of 1.7hp.

(11) Unlike the conventional hover performance flight-testing method, the

CVSDR approach is capable of representing non-linear phenomena such as
compressibility and drag divergence in its empirical model. The CVSDR method does
not determine beforehand which predictors must be used in the empirical model.
Instead, it chooses from a list of 15 corrected variables (derived from dimensional
analysis) the most essential and effective predictors to represent the specific flight test
data analysed. This approach, by itself, provides more flexibility and allows for more
accurate empirical modelling, as compared to the conventional method.

223
7 | CONCLUSIONS AND RECOMMENDATIONS

(12) The novel CVSDR method for OGE hover performance requires no
changes to the manner current OGE hover flight test sorties are carried out. The
modification is with the data analysis procedure only.

Table 7.1. A step-by-step guidance for CVSDR OGE hover testing.

Step Task Description & Instruction

Phase One – Establish an applicable list of CVs to represent the hover performance. This
phase is described in Subsection 5.4.1.

1 Compute all 15 CVs (Table 5.3) for each stabilized OGE hover data point
measured.
2 Arrange the computed CVs in a matrix form (this is matrix Z). The rows of Z
should represent the different data points and columns of Z should represent the
various CVs.

Phase Two – Screening for the most effective CVs using dimensionality reduction. This
phase is thoroughly explained in Subsection 5.4.2.

3 Normalize all columns of matrix Z as per Eq.(5.18) to have a zero mean and a
variance equals one.
4 Decompose the normalized matrix Z into its three unique matrices (U,Σ and V)
using a Singular Value Decomposition (SVD) algorithm. Matrix U is also referred
to as the Left Singular Vectors (LSV), matrix Σ is called the singular values and
matrix V is called the Right Singular Vectors (RSV).
5 Normalize all singular values (entries along the main diagonal of matrix Σ) as per
Eq.(5.19). The normalized values represent the relative strength of the various
dimensions exist in the data. Determine the number of significant dimensions
involved in the specific hover performance data, based on the cumulative sum of
the normalized singular values (as presented in Fig.5.5).
6 Normalize the rows of matrix VT (RSV) as per Eq.(5.20). This normalization calls
for the absolute value of each element along the rows of RSV to be divided by
the sum of all elements absolute values along the corresponding row of RSV.
7 Identify the most significant CVs of the specific hover performance analysed.
The level of correspondence between each CV and an abstract dimension of the
hover problem is illustrated in Fig. 5.4. Note that only the first significant rows
of the normalized RSV should be evaluated. The number of significant rows of
RSV equals the number of significant dimensions retrieved in sequential step 5
above. Example for this step is presented in Fig.5.6.

Phase Three – Forming a practical empirical model (Subsection 5.4.3)

8 Use the most significant CVs identified in sequential step 7 to form a practical
polynomial that uses the relevant CVs as regressors in this empirical model.

224
 7.2 | C O N C L U S I O N S

7.2.3 Flight Testing for Power Required in

Level Flight

This subsection addresses the five Research Questions (RQ3, RQ4, RQ5, RQ6,
and RQ7) which are specified in the problem statement of this thesis (Section 1.4).
These five research questions relate to the deficiencies associated with the current
level-flight performance flight-testing of a conventional helicopter. The novel CVSDR
flight-test method for power required in level flight was developed specifically for
addressing the five RQ’s (RQ3, RQ4, RQ5, RQ6, and RQ7). This novel method is
thoroughly discussed and demonstrated in Chapter 6 and its explicit steps are
summarized in Table 7.2 presented hereinafter. Abstractly, the CVSDR method for
level flight can be regarded as a rigours expansion of the hover CVSDR method into
a higher dimensional-space. The following ten conclusions, Conclusion (13) to
Conclusion (22), were drawn while developing the CVSDR method for level flight
performance. The detailed mapping of these ten conclusions to the particular five RQ’s
is presented in Fig. 7.1.

(13) The power required for level flight of a conventional helicopter can be
adequately described by a multivariable first order polynomial in corrected variables
(predictors) retrieved from a rigorous dimensional analysis. The list of corrected
variables includes predictors that represent various coefficient-of-weight and account
for non-linear effects. This conclusion relates directly to two research questions; the
non-linear effects of RQ3 and the various coefficient-of-weight of RQ4.

(14) The identity and quantity of corrected variables required for the level flight
multivariable empirical model (‘conceptual empirical model’) are established by the
CVSDR method. For the example MBB BO-105 helicopter and the specific level flight
test data base presented in Chapter 6, the CVSDR method proposed a list of seven
corrected variables that represent 96.5% of the measured variance in the data.

(15) The power predictions accuracy achieved using the CVSDR method for
level-flight was nearly 21% better (on average and at the 95% confidence level), as

225
7 | CONCLUSIONS AND RECOMMENDATIONS

compared to the prediction accuracy yielded from the conventional method. Note this
conclusion relates directly to the improved prediction accuracy of the novel CVSDR
method as required by RQ3.

(16) The novel CVSDR method for level flight made planning and execution
of flight-test sorties more efficient and time conserving. It is estimated to reduce flight-
time for data gathering by at-least 60%. Note this conclusion relates directly to the
efficiency of the novel CVSDR method as required by RQ4.

(17) The novel CVSDR method for level flight does not require a continuous
and accurate adjustment of the flight altitude, as mandated by the conventional
method. Renouncing this burdensome requirement can free up valuable crew
resources and promote flight safety. Note this conclusion relates directly to the
efficiency of the novel CVSDR method as required by RQ4

(18) The novel CVSDR method for level flight does not require to keep the
main-rotor angular speed constant throughout the test, as required by the conventional
method. This makes the CVSDR method more versatile and relevant for helicopter
types, which do not enable pilot-initiated main rotor speed adjustments under standard
flight procedures. This conclusion relates directly to RQ7.

(19) The novel CVSDR method for level flight is not restricted by the high-
speed approximation like the conventional method. This makes the CVSDR an
appropriate method for the low-airspeed regime, and can potentially bridge the
empirical modelling gap between the hover and level-flight domains. This conclusion
relates directly to RQ5.

(20) The power predication errors yielded by the CVSDR method were not
significantly correlated to the advance ratio, as opposed to the prediction errors
returned from the conventional method. This might suggest that the CVSDR method
is capable of identifying a latent advance-ratio related phenomenon, completely
overlooked by the conventional method.

226
 7.2 | C O N C L U S I O N S

(21) The novel CVSDR flight test method for level flight comprises the effect
of center-of-gravity location on the power required. This significant competence adds
much value to the CVSDR method over the conventional method. This conclusion
relates directly to RQ6.

(22) The soundness of the conventional flight-test method for level flight
performance is seriously questionable in light of the research level-flight test sorties.
The theoretical uniqueness of the coefficient-of-power (Cp) to advance ratio (μ) curve
for four sorties executed at a nominal constant coefficient-of-weight (CW) was found
inaccurate. The measured 11% variance in Cp cannot be entirely explained by the
inaccurate flight test execution which resulted in only 1% variance in CW. Note this
conclusion is not related directly to any RQ’s but refers to the drawbacks of the
conventional level-flight performance method.

227
7 | CONCLUSIONS AND RECOMMENDATIONS

Table 7.2. A step-by-step guidance for CVSDR level-flight testing.

Step Task Description & Instruction

Phase One – Establish an applicable list of CVs to represent the level-flight performance.
This phase is described in Subsection 6.4.1.

1 Compute all 36 CVs (Table 6.3) for each stabilized level-flight data point
measured. There should be 136 stabilized data points, If all sorties of Fig. 6.12
are closely executed.
2 Arrange the computed CVs in a matrix form (this is matrix Z). The rows of Z
should represent the different data points and columns of Z should represent the
various CVs. If all sorties of Fig. 6.12 were closely executed, matrix Z should be
of size 136x36.

Phase Two – Screening for the most effective CVs using dimensionality reduction. This
phase is described in Subsection 6.4.2.

3 Normalize all columns of matrix Z as per Eq.(6.24) to have a zero mean and a
variance equals 1.
4 Decompose the normalized matrix Z into its three unique matrices (U,Σ and V)
using a Singular Value Decomposition (SVD) algorithm. Matrix U is also referred
to as the Left Singular Vectors (LSV), matrix Σ is called the singular values and
matrix V is called the Right Singular Vectors (RSV).
5 Normalize all singular values (entries along the main diagonal of matrix Σ) as per
Eq.(6.25). The normalized values represent the relative strength of the various
dimensions exist in the data. Determine the number of significant dimensions
involved in the specific level-flight performance data, based on the cumulative
sum of the normalized singular values (as presented in Fig.6.10).
6 Normalize the rows of matrix VT (RSV) as per Eq.(6.26). This normalization calls
for the absolute value of each element along the rows of RSV to be divided by
the sum of all elements absolute values along the corresponding row of RSV.
7 Identify the most significant CVs of the specific level-flight performance
analysed. The level of correspondence between each CV and an abstract
dimension of the level-flight problem is illustrated in Fig. 6.9. Note that only the
first significant rows of the normalized RSV should be evaluated. The number of
significant rows of RSV equals the number of significant dimensions retrieved in
sequential step 5 above. Example for this step is presented in Fig.6.11.

Phase Three – Forming a practical empirical model (Subsection 6.4.3)

8 Use the most significant CVs identified in sequential step 7 to form a practical
polynomial that uses the relevant CVs as regressors in this empirical model.

228
 7.3 | R E C O M M E N D A T I O N S

7.3 R ECOMMENDATIONS

The following are recommendations concerning possible future expansion of the

research.

(1) One of the main advantages of the CVSDR method for level flight is that it is
unrestricted by the high-speed approximation (Conclusion 19), therefore is applicable
to the low-airspeed regime, unlike the conventional method. This opens up an
opportunity to provide a unified empirical model to describe the power required from
hover to the maximum horizontal airspeed of the helicopter. Under the current
research limitations for number of flight-hours, availability of aircraft and special
flight-test instrumentation it was not feasible to employ the CVSDR method for level
flight, from a hover, through the low airspeed regime to the maximum airspeed for
level flight. Future research should focus on the applicability and accuracy of the
CVSDR method when used as a unified empirical model for the power required from
hover to maximum airspeed in level flight.

(2) The current research was limited to out of ground effect (OGE) hover only.
Although performance flight testing for in-ground-effect (IGE) was excluded, the
derivation of the proposed CVSDR method in Chapter 5 includes provisions to also
address the IGE hover. The applicability and accuracy of the CVSDR method for
power required to IGE hover should be evaluated in future research.

(3) Future research should expand the CVSDR flight-testing method to include
more areas of helicopter performance. These are the power required in a climb (vertical
climb and forward climb), and partial-power and unpowered descent performance
(‘Autorotation’).

(4) In recent years we have witnessed an increasing number of vertical lift aircraft
types that combine both fixed-wing (FW) and rotary-wing (RW) characteristics. This
duality also affects the performance flight-test methods, especially for the transition
from RW to FW envelope. The general approach presented in this thesis, of

229
7 | CONCLUSIONS AND RECOMMENDATIONS

establishing a generic list of corrected-variables using dimensional analysis, followed

by an elimination procedure based on dimensionality reduction to identify the most
essential for the specific performance problem analysed and then to establish an
empirical multivariable model, can work better for this type of a dual-characteristic
aircraft. Future research should be focused on evaluation of the applicability and
efficiency of the performance flight-testing method developed in the current research,
to relevant vertical-lift aircraft that combine both RW and FW characteristics.

(5) The current research shows that power prediction accuracy for hover and
level-flight is better with the proposed empirical models, as compared to the empirical
models yielded by the conventional methods. Part of the improved prediction accuracy
can be attributed to the increased number and improved quality of the predictors
(corrected-variables) used. Using more appropriate and effective degrees of freedom in the
empirical model surely promotes prediction accuracy. That being said, the hover and
level-flight research sorties were conducted under relatively moderate flight-
conditions. It is believed the full potential of the proposed performance flight-testing
methods was not entirely exposed by the relatively moderate flight conditions tested.
Future research should apply the novel CVSDR method for hover and level flight
under extreme conditions of atmosphere and configuration. This includes high altitude
and low ambient temperatures to expose the helicopter to severe compressibility
effects, and for all corners of gross-weight/center-of-gravity envelope.

(6) Future research should look into the potential and feasibility of employing the
CVSDR method for empirical modelling by the Health and Usage Monitoring Systems
(HUMS) installed in helicopters. For example, the novel CVSDR method for
performance flight testing could be entirely automated and integrated into HUMS to
provide real time performance empirical modelling for and prediction of the specific
helicopter (not just the type). The CVSDR algorithm can also be used to flag
exceptional prediction discrepancies that might be indicative of potential helicopter
malfunctions and hazards.

230
 7.4 | C L O S I N G R E M A R K S

7.4 C LOSING R EMARKS

The performance flight-tester needs practical and efficient flight-test methods

with the purpose of producing accurate empirical models for aircraft performance
prediction. The novel flight-test method presented in this dissertation fits into this
need for a practical, efficient and accurate performance flight-testing method. The
novel method developed and tested in this research, yielded better prediction accuracy
as compared to the accuracy level of conventional methods. The available power
prediction using the novel method was on average 5.75 times more accurate than the
conventional flight-test method. The novel method predicted power required for
OGE hover, about 1.9 times more accurately than the conventional method. Superior
performance prediction of the novel method was also demonstrated for power
required for level flight (about 21% more accurate). It is believed the main reason for
superior prediction accuracy is mainly attributed to the flexibility of the novel method.
It is capable to adjust and to adapt to specific flight-test data reflecting various
helicopter types, aircraft configurations and ambient conditions. This flexibility does
not exist to the same extent within the conventional flight-test methods. The efficiency
and practicality of the novel method is mostly demonstrated in level flight performance
flight-testing. As thoroughly discussed in Chapter 6, the novel method is estimated to
reduce flight-time for data gathering by at-least 60%, while at the same time decreasing
the complexity in flight-test execution, hence promoting safety of flight.

Alongside the many advantages of the novel flight-test method, there is one
drawback. At the core of this novel method lies an analytical procedure that starts
with fundamental dimensional analysis, followed by dimensionality reduction
procedure, based on the concept of singular-value-decomposition (SVD). This
dimensionality reduction procedure requires for more analysis effort, both in terms of
flight-test data reduction and a supporting software package, capable of performing
the SVD algorithm.

The scope of this research was limited to the power available of gas-turbine
engines and the power required for a conventional helicopter in OGE hover and in

231
7 | CONCLUSIONS AND RECOMMENDATIONS

level-flight. Despite focusing on limited flight-testing areas only, it is believed the novel
method presented in this dissertation is applicable to other areas of performance flight-
testing and can be employed outside of the conventional helicopter configuration. It
is recommended the novel performance flight-test method be evaluated in the future
under extreme flight and ambient conditions, for other areas of performance flight-
testing and also for unconventional helicopters, those which do not conform to the
single main-rotor and a single tail-rotor configuration.

232
 A. G AS -T URBINE E NGINE
D IMENSIONAL A NALYSIS

T his appendix provides the generic gas-turbine engine dimensional analysis. The
classical Buckingham Pi theorem [66] is used to demonstrate how the gas-turbine
engine dimensional physical problem is converted into a non-dimensional (ND)
problem and how should the dimensional variables involved in the gas-turbine engine
performance problem be non-dimensionalized or ‘corrected’ in order to reduce the
number variables involved in the problem. The method of parameter correction
presented here is the classical approach that follows from the theorem. Note there are
few turbine engine manufacturers that tweak or ‘fine-tune’ these classical correction
factors to better work with their specific engine types.

Consider first the dimensional variables that affect the generic gas turbine engine
output power problem. The practice would suggest that these should be the ambient
static pressure, Pa, ambient static temperature, Ta, engine compressor speed, Ng,
engine temperature, TGT, fuel-flow (weight flow), Wf and the physical size of the
engine. Consider a descriptive cross section area, Ae, to represent the physical size of
the engine. The engine output power, SHP, can be mathematically represented as in
Eq.(A.1) or equivalently implicitly as in Eq.(A.2).

SHP  f ( Pa , Ta , Ng , TGT ,W f , Ae) (A.1)

f ( SHP, Pa , Ta , Ng , TGT ,W f , Ae)  0 (A.2)

233
A | GAS-T URBI NE ENGI NE DIME NSIONAL ANA LY SIS

The dimensions involved in this physical problem are presented in Table A.1. The
notation M represents mass, L represents length and T represents time.

Table A.1 – Gas turbine engine - summary of variables and dimensions involved.

# Physical Variable Notation Dimension

1 Shaft Output Power SHP [M][L]2[T]-3
2 Ambient Air static Pressure Pa [M][L]-1[T]-2
3 Ambient Air static Temperature Ta [L]2[T]-2
4 Engine Compressor Speed Ng [T]-1
5 Engine Temperature TGT [L]2[T]-2
6 Engine Cross Section Area Ae [L]2
7 Fuel (Weight) Flow 𝑤𝑓 [M][L][T]-3

There are seven dimensional variables involved in the problem with three
dimensions (L, M, and T). According to the Buckingham Pi Theorem, the complexity
of the physical problem can be reduced from seven dimensional variables to four Non-
Dimensional (ND) variables. This is the number of dimensional variables minus the
number of dimensions involved in the problem, i.e., seven minus three equals four.

The next phase is to build those four ND variables as products of the dimensional
variables. The four ND variables will be denoted by 𝜋𝑖 (hence the origin of the name
of this Pi theorem). Since there exist seven dimensional variables to be used for the
construction of the four ND variables, three dimensional variables out of the seven
must be used as repeating variables in the ND products.

There are 35 different options to choose three variables out of seven for the case
where the order does not matter (combinations) as shown by Eq.(A.3), for which N
represents the number of options.

7 7!
N    35 (A.3)
 3   3! 4!

234
 Appendix A

It is a fairly tedious task of screening between 35 different combinations. This

phase usually involves some trial and error rounds before obtaining the most suitable
configuration. The following is a demonstration of only one combination out of the
35 options available. In this particular example presented in Eq.(A.4), the three
repeating variables were chosen as the ambient air static pressure, Pa, the ambient air
static temperature, Ta, and the engine cross-sectional area, Ae.

   P a1 T b1  Ae c1  SHP  

 1 a a

   P a2 T b2  Ae c2  Ng  
 2 a a 
  (A.4)
 3   Pa  Ta   Ae  TGT  
a3 b3 c3

 

b
 
 4   Pa  4 Ta  4  Ae  4 W f 
a c


According to Buckingham [66], the repeating variables should be raised to some

arbitrary powers, those are denoted as a1,b1,c1,…,c4 in Eq.(A.4). As demonstrated
hereinafter, these arbitrary powers are identified as those numeric values that make the
𝜋𝑖 products non-dimensional.

The next step in this process of non-dimensionalizing the problem is to enforce

all four 𝜋𝑖 products to be non-dimensional. First, each variable is replaced with its
equivalent dimensions representation, then a system of linear equations is produced.
Each non-dimensional variable produced a set of three equations in three unknowns
as presented in Eq.(A.5).

235
A | GAS-T URBI NE ENGI NE DIME NSIONAL ANA LY SIS

 a1
 2 1
b
 ML2  
1   M   L  L 
c1
 
2
 3
a 1  a  2b  2 c  2 2 a  2b 3
  M 1 L 1 1 1 T 1 1 M 0 L0T 0 
  LT 2   T 2   T  
 
 M  2  L2 
a b2

 2      
c2
   L2 T 1  M a2 L a2  2b2  2c2 T 2 a2  2b2 1 M 0 L0T 0 
  LT 2   T 2  
 b3
 (A.5)
  M 3L 
a
c3  L2  
 
2
 3      L2  2   M a3 L a3  2b3  2c3  2T 2 a3  2b3  2 M 0 L0T 0 
  LT 2   T 2  T 
  
 b4 
  M 4L  
a

 
2 c4  ML 
       
 4   L2  3   M 4 L 4 4 4 T 4 4
a 1 a 2 b 2 c 1 2 a 2 b 3
   M 0 L0T 0 
  LT 2   T 2  T  

Realizing the exponents for each ND variable 𝜋𝑖 is done through solving the
four sets of three equations with three unknowns. This procedure is demonstrated for
for 𝜋1 in Eq.(A.6):

 M  : a1  1  0 
 
   1 0 0  a1   1   a1   1 
          
 L  : a1  2b1  2c1  2  0   1 2 2  b1    2   b1     1 2  (A.6)
   2 2 0  c   3   c   1 
    1     1  
T  : 2a1  2b1  3  0 
 

Once the exponents of 𝜋1 are found, the first ND variable is revealed (Eq.)

1 SHP 1
1   Pa  1 Ta  1  Ae  1  SHP    Pa 
a b c
Ta 1/2  Ae 1  SHP   (A.7)
Pa Ta Ae

The static ambient pressure, Pa, and the static ambient temperature, Ta, can be
replaced with the pressure and temperature ratios as respectively appears in Eq.(A.8)
and (A.9) to constitute Eq.(A.10)

pa
 , po  14.7 psi (A.8)
po

236
 Appendix A

Ta
 , T0  288K (A.9)
T0

SHP 1 SHP
1     K1 (A.10)
  Po Ae To  

Since 𝐾1 in Eq.(A.10) is a constant for a specific helicopter (the engine cross

sectional area does not change and P0 and T0 represent the standard sea level values
for static air pressure and static air temperature respectively) it is evident that the first
ND variable involved in the problem (𝜋1 ) is the expression presented in Eq.(A.11).

SHP
1*  (A.11)
 

Note that the term defined in Eq.(A.11) is not a pure ND parameter and it
carries units. For this it is better defined as a “corrected-variable” and is symbolized
with an asterisk (*) to differentiate it from the true ND variable in Eq.(A.10).

From the second equation in Eq.(A.4), one can extract the following set of
equations (Eq.(A.12)) to yield the second ND (Eq.(A.13),(A.14)) and corrected
variable (Eq.(A.15)) by following the same procedure presented above for 𝜋1 .

 M  : a2  0 
 
  1 0 0  a2   0   a2   0 
          
 L  : a2  2b2  2c2  0   1 2 2  b2    0   b2    1 2  (A.12)
   2 2 0  c   1  c   1 2
    2     2  
T  : 2a2  2b2  1  0 
 

237
A | GAS-T URBI NE ENGI NE DIME NSIONAL ANA LY SIS

Ng
 2   Pa 
a2
Ta b  Ae c  Ng    Pa 0 Ta 1/2  Ae 1/2  Ng  
2 2
Ae (A.13)
Ta

Ng Ae Ng
2     K2 (A.14)
 T0 

Ng
 2*  (A.15)


From the third equation in Eq.(A.4), one can extract the following set of
equations (Eq.(A.16)) to yield the third ND (Eq.(A.17),(A.18)) and corrected variable
(Eq.(A.19)) by following the same procedure presented above for 𝜋1 , 𝜋2 .

 M  : a3  0 
 
   1 0 0   a3   0   a3   0 
          
 L  : a3  2b3  2c3  2  0   1 2 2   b3    2   b3    1 (A.16)
   2 2 0   c   2  c   0
    3     3  
T  : 2a3  2b3  2  0 
 

1 TGT
 3   Pa  3 Ta  3  Ae  3 TGT    Pa  Ta   Ae 0 TGT  
a b c 0
(A.17)
Ta

TGT 1 TGT
3    K3 (A.18)
 T0 

TGT
 3*  (A.19)


From the fourth equation in Eq.(A.4), one can extract the following set of
equations (Eq.(A.20)) to yield the third ND (Eq.(A.21),(A.22)) and corrected variable
(Eq.(A.23)) by following the same procedure presented above for 𝜋1 , 𝜋2 and 𝜋3 .

238
 Appendix A

 M  : a4  1  0 
 
  1 0 0  a4   1   a4   1 
          
 
L :  a4  2b4  2c4  1  0    1 2 2 b 
 4    1   b4     1 2  (A.20)
   2 2 0  c   3   c   1 2 
    4     4  
T  : 2a4  2b4  3  0 
 

W f    Pa 1 Ta 1/2  Ae 1/2 W f   P

Wf 1
 4   Pa  Ta b  Ae c
a4
(A.21)
4 4

a Ta Ae

Wf 1 Wf
4     K4 (A.22)
  Po To Ae  

Wf
 4*  (A.23)
 

Finally, Eq.(A.2) can be rewritten in its corrected form (ND for a specific
engine) as in Eq.(A.24). The gas-turbine engine output power problem is now reduced
to be a function of only four corrected variables, π1*, π2*, π3* and π4*. Another way to
present Eq.(A.24) is in its explicit form as Eq.(A.25). The corrected output power of
the engine is a function of three corrected variables i.e., the corrected compressor
speed, the corrected engine temperature and the corrected engine fuel flow.

 SHP Ng TGT W f 
f * (1* ,  2* ,  3* ,  4* )  f *  , , ,   0 (A.24)
      

SHP  Ng TGT W f 
1*  f ** ( 2* ,  3* ,  4* )   f **  , ,  (A.25)
       

239
 B. H IGH S PEED A PPROXIMATION ,
10K F T ., S TANDARD D AY

This appendix provides a graphical tool to assess the inaccuracy introduced by

using Glauert’s high-speed approximation, as compared to the CMIV under standard
day conditions, 10K ft. pressure altitude. A similar graph for Standard Sea Level (SSL)
conditions is presented in Chapter 2 (Subsection 2.2.1) as Fig. 2.12.

241
Appendix B

242
 C. R ESEARCH H ELICOPTERS
D ESCRIPTION

This appendix provides a more detailed description of the two helicopters used
for this research; the Bell Jet Ranger and the MBB BO-105 helicopters used for training
at the National Test Pilot School (NTPS) in Mojave, California.

T HE B ELL J ET -R ANGER H ELICOPTER

The Bell Jet-Ranger helicopter used for the research is a single-engine light
observation helicopter, designed for day and night, visual flight rules (VFR) and
instrument flight rules (IFR) operations. The helicopter is designed for landing and
take-off from prepared or unprepared surfaces with a skid-type landing gear. The
helicopter can be flown by a single pilot from the right-hand seat and has a place for
three passengers in the back. The helicopter has an overall length of 12.5 m (main
rotor fore to aft end of tail) and a maximum allowed take-off gross weight of
3,200 lbs. (1,452 kg.). The helicopter conforms to the definition of a ‘conventional’
helicopter, as it has a single main-rotor (M/R) and a single tail-rotor (T/R). The M/R
assembly is a two-bladed, semi-rigid, teetering type also known as underslung
feathering axis hub. The M/R rotates counter-clockwise, when viewed from above, at
a standard angular speed of 354 RPM. An audio warning tone and a RPM warning
light are designed to come on and alert the pilot when the M/R angular speed decreases
below 335 RPM. The M/R blades are all metal and consist of extruded aluminium

243
C | RESEARCH HELICOPTERS DESCRIPTION

alloy nose block and trailing edge, filled with aluminium honeycomb structure filler.
The T/R configuration is two bladed teetering with 30° delta-three flapping hinge
offset. The T/R is mounted on the left side of the tail-boom structure, and rotates
with the bottom blade traveling forward. The T/R operates as a ‘pusher’ type, i.e., it
generates an anti-torque force which pushes the tail structure of the helicopter to the
right.

The helicopter is powered by a single Allison T63-A-720 gas-turbine engine,

which is installed immediately behind the main transmission as shown in Fig. C.1. The
engine uninstalled maximum output power is rated at 420 shaft horsepower (shp)
under standard day, sea level conditions. Once installed in the Jet-Ranger helicopter,
the engine performance is de-rated due to drivetrain limitations to a maximum output
power of 317 hp (5 minutes take off rating limit) and of 270 hp (continuous operation).
The dry weight of this engine is 158 lbs. The engine power is transmitted through a
freewheeling unit to the main transmission, which drives both the main and the tail
rotors. The freewheeling unit is designed to disconnect both rotors drive shafts from
the engine, enabling rotation of the main and the tail rotors through the main
transmission in case of an emergency engine-out auto-rotational descent. The
maximum output power (take-off rating) is attained with a 100% engine indicated
torque and 100% N2 (power turbine angular speed). The engine incorporates a
compressor section with six axial stages followed by a centrifugal stage, a gas-turbine
section with two axial stages and a two axial stages power-turbine. This engine
configuration is of a reverse-flow annular type, for which the exhaust gases follow a
reverse path to the compressor gasses, before passing through the turbine section.

244
 THE BELL JET-RANGER HELICOPTER

Figure C.1. The Allison T63-A-720 gas turbine engine. This photo shows the engine, as
installed on the Jet Ranger helicopter.

The output power of the engine is controlled by the turbine power governor
through means of varying the amount of fuel that flows to the engine. The power
turbine speed (N2) is selected by the pilot, and the output power required to maintain
this speed is commanded by the power-turbine hydro-mechanical governor. With the
throttle rotated to full-open position, the power-turbine governor works to maintain a
constant N2 speed. Rotating the throttle towards the idle position causes the N2 speed
to be manually selected, instead of automatically controlled. A droop-compensator
unit, within the power-turbine governor, maintains a constant N2 speed as power
demand is increased (or decreased) by main-rotor collective stick manipulation. This
function is implemented through a mechanical linkage that connects between the
collective stick to the speed selector lever on the N2 governor.

The dual flight controls of the helicopter (for the pilot and for another crew
member) consist of mechanical type non reversible cyclic-stick, collective-stick, and
reversible directional pedals to control the helicopter in yaw. Figure C.2 shows the
pilot station (right-hand side) flight controls. The flight controls pass-on the pilot
inputs to the three main-rotor hydraulic servo cylinders and to the tail-rotor forged

245
C | RESEARCH HELICOPTERS DESCRIPTION

aluminium alloy yoke. An electrically operated mechanical unit provide cyclic stick
force trim function which helps in providing artificial force gradient to the pilot and
to hold the cyclic stick in place and prevent it from migrating due to vibrations and
mass imbalance. The collective stick incorporate no means of ‘trimming’ capabilities
besides a simple adjustable mechanical friction.

The helicopter features a rectangular fixed-position horizontal stabilizer. This

stabilizer is mounted approximately at the middle of the tail-boom and has a span of
6 ft. and 5 inches (195.6 cm.) and a chord of 45cm. This rectangular shaped horizontal
stabilizer (shown in Fig. C.3) is designed to generate a down force with forward
airspeed. A fixed position vertical fin is located to the right of the tail-boom.

Figure C.2. The Jet-Ranger flight-controls. This photo shows the pilot seat (right-hand side)
flight-controls; the cyclic-stick, the tail-rotor pedals (only the right pedals is shown) and the
collective stick mounted to the left of the pilot seat.

246
 THE BELL JET-RANGER HELICOPTER

The pitot-static system is used to measure airspeed, pressure-altitude and vertical

rate of climb and descent. This system is made-up of two ports (static pressure ports)
mounted on either side of the helicopter, forward of the pilot and the co-pilot doors,
a single Pitot port mounted on the nose of the aircraft, plastic tubing and analogue
instruments of (1) airspeed indicator (ASI), (2) altimeter and (3) vertical velocity
indicator (VVI) all three located on the instrument panel as shown in Fig. C-4. The
Pitot tube which is a heated type and is mounted approximately four inches to the right
of the helicopter centreline. The static pressure is fed by means of plastic tubing to all
three pneumatic analogue instruments. The total pressure is fed from the Pitot tube to
the airspeed indicator only.

Figure C.3. The horizontal stabilizer. The horizontal stabilizer is designed to generate a
down-loaded aerodynamic force in forward flight.

247
C | RESEARCH HELICOPTERS DESCRIPTION

Figure C.4. The Jet-Ranger flight instruments fed by the Pitot system. The pneumatic
flight instruments that present to the flight crew the airspeed (ASI), pressure altitude (Altimeter)
and the vertical velocity (VVI).

The Jet-Ranger’s important specifications pertaining to its performance are

summarized in Table C.1.

Table C.1 – The Bell Jet Ranger performance specifications.

Parameter Value
M/R (T/R) diameter [ft.] 35.3 (5.2)
M/R (T/R) standard angular speed (RPM) 354 (2,670)
M/R (T/R) number of blades (b) 2 (2)
M/R (T/R) blade chord [ft.] 1.08 (0.43)
M/R (T/R) solidity ratio (𝜎𝑅 ) 0.039 (0.105)
Take-off (continuous) power rating [hp.] 317 (270)
Take-off (continuous) max. TGT [°C] 810 (738)
Max. take-off gross weight [lbs.] 3,200
Never Exceed Airspeed, VNE [KCAS] 120

248
T HE MBB BO-105 H ELICOPTER

The Messerschmitt-Bölkow-Blohm (MBB) BO-105 helicopter used for the

research is a dual-engine light observation and utility helicopter, designed for day and
night, visual flight rules (VFR) and instrument flight rules (IFR) operations. The
helicopter is designed for landing and take-off from prepared or unprepared surfaces
with a skid-type landing gear. The helicopter can be flown by a single pilot from the
right-hand seat and has a place for three passengers in the back. The helicopter has an
overall length of 11.86 m (main rotor fore to aft end of tail) and a maximum allowed
take-off gross weight of 5,512 lbs. (2,500 kg.). The helicopter conforms to the
definition of a ‘conventional’ helicopter, as it has a single main-rotor (M/R) and a
single tail-rotor (T/R). The M/R assembly is a four-bladed, hinge less, rigid type with
an effective flapping offset of 14%. The M/R blades motion in flapping in lead/lagging
was enabled through a flexible blade root. The M/R rotates counter-clockwise, when
viewed from above, at a standard angular speed of 423 RPM. The M/R blades are
constructed of fiberglass reinforced plastic with a NACA 23012 aerofoil and a
geometric pitch angle twist of -8° between the root and the tip. As shown in Fig. C.5,
each one of the four M/R blades is fitted with appended pendulum absorber (‘pendab’)
designed to reduce the 4/rev. vertical vibration transferred to the fuselage. These
pendabs are installed on the roots of the M/R blades, at about 16.2% of the blade
length (M/R radius). The T/R configuration is a two-bladed semi-rigid, teetering with
45° delta-three flapping hinge offset. The tail-rotor is mounted on the left side of the
tail-boom structure, and rotates at a standard 2,219 RPM with the bottom blade
traveling forward. The tail-rotor operates as a ‘pusher’ type, i.e., it generates an anti-
torque force which pushes the tail structure of the helicopter to the right.

249
C | RESEARCH HELICOPTERS DESCRIPTION

Figure C.5. The main-rotor assembly of the BO-105 helicopter. Each one of the blade is
fitted with a pendulum absorber designed to alleviate vertical vibrations.

The BO-105 helicopter is powered by two Allison 250-C20B gas-turbine engines,

each with independent drive-train systems. These engines are very similar to the Bell
Jet-Ranger gas-turbine engine. The maximum (uninstalled) output power of each
engine is rated at 420 shaft horsepower under standard-day sea-level conditions. Due
to drive-train limitations of the helicopter, each one of the 158 lbs. (dry) engines, is de-
rated to a maximum output power of 400 hp (5 minutes take off rating limit) and of
344 hp (continuous operation). Each engine incorporates a compressor section with
six axial stages followed by a centrifugal stage, a gas-turbine section with two axial
stages and a two axial stages power-turbine. This engine configuration is of a reverse-
flow annular type, for which the exhaust gases follow a reverse path to the compressor
gasses, before passing through the turbine section

The dual flight controls of the helicopter (for the pilot and for another crew
member) consist of mechanical type non-reversible cyclic-stick, collective-stick, and
reversible directional pedals to control the helicopter in yaw. The helicopter is
equipped with a dual-redundant, 103 bar hydraulic power supply system that is used

250
 T H E MBB BO - 1 05 H E L I C O P T E R

to boost the cyclic-stick and the collective stick controls. The T/R pedals are not
hydraulically boosted. The helicopter features a trim-system designed to provide
artificial cyclic-stick force gradient and to reduce opposing control forces to zero. This
is activated by a four-way switch located on top of the cyclic stick grip.

The airframe consists of the fuselage, the cabin and the cargo compartment. The
fuselage is a conventional “semi-monocoque” structure. The floor of the helicopter
runs through both the cabin and the cargo compartment at the same level. The engines
deck forms the roof of the cargo compartment and acts as a firewall for the engines.
The engines deck also provides mounting for the main transmission, the two engines
and the dual-redundant hydraulic boost system. The two cabin doors are sliding doors
which allow them to be opened during flight. As shown in Fig. C.6, access to the cargo
compartment is via two clam-shell doors, opening sideways at the rear end of the
fuselage. The aft fuselage incorporates a spoiler designed to assist in increasing stability
by imposing a flow separation over the clamshell doors.

Figure C.6. The rear end of the BO-105 fuselage. Access to the cargo compartment in via
two clamshell doors.

251
C | RESEARCH HELICOPTERS DESCRIPTION

The BO-105 helicopter incorporates a horizontal stabilizer with vertical endplates

installed symmetrically on both sides of the tail boom as shown in Fig. C.7. The
horizontal stabilizer has a span of 2.5 m with an aspect ratio of 5.2 and extends beyond
the maximum width of the fuselage.

Figure C.7. The tail section of the BO-105 helicopter. The 2.5 m span horizontal stabilizer
extends beyond the fuselage width.

The pitot-static system is used for measuring airspeed, pressure-altitude and

vertical rate of climb and descent. This system is made-up of two static ports mounted
on either side of the helicopter, below the pilot and the co-pilot doors, a single Pitot
port mounted in proximity and below the right static port, plastic tubing and analogue
instruments of airspeed indicator, altimeter and vertical velocity indicator located on
the instrument panel (see Fig. C.8). The BO-105 helicopter basic specifications
pertaining to its performance are summarized in Table C.2.

252
 T H E MBB BO - 1 05 H E L I C O P T E R

Table C.2 – The MBB BO-105 performance specifications.

Parameter Value
M/R (T/R) diameter [ft.] 32.3 (6.28)
M/R (T/R) standard angular speed (RPM) 423 (2,219)
M/R (T/R) number of blades (b) 4 (2)
M/R (T/R) blade chord [ft.] 0.86 (0.59)
M/R (T/R) solidity ratio (𝜎𝑅 ) 0.07 (0.12)
Take-off (continuous) power rating [hp.] 800 (688)
Take-off (continuous) max. TGT [°C] 810 (738)
Max. take-off gross weight [lbs.] 5,512
Never Exceed Airspeed, VNE [KCAS] 145

Figure C.8. The BO-105 Instrument Panel. The pneumatic flight instruments are located first
from the right (pressure altitude on the top row and vertical rate of climb/descent on the
bottom). The airspeed indicator is located on the top row, third from the right.

253
REFERENCES

R EFERENCES

[1] Pavlok, K. M., Flight Test Engineering, Edwards California: NASA Dryden
[Armstrong] Flight Research Center, NASA Report #DRFC-E-DAA-TN11035,
September 19, 2013

[2] Knoth, Florian, and Christian Breitsamter. “Aerodynamic Testing of Helicopter

Side Intake Retrofit Modifications.” Aerospace 4, no. 3 (2017): 33–33.

https://doi.org/10.3390/aerospace4030033.

[3] M. Nowakowski. 2011. “Flight Tests of Upgraded Helicopters.” Journal of Kones 18

(4): 317–24.

[4] Flemming, Robert J, Kimberly W Hanks, M. Lynn Hanks, 2007 SAE Aircraft and

Engine Icing International Conference Seville, Spain 2007-09-24. “US Army UH-60M

Helicopter Main Rotor Ice Protection System.” SAE Technical Paper (20070924) (2007).

https://doi.org/10.4271/2007-01-3301

[5] M. Wojtas, Ł. Czajkowski, and A. Sobieszek. “The Influence of the Blades Leading

Edge Anti-Erosion Protection on Main Rotor Performances.” Journal of Kones 25, no.

2 (2018). https://doi.org/10.5604/01.3001.0012.2866.

[6] Buckanin, R.M. et. al., ‘Level Flight Performance Evaluation of the UH-60A

Helicopter with the Production External Stores Support System and Ferry Tanks

254
 REFERENCES

Installed’, US Army Aviation Engineering Flight Activity, Project 86-01, September

1986.

[7] (ASA), Federal Aviation Administration (FAA)/Aviation Supplies & Academics.

2016. Pilot's Handbook of Aeronautical Knowledge: FAA-H-8083-25b. FAA Handbooks
Series. Newcastle: Aviation Supplies and Academics.

[8] Prouty RW. Helicopter Performance, Stability, and Control. Boston: PWS Engineering;
1986, Chapter 4, “Performance Analysis”.

[9] Gessow A., Myers GC, Aerodynamics of the Helicopter. Third printing, Frederick
Ungar Publishing Company, New York, 1967, Chapter 4, “Hovering and Vertical
Flight Performance Analyses”.

[10] Leishman JG. Principles of Helicopter Aerodynamics. 2nd ed. Cambridge: Cambridge
University Press; 2006, Chapter 5, “Helicopter Performance”.

[11] Anonymous "Sikorsky Introduces 'Project Firefly(TM)' Electric Helicopter

Demonstrator," PR Newswire, 2010. Available: https://www.proquest.com/wire-
feeds/sikorsky-introduces-project-firefly-tm-electric/docview/613400680/se-
2?accountid=27026.

[12] Moon K, et al. “A Comparative Statistical Analysis of Global Trends in Civil

Helicopter Accidents in the U.S., the EU, and the CIS.” Iop Conference Series: Materials
Science and Engineering, vol. 868, no. 1, 2020, doi:10.1088/1757-899X/868/1/012020.

[13] Turczeniuk, Bohdan. 1982. “Exhaust Gas Reingestion Measurements.” Journal of

the American Helicopter Society 27 (3): 4–10. https://doi.org/10.4050/JAHS.27.4.

[14] Jackson, Michael E, and Richard L House. 1982. “Exhaust Gas Reingestion during
Hover in‐Ground‐Effect.” Journal of the American Helicopter Society 27 (3): 74–79.
https://doi.org/10.4050/JAHS.27.3.74.

255
REFERENCES

[15] Taslim, M. E, and S Spring. 2010. “A Numerical Study of Sand Particle

Distribution, Density, and Shape Effects on the Scavenge Efficiency of Engine Inlet

Particle Separator Systems.” Journal of the American Helicopter Society 55 (2): 22006–

220069. https://doi.org/10.4050/JAHS.55.022006.

[16] M. Paszko. 2017. “Infrared Signature Suppression Systems in Modern Military

Helicopters.” Prace Instytutu Lotnictwa 3 (248).

[17] Nidhi Baranwal, and Shripad P. Mahulikar. “Review of Infrared Signature

Suppression Systems Using Optical Blocking Method.” Defence Technology 15, no. 3:

432–39. Accessed July 31, 2022.

[18] Cooke AK, and Fitzpatrick EWH., Helicopter Test and Evaluation, 1st ed. AIAA

Education Series, Wright Patterson Air Force Base, Ohio, USA, 2002.

[19] National Test Pilot School, Professional Course Textbook Series, Chapter 5, Vol.

VII, Rotary Wing Performance Flight Testing, Mojave, 2017.

[20] U.S. Naval Test Pilot School, Flight Test Manual No. 106, Rotary Wing

Performance, Naval Air Warfare Center, Patuxent River, Maryland, USA, 1996.

[21] US Army Material Command (AMC), Engineering Design Handbook, Helicopter

Performance Testing, AMCP 706-204, August 1974.

[22] Leishman, J. G., Principles of Helicopter Aerodynamics, 2nd ed., Cambridge University

Press, 2006, Chapter 1, “Introduction- A History of Helicopter Flight”.

[23] Gessow A, and Myers G., Aerodynamics of the Helicopter, Third printing, Frederick

Ungar Publishing Company, New York, 1967, Chapter 1, “The Development of

Rotating-Wing Aircraft”.

256
 REFERENCES

[24] Richards RB, Naval Test Pilot School Textbook, USNTPS-T-No.1, Principles of

Helicopter Performance, Naval Air Test Center, Patuxent River, Maryland, USA, 1968.

[25] Lewicki D. G., Coy J. J., “Helicopter Transmission Testing at NASA Lewis

Research Center”, NASA Technical Memorandum 89912, US Army Aviation

Research and Technology Activity, AVSCOM Technical Report 87-C-10, Lewis

Research Center, Cleveland, Ohio, June 1987.

[26] Coy J. J., Townsend D. P. et al., “Helicopter Transmission Testing at NASA Lewis

Research Center”, NASA Technical Memorandum 100962, US Army Aviation

Research and Technology Activity, AVSCOM Technical Report 88-C-003, Lewis

Research Center, Cleveland, Ohio, November 1988.

[27] Rankine, W. J. M. 1865. “On the Mechanical Principles of the Action of

Propellers”, Transactions of the institute of Naval Architects, 6, pp. 13-39.

[28] Froude, W. 1878. “On the Elementary Relation between Pitch, Slip and Propulsive

Efficiency”, Transactions of the institute of Naval Architects, 19, pp. 47-57.

[29] Glauert, H. 1935. “Airplane Propellers”, in division L of Aerodynamic Theory,

edited by W. F. Durand, Springer Verlag, Berlin Germany. Reprinted by Peter Smith,

Glouster, MA, 1976.

[30] Prandtl, L., “Applications of Modern Hydrodynamics to Aeronautics”, NACA

116, June 1921.

[31] Gessow, Alfred, Langley Aeronautical Laboratory, and United States. National

Advisory Committee for Aeronautics. 1948. Effect of Rotor-Blade Twist and Plan-

Form Taper on Helicopter Hovering Performance. National Advisory Committee for

257
REFERENCES

Aeronautics Technical Note, 1542. Washington, D.C.: National Advisory Committee

for Aeronautics. https://digital.library.unt.edu/ark:/67531/metadc54535.

[32] Prouty RW. Helicopter Performance, Stability, and Control. Boston: PWS Engineering;

1986. Chapter 10, “Preliminary Design”.

[33] Prouty RW. Helicopter Performance, Stability, and Control. Boston: PWS Engineering;

1986. Chapter 1, “Aerodynamics of Hovering Flight”.

[34] Leishman, J. G., Principles of Helicopter Aerodynamics, 2nd ed., Cambridge University

Press, 2006, Chap. 2, “Fundamentals of Rotor Aerodynamics”.

[35] Gessow A, and Myers G., Aerodynamics of the Helicopter, Third printing, Frederick

Ungar Publishing Company, New York, 1967, Chapter 3, “An Introduction to

Hovering Theory”.

[36] Shahmiri, Farid, Maryam Sargolzehi, and Mohammad Ali Shahi Ashtiani.

“Systematic Evaluation of the Helicopter Rotor Blades: Design Variables and

Interactions.” Aircraft Engineering and Aerospace Technology 91, no. 9 (2019): 1223–37.

https://doi.org/10.1108/AEAT-06-2018-0163.

[37] Harris, Franklin D, and Michael A McVeigh. “Uniform Downwash with Rotors

Having a Finite Number of Blades.” Journal of the American Helicopter Society 21, no. 1

(1976): 9–20. https://doi.org/10.4050/JAHS.21.9.

[38] Carpenter, Paul J. Lift and Profile-Drag Characteristics of an Naca 0012 Airfoil Section

As Derived from Measured Helicopter-Rotor Hovering Performance. National Advisory

Committee for Aeronautics Technical Note, 4347. Washington: NACA, 1958.

258
 REFERENCES

[39] Boatwright, “Measurements of Velocity Components in the Wake of a Full-Scale

Helicopter Rotor in Hover”, United States Army Air Mobility Research and

Development Laboratories (AMRDL), Technical Report 72-33, 1972.

[40] Cheeseman, I.C. and Bennett, W.E. The Effect of the Ground on a Helicopter Rotor in

Forward Flight. A.A.E.E. Report Ray./288 3021, Aeronautical Research Council (1955).

[41] Hayden, J.S., The effect of the ground on helicopter hovering power required. In,

proceedings of the American Helicopter Society, 32nd annual forum, 1976.

[42] Kutz, Benjamin M, Manuel Keßler, and Krämer Ewald. “Experimental and

Numerical Examination of a Helicopter Hovering in Ground Effect.” Ceas Aeronautical

Journal: An Official Journal of the Council of European Aerospace Societies 4, no. 4 (2013): 397–

408. https://doi.org/10.1007/s13272-013-0084-x.

[43] Griffiths, Daniel A, Shreyas Ananthan, and J. Gordon Leishman. “Predictions of

Rotor Performance in Ground Effect Using a Free-Vortex Wake Model.” Journal of the

American Helicopter Society 50, no. 4 (2005): 302–14.

https://doi.org/10.4050/1.3092867.

[44] Porterfield, J. D., and Alexander W. T., “Measurements and Evaluation of

Helicopter Flight Loads Spectra Data”, Journal of the American Helicopter Society,

Vol. 15, Number 3, 1 July 1970, pp. 22-34(13).

[45] U.S. Department of Transportation, Federal Aviation Administration,

Certification of Normal Category Rotorcraft, AC 27-1B, Change 3, Subpart C,

Strength Requirements, AC 27.571, September 2008.

[46] Glauert, H., “On the Horizontal Flight of a Helicopter”, Aeronautical Research

Committee, Reports & Memoranda 1157, 1928.

259
REFERENCES

[47] Prouty RW. Helicopter Performance, Stability, and Control. Boston: PWS Engineering;

1986, Chapter 3, “Aerodynamics of Forward Flight”.

[48] Vil'dgrube, L.S, Vozhdayev Y.S, and National Aeronautics and Space

Administration. Vertol Division. 1976. Theory of the Lifting Airscrew. NASA Technical

Translation, F-823. Washington: NASA.

[49] Stepniewski, W.Z., and Keys, C.N., Rotary Wing Aerodynamics, Two Volumes

Bounded as One, Dover Publication Inc., New York, 1984, Chapter 3,”Blade Element

Theory- Forward Flight”.

[50] Leishman JG. Principles of Helicopter Aerodynamics. 2nd ed. Cambridge: Cambridge

University Press; 2006. Chapter 6, “Aerodynamic Design of Helicopters”.

[51] Noonan KW and Bingman GJ, Two-Dimensional Aerodynamic Characteristics

of Several Rotorcraft Airfoils at Mach Numbers from 0.35 to 0.9, NASA TM X-73990,

Langley Research Center, Virginia, January 1977.

[52] McCloud, John L, James C Biggers, Robert H Stroub, and National Aeronautics

and Space Administration. An Investigation of Full-Scale Helicopter Rotors at High Advance

Ratios and Advancing Tip Mach Numbers. Nasa Technical Note, D-4632. Washington:

NASA, 1968

[53] Caradonna, Franck X, Jean-Jacques Philippe, “The Flow Over a Helicopter Blade Tip

in the Transonic Regime”, Proceedings of the 2nd European Rotorcraft and Powered Lift

Aircraft Forum, Buckeburg, Germany, 1976.

[54] Paul, William F. “A Self‐Excited Rotor Blade Oscillation at High Subsonic Mach

Numbers.” Journal of the American Helicopter Society 14, no. 1 (1969): 38–48.

https://doi.org/10.4050/JAHS.14.1.38.

260
 REFERENCES

[55] Abbott, Ira H, and Abert E Von Doenhoff. 1999. Theory of Wing Sections:

Including a Summary of Airfoil Data. Mineola: Dover Publications.

[56] Thwaites, Bryan, ed. Incompressible Aerodynamics: An Account of the Theory

and Observation of the Steady Flow of Incompressible Fluid Past Aerofoils, Wings,

and Other Bodies. Fluid Motion Memoirs. Oxford: Clarendon Press, 1960.

[57] McCroskey, W. J, K. W McAlister, L. W Carr, S. L Pucci, O Lambert, and R. F

Indergrand. “Dynamic Stall on Advanced Airfoil Sections.” Journal of the American

Helicopter Society 26, no. 3 (1981): 40–50. https://doi.org/10.4050/JAHS.26.40.

[58] Carr, Lawrence W, Kenneth W Mcalister, and William J McCroskey. Analysis of

the Development of Dynamic Stall Based on Oscillating Airfoil Experiments. NASA

Technical Note, D-8382. Washington: NASA, 1977.

[59] Martin, J. M, R. W Empey, W. J McCroskey, and F. X Caradonna. “An

Experimental Analysis of Dynamic Stall on an Oscillating Airfoil.” Journal of the

American Helicopter Society 19, no. 1 (1974): 26–32.

https://doi.org/10.4050/JAHS.19.26.

[60] Merz, Christoph B, et al. “New Results in Numerical and Experimental Fluid

Mechanics X : Contributions to the 19th Stab/Dglr Symposium Munich, Germany,

2014.” Experimental Investigation of Dynamic Stall on a Pitching Rotor Blade Tip, Cham:

Springer International Publishing: Springer, 2016, pp. 339–348.

[61] Johnson, Wayne, and Norman D Ham. “On the Mechanism of Dynamic

Stall.” Journal of the American Helicopter Society, vol. 17, no. 4, 1972, pp. 36–45.,

https://doi.org/10.4050/JAHS.17.36.

261
REFERENCES

[62] Dommasch, D.O., Sherby S.S. and Connoly T.F., “Airplane Aerodynamics”,

Fourth Edition, Pitman Publishing Corporation, 1967. Chapter 2, “Fundamental

Dynamics, and Thermodynamics of Air”.

[63] Rosenstein, H., Stanzione, K. 1981. “Computer-Aided Helicopter Design”, 37th

Annual Forum of the American Helicopter Society, Washington D.C., May 17-20.

[64] Sheehy, Thomas W. 1977. “A General Review of Helicopter Rotor Hub Drag

Data.” Journal of the American Helicopter Society 22 (2): pp. 2–10.

https://doi.org/10.4050/JAHS.22.2.2.

[65] Arnold, J.R, and Skinner, G.L. ‘Army Preliminary Evaluation YOH-58A Helicopter

with a Flat-Plate Canopy: Final Report’, US Army Aviation Engineering Flight Activity,

Project 75-20, December 1975.

[66] Buckingham, E., “On Physically Similar Systems; Illustrations of the Use of

Dimensional Equations”, Physical Review, Vol IV, No. 4, 1914, pp. 345 – 376,

doi:10.1103/PhysRev.4.345.

[67] Evans, HJ, “Dimensional Analysis and the Buckingham PI Theorem”, American

Journal of Physics 40, 1815 (1972); doi: 10.1119/1.1987069.

[68] Gratton, J, “Applying Dimensional Analysis to Wave Dispersion”, American

Journal of Physics 75, 158 (2007); doi: 10.1119/1.2372471.

[69] Jensen, JH, “Introducing Fluid Dynamics using Dimensional Analysis”, American

Journal of Physics 81, 688 (2013); doi: 10.1119/1.4813064.

[70] Robinett, RW, “Dimensional Analysis as the other Language of Physics”,

American Journal of Physics 83, 353 (2015); doi: 10.1119/1.4902882.

262
 REFERENCES

[71] G. Maheedhara Reddy, V. Diwakar Reddy, Theoretical Investigations on

Dimensional Analysis of Ball Bearing Parameters by Using Buckingham Pi-Theorem,

Procedia Engineering, Volume 97, 2014, Pages 1305-1311,ISSN 1877-7058,

https://doi.org/10.1016/j.proeng.2014.12.410.

[72] Lawal, Abiodun Ismail, Seun Isaiah Olajuyi, Sangki Kwon, and Moshood Onifade.

2021. “A Comparative Application of the Buckingham Π (Pi) Theorem, White-Box

Ann, Gene Expression Programming, and Multilinear Regression Approaches for

Blast-Induced Ground Vibration Prediction.” Arabian Journal of Geosciences 14 (12).

https://doi.org/10.1007/s12517-021-07391-x.

[73] Knowles, P. The Application of Non-dimensional Methods to the Planning of Helicopter

Performance Flight Trials and Analysis Results, Aeronautical Research Council ARC CP

927, 1967.

[74] Zohuri B, Dimensional Analysis beyond the Pi Theorem, Chap. 1, Cham, Switzerland:

Springer; 2017. doi:10.1007/978-3-319-45726-0.

[75] Jackson ME, House RL. Exhaust gas reingestion during hover in‐ground‐

effect. Journal of the American helicopter society. 1982;27(3):74-79.

doi:10.4050/JAHS.27.3.74.

[76] U.S. Naval Test Pilot School, Flight Test Manual No. 106, Rotary Wing

Performance, Chapter 7, Naval Air Warfare Center, Patuxent River, Maryland, USA,

1996.

[77] Nagata JI, Piotrowski JL, Young CJ, et al. Baseline Performance Verification of

the 12th Year Production UH-60A Black Hawk Helicopter. Final Report, US Army

Aviation Engineering Flight Activity, Edwards AFB, California, USA, January 1989.

263
REFERENCES

[78] Belte D, Stratton MV. Fuel Conservation Evaluation of U.S. Army Helicopters,

Part 4, OH-58C Flight Testing. Final Report, US Army Aviation Engineering Flight

Activity, Edwards AFB, California, USA, August 1982.

[79] Advisory Group for Aerospace Research & Development (AGARD), Flight Test
Techniques Series. Vol 14 on Introduction to Flight Test Engineering. AGARD-AG-300,
September 1995.

[80] Ulbrich, N., Regression Model Optimization for the Analysis of Experimental
Data, AIAA 2009–1344, paper presented at the 47th AIAA Aerospace Sciences
Meeting and Exhibit, Orlando, Florida, January 2009.

[81] Ulbrich, N., Optimization of Regression Models of Experimental Data using

Confirmation Points, AIAA 2010–930, paper presented at the 48th AIAA Aerospace
Sciences Meeting and Exhibit, Orlando, Florida, January 2010.

[82] Zhao, D. and Xue, D., A multi-surrogate approximation method for

metamodeling, Engineering with Computers, Vol. 27, No. 2, pp. 139-153, 2011.

[83] Strang, G., Introduction to Linear Algebra, 5th ed., Wellesley-Cambridge Press,
Wellesley MA, 2016, Chap. 4.

[84] Guttman, I., Wilks, S., and Hunter, J., Introductory Engineering Statistics, 2nd ed.,
John Wiley & Sons, Inc., New York, 1971, Chap. 10, ‘Statistical Tests’.

[85] Singiresu, R., Engineering Optimization Theory and Practice, 4th ed., John Wiley
& Sons, Inc., New Jersey, 2009, Chap. 2.

[86] Benson TP, Buckanin RM, Mittag CF, et al. Evaluation of a OH-58A Helicopter
with an Allison 250 C-20B Engine. Final Report, US Army Aviation Engineering Flight
Activity, Edwards AFB, California, USA, April 1975.

264
 REFERENCES

[87] Kreyszig E. Advanced Engineering Mathematics, 3rd ed., John Wiley & Sons, Inc.,
New York, 1972, Chap. 19.

[88] Strang, G. Introduction to Linear Algebra, 4th ed., Belmont, CA: Thomson
Brooks/Cole, 2006, Chap. 6.

[89] Gomez A, Arantes J, De Andrade D, et al. Helicopter Engine Performance

Determination using Analysis of Variance. In: AHS 71st Annual Forum, Virginia
Beach, Virginia, USA, May 5-7 2015.

[90] MacMillan WL. Development of a modular-type computer program for the

calculation of gas-turbine of design performance, Ph.D. Dissertation, School of
Engineering, Cranfield University, Cranfield, Bedford, 1974.

[91] Goulos I, Giannakakis P, Pachidis V. et al. Mission Performance Simulation of

Integrated Helicopter-Engine Systems using an Aeroelastic Rotor Model, Journal of
Engineering for Gas Turbines and Power, Volume 135, September 2013, Vol. 135.

[92] Heng W, Shufan Z, Jijun Z, et al. Gas Turbine Power Calculation Method of
Turboshaft based on Simulation and Performance Model, MATEC Web of
Conferences 189, 02003, 2018. DOI:10.1051/matecconf/201818902003.

[93] Zhang C, Gummer V. The potential of helicopter turboshaft engines

incorporating highly effective recuperators under various flight conditions. Aerospace
Science and Technology, 6 March 2019. http://doi.org/10.1016/j.ast.2019.03.008.

[94] Savelle SA, Garrard GD. Application of transient and dynamic simulations to
the US Army T55-L-712 helicopter engine. In: International Gas Turbine and
Aeroengine Congress and Exhibition, Birmingham, UK, June 10-13 1996.

[95] Yeo H, Bousman WG and Johnson W. Performance analysis of a utility helicopter

with standard and advanced rotors. In: AHS Aerodynamics, Acoustics and Test and

265
REFERENCES

Evaluation Technical Specialist Meeting, San Francisco, California, USA, January

23 – 25 2002.

[96] Bousman, W. G., “Out-of-Ground Effect Hover Performance of the UH-60A”,

UH-60A Airloads Program Occasional Note 2001-01, US Army Aeroflight dynamics
Directorate (AMCOM) Ames Research Center, February 2001.

[97] Siva, C., Murugan, M. S., Ganguli, R., “Uncertainty Qualification in Helicopter
Performance Using Monte Carlo Simulations”, Journal of Aircraft, Vol. 48, No. 5,
2011, pp. 1503-1511, DOI: 10.2514/1.C000288.

[98] Jacobson, K. E., Smith, M., J., “Carefree Hybrid Methodology for Rotor Hover
Performance Analysis”, Journal of Aircraft, Vol. 55, No. 1, 2018, pp. 52-65, DOI:
10.2514/1.C034112.

[99] Garcia, A. J., Barakos, G., N., “Accurate Predictions of Rotor Hover Performance
at Low and High Disc Loadings”, Journal of Aircraft, Vol. 55, No. 1, 2018, pp. 89-
110, DOI: 10.2514/1.C034144.

[100] Matayoshi, N., Asaka, K., Okuno, Y., “Flight-Test Evaluation of a Helicopter
Airborne Lidar”, Journal of Aircraft, Vol. 44, No. 5, 2007, pp. 1712-1720, DOI:
10.2514/1.28338.

[101] Boirun, B. H., “Generalizing Helicopter Flight Test Performance Data

(GENFLT), The 34th Annual National Forum of the American Helicopter Society,
May 1978.

[102] Abraham, M., Costello, M., “In-Flight Estimation of Helicopter Gross Weight
and Mass Center Location”, Journal of Aircraft, Vol. 46, No. 3, 2009, pp. 1042-1049,
DOI: 10.2514/1.41018.

266
 REFERENCES

[103] Scmitz S., Bhagwat M., Moulton M. A., et al. The Prediction and Validation of
Hover Performance and Detailed Blade Loads, Journal of the American Helicopter
Society 54, 032004, May 2009.

[104] Wang, Q., and Zaho Q., Rotor aerodynamic shape design for improving
performance of an unmanned helicopter. Aerospace Science and Technology, 3 March
2019, DOI: 10.1016/j.ast.2019.03.006.

[105] Peterson, R. L., and Warmbrodt, W., Hover Performance and Dynamics of a
Full-Scale Hingeless Rotor. In: The Tenth European Rotorcraft Forum, The Hague,
The Netherlands, August 1984.

[106] Horn, RA., and Johnson CR., Matrix Analysis, 2nd edition, Cambridge
University Press, New York, NY, 2012, Chap. 5.

267
 ACK NOW LEDGEMENTS

A CKNOWLEDGEMENTS

In April 2010, I started to work at the National Test Pilot School (NTPS) in
Mojave, California as a rotary wing flight test engineer instructor. I would like to thank
Mr. Sean C. Roberts (RIP) for offering me this position and giving me the privilege of
educating and training many remarkable individuals who belong to the international
community of flight-testers. The idea of pursuing a Ph.D. research in the field of
helicopter performance flight testing has been a goal of mine for several years and
became a reality in February of 2018. I would like to thank Dr. Lester Ingham who
together with Dr. Marilena Pavel made this Ph.D. research possible as a collaboration
between the NTPS and TU Delft.

Many scholars have already said before, that you never really know something
until you teach it to someone else. To all my past and future students, thank you for
everything you taught me so far, and for the things you are about to teach me in the
future. Spending valuable time with you in class and in the air fills me with scholastic
satisfaction and a sense of fulfilment. Thanks to you, I have found my own fountain
of youth. A special appreciation goes to test pilot Stefan Hanekom, an NTPS class of
2011 student member who became an instructor colleague and a true friend in 2016.
Thank you Stefan for your support in flying data points during this Ph.D. research. It
has been great working with you and I hope our professional paths merge again in the
future.

I wish to extend my gratitude and appreciation to my two extraordinary

promotors, Dr. Marilena Pavel and Professor Max Mulder. When I started the Ph.D.
research, Marilena resided in California, just a few miles away from me. Frequent
meetings took place in the neighbourhood ‘Starbucks’ where fruitful professional
discussions were accomplished and resolutions were made. The first paper was written
and the frame for the second one was laid down. Those were the early days that built

269
ACK NOW LEDGEMENTS

a strong foundation which allowed for the continuation of the research, following
Marilena’s return to the Netherlands. The in-person meetings were replaced with
coffee-less Skype conferences. Thank you Marilena for your intelligent and
compassionate guidance throughout this research. You have always practiced one of
Newton’s famous quotes: ‘Tact is making a point without making an enemy’. My first
in-person encounter with Professor Max Mulder took place in September 2018, while
I came to present a paper during the European Rotorcraft Forum that took place in
TU Delft. I met an impressive professor who listens to you very carefully, quickly
processes the information, and communicates back in the most efficient manner. Max,
your exceptional ability to say so much in a few words is remarkable. Your ratio of
content to number of words used, can only be surpassed by the lift to drag ratio of a
modern competition glider…Thank you very much for sharing your wisdom with me.
Although this research was done remotely being an external Ph.D. student, Marilena
and Max’s support and guidance have travelled over the Atlantic Ocean and most of
the US continent, all the way to California.

Many thanks are in order to the helpful staff of the Aerospace Engineering
faculty, especially Ms. Bertine Markus, and of the graduate school of TU Delft. You
were always happy to assist and you have made my personal experience as pleasant as
possible.

I was born and raised in Israel, the second child among three. The importance
of good education and striving for excellence in this walk of life, are qualities that both
my parents worked to instil in me. I would like to express my infinite gratitude to my
first ever two teachers, my mother Raymonde and my father Armand of blessed
memory. Much of my motivation to pursue a Ph.D. in this later phase of my life, and
the need to ‘finish the job’ was fuelled by these two greatest supporters. I would also
like to thank my older brother David, who was my first successful Mathematics
teacher. My personal Mathematics ‘light-bulb’ lit up thanks to you and has remained
on ever since.

270
 ACK NOW LEDGEMENTS

I would like to thank the educational institution I consider as my ‘alma mater’.

It is the Israeli Institute of Technology (‘Technion’). This excellent institution has
prepared me well to the many engineering and scientific challenges I have faced to
date. The high standards and knowledgeable professors have made me suffer… but
also made me proud to be a graduate of this prestigious institute. I can’t mention my
aeronautical engineering alma mater without including my flight-testing alma mater.
Much appreciation and many thanks go to the US Naval Test Pilot School (USNTPS)
in Patuxent River, Maryland. I still consider the yearlong rigorous training back in 1993,
as the most challenging and rewarding year of my life.

Last but definitely not least, my upmost appreciation and many thanks go to
my immediate family. This Ph.D. research is dedicated to you. To my son and
daughter, Ofek and Maya, thank you for challenging me with the most fundamental
question of what do I need this Ph.D. for, especially in this advanced phase of my life?
Thank you both for keeping me current with the fundamental and advanced concepts
of Calculus while seeking for help, even in the middle of the night... This helped me
during my research and promoted few ideas. To my wife and life-long partner, Shoshi,
thank you very much for your endless support during this unorthodox quest I have
embarked on. You have always been there to support me pursuing this dream of mine,
although many of our weekends and holidays were consumed by this research. My
dearest wife and friend, this Ph.D. research is yours as much as it is mine. THANK
YOU!

Ilan Arush

Lancaster, California
September 2023

271
 CURRICULUM VITÆ

C URRICULUM V IT Æ

Ilan Arush was born on November 11th, 1966, in the town of Rehovot, Israel.
After graduating with a Bachelor’s in Science in aeronautical engineering from the
Israel institute of technology (“Technion”) in 1989, he joined the Israeli Air Force.

Ilan served in the IAF for 21 years in a wide range of positions and roles. The
majority of his time in the IAF was spent in the flight test center where he planned,
conducted and analysed numerous flight test campaigns on both fixed wing and rotary
wing aircraft. In 1993 he attended a yearlong course at the US naval test pilot school
in Patuxent River, Maryland where he was qualified as a flight test engineer (FTE).
Ilan accumulated over 1200 flight hours in various helicopters and jet fighters in a role
of a FTE. Amongst his various positions within the IAF, he was the head of rotary
wing section in the flight test centre, the program officer for the Sikorsky CH-53
upgrade program and the lead liaison officer for the Sikorsky S70A-55 program. For
the later position he was officially commended by the US Army for exemplary
performance of duties. Ilan holds a Master’s in science in industrial and management
engineering from the Ben-Gurion University in Israel.

Since April 2010, Ilan works as a flight test engineer instructor at the National
Test Pilot School (NTPS) in Mojave, California. He teaches a variety of graduate level
classes in helicopter performance, stability and flying qualities, vibrations, system
command and control, and fundamental topics in Mathematics and Physics. He is also
the NTPS chief academic officer, a position he was appointed in 2018. Ilan has
instructed and trained hundreds of international test pilots and FTEs since 2010.

When not working, Ilan enjoys road and mountain biking, and long hikes. He
is married to Shoshi and the couple has two children; Ofek (born March 1998) and
Maya (born May 2002).

273
 LIST OF PUBLICATIONS

L IST OF P UBLICATIONS

Journal Publications

1. Arush, I., and Pavel, M. D., “Helicopter Gas Turbine Engine Performance
Analysis: A Multivariable Approach”, Proceedings of the Institute of Mechanical
Engineers, Part G: Journal of Aerospace Engineering, Vol. 223, No. 3, March
2019. https://doi.org/10.1177/0954410017741329

2. Arush, I., and Pavel, M. D., and Mulder, M., “A Singular Value Approach in
Helicopter Gas Turbine Engine Flight Testing Analysis”, Proceedings of the
Institute of Mechanical Engineers, Part G: Journal of Aerospace Engineering,
April 2020. https://doi.org/10.1177/0954410020920060

3. Arush I., Pavel M.D., and Mulder M., “A Dimensionality Reduction Approach in
Helicopter Hover Performance Flight Testing”, Journal of the American
Helicopter Society 67, no. 3 (2022): 129–41.
https://doi.org/10.4050/JAHS.67.032010

4. Arush I., Pavel M.D., and Mulder M. “A Dimensionality Reduction Approach in

Helicopter Level Flight Performance Flight Testing”, Journal of the Royal
Aeronautical Society, published as First View 13 July 2023.
https://doi.org/10.1017/aer.2023.57

Conference Publications

1. Arush, I., “Lateral Center of Gravity Envelope Development for the UH-1N
Helicopter”, 39th Israel Annual Conference on Aerospace Sciences, Israel, March
1999.

2. Arush, I., & Pavel, M.D., “Flight testing and analysis of gas turbine engine
performance: A multivariable approach.” In C. Hermans (Ed.), Proceedings of the
44th European Rotorcraft Forum: Delft, The Netherlands, September 2018.

275