Experimental Investigations into
High-Altitude Relight
of a Gas Turbine
Robert William Read
Homerton College
University of Cambridge
This dissertation is submitted for
the degree of Doctor of Philosophy
2008
To my mother and father
Declaration
I hereby declare that this dissertation is the result of my own work and contains
nothing that is the outcome of work done in collaboration except as declared in the
preface and specified in the text. Furthermore, this dissertation is not substantially
the same as any that I have submitted or will be submitting for a degree, diploma,
or other qualification at this or any other University. The word and figure limit
prescribed by the Engineering Degree Committee has not been exceeded.
Robert William Read
January 3, 2009
Abstract
This thesis describes experiments to investigate high-altitude relight of a lean direct
injection (LDI) combustor. The features that make LDI technology less polluting in
terms of NOx compared to conventional combustors are expected to impede relight
performance. Therefore an improved understanding of ignition behaviour is required
to ensure that stringent relight requirements can be satisfied.
Realistic operating conditions are simulated in a ground-based test facility. The
application of laser diagnostics presents particular difficulties due to the large quan-
tities of liquid fuel that impinge on the combustor walls during relight. Advances are
made in the application of planar laser-induced fluorescence (PLIF) to monitor fuel
placement in a combustor under these conditions. A novel apparatus is developed
to deliver a laser sheet to the combustion chamber while protecting all optical sur-
faces from contamination. The PLIF images are compared with the cold flow field
obtained from CFD modelling. These results indicate that fuel becomes trapped
inside the central recirculation zone in high concentrations.
High-speed flame imaging performed simultaneously with the PLIF measure-
ments provides important insights into the motion and breakup of flame during
relight. An algorithm developed to track the flame activity reveals that the initial
spark kernel is convected downstream, before breaking apart and moving upstream
towards a recovery origin close to the fuel injector. Analysis of many ignition events
has revealed several distinct modes of ignition failure.
Keywords: altitude relight, planar laser-induced fluorescence (PLIF), gas turbine,
lean direct injection, spark ignition.
Acknowledgements
I would like to thank my supervisor, Simone Hochgreb, for her support and guidance.
Thanks also to my adviser, Nondas Mastorakos, particularly for the generous loan
of his high-speed camera.
I owe a large debt of gratitude to my colleague and friend Jim Rogerson whose
technical contribution, most notably to the development of the flame tracking algo-
rithm, has generated key insights into the relight process. His editorial suggestions
have improved the quality of this dissertation significantly, and his gentle encour-
agement has been much appreciated.
This work has been supported by Rolls-Royce plc and I would like to express
my gratitude to the engineers at Derby, who have provided considerable technical
assistance. My particular thanks go to Ken Young, Peter Schober, Mike Whiteman,
Kevin Ronan, and Steve Harding. Chris Goddard and Marco Zedda generated
valuable CFD data, and discussions with Arthur Rowe provided insight into gas
turbine performance at altitude relight conditions. I am also grateful to the technical
staff at the Strategic Research Centre, particularly Allan Collard and Peter Athey,
who offered extensive help during experimental testing.
I thank the members of the Heat Gallery for providing the most stimulating
of intellectual and social environments. Special thanks are due to Bryn Jones for
sharing his knowledge of gas turbine design with both patience and alacrity, and
to James Dawson, who provided many fruitful research ideas. The assistance of
the workshop staff in the Engineering Department was also much appreciated with
special thanks to Alistair Ross, Neil Houghton, and Davor Dukic for their expert
design advice and uplifting good humour.
My mother and father have provided a continual source of support during my
life and I thank them for giving me the curiosity to embark on a PhD, and the
determination to see it through. My final thanks to Kathryn for her limitless love
and encouragement.
Contents
List of tables ix
List of figures xi
Nomenclature xiv
1 Introduction 1
2 Background 6
2.1 The combustion system . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Altitude restart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Restart conditions . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.2 The restart process . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.3 Design for altitude restart . . . . . . . . . . . . . . . . . . . . 19
2.3 Review of ignition and flame stabilisation . . . . . . . . . . . . . . . . 23
2.3.1 Paradigms of ignition research . . . . . . . . . . . . . . . . . . 24
2.4 Fundamental experiments to study ignition . . . . . . . . . . . . . . . 33
2.4.1 Ignition of gaseous blends . . . . . . . . . . . . . . . . . . . . 34
2.4.2 Quiescent droplet suspensions . . . . . . . . . . . . . . . . . . 41
2.4.3 Flowing heterogeneous mixtures . . . . . . . . . . . . . . . . . 47
2.4.4 Laboratory-scale relight experiments . . . . . . . . . . . . . . 51
v
Contents vi
3 Experimental Method 55
3.1 The altitude test facility . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.1.1 Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.1.2 Operating condition set points . . . . . . . . . . . . . . . . . . 58
3.2 PLIF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.2.2 Beam delivery configuration . . . . . . . . . . . . . . . . . . . 62
3.2.3 Optical table . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.2.4 Periscope and sheet expansion assembly . . . . . . . . . . . . 66
3.2.5 Air purges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.3 High-speed flame imaging . . . . . . . . . . . . . . . . . . . . . . . . 79
3.3.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.3.2 Viewing angle . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.4 Timing and data acquisition . . . . . . . . . . . . . . . . . . . . . . . 83
3.4.1 General setup . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.4.2 Cold fuel imaging . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.4.3 Fuel and flame imaging at unsafe condition . . . . . . . . . . . 87
3.4.4 Fuel and flame imaging at safe condition . . . . . . . . . . . . 89
3.4.5 Test numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4 Results I: Preliminary Analysis 91
4.1 Operating conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.2 Properties of laser beam and sheet . . . . . . . . . . . . . . . . . . . 92
4.2.1 Dye fluorescence . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2.2 Slit traverse . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.3 Characteristic times . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.3.1 Recovery and failure times . . . . . . . . . . . . . . . . . . . . 96
4.3.2 Classification of PLIF images . . . . . . . . . . . . . . . . . . 99
4.4 Spark effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Contents vii
4.4.1 Measured spark parameters . . . . . . . . . . . . . . . . . . . 102
4.4.2 Interaction between successive sparks . . . . . . . . . . . . . . 105
5 Results II: Fuel Imaging 109
5.1 Analysis of uncorrected PLIF images . . . . . . . . . . . . . . . . . . 109
5.1.1 Individual cold images . . . . . . . . . . . . . . . . . . . . . . 109
5.1.2 Mean cold images . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.1.3 Influence of flame . . . . . . . . . . . . . . . . . . . . . . . . . 120
5.2 PLIF image correction . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.2.1 Scattering and reflections . . . . . . . . . . . . . . . . . . . . . 126
5.2.2 Sheet thickness . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.2.3 Frontal area . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.2.4 Beam energy distribution . . . . . . . . . . . . . . . . . . . . . 131
5.2.5 Interpretation of corrected images . . . . . . . . . . . . . . . . 136
5.3 CFD comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6 Results III: Flame Imaging 141
6.1 Flame tracking software . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.1.1 Search principle . . . . . . . . . . . . . . . . . . . . . . . . . . 142
6.1.2 Algorithm features . . . . . . . . . . . . . . . . . . . . . . . . 143
6.2 Developing flame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
6.2.1 Flame trajectories . . . . . . . . . . . . . . . . . . . . . . . . . 149
6.2.2 Flame size distributions . . . . . . . . . . . . . . . . . . . . . 155
6.2.3 Spark kernel velocities . . . . . . . . . . . . . . . . . . . . . . 159
6.3 Successful ignition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.3.1 Flame velocities . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.3.2 Recovery origin . . . . . . . . . . . . . . . . . . . . . . . . . . 165
6.3.3 Frequency content . . . . . . . . . . . . . . . . . . . . . . . . 168
6.4 CFD comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
Contents viii
7 Conclusion 173
7.1 Operation of test facility . . . . . . . . . . . . . . . . . . . . . . . . . 173
7.1.1 Apparatus modifications . . . . . . . . . . . . . . . . . . . . . 174
7.1.2 Apparatus characterisation . . . . . . . . . . . . . . . . . . . . 175
7.2 Main findings of fuel and flame imaging . . . . . . . . . . . . . . . . . 176
7.3 Recommendations for future work . . . . . . . . . . . . . . . . . . . . 178
7.3.1 Measures to improve LDI relight performance . . . . . . . . . 178
7.3.2 Further investigations . . . . . . . . . . . . . . . . . . . . . . . 179
Appendices 180
A Viewing Angle Corrections 181
B Geometry of Laser Sheet 185
B.1 Lens calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
B.2 Frontal area calculation . . . . . . . . . . . . . . . . . . . . . . . . . . 188
References 191
List of Tables
2.1 Experiments investigating ignition of gaseous mixtures . . . . . . . . 35
2.2 Experiments investigating ignition of quiescent aerosols . . . . . . . . 36
2.3 Experiments investigating ignition of flowing droplet mixtures . . . . 37
3.1 Safe and unsafe condition set points for relight testing . . . . . . . . . 59
3.2 Beam dimensions at principal point of each lens . . . . . . . . . . . . 71
3.3 Number of tests recorded with high-speed flame camera . . . . . . . . 90
3.4 Number of tests recorded with PLIF camera . . . . . . . . . . . . . . 90
4.1 Experimental operating conditions during ignition testing . . . . . . . 92
4.2 Classification of PLIF images . . . . . . . . . . . . . . . . . . . . . . 100
4.3 Mean measured spark parameters . . . . . . . . . . . . . . . . . . . . 103
4.4 Standard deviation of measured spark parameters . . . . . . . . . . . 103
5.1 Region A PLIF intensity statistics . . . . . . . . . . . . . . . . . . . . 118
5.2 Region B PLIF intensity statistics . . . . . . . . . . . . . . . . . . . . 127
6.1 Ignition activity in three regions of the combustor . . . . . . . . . . . 153
6.2 Flame-centroid and CFD velocities at safe condition . . . . . . . . . . 159
6.3 Flame-centroid and CFD velocities at unsafe condition . . . . . . . . 159
A.1 Calibration target vertex coordinates . . . . . . . . . . . . . . . . . . 184
A.2 Viewing angle transformation matrices and scaling factors . . . . . . 184
ix
List of Tables x
B.1 Dimensions and focal distances of sheet-formation lenses . . . . . . . 187
List of Figures
2.1 Aviation gas turbine combustor concepts . . . . . . . . . . . . . . . . 8
2.2 The standard atmosphere . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Combustor entry conditions . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Ignition and stability loops . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5 Temperature profiles after point energy deposition . . . . . . . . . . . 26
3.1 Photograph of experimental apparatus . . . . . . . . . . . . . . . . . 56
3.2 The modified test section . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3 Fluorescence spectrum of Jet A fuel . . . . . . . . . . . . . . . . . . . 61
3.4 The optical table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.5 Laser beam path through modified test section . . . . . . . . . . . . . 67
3.6 The periscope assembly . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.7 Excitation volume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.8 The combustor window purge system . . . . . . . . . . . . . . . . . . 74
3.9 Air purge characteristic curves . . . . . . . . . . . . . . . . . . . . . . 77
3.10 The periscope and sheet expansion purge systems . . . . . . . . . . . 78
3.11 Spectral sensitivity of high-speed camera . . . . . . . . . . . . . . . . 81
3.12 Plan view of test section including cameras . . . . . . . . . . . . . . . 82
3.13 Timing and data acquisition hardware . . . . . . . . . . . . . . . . . 85
3.14 Timing of data acquisition . . . . . . . . . . . . . . . . . . . . . . . . 86
4.1 Dye fluorescence measurements of beam energy profile . . . . . . . . . 94
xi
List of Figures xii
4.2 Slit traverse measurements of beam and sheet energy profiles . . . . . 96
4.3 Definition of recovery and failure times . . . . . . . . . . . . . . . . . 98
4.4 Cumulative frequency plot of recovery and failure times . . . . . . . . 99
4.5 Image of spark discharge . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.6 Histograms of measured spark parameters . . . . . . . . . . . . . . . 104
4.7 Distributions of successful spark number . . . . . . . . . . . . . . . . 106
5.1 Uncorrected, individual, safe, cold PLIF image . . . . . . . . . . . . . 110
5.2 Uncorrected, individual, unsafe, cold PLIF image . . . . . . . . . . . 111
5.3 Cold PLIF details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.4 Cold PLIF details with logarithmic scale . . . . . . . . . . . . . . . . 113
5.5 High speed images of cold flow . . . . . . . . . . . . . . . . . . . . . . 114
5.6 Uncorrected, mean, cold PLIF images . . . . . . . . . . . . . . . . . . 116
5.7 Uncorrected, mean, cold PLIF image with logarithmic scale . . . . . . 117
5.8 Histograms of cold PLIF signal in region A . . . . . . . . . . . . . . . 119
5.9 Scatter plot of PLIF intensity versus flame intensity . . . . . . . . . . 120
5.10 Simultaneous images of fuel and flame . . . . . . . . . . . . . . . . . 122
5.11 Simultaneous images of fuel and high-intensity flame . . . . . . . . . 123
5.12 Fuel droplets following flame stabilisation . . . . . . . . . . . . . . . . 124
5.13 Uncorrected, individual, cold PLIF image with logarithmic scale . . . 127
5.14 Histograms of cold PLIF signal in region B . . . . . . . . . . . . . . . 128
5.15 Radial correction effects . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.16 Comparison of beam energy profile and fluorescence signal . . . . . . 134
5.17 Angular correction effects . . . . . . . . . . . . . . . . . . . . . . . . 135
5.18 Corrected, mean, cold PLIF images . . . . . . . . . . . . . . . . . . . 137
5.19 Time-averaged velocity in sheet plane predicted by CFD . . . . . . . 139
6.1 Flame tracking process . . . . . . . . . . . . . . . . . . . . . . . . . . 142
6.2 Kernel search process . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
6.3 Expanded search process . . . . . . . . . . . . . . . . . . . . . . . . . 146
List of Figures xiii
6.4 Early high-speed images and corresponding kernel outlines . . . . . . 147
6.5 Trajectory plot example . . . . . . . . . . . . . . . . . . . . . . . . . 148
6.6 Safe and unsafe flame trajectory plots . . . . . . . . . . . . . . . . . . 150
6.7 Trajectory plots of safe failures . . . . . . . . . . . . . . . . . . . . . 152
6.8 Flame detection times in fuel-injector region . . . . . . . . . . . . . . 154
6.9 Points of last activity during failed ignition events . . . . . . . . . . . 155
6.10 Variation in imaged flame area with time . . . . . . . . . . . . . . . . 156
6.11 Distributions of measured kernel size . . . . . . . . . . . . . . . . . . 158
6.12 Spark stream-wise velocity at safe condition . . . . . . . . . . . . . . 161
6.13 Flame velocity development maps . . . . . . . . . . . . . . . . . . . . 163
6.14 Average flame velocity development map . . . . . . . . . . . . . . . . 164
6.15 Flame intensity variations during successful ignition . . . . . . . . . . 165
6.16 Kernel number density map example . . . . . . . . . . . . . . . . . . 166
6.17 Kernel number density maps . . . . . . . . . . . . . . . . . . . . . . . 167
6.18 Stabilised flame oscillations . . . . . . . . . . . . . . . . . . . . . . . 169
6.19 Spectral analysis of measured flame intensity signal . . . . . . . . . . 170
6.20 Phase shift of stabilised flame oscillations relative to spark . . . . . . 171
6.21 CFD streamlines passing below igniter tip . . . . . . . . . . . . . . . 172
A.1 Rotated square . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
B.1 Sheet-formation optics . . . . . . . . . . . . . . . . . . . . . . . . . . 186
B.2 Laser sheet geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
Nomenclature
Terms
safe test condition conducive to successful ignition
}
see Table 3.1
unsafe test condition that prevents successful ignition
Abbreviations
AFR air-to-fuel ratio
ATF altitude test facility
CRZ central recirculation zone
DNS direct numerical simulation
FAR fuel-to-air ratio
LDI lean direct injection
LES large-eddy simulation
PAH polycyclic aromatic hydrocarbons
PLIF planar laser-induced fluorescence
RANS Reynolds-averaged Navier-Stokes
SMD Sauter mean diameter
xiv
Chapter 1
Introduction
Air safety regulations demand that gas turbine manufacturers provide a means of
restarting an engine at high altitude in the event of a flame-out (Federal Aviation
Adminisatration 2000; European Aviation Safety Agency 2003). Future-generation
low-NOx combustors currently in development create a high-velocity, fuel-lean zone
close to the spark igniter that has the potential to inhibit restart performance. Little
research has been conducted to investigate this problem. The experimental study
described here aims to increase understanding of ignition and flame stabilisation in-
side a lean direct injection (LDI) combustor operating at altitude restart conditions.
To understand the circumstances which have led to the development of the LDI
concept, let us consider the three factors that dominate the design of modern avia-
tion gas turbines: fuel consumption, noise, and emissions. In today’s engines, fuel
consumption and noise are largely controlled by overall engine design parameters
such as the pressure ratio, turbine inlet temperature, and bypass ratio. They are also
influenced by the detailed design of the compressor and turbine. The combustion
system liberates virtually all of the heat present in the fuel and generates com-
paratively little noise directly, but is the major factor governing engine emissions.
Hence the combustor design is heavily influenced by regulations governing emissions
of carbon monoxide (CO), unburned hydrocarbons (UHC), smoke, and nitrogen
1
1. Introduction 2
oxides (NOx) emitted during a standard landing-takeoff cycle (International Civil
Aviation Organisation 2006).
For conventional combustion systems, the reduction of NOx emissions presents
the greatest challenge. Oxides of nitrogen are toxic to human health when emitted
at ground level (Kampa and Castanas 2008), and react to increase ozone levels in the
upper troposphere and lower stratosphere when deposited at altitude (Penner et al.
1999). Ozone is a greenhouse gas that also leads to the generation of photochemical
smog in the lower atmosphere. As a consequence of these effects, nitrogen oxide
emissions have been regulated with increasing stringency during recent years, and
ambitious goals have been set for further reductions in the future (Advisory Council
for Aeronautics Research in Europe 2002). While NOx refers to both nitric oxide
(NO) and nitrogen dioxide (NO2), NO produced via the extended Zeldovich or
thermal mechanism constitutes the majority of the nitrogen oxides generated in a
non-premixed combustion system (Glassman 1996). The formation rate of thermal
NO increases rapidly when the flame temperature exceeds approximately 1800 K
(Turns 1996).
Today’s production engines restart readily at altitudes of up to 30000 ft (9.1 km)
above sea level using conventional combustor designs, thanks to the stoichiometric,
low-velocity conditions generated in the region between the fuel injector and the
spark igniter. However, the high temperatures associated with stoichiometric com-
bustion generate NOx in concentrations that cannot be reduced below a certain
level using conventional combustor technology. Changes in stoichiometry may limit
formation of NO and NO2, but the associated reduction in flame temperature tends
to increase production of CO. The central objective of all low-emissions technology
is therefore to maintain the combustion zone temperature within a narrow band
between approximately 1670 K and 1900 K across the full engine power range.
A variety of alternative low-NOx combustor design concepts are presently being
evaluated for aeronautical use. The lean premixed prevapourised (LPP) combustor
1. Introduction 3
is a promising concept for low-NOx combustion of liquid fuels (Canepa et al. 2006;
Allouis et al. 2008). The principle of operation is to create a fully homogeneous
fuel-air mixture prior to combustion with an equivalence ratio close to the lean
flammability limit (φ ≈ 0.6). The comparatively low flame temperature that results
produces a low concentration of NOx that does not increase with residence time. The
thorough premixing of fuel and air coupled with extended residence times ensures
that emissions of CO and UHC are also low. However, the principal difficulty with
this approach is that the time required for vapourisation and mixing in the fuel
preparation duct can exceed the autoignition delay time at high power (Brandt
et al. 1997). The resulting migration of flame poses a safety hazard. Furthermore,
LPP systems are prone to damaging thermoacoustic instabilities that result from
a feedback between pressure and heat release (Dowling and Hubbard 2000). These
concerns have so far limited the use of LPP combustors to ground-based gas turbines.
The rich-burn/quick-quench/lean-burn (RQL) combustor is a low-NOx design
concept that is also being developed (Anacleto et al. 1996; Blomeyer et al. 1999).
A fuel-rich primary combustion zone is created where NOx formation is limited by
the low reaction temperatures and a lack of oxygen. The additional air required
to ensure full oxidation and to reduce the exit gas temperature is injected into a
secondary zone in a manner that ensures rapid and uniform mixing with the primary
zone gases. Compared to a conventional design, the RQL combustor has inherently
stronger ignition and lean blowout performance. However, improvements in the
effectiveness of the secondary mixing are required to deliver the full benefits of this
concept (Lefebvre 1995).
The principle of modifying the equivalence ratio in two or more combustion zones
according to the engine power setting has been investigated more generally. A vari-
able geometry arrangement can be used to adjust the distribution of air between
the primary and dilution zones of a conventional combustor to maintain a consistent
primary zone flame temperature (Gupta et al. 1991; Li and Hales 2003). Alterna-
1. Introduction 4
tively, the added weight and complexity of a variable geometry mechanism may be
avoided by using staged combustion systems such as lean direct injection. An ex-
ample of a LDI design is the low-NOx combustion system developed by Rolls-Royce
plc as part of the ANTLE (advanced near-term low-emissions) programme. In this
combustor, a central pilot atomiser creates a comparatively rich mixture close to the
combustor centre-line, providing low-power stability. The additional fuel required
for high-power operation is introduced through a larger diameter main atomiser,
mounted concentrically with the pilot. At take-off and cruise, a lean flame is cre-
ated around the pilot by admitting a large proportion of the engine air-flow through
the fuel injector. At the pilot-only altitude restart condition, a comparatively high-
velocity, fuel lean region is thus generated between the igniter and the fuel injection
point. These features hinder the processes of ignition, flame propagation and flame
stabilisation required to successfully restart the engine.
This thesis describes an investigation into ignition and flame stabilisation inside
a LDI combustor sector operating under altitude restart conditions. Chapter 2
presents background information concerning the altitude restart requirement and its
context in the overall engine design process. A survey of fundamental combustion
research describing ignition of gaseous and droplet mixtures and flame development
in simplified flow geometries is also included.
Experimental testing was conducted using the altitude test facility (ATF) at
the Rolls-Royce Strategic Research Centre, Derby, UK. The fuel and flame distri-
butions inside the combustion chamber were recorded using kerosene planar laser-
induced fluorescence (PLIF) and high-speed flame emission imaging respectively, as
described in Chapter 3. Also presented in this chapter are novel measures to enable
high-resolution imaging at the restart condition, where the presence of liquid fuel
impinging on optical surfaces makes PLIF unusually difficult.
Having established the experimental apparatus, preliminary results are used in
Chapter 4 to characterise the operational variability of the facility including such
1. Introduction 5
factors as flow rates, spark performance, and laser energy. The results of the PLIF
measurements are analysed in detail in Chapter 5, where the fuel placement prior
to the spark is identified, and the effect of flame on the PLIF signal is also assessed.
Chapter 6 is concerned with interpretation of the high-speed flame recordings. The
development of image processing software to analyse the motion and breakup of
flame is described in some depth. The findings from the simultaneous PLIF and
flame imaging experiments are summarised in Chapter 7, where recommendations
for further work are also given.
Chapter 2
Background
2.1 The combustion system
The role of the combustion system in a gas turbine is to heat the compressor outlet
air before it flows to the turbine. This is invariably achieved in aviation applications
by generating a kerosene-fuelled flame that releases stored chemical energy as it
burns. To maximise the efficiency of the gas turbine cycle, the fuel must be fully ox-
idised whilst ensuring that the air-flow through the combustion chamber experiences
the smallest possible stagnation pressure drop.
This task is complicated, however, by various additional design requirements.
Ignition of the fuel-air mixture must occur reliably, both on the ground during
start-up and at high altitude following flame extinction. Once established, the flame
must remain stable and free from combustion-induced oscillations when subjected to
the wide-ranging inlet conditions and fuel-to-air ratios that occur during different
phases of the flight cycle. A uniformly distributed hot gas must be delivered to
the turbine, containing minimal concentrations of CO, UHC, smoke, and NOx. In
addition, the combustion system as a whole must be small, light, durable, and simple
to manufacture and maintain. This set of conflicting requirements constitutes a
formidable design challenge that requires a large degree of technical compromise.
6
2. Background 7
Figure 2.1(a) illustrates the concept most widely employed to satisfy these de-
sign requirements in production aviation gas turbines (Nicholas 2005). Air leaves
the compressor with a velocity of approximately 150 m/s. Since the pressure drop
associated with combustion is proportional to the square of the air velocity, an inlet
diffuser slows the flow velocity to around 100 m/s. To further reduce the velocity
to a level that can sustain a hydrocarbon flame, less than half of the air-flow is
introduced into an upstream region of the flame tube known as the primary com-
bustion zone. Roughly 10% of the flow is directed into this zone through one or more
swirlers situated close to the fuel injection point. Vortex breakdown associated with
the induced swirl generates a highly-turbulent region of strongly recirculating flow,
referred to as the central recirculation zone (CRZ).
Liquid kerosene fuel is injected into the primary combustion zone in the form of
a finely atomised spray. Droplets are formed either by forcing the liquid through a
small orifice under pressure, or by using the momentum of the air to create shear
between the liquid and gas phases, thereby inducing atomisation. The fuel-air reac-
tion occurs in the gas phase on a molecular scale, and the highly turbulent air-flow
in the primary zone is therefore designed to cause rapid evaporation of the liquid
kerosene and fast fuel-air mixing.
The comparatively low-velocity recirculating flow in this region is conducive to
flame development and stabilisation, and the tip of the spark igniter that provides
the ignition energy is therefore often situated at the edge of the primary zone.
Following successful ignition, recirculation of the combustion products stabilises
the flame by providing a high-temperature source to ignite the incoming fuel-air
mixture. This feature ensures stable flame operation across a wide range of flow
conditions and air-to-fuel ratios (AFR). Moreover, the turbulent diffusivity increases
in direct proportion to the air flow rate, producing an increase in the fuel-air mixing
rate (Cant and Mastorakos 2008). The size of flow structures such as the CRZ
are thus insensitive to changes in bulk velocity, allowing operation of fixed volume
2. Background 8
...........................
.....
...........................
.....
air
fuel
(a)
........................................................
.............
.....
...
...
...
...
...
...
...
...
...
...
...
...
..
..............
......
....
...
...
...
...
...
...
...
.
...
...
...
...
...
...
...
...
.....
.........
.......
....................
........
......
.....
....
....
...
...
...
...
...
...
...
...
...
....
....
.....
.......
.........
..................
........................................................................................................................ ..............
......
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.....
........
........
.......................................
....
.........................................................................
....
.........................................................................
....
...
...
...
...
...
...
...
...
.....
...................................
....
................................................................
....
..........................
...
..
...
..
...............................
....
..........................
...
..
...
..
...
...
...
...
...
...
...
...
......
....
...
...
...
...
...
..............
................
.............
.....
...
....................................................................
.......................
...
...
...
...
...
...
...
...
...
...
.......
........... CRZ
dilution
zone
p
ri
m
a
ry
zo
n
e
igniter cooling air air casing
flame tube
fuel injector
.....................................
(b)
..............
......
....
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.....
.........
.......
...................
.........
.......
.....
....
....
...
...
...
...
...
...
...
...
...
...
....
.....
.......
.........
.................
......................................................................................................................... ..............
......
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.....
........
........
.......................................
....
.......................................
....
.......................................
....
...........................................................................................................
.....
...........................................................................................................
.....
........
.....
...
...
...
...
...
...
...
.....
.........
...............................................
.....
.................................................................................................
.....
...............................
...
................
..............
.....
...
...
...
...
...
...
...
...
...
...
...
...
...
....
.......
................................ .
...
...................................
.....
...
...
...
...
...
...
...
...
...
...
...
...
...
...
............
igniter
cooling air
fuel injector
......................................
Fig. 2.1: Gas turbine combustor concepts: (a) conventional; (b) LDI ANTLE.
2. Background 9
combustors across a range of flow rates without any increase in flame length. At
high-power conditions, fuel and air are introduced into the primary combustion
zone in roughly stoichiometric proportions, and the high temperature, pressure, and
turbulence intensity are sufficient to achieve essentially complete combustion of the
fuel-air mixture.
The high air pressure produced by the compressor coupled with the intense fuel-
air mixing resulting from an extremely high turbulence intensity lead to remarkable
rates of heat release. In the primary zone of a typical aviation gas turbine combustor
the heat release rate at take-off is approximately 500 MW/m3 – 100 times more
intense than that generated in the furnace of a large coal-fired power-plant (Hill
and Peterson 1992). The temperature of the hot gases produced by such a flame
is approximately 2400 K, and this is sufficiently high to damage the walls of the
flame tube. Approximately 40% of the combustor air-flow is therefore employed to
protect these surfaces. The remaining 20% of the flow is introduced into the dilution
zone, where it mixes with the primary zone gases to generate an exit flow with a
temperature distribution acceptable to the turbine.
A comparison of Figures 2.1(a) and 2.1(b) indicates that the flow conditions and
fuel placement in the low-NOx LDI ANTLE combustor introduced in Chapter 1 are
substantially different from those generated inside the conventional combustor de-
scribed above. The principal difference is the large increase in the proportion of the
air-flow passing through the fuel injector – some 65% compared to the 10% typical
for a conventional design. The velocities in the upstream portion of the ANTLE
combustion chamber are thus comparatively high, and a conventional primary com-
bustion zone cannot be identifed. The AFR adjacent to the igniter tip also varies
with the configuration. In a conventional combustor, the spark igniter and the fuel
injector are separated by a small distance, with a comparatively rich fuel-air mixture
created in the space between them. In contrast, the distance between the ANTLE
igniter and injector is large, with a particularly lean mixture expected immediately
2. Background 10
below the igniter tip. This combination of increased velocity and reduced equiva-
lence ratio has the potential to hinder the combustion processes required to achieve
altitude restart. It is this concern that has prompted the present experimental study
of the processes occurring inside the ANTLE combustion chamber during altitude
restart.
2.2 Altitude restart
High-altitude flame extinction inside a gas turbine can be caused by transient dis-
turbances to the air-flow through the engine, or by severe ingestion of ice, water,
or dust. Though such an event is rare, the capability to perform an in-flight engine
restart is mandatory. American and European air safety authorities presently re-
quire the submission and validation of a flight envelope defining restart capability
(Federal Aviation Adminisatration 2000; European Aviation Safety Agency 2003).
This section considers the conditions generated inside the combustor following flame-
out, and the influence of the engine restart requirement on the overall combustor
design.
2.2.1 Restart conditions
The conditions inside a gas turbine combustor following a high-altitude flame-out
vary with both the altitude and the flight speed of the aircraft. The International
Civil Aviation Organisation has formulated a standard atmosphere to represent the
variation in air temperature, pressure, and density as a function of altitude (In-
ternational Civil Aviation Organisation 1993). Subsonic civil aircraft fly in the
troposphere and lower stratosphere at an average cruise altitude of between 35000
and 38000 ft (10.5 and 11.5 km) (Penner et al. 1999). The atmospheric tempera-
ture, pressure and density at altitudes up to 20 km above sea level are plotted in
Figure 2.2. These data are expressed as a fraction of their sea-level values.
2. Background 11
altitude (km)
altitude (ft)
fr
a
ct
io
n
o
f
va
lu
e
a
t
se
a
le
v
el
×103......................................................................................................................................................................................................................................................................
....
......
......
......
......
.....
.....
.....
.....
.....
.....
.....
....
....
.....
....
....
....
....
....
....
....
....
.....
.....
.....
...... ...... ...... ...... ...... ...... ...... ......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . . . . . . . . . . . . . .
temperature
pressure
density
stratospheretroposphere
0 4 8 12 16 20
0 15 30 45 60
0.0
0.2
0.4
0.6
0.8
1.0
Fig. 2.2: The standard atmosphere. Each parameter is plotted as a fraction of its
value at sea level (Tsl = 288.15 K, Psl = 101.3 kPa, and ρsl = 1.23 kg/m
3).
The standard atmosphere is based on average values recorded at a latitude of
45◦ N, and significant excursions from the plotted data occur with location and
season. The temperature varies more than the pressure in this respect, and these
fluctuations are greatest close to the ground. For example, it is not uncommon
for the ground temperature at an airport to range from −40 ◦C to 40 ◦C. Suffi-
cient margin must be provided for these variations to ensure engine ground start-up
capability.
Figure 2.2 shows that the temperature, pressure, and density all decrease with
altitude. It is assumed that below the tropopause (11 km above sea level) the
temperature falls linearly at a rate of 6.5 K/km, while above the tropopause the
temperature remains constant at 216.65 K. The pressure at a given altitude is given
2. Background 12
by
P = Psl
(
T
Tsl
)g/kR
, (2.1)
where g is the acceleration due to gravity at sea level, k is the rate of decrease of
temperature with altitude, and R is the gas constant for dry air. From the ideal gas
law, the density is then
ρ =
P
RT
. (2.2)
Though strongly influenced by the temperature and pressure outside the aircraft,
the thermodynamic conditions inside the combustor also vary with flight speed. In
addition, the conditions are affected by various engine design parameters including
the bypass ratio, power offtake, and exhaust configuration (Braig et al. 1999).
In this study, a simplified analysis has been performed to illustrate changes in the
combustor conditions with flight condition. Software has been developed to calculate
the combustor inlet air temperature, pressure and mass flow rate corresponding
to a range of flight Mach numbers and altitudes. These results are presented in
Figure 2.3.
Calculations of the combustor inlet conditions were based on several assumptions
concerning the air-flow through the engine. Following an in-flight flame-out, the gas
turbine cools rapidly and the engine spools decelerate to a comparatively low speed
which is maintained by the ram pressure at the compressor inlet. Immediately prior
to a restart attempt, the heat and work transfers from the compressor to the air-flow
are therefore negligible. In addition, the velocity inside the combustion chamber is
only a small fraction of the flight speed. It is therefore valid to assume that the
combustor inlet temperature is equal to the stagnation temperature defined by the
speed and altitude of the aircraft. Thus, the combustor temperature is given by
Tc = T
[
1 +
(
γ − 1
2
)
M2
]
, (2.3)
where T is the atmospheric temperature at the flight altitude, γ is the specific heat
ratio of dry air, andM is the flight Mach number. While the stagnation temperature
2. Background 13
of the air is not significantly modified as it passes through the compressor, the
rotating blades generate highly irreversible, non-isentropic flow processes, leading to
a significant drop in stagnation pressure between the engine inlet and the combustion
chamber. It is reasonable to assume that half of the engine inlet dynamic head is
lost in the compressor (Rowe 2005), and in this case the combustor inlet pressure is
given by
Pc =
P0 + P
2
, (2.4)
where P0 is the engine inlet stagnation pressure and P is the atmospheric pressure
at the flight altitude. Since the flow velocity inside the combustion chamber is com-
paratively low, the static and stagnation values of both temperature and pressure
are approximately equal.
The total mass flow rate of air through the engine is controlled by the throat
area of the nozzle guide vanes (NGV) situated at the entrance to the high-pressure
turbine. A one-dimensional compressible flow calculation has been performed to
calculate the mass flow rate. This calculation assumes that the air-flow through the
NGV is steady, adiabatic, and isentropic. If it is also assumed that the pressure drop
across the NGV represents two thirds of the drop between the combustion chamber
and the engine exhaust, then the Mach number at the nozzle throat is
M =
[
2
γ − 1
((
Pc
Pt
)γ−1/γ
− 1
)]1/2
, (2.5)
where Pt is the static pressure at the throat. This may be expressed in terms of the
combustion chamber pressure as follows
Pt =
Pc + 2P
3
. (2.6)
The air mass flow rate through the engine is given by
m˙ = AMPc
√
γ
RTc
[
1 +
(
γ − 1
2
)
M2
]
−(γ+1)/2(γ−1)
, (2.7)
where A is the total throat area of the NGV.
2. Background 14
(a)
flight Mach number
a
lt
it
u
d
e
(f
t)
a
lt
it
u
d
e
(k
m
)
×103
..................................................
...................................................................................................................................................
.........................................................................................................................................................
..............................
............
..........
........
.......
.......
.......
......
......
.....
......
.....
.....
.....
.....
....
....
.....
.....
....
....
....
....
....
....
...
....
..
............
.........
........
.......
.......
......
......
.....
.....
.....
.....
....
.....
....
....
....
.....
....
....
....
....
...
....
...
...
....
...
...
...
....
.....
....
....
....
...
....
...
....
...
....
....
....
....
...
...
...
...
.
...
...
...
...
...
...
...
...
...
..
...
..
..
...
...
...
...
...
...
...
...
...
...
..
..
...
...
...
...
...
...
...
...
...
..
...
...
...
..
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..
...
...
...
...
.
230
250
270
290
310
330
0.0 0.2 0.4 0.6 0.8 1.0
0
10
20
30
40
50
0
3
6
9
12
15
(b)
flight Mach number
a
lt
it
u
d
e
(f
t)
a
lt
it
u
d
e
(k
m
)
×103
.........................................................
.......................
..................
...............
.............
............
...........
...........
.........
........
.....................................................
......................
.................
...............
.............
...........
...........
.........
.........
.........
.......
...................................................
......................
................
..............
............
...........
..........
..........
.........
........
........
.....
...................................................
.....................
...............
..............
...........
...........
..........
..........
.........
........
........
.......
..
..................................................
.....................
................
.............
...........
...........
..........
.........
.........
........
........
........
....
.........
..........
........
........
........
........
.......
........
....
...
..
...
..
...
...
...
..
.
...
...
..
...
...
...
...
...
..
...
...
...
...
...
...
...
...
..
...
...
...
...
..
...
...
...
...
.
...
...
...
...
...
...
...
..
...
...
...
...
...
...
...
...
..
0.2
0.4
0.6
0.8
1.0
1.2
0.0 0.2 0.4 0.6 0.8 1.0
0
10
20
30
40
50
0
3
6
9
12
15
(c)
flight Mach number
a
lt
it
u
d
e
(f
t)
a
lt
it
u
d
e
(k
m
)
×103
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..
...
...
...
...
...
...
...
....
....
....
...
....
....
....
....
....
..
...
...
...
...
...
...
...
...
...
..
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....
...
...
....
....
...
....
....
....
.....
....
.....
.....
.....
......
.....
.
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....
....
...
....
....
....
.....
....
.....
.....
.....
.....
.....
...
...
...
...
...
...
...
...
...
...
...
...
...
....
....
...
....
....
.....
....
....
.....
....
...
...
...
....
...
...
....
...
....
....
....
....
....
.....
..
...
...
....
....
....
...
....
....
.
....
....
...
...
..
...
..
...
...
..
...
...
...
...
..
...
...
...
...
...
...
...
..
...
...
...
...
...
...
...
...
..
...
.
...
...
...
...
...
...
...
...
.
...
...
...
...
...
...
...
...
...
...
...
...
0 50
1 25
2 00
2 75
3 50
4 25
0.0 0.2 0.4 0.6 0.8 1.0
0
10
20
30
40
50
0
3
6
9
12
15
Fig. 2.3: Combustor entry conditions: (a) temperature (K); (b) pressure (bar); (c)
mass flow rate (kg/s). A typical flight envelope is shown in red.
2. Background 15
Figure 2.3 illustrates that an increase in altitude at a fixed flight Mach number
leads to a decrease in combustion chamber air temperature, pressure and mass flow
rate. Conversely, an increase in flight Mach number at a fixed altitude increases the
temperature, pressure and mass flow rate inside the combustion chamber.
Airframe and engine manufacturers agree a maximum restart altitude that typ-
ically ranges from 20000 to 30000 ft (6.1 to 9.1 km) above sea level. Following a
flame-out at 30000 ft, the combustor inlet temperature and pressure are typically
265 K and 0.4 bar respectively. Furthermore, the air mass flow rate through the
combustion chamber is much lower than that generated at cruise, resulting in lower
combustor flow velocities. Despite the reduced velocities the Reynolds number re-
mains extremely high, and the flow is consequently highly turbulent.
These operating conditions impede several key physical and chemical mecha-
nisms required to achieve high-altitude restart. Perhaps foremost amongst these is
atomisation of the liquid kerosene fuel. As noted in Section 2.1, the air-blast atom-
iser that is often employed in modern gas turbine combustion systems operates by
creating a high-velocity airstream to fragment a sheet of fuel discharged from the
edge of a pre-filming surface. The comparatively low density, velocity, and in turn
momentum of the air-flow, results in a significant degradation in atomisation quality
at restart conditions (Beck et al. 1991). Indeed Caines et al. (2001) have observed
droplets as large as 300 µm in diameter at sub-atmospheric operating points.
The formation of fine droplets may be further hampered by low fuel tempera-
tures. The temperature of fuel in the aircraft storage tanks decreases during flight to
a minimum of around −35 ◦C (Nicholas 2005), and for this reason the freezing point
of Jet A kerosene must not exceed −45 ◦C (ASTM 2008). During cruise operation
the fuel is heated as it passes through the injector feed arm. Since the engine cools
rapidly following flame-out, this warming does not occur and the viscosity of the
fuel increases appreciably as a result (Coordinating Research Council 2004; Atkins
et al. 2005), further impeding atomisation (Lefebvre 1989).
2. Background 16
Ballal and Lefebvre (1979) have shown that the minimum energy required to
ignite a flowing heterogeneous mixture varies with the fourth power of mean droplet
diameter (see Section 2.4.3). This relationship results from the slower evaporation of
larger droplets by virtue of their smaller surface area to volume ratio. Furthermore,
Richards and Lefebvre (1989) have demonstrated that for droplet diameters greater
than 40 µm, the speed of an ambient kerosene-air flame is inversely proportional
to the Sauter mean diameter (SMD - the diameter of a drop with a surface area
to volume ratio equal to that of the spray as a whole). This supports independent
observations that the combustion efficiency is also adversely affected by increasing
droplet size (Lefebvre 1999).
Following injection, the evaporation of fuel is further limited by the low gas and
liquid-phase temperatures. As described above, the liquid fuel temperature is likely
to be low, with a correspondingly low vapour pressure. Prior to restart, the greater
mass of fuel is therefore likely to be contained in liquid droplets that are considerably
larger than those produced at any other engine operating point. Fuel-air mixing is
thus comparatively poor. The inertia of these large liquid droplets often causes them
to deviate significantly from the streamline of the surrounding air. Coupled with
low evaporation rates, this results in significant impingement of liquid fuel on the
internal surfaces of the combustor.
In principle, the quantity of fuel injected into a LDI combustion chamber can be
modified to provide the optimum AFR for successful restart. In practice however
the temperature rise inside the combustion chamber must be limited to avoid com-
pressor surge and high-temperature damage to the turbine, and the global AFR
is consequently set leaner than stoichiometric. However, large fluctuations and
point-to-point differences in equivalence ratio undoubtedly exist due to high rms
turbulence levels and significant spatial and temporal variations in gas velocity and
droplet size.
The ignition, stabilisation, and efficiency of kerosene-air flames generated under
2. Background 17
altitude conditions are all adversely affected by the comparatively low air tempera-
ture and pressure inside the combustion chamber. To achieve ignition, the temper-
ature of the reactants must be raised to a point where the thermal energy release
is sufficient to sustain a chemical reaction. Ignition of low-temperature reactants
therefore requires more energy than at ambient conditions. The temperature of the
reacting gases is also reduced by the cold fuel and excess air. The rate of most
chemical reactions varies with temperature according to the Arrhenius relationship
k = A exp
(
−EA
RT
)
, (2.8)
where A is known as the pre-exponential factor, EA is the activation energy, and R
is the universal gas constant (Glassman 1996). The reaction rate decreases expo-
nentially with reductions in temperature.
The rate of chemical reaction is also adversely affected by reductions in pressure;
at low pressures the reaction rate is approximately proportional to the square of
pressure (Kerrebrock 1977). These temperature and pressure dependencies result
in a dramatically reduced reaction rate under restart conditions. This extended
reaction time limits combustion efficiency and reduces the stability limits of the
flame (Hill and Peterson 1992).
2.2.2 The restart process
Lefebvre (1999) has proposed that the in-flight engine restart process can be sepa-
rated into three phases:
1. a spark discharge giving rise to the formation of a flame kernel containing
sufficient heat to propagate successfully;
2. migration of this kernel to the fuel injector face and stabilisation of the flame
at a single fuel injector;
3. flame spread around the combustor annulus to all fuel injectors (known as
‘light-round’).
2. Background 18
To this list one final phase might be added:
4. the generation of sufficient heat release to accelerate the gas turbine to its
operational speed (known as ‘pull-away’).
Pull-away generally requires a combustion efficiency exceeding 80% (Lefebvre 1999).
All four phases must be successful in order to restart the engine. In the current
study, phases 1 and 2 of the engine restart process are considered, and these will be
collectively referred to as relight from now on.
Though ignition of a combustible mixture may be accomplished using a variety
of methods, a spark discharge is a simple and effective means of generating the large
temperature increase necessary to create a viable flame kernel. For this reason spark
ignition has been universally adopted in aviation gas turbine applications. A modern
ignition system consists of an exciter unit connected to a surface-discharge igniter
via a lead. Typically two of these systems are installed in each engine. A reservoir
capacitor inside the voltage generator unit is charged by means of a transistor, and
discharges through the igniter when the breakdown voltage is exceeded.
The two igniters are mounted radially on opposite sides of the combustor annulus.
In a conventional combustion system, the tip of each igniter is generally situated at
the upstream edge of the primary zone, flush with the external surface of the flame
tube. The igniter tip is usually located further downstream in a LDI combustor,
above the central recirculation zone. The surface-discharge igniter used in aircraft
engines consists of two concentric electrodes separated by a ceramic insulator. The
outer electrode is grounded, and the open end of the insulator is coated with a thin
layer of silicon carbide. This semiconducting layer promotes ionisation between the
electrodes, allowing the generation of sparks at comparatively low voltages (≈ 2 kV).
When the breakdown voltage is exceeded, the current that begins to flow through
the semiconductor is concentrated into a fine filament. Having created an ionisation
path between the electrodes, the main discharge follows as a powerful arc. This
process is unaffected by the pressure of the surrounding air.
2. Background 19
For large turbofan engines the stored energy is normally between 4 and 12 J with
a typical sparking rate of 1 Hz. Lefebvre (1999) has reported that only a quarter of
the energy discharged from the capacitor reaches the igniter tip. Furthermore, only
a third of this energy directly heats the fuel-air mixture, the remainder being either
conducted to the igniter face or dissipated as light and sound. Large quantities of fuel
deposited on the igniter tip can further reduce the spark energy (Odgers and Coban
1977), in addition to significantly shortening the igniter operating life (Burland et al.
1984). As detailed in Section 2.4 below, previous studies have demonstrated that
spark ignition of a droplet mixture is influenced by various attributes of the spark
discharge. Ignition probability has, for example, been shown to increase with spark
size, energy, duration and frequency. While these effects have been quantified in
well-characterised mixtures, their significance during gas turbine restart has received
little attention.
2.2.3 Design for altitude restart
As discussed above in Section 2.2.1, the fuel preparation, mixing, ignition and heat
release processes that are necessary for successful restart are all slowed by the low air
temperatures and pressures inside the combustion chamber following flame-out. The
residence time of the reactants inside the combustor must therefore be comparatively
long to ensure that the heat released is sufficient to create and sustain a flame. For
this reason, the overall combustion chamber volume is determined by the restart
requirement (Nicholas 2005). However, an unnecessarily large combustor volume
must be avoided, both to limit the time available for NOx formation at high power,
and to minimise the size and weight of the combustion subsystem. From an engine
design perspective, it is therefore necessary to develop a good understanding of the
combustor volume required to ensure ignition, flame stabilisation and the required
combustion efficiency.
At present, the combustion chamber volume is determined using a set of empirical
2. Background 20
correlations developed during the design of conventional combustion systems. These
design rules are proving inappropriate for the development of future-generation LDI
combustors, as the combustor aerodynamics and fuel placement generated by these
systems are radically different from those produced by conventional designs. In
addition, the use of computational modelling as a design tool is severely limited
by the complex turbulence-chemistry interactions that occur in gas turbine flames
(Cant and Mastorakos 2008).
The lack of design data for LDI combustion systems coupled with the limited
modelling capability described above has led to increased use of ground-based al-
titude testing to determine restart performance. Altitude test facilities have been
constructed that can simulate the conditions inside the combustion chamber follow-
ing flame-out. These facilities provide a flexible means of evaluating and developing
combustor designs. Though costly in comparison to computational modelling work,
ground-based testing is significantly less expensive than flight testing on board an
aircraft.
The objective of ground-based relight testing of the kind conducted using the
Rolls-Royce ATF (see Section 3.1), is to identify those combinations of air tempera-
ture, pressure, mass flow rate, and fuel-air ratio (FAR) that allow ignition and flame
stabilisation at a single fuel injector. While maintaining a constant temperature,
pressure and mass flow rate, relight is attempted across a range of FAR. A relight
attempt is regarded as successful if ignition occurs within a predefined period of the
first spark, and the flame sustains itself for a further interval following termination
of the spark (both periods are typically 10 s). The procedure is repeated for a range
of air mass flow rates.
As illustrated schematically in Figure 2.4, these results are presented in the form
of an ignition loop. This loop separates regions of successful and unsuccessful relight
at a given temperature and pressure. Relight is successful inside the area enclosed by
the loop, and the point identified as ‘safe’ is therefore conducive to ignition. Above
2. Background 21
air mass flow rate
fu
el
-t
o
-a
ir
ra
ti
o
..........................................................................................................................................................................................................................................................................................
•
•
.............................................................................
...............
...........
.......
......
....
...
..
...
...
...
...
...
...
...
...
...
....
....
...
....
....
.....
.....
.....
.....
....
....
....
....
....
....
....
....
....
....
....
....
.....
.....
...
...... ...... ...... ...... ...... ...... ...... ...... ...... ......
...... ......
...... .
.....
.....
...
.
...
...
...
...
...
.
...
.
...
.
...
.
....
....
....
.....
.....
.....
.....
.....
.....
.....
......
...
......
.....
safe unsafe
ignition possible
within loop
ignition loop
stability loop
Fig. 2.4: Ignition and stability loops. A line of constant fuel flow rate is shown in
grey.
and below this region the mixture is too rich and too lean respectively to ignite.
As the air mass flow rate increases, the relight limits narrow until the flame can no
longer stabilise. Ignition cannot therefore occur at the point marked ‘unsafe’. In
general, several ignition loops obtained at different temperatures and pressures are
sufficient to determine the relight behaviour of the engine. Given knowledge of the
maximum permissible fuel flow rate at restart, it is then possible to estimate the
aircraft relight envelope as a function of flight altitude and speed.
Similar loops can be plotted to represent the stability limits of established flame.
Following successful ignition at a given operating point, the fuel flow rate is reduced
until flame extinction occurs. After recording the FAR at this lean blowout point,
combustion is again established, and the fuel flow rate incrementally increased until
rich extinction occurs. As before, this process is repeated at a variety of air mass flow
rates. Since ignition and flame stability are influenced by similar factors, we may
expect the two loops to coincide. As Lefebvre (1999) has noted however, the ignition
2. Background 22
kernel has a comparatively large surface area to volume ratio and is surrounded by
cold air, while a stabilised flame is larger and is bounded by the hot combustion
chamber. The higher heat losses during ignition therefore make the developing
flame kernel more susceptible to extinction, and for this reason the ignition loop
always lies inside the stability loop. The relative position of these loops indicates
to the combustor designer the measures required to improve relight performance.
For instance, if the ignition loop lies well within the stability loop then relight
performance may be improved by increasing the spark energy. On the other hand,
if the ignition and stability loops are closely spaced and there still exists a relight
problem, then wider measures are required to encourage flame stabilisation, such
as changes to the combustor aerodynamics to ensure appropriate transport of the
developing kernel.
Though constructed primarily for development testing, ground-based altitude
facilities also provide a means of more closely exploring the physical and chemical
processes occurring during restart. For example, optical access to these facilities al-
lows the use of advanced non-intrusive laser techniques to understand flow structure,
fuel placement, and flame development. From a design perspective, identifying the
role and significance of the spark in the relight process may also lead to beneficial
design changes. For example, the spark energy provided by the igniter exceeds the
minimum ignition energies recorded under similar operating conditions (Ballal and
Lefebvre 1978b; Ballal and Lefebvre 1979), suggesting that ignition success or fail-
ure is determined during the later stages of flame development. If the spark energy
requirement could be reduced by more favourable spark placement in the flow field,
then it may be possible to minimise the igniter hardware bulk and extend igniter
life.
2. Background 23
2.3 Review of ignition and flame stabilisation
Ignition is defined as the initiation of a self-sustaining exothermic reaction in an
inflammable mixture. This is typically achieved by mixing the fuel and oxidant in
appropriate proportions, and raising the temperature of the mixture to generate
chemical radicals that react to initiate a flame. The initial temperature increase
may be generated by a global heating process such as that occurring during the
compression stroke in a diesel engine, or by a localised deposition of energy such
as that produced by an electric spark. A spark generates an ignition kernel; a
confined region of high chemical reactivity and heat release. The probability of this
kernel evolving successfully into a self-propagating flame is governed by features of
the spark coupled with parameters describing the state of the local mixture. The
influence of these parameters is assessed by determining, either experimentally or
theoretically, the heat addition required to generate a self-sustaining flame. This
quantity of heat is referred to as the minimum ignition energy.
The ability of a spark to provide a rapid, highly-localised deposition of energy
to a flowing fuel-air mixture has led to the universal adoption of surface-discharge
igniters on aviation gas turbines. Numerous studies have been conducted examin-
ing the ground-level and high-altitude ignition performance of specific combustion
system configurations (Xiong et al. 1981; Naegeli and Dodge 1991; Pucher and Al-
lan 2004). However, as noted by Rao and Lefebvre (1976) the heterogeneous flow
through a gas turbine combustion chamber is highly complex, producing marked
fluctuations in the equivalence ratio, flow velocity, and turbulence properties below
the igniter tip. Understanding derived from these investigations, while contributing
to the development of a particular combustor, fails to provide generic information
concerning ignition energy requirements. To achieve a sounder understanding of the
underlying mechanisms governing flame development, it is necessary to evaluate the
minimum ignition energy under well controlled conditions.
2. Background 24
2.3.1 Paradigms of ignition research
Glassman (1996) notes that the theory of thermal ignition is based on a simple
concept: an explosive condition exists if the rate of thermal energy release within a
system exceeds the rate of thermal energy transfer from the system. The generation
of a radially-propagating combustion wave following a spark discharge has thus been
viewed as a process regulated by the competition between chemical energy release
within the flame kernel and heat loss across the wave boundary. Deflagration models
developed to characterise the nature of this thermal balance share several conceptual
principles which are outlined below.
The spark discharge generates a small volume of high temperature gas. It has
been suggested that in order to propagate through the mixture, the inflamed kernel
must attain a minimum size before the internal temperature falls below the normal
flame temperature. At this critical diameter, the surface area to volume ratio of
the kernel is just small enough to ensure growth. The minimum ignition energy is
defined as the quantity of energy required to heat the contents of this critical sphere
to the mixture flame temperature. As such, the critical diameter and minimum
ignition energy are closely related. If the supplied spark energy is less than the
minimum ignition energy, the flame fails to become established as the mixture is
quenched by the surrounding ambient gas and the electrodes.
During experimental ignition studies, heat transfer from the flame kernel to the
electrodes causes an increase in the recorded minimum ignition energy when the
electrode separation is less than a minimum value known as the quenching distance.
Though the critical diameter and quenching distance have been viewed by some
authors as directly equivalent (Lewis and von Elbe 1961), their distinct definitions
should be noted.
Treating the spark discharge as a point source of heat, Glassman (1996) recom-
mends the use of Zeldovich’s thermal model for spark ignition. This theory proposes
that for a quiescent premixed system, flame propagation will proceed provided that
2. Background 25
the characteristic cooling time exceeds the combustion reaction time. The time-
dependent temperature distribution generated by a point source of heat can be
calculated from the heat diffusion energy equation. Expressed in spherical polar
coordinates this is
∂T
∂t
=
α
r2
∂
∂r
(
r2
∂T
∂r
)
, (2.9)
where T is the mixture temperature, α is the thermal diffusivity of the mixture
(α ≡ k/ρcp), and r is the radial distance from the energy deposition point. The
boundary conditions are
T → T0 as r →∞ and ∂T
∂r
= 0 at r = 0. (2.10)
The solution satisfying these boundary conditions has the form
T = T0 + A (t) exp
(−r2/4αt) . (2.11)
Energy is conserved, and thus
E = 4piρcp
∫
∞
0
(T − T0) r2dr , (2.12)
where E is the deposited heat, ρ is the mixture density, and cp is the specific heat
at constant pressure of the mixture. Substituting Eq. (2.11) into the integral and
integrating by parts, we find∫
∞
0
(T − T0) r2dr = A (t)
∫
∞
0
r2 exp
(−r2/4αt) dr
= 2αtA (t)
∫
∞
0
exp
(−r2/4αt) dr
= 2pi1/2 (αt)3/2A (t) . (2.13)
Combining Eqs. (2.12) and (2.13) gives
A (t) =
E
ρcp (4piαt)
3/2
, (2.14)
and so the solution to Eq. (2.9) is
T = T0 +
[
E
ρcp (4piαt)
3/2
]
exp
(−r2
4αt
)
. (2.15)
2. Background 26
radius, r
te
m
p
er
a
tu
re
,
T
............................................................................................................................................................................................................................................................................................................
....................................................................................................................... .......................................................................................................................
............................................................................................................................ ...............
.....................................................................................
...............
...........................................
................
...
..
...
..
increasing time, t
T0
Tm
Fig. 2.5: Temperature profiles after point energy deposition (see Eq. (2.15)).
The spatial and temporal form of this solution is illustrated in Figure 2.5.
The maximum temperature, Tm, occurs at r = 0 and is therefore
Tm = T0 +
E
ρcp (4piαt)
3/2
. (2.16)
The characteristic cooling time, τc, is defined as the period required for the temper-
ature at r = 0 to decrease by a small value
θ =
RT 2m
E
, (2.17)
where R is the universal gas constant. The cooling time is thus approximated by
τc =
θ
(dTm/dt)Tm=Tf
. (2.18)
The rate of change of temperature is evaluated when the temperature at r = 0 is close
to the adiabatic flame temperature of the mixture, Tf . Differentiating Eq. (2.16) with
respect to time and substituting into Eq. (2.18) gives
τc =
θ
6piα (Tf − T0)
[
E
ρcp (Tf − T0)
]2/3
. (2.19)
2. Background 27
Supposing that the input energy E is equal to the energy required to heat a spherical
volume of gas with diameter dcr from temperature T0 to temperature Tf , then
E =
pi
6
d3crρcp (Tf − T0) . (2.20)
Substituting Eq. (2.20) into Eq. (2.19)
τc = 0.034
θ
(Tf − T0)
(
d2cr
α
)
. (2.21)
The thermal flame theory proposed by Semenov and Frank-Kamenetskii provides
the approximate reaction duration time associated with the combustion zone of a
laminar flame
τr ≈ 2 θ
(Tf − T0)
(
α
S2L
)
, (2.22)
where SL is the laminar flame speed. The thermal flame thickness is
δ =
α
SL
. (2.23)
Combining Eqs. (2.21), (2.22), and (2.23), and applying the condition that τc ≥ τr
gives the condition for ignition as
dcr ≥ 7.6δ . (2.24)
This implies that the critical diameter must exceed the thermal flame thickness by a
factor of approximately 8 for successful ignition to occur. Experimental confirmation
of this relationship has proved challenging due to difficulties in determining the exact
heat input generated by an air-gap spark. Despite this, experimental results suggest
that for ignition to occur the critical diameter must be at least 6 times the flame
thickness (Blanc et al. 1949).
Substituting Eqs. (2.23) and (2.24) into Eq. (2.20), and using the ideal gas law to
re-express density in terms of temperature and pressure, gives a relationship between
the minimum ignition energy, Emin, the far-field temperature, T0, and the mixture
pressure, P
Emin ∝
(
k
SL
)3(
T0
Pcp
)2
(Tf − T0) . (2.25)
2. Background 28
Though temperature and pressure are clearly influential, the highest order depen-
dences are associated with the thermal conductivity of the mixture and the laminar
flame speed. Mixtures with a low thermal diffusivity that generate high flame speeds
require less ignition energy by virtue of their thin thermal flame fronts.
Following an extensive series of experimental measurements, Lewis and von Elbe
(1961) presented an alternative description of gaseous spark ignition in terms of
adiabatic combustion wave theory. These authors proposed that kernel propagation
occurs provided that the temperature gradient between the reactants and the com-
bustion products is similar to that generated in the steady-state wave. They argue
that exceptionally steep temperature gradients lead to a fast reduction in tempera-
ture throughout the combustion volume, with the ambient unburned gas quenching
the flame. The quenching phenomenon is attributed to flame stretch.
To investigate this hypothesis the authors estimated the fractional increase in
flame area when a combustion wave propagates a distance equal to the thermal
flame thickness. Assuming that the ratio of the critical diameter to flame thickness
is large i.e. dcr ≫ δ, the critical stretch factor is
K = 4
δ
dcr
ρu
ρb
, (2.26)
where δ is the thermal flame thickness (see Eq. (2.23)), dcr is the critical diameter
of the flame sphere, ρu is the density of the unburned mixture, and ρb is the density
of the burned products. By identifying dcr as the experimental quenching distance,
the data reported by Blanc et al. (1949) were used to calculate the critical stretch
factor for a wide range of fuel-oxidant mixtures.
Stretch factors of between 0.5 and 1 were consistently reported, implying a degree
of correlation between the measured quenching distances and the burning velocities,
densities, and specific heats of the explosive gases. This finding is consistent with
the Zeldovich analysis described above since the critical stretch factor, and thus the
ratio of flame thickness to critical diameter, must be less than a threshold value to
ensure flame propagation.
2. Background 29
Lewis and von Elbe (1961) also evaluated the relationship between measured
values of the minimum ignition energy, and two reference energy parameters, denoted
E ′ and E ′′. The increase in sensible heat contained in the minimum flame sphere
following a complete conversion of reactants to combustion products is
E ′ =
pi
6
d3crρbcp (Tb − Tu) , (2.27)
where Tb and Tu are the temperatures of the burned and unburned gas mixtures
respectively. Assuming that cp does not vary significantly with temperature, and
that the variation in density with temperature is given by ρT = ρbTb, then the
sensible heat contained in the minimum flame can be approximated by∫ dcr/2
0
4piρcp (T − Tu) r2dr ≈ 4piρbcp
∫ dcr/2
0
Tb
T
(T − Tu) r2dr
< 4piρbcp
∫ dcr/2
0
(Tb − Tu) r2dr
=
pi
6
d3crρbcp (Tb − Tu)
= E ′ (by Eq. (2.27)). (2.28)
Although E ′ is thus an overestimate of the sensible heat contained in the minimum
flame sphere, it is within an order of magnitude of the true value. Considering the
energy balance in a plane combustion wave, the total heat per unit area that is
conducted through the flame thickness from hotter to cooler layers is approximately
given by k (Tb − Tu) /SL. Treating the minimum flame as planar, the total conducted
heat stored within the kernel is thus
E ′′ = pid2cr
k
SL
(Tb − Tu) . (2.29)
The difference between the volume integrals of sensible and conducted heat inside the
minimum flame kernel, (E ′−E ′′), represents the heat released by chemical reaction.
Despite an order of magnitude correspondence, the measured values of Emin are
generally less than both E ′ and E ′′, indicating that the heat in the minimum flame
derives partly from the ignition source and partly from early chemical heat release.
2. Background 30
The authors also suggested that all combustion waves possess excess enthalpy
that maintains the correct temperature profile to ensure a balance between heat
flow into the preheat zone and heat release in the reaction zone. A spherically
propagating combustion wave thus requires a continuous source of excess enthalpy
from an external source in proportion to the surface area of the flame front. To
enable the flame to grow to a viable size, the ignition source must supply a minimum
ignition energy equal to the excess enthalpy requirement of the minimum flame
sphere. Though this hypothesis remains unproven, the authors maintain that it is
consistent with the observation that measured values of Emin are generally smaller
than the corresponding approximate values of the conducted heat, E ′′.
A characteristic time approach has been developed by Peters and Mellor (1980)
to examine the more complex phenomenon of spray ignition. This theory assumes
that mixing and chemical reaction occur infinitely fast, and that the rate of heat
release in the developing kernel is limited by the fuel evaporation rate. Having
formulated order of magnitude expressions for quenching and evaporation timescales,
these authors proposed that the kernel cooling period must exceed the total fuel
evaporation time if successful flame propagation is to occur.
Assuming that heat is removed from the developing flame kernel by conduction,
the cooling time is
τc =
pid2
α
, (2.30)
where d is the diameter of the kernel. In the absence of fuel vapour, the evaporation
time of a monodisperse spray is defined as the ratio of the fuel mass contained in
the kernel to the fuel evaporation rate
τe =
ρfcpD
2
8kaln (1 +B)
, (2.31)
where ρf is the liquid fuel density, cp is the specific heat capacity at constant pressure
of the air, D is the drop size, ka is the thermal conductivity of air, and B is the fuel
transfer number (physically, the ratio of the energy available for evaporation to the
energy required).
2. Background 31
Identifying d as the quenching distance, and using the experimental data of Bal-
lal and Lefebvre (1978a), a linear relationship was observed between the evaporation
time and the kernel cooling time for a wide range of fuel transfer numbers, equiv-
alence ratios, and droplet diameters. Based on this finding, the critical diameter
could be identified for a defined set of system parameters. Having identified this
diameter, the minimum ignition energy was calculated according to Eq. (2.20).
The values of minimum ignition energy derived from this model showed reason-
able agreement with experimental results. Discrepancies were however noted in cases
where the test conditions produced rates of chemical reaction and fuel evaporation
that were comparable. This occurs in lean, low pressure sprays containing small,
highly volatile fuel droplets. This deficiency, together with the possibility of fuel
vapourisation prior to the spark, led Ballal and Lefebvre (1981) to devise a general
model of fuel-air ignition that applies to both quiescent and flowing mixtures of air
with any combination of gaseous or liquid fuel.
In addition to the cooling and evaporation times identified by Peters and Mellor
(1980), a chemical reaction timescale, τr was included. Thus for a flame kernel with
the critical diameter,
τc = τe + τr . (2.32)
For a gaseous fuel this simplifies to
τc = τr . (2.33)
If the right-hand side of Eqs. (2.32) or (2.33) is less than the cooling time then
successful flame propagation will occur. Conversely, if they are greater than the
cooling time then the ignition attempt will be unsuccessful. The general expression
for the cooling time is
τc ∝ d
2
α+ Au′d
, (2.34)
where A is a constant and u′ is the rms value of fluctuating velocity. The fuel
2. Background 32
evaporation time is given by
τe ∝ (1− Ω) ρfcpD
2
kaφ ln (1 +B)
(
1 + CRe
1/2
D
) , (2.35)
where Ω is the prevapourised fraction of the fuel, φ is the equivalence ratio, C is
a constant, and ReD is the Reynolds number based on the droplet diameter and
the rms value of fluctuating air velocity. The chemical reaction times in quiescent,
low-turbulence, and high-turbulence conditions respectively are
τr ∝ α
S2L
, τr ∝ α
(SL −Eu′)2
, and τr ∝ α
u′ (ST − Fu′) , (2.36)
where E and F are constants and ST is the turbulent flame speed.
Each characteristic time is influenced by the flow Reynolds number. The evap-
oration time decreases with Reynolds number, while induced turbulence reduces
both the cooling and reaction timescales. Again assuming the critical diameter and
the quenching distance to be identical, the critical diameter and minimum ignition
energy were calculated using Eqs. (2.20), (2.32), and (2.33).
Theoretical predictions produced by this model correlated well with experimental
data for a wide range of fuel, flow and droplet parameters. As expected, chemical
kinetic effects proved increasingly important with reductions in equivalence ratio
and pressure. Inclusion of the chemical reaction time for these conditions produced
a marked improvement in correspondence between theory and experiment. Using a
virtually identical approach to analyse ignition of an n-decane spray, Dietrich et al.
(1990) demonstrated good agreement with experimental results by using a chemical
reaction timescale based on the Arrhenius reaction rate law.
In recent years the theoretical and empirical approaches to ignition research
outlined above have been complemented by computational fluid dynamics (CFD)
approaches that solve the fundamental transport equations for mass, momentum,
and energy. Advances in numerical methods and parallel computing have enabled
the application of direct numerical simulation (DNS) to the ignition of turbulent
gaseous mixtures under conditions of both homogeneous (Klein et al. 2006) and
2. Background 33
heterogeneous (Chakraborty et al. 2007) fuel distribution. Some agreement between
DNS results and experimental data has also been demonstrated. For example, a
study undertaken by Kaminski et al. (2000) of a turbulent premixed methane-air
spark kernel showed reasonable agreement between the OH mole fraction predicted
by DNS and that measured by high-speed PLIF. However, the comparison of DNS
with experiments, or the use of DNS in place of experiments to validate theoretical
ignition models is restricted by the computational difficulty of simulating realistic
Reynolds numbers and complex chemical reactions.
Ignition modelling in practical combustion devices has been demonstrated by
Boileau et al. (2008a, 2008b), who performed large-eddy simulations (LES) of igni-
tion and light-round in a helicopter gas turbine. Flagship simulations such as these
point to the future use of LES and DNS as design tools for improving the restart
performance of gas turbines. However, the massive computational effort involved
(700 parallel processors in the case of Boileau et al. (2008b)), currently limits their
applicability in general. Ignition and flame development in a real gas turbine is a
highly stochastic process, sensitive to initial conditions. To identify a representative
range of ignition behaviour inside a gas turbine combustion chamber would therefore
require many simulations; an approach that is prohibitively expensive at present.
This difficulty is not encountered in experiments where a large sample of results can
be obtained relatively easily. Experiments investigating ignition are reviewed in the
next section.
2.4 Fundamental experiments to study ignition
As Warnatz et al. (2006) have highlighted, experimental research into spark-induced
ignition is hampered by the necessary proximity of the spark electrode to the devel-
oping flame kernel, raising questions concerning surface chemistry and conductive
heat losses. Revealing the relative proportions of spark energy responsible for local
temperature increases and independent radical generation is similarly fraught with
2. Background 34
difficulty. Despite these obstacles, numerous experimental studies have been under-
taken to identify how the spark energy required to ignite a fuel-oxidant mixture is
influenced by a variety of system parameters. These investigations have provided
quantitative assessments of the independent effects on ignition performance of vari-
ations in the spark gap and timing, reactant temperature and pressure, mixture
chemical composition and fuel phase, as well as flow velocity and turbulence char-
acteristics. Lengthy description of the system parameters for each of the reviewed
studies has been avoided by providing details of individual experimental configu-
rations in tabular format. Tables 2.1, 2.2, and 2.3 provide information relating to
ignition studies of gaseous blends, quiescent droplet suspensions, and flowing het-
erogeneous mixtures respectively.
It has been widely reported that, other things being equal, the measured min-
imum ignition energy varies markedly with both electrode separation distance and
discharge duration. Endeavouring to report the true minimum energy requirement,
the majority of the highlighted studies included optimisation procedures for both
spark duration and gap width. This practice was not universally adopted however,
with some researchers electing to maintain constant electrode separation and spark
duration throughout the experimental data collection.
2.4.1 Ignition of gaseous blends
Blanc et al. (1947, 1949) developed apparatus to investigate the spark ignition of
homogeneous combustible gas mixtures. Two glass-flanged electrodes were mounted
near the centre of a spherical test bomb. Accurate adjustment of the spark gap was
provided by a micrometer. Several fixed and variable capacitors and a static volt-
meter were connected in parallel to provide a reservoir of known voltage. Initially,
a rotary charger was used to transfer charge from a high voltage source to the elec-
trode capacitor system, but this was later replaced by a paper rod impregnated with
Bakelite. Fine adjustment of the system capacitance was achieved using numerous
2
.
B
a
ck
g
ro
u
n
d
35
Table 2.1: Experiments investigating ignition of gaseous mixtures.
Blanc et al. (1947, 1949) Ballal and Lefebvre (1977a) Ballal and Lefebvre (1977b)
working section
spherical bomb, horizontal box chamber, horizontal box chamber,
127 mm internal diameter 90 mm× 90 mm× 310 mm 90 mm× 90 mm× 310 mm
electrodes
glass-flanged, unflanged, unflanged,
1.6 mm diameter 2 mm diameter 2 mm diameter
fuel specification numerous, mainly alkanes propane and methane propane and methane
oxidant ratio
O2
O2+N2
= 0.21-1, O2
O2+x
= 0.21, O2
O2+N2
= 0.21-1, O2
O2+x
= 0.21, O2
O2+N2
= 0.21-1, O2
O2+x
= 0.21,
where x = He,Ar,CO2 where x = He,Ar,CO2 where x = He,Ar,CO2
mixture temperature ambient ambient ambient
mixture pressure (bar) 0.2-1.0 0.08-0.50 0.08-0.35
mixture velocity (m/s) – 0-30 0-15.3
turbulence intensity (%) – 2.5-15 2.5-15
turbulence scale (mm) – ≤ 10 ≤ 12
quenching distance (mm) 0.2-20 1.5-40 4-30
spark duration (µs) not reported not reported ≈ 60, optimised
Emin (mJ) 0.002-100 0.4-90 0.02-60
2
.
B
a
ck
g
ro
u
n
d
36
Table 2.2: Experiments investigating ignition of quiescent aerosols.
Ballal and Lefebvre (1978a) Singh et al. (1986) Danis et al. (1988) Dietrich et al. (1990)
working section
vertical box chamber, vertical nozzle, vertical tube, vertical tube,
90 mm× 90 mm× 310 mm 10 mm diameter 16 mm diameter 16 mm diameter
electrodes
glass-flanged,
1 mm diameter
flattened cathode, flattened cathode,
2 mm diameter 1 mm diameter 1 mm diameter
fuel specification six liquid fuels Tetralin methanol and n-heptane n-decane
equivalence ratio, φ 0.4-0.9 0.7-1.2 0.44-1.8 0.75-2.2
prevapourised fraction, Ω neglected neglected 0.35, 0.45 < 0.07
air temperature (K) 290 ambient ambient ambient
air pressure (bar) 0.2-1.0 ambient ambient ambient
flow details air velocity ≈ 0.2 m/s nozzle Re = 300 tube Re = 100-300
air flow rate =
0.27−0.45 m3/min
means of atomisation spinning cup atomiser controlled condensation
vibrating orifice twin vibrating orifice
aerosol generator aerosol generators
droplet size parameters
SMD = 30-180 µm SMD = 6.7-40 µm
SMD = 30-57 µm
bidisperse, with peaks
Rosin-Rammler ind. = 8-11 std. dev. = 6-13% at 30 µm and 90 µm
quenching distance (mm) 2-15 3-4 2-7 not determined
spark duration (µs) 60-100, optimised 40-130, optimised 30-100, optimised 60, constant
Emin (mJ) 0.1-1000 1-50 0.2-100 0.4-40
2
.
B
a
ck
g
ro
u
n
d
37
Table 2.3: Experiments investigating ignition of flowing droplet mixtures.
Rao and Lefebvre (1976) Ballal and Lefebvre (1978b) Ballal and Lefebvre (1979)
working section
horizontal box chamber, horizontal box chamber, horizontal box chamber,
75 mm× 75 mm× 350 mm 75 mm× 75 mm× 350 mm 75 mm× 75 mm× 350 mm
electrodes
tapered tungsten, tapered stainless steel, tapered stainless steel,
1 mm diameter 2 mm diameter 2 mm diameter
fuel specification aviation kerosene aviation kerosene six liquid fuels
equivalence ratio, φ 0.35-1.05 0.3-0.9 0.4-1.9
prevapourised fraction, Ω neglected 0.15 neglected
air temperature (K) ambient 228-350 290
air pressure (bar) ambient 0.2-0.9 0.2-1.0
air velocity (m/s) 19.0-49.5 9-38 15-40
turbulence intensity (%) not reported not reported 2.5-15
means of atomisation five simplex atomisers
simplex, acoustic, and
twelve simplex atomisers
spinning cup atomisers
SMD (µm) 30-80 20-180 40-150
quenching distance (mm) not determined not reported 2-22
optimum spark duration (µs) 30-80 30-100 40-100
Emin (mJ) 3-90 2-800 0.02-1000
2. Background 38
detachable rod and ball capacitors mounted on the high-voltage terminal. Ignition
tests were performed with mixtures of oxygen, nitrogen and a variety of hydrocarbon
fuels, mainly alkanes. The results of hydrogen ignition tests were also published at
a later date (Lewis and von Elbe 1961). By replacing the nitrogen component of air
with individual inert gases, it was possible to assess the flame-quenching effects of
helium, argon and carbon dioxide.
Following adjustment of the spark gap to the required distance, the bomb was
filled with an explosive mixture of defined composition and pressure. The electrode
system was slowly charged and the breakdown voltage, V , noted. Ignition was
confirmed by visual observation of a flame. If flame propagation did not occur, then
the capacitance, C was incrementally increased until a developing flame kernel was
produced. The minimum ignition energy was then identified as the energy stored in
the ignition circuit prior to the spark discharge given by
Emin =
1
2
CV 2. (2.37)
Typical plots of minimum ignition energy versus electrode separation indicated
that for large electrode separations, increasing spark gap distances require greater
ignition energies. For well separated electrodes, heat generation occurs throughout
a comparatively large volume, and increasing energies are therefore required to pro-
duce the high-temperature, initial kernel necessary to ensure flame development. As
the spark gap width is reduced below the quenching distance, an increasing propor-
tion of the deposited energy is drained from the kernel by conductive heat transfer
to the electrodes. With the flanged electrodes mounted, this produced a sharp rise
in ignition energy requirements. Free electrode tips showed a more gradual increase
in minimum ignition energy as the spark-gap width was reduced below the critical
separation.
For mixture pressures exceeding 0.33 bar, the minimum ignition energy is con-
stant for mid-range values of electrode separation. Though the cause of these energy
plateaux remains unclear, they may represent a fundamental lower energy limit as-
2. Background 39
sociated with chemical and physical properties of the system, independent of elec-
trode quenching characteristics. If true, this implies that electrode quenching may
result in overestimates of ignition energy for configurations producing no observable
plateaux.
Minimum ignition energies corresponding to the lowest points on each energy-
separation curve were recorded for a variety of gaseous mixtures and total pres-
sures. The results are generally consistent with the trends observed by Calcote
et al. (1952). Both the quenching distance and minimum ignition energy increase
with reductions in pressure and oxygen concentration. Remarkably, the energy curve
minima for a variety of fuel-air mixtures at atmospheric pressure consistently fell
within a narrow range between 0.23 and 0.28 mJ. It was also noted that the minima
shifted to fuel-rich mixtures as the number of carbon atoms in the fuel increased.
These features reflect variations in mixture flame speed with equivalence ratio.
As is true for surface-discharge sparks, much of the electrical energy delivered
between the electrodes of an air-gap spark does not produce heating of the com-
bustible mixture, but is instead transferred away by radiation, shock wave forma-
tion, and conductive heat losses along the electrodes. Rose and Priede (1959) have
demonstrated that parameters such as the spark-gap geometry, electrode material
and discharge interval play a significant role in determining the magnitude of these
losses.
Ballal and Lefebvre (1977a, 1977b) extended this research to examine the ignition
behaviour of flowing, turbulent gaseous mixtures. The ignition test section was
mounted in a closed-circuit wind tunnel, allowing sub-atmospheric circulation of the
combustible mixtures at ambient temperature. Several interchangeable perforated
plates mounted upstream of the working section generated near isotropic turbulence
in the ignition zone. The fuels studied were methane and propane, with nitrogen
in the air partially replaced by oxygen or totally replaced by various inert gases
(CO2, He, Ar). A wide variation was thus achieved in the physical and chemical
2. Background 40
properties known to affect quenching distance and minimum ignition energy, namely
thermal conductivity, pressure, specific heat capacity, laminar flame speed and flame
temperature (see Eq. (2.25)). Two high voltage probes recorded traces of electrode
current and voltage as a function of time. The spark energy was then evaluated by
integrating the product of these values over the spark duration.
The required mixture was circulated at the appropriate pressure and velocity and
the spark energy was gradually increased until the onset of ignition, confirmed by the
appearance of flame. The spark-gap width was incrementally adjusted to generate a
plot of ignition energy versus electrode separation. The plateaux observed by Blanc
et al. (1947) at the base of each curve was not evident for the low pressure data
presented. The quenching distance was therefore defined as the electrode separation
at the point of minimum ignition energy. The spark duration was also optimised at
each test condition.
Initial experiments to confirm the ignition behaviour of quiescent propane mix-
tures at sub-atmospheric pressures showed quenching distances in good agreement
with previous data. However, the minimum ignition energies reported were sub-
stantially lower than suggested by earlier data, possibly due to the authors’ efforts
to optimise spark duration. All other parameters being equal, larger flow velocities
increased the minimum ignition energy requirements. This was attributed to in-
creasing turbulence intensity rather than to the mean flow rate; though turbulence
raises the burning velocity, the rate of heat loss from the developing spark kernel
also increases, and it is this effect that dominates. In line with trends observed for
quiescent gases, the minimum ignition energy of flowing mixtures was significantly
reduced by relatively small increases in pressure and oxygen concentration. Further-
more, ignition energy minima coincided with points of maximum burning velocity
for a range of fuel and oxygen mass fractions, suggesting that for flowing mixtures,
transport processes are more significant than preferential diffusion effects.
In addition to laying foundations later incorporated into a general model of spark
2. Background 41
ignition, this work provided two informative insights into the spark ignition process.
Firstly, having derived a theoretical expression for the kernel critical diameter as-
suming a linear variation with flame thickness, empirical minimum ignition energy
data were observed to vary with the cube of critical diameter. This relationship was
demonstrated for a wide range of mixture composition and turbulence intensity.
Secondly, contrary to the earlier assumption of Lewis and von Elbe (1961), Ballal
and Lefebvre demonstrated that the kernel critical diameter and quenching distance
are not equal, but proportional to each other. For the gaseous mixtures examined,
the critical diameter corresponded to approximately 20% of the quenching distance.
2.4.2 Quiescent droplet suspensions
A central feature of heterogeneous ignition is that flame propagation may result from
one, or a combination of several distinct physical mechanisms. Aggarwal (1998) has
identified three ignition modes for two-phase fuel-air mixtures. Individual droplet
ignition (mode 1) is characterised by the development of a diffusion flame surround-
ing, and of similar characteristic dimension to an isolated droplet. In contrast, spray
ignition (mode 3) is associated with the propagation of a sheath flame, several or-
ders of magnitude larger than any drop. The droplet cluster mechanism (mode 2)
represents an intermediate regime, providing a link between the local droplet and
global sheath ignition modes. While the dominant burning mechanism has impor-
tant effects on pollutant formation (Rah et al. 1986), the variation in flame speed
between these modes is highly significant when assessing ignition energy trends.
In a seminal paper, Burgoyne and Cohen (1954) examined the influence of droplet
size on both flame propagation modes and rates in aerosol mixtures. A novel ap-
paratus was developed to generate monodisperse suspensions of Tetralin (tetrahy-
dronaphthalene) by controlled condensation. Following dilution of the aerosol with
appropriate quantities of air, a conical inverted flame was established at the base of
a cooled burner tube. The burning velocity was calculated for a range of drop sizes
2. Background 42
by measuring the suspension flow rate and the flame geometry.
The authors observed a transition in the flame propagation process as the droplet
size was increased from 7 to 55 µm. For droplet diameters below 10 µm, vaporisation
and mixing of fuel with the surrounding air occurred prior to the reaction zone,
producing a coherent blue flame front. As the droplet size increased beyond the full
vaporisation limit, white combustion centres became evident, resulting from diffusion
burning of individual droplets. Above 20 µm there appeared insufficient preliminary
vaporisation to sustain a coherent front, with the flame forming a brush-like spray
of yellow centres, lacking visible interconnection.
In light of these observations the authors proposed that convection processes
facilitate the particulate mode of propagation, generating a solid network of flame
passing through the mixture, whilst leaving appreciable cells of relatively cool excess
air. For droplet dimensions exceeding 30 µm, experimental results indeed confirmed
that the transmission of flame between droplets was possible over distances of up
to 30 diameters. Experimental data also suggested that a significant increase in
flame propagation speed with droplet diameter occurs during the transition from a
fully-vaporised flame to combustion of isolated droplets. For instance, maintaining
the same mass concentration of fuel, droplet enlargement from 9 to 30 µm produced
a two-fold increase in the burning velocity.
Polymeropoulos and Das (1975) have reported a similar trend for polydisperse
kerosene-air mixtures. A peak in burning velocity was noted for a droplet size dis-
tribution centred at 30 µm, with a decrease in propagation speed for larger drops.
Richards and Lefebvre (1989) also noted an optimum droplet diameter while investi-
gating flame development in monodisperse, rich, kerosene-air mixtures with droplet
sizes of less than 110 µm. At an overall equivalence ratio of 1.46, the authors re-
ported a peak in burning velocity corresponding to a mixture droplet diameter of
approximately 45 µm.
To measure the effect of droplet size on minimum ignition energy, a fuel air mix-
2. Background 43
ture must be generated containing droplets that are the same size, or monodisperse.
Considerable experimental challenges have been overcome to create monodisperse
aerosols containing droplets ranging from 10 to 200 µm in diameter. The droplet
size parameter most widely employed to characterise liquid sprays is the Sauter
mean diameter. This dimension represents the diameter of a drop with a volume to
surface area ratio equal to that of the whole spray. Thus
SMD =
∑N
i=1D
3
i∑N
i=1D
2
i
, (2.38)
where N is the total number of droplets and Di is the diameter of an individual
droplet. Though a range of mean diameters can be formulated by averaging droplet
diameters, surface areas, and volumes, Lefebvre (1989) notes that the SMD alone
offers a reliable indication of atomisation quality for combustion applications. In-
vestigating the ignition behaviour of a bidisperse spray, Dietrich et al. (1990) have
shown that for a polydisperse mixture the SMD is the only droplet size parameter
that reflects the increase in ignition energy with droplet diameter expected for an
evaporation-controlled process.
Clearly no mean dimension alone can uniquely represent the partition of mass
in a fuel spray. However, a single additional parameter representing the drop size
dispersion can often be combined with a representative diameter to describe the
distribution more completely. The most widely employed expression for drop size
distribution was devised by Rosin and Rammler (1933) and may be expressed as
Q = 1− exp
[
−
(−D
X
)q]
, (2.39)
where Q is the fraction of the total fuel volume contained in droplets of diameter
less than D, X is the representative diameter, and q is the dispersion parameter. As
the dispersion parameter increases, the distribution of droplet size becomes more
uniform. This dispersion parameter lies between 1.5 and 4 for most atomisers,
though values exceeding 11 have been achieved using some rotary designs.
2. Background 44
Having demonstrated means of creating heterogeneous fuel-air mixtures con-
taining droplets of well defined size and distribution, Ballal and Lefebvre (1978a)
investigated the influence of equivalence ratio, pressure and droplet size on the min-
imum ignition energy of quasi-stationary fuel mists. Fuel volatility effects were also
measured following earlier work highlighting the significance of fuel properties in de-
termining ignition performance. Postulating that successful ignition occurs when the
mass of fuel vapour generated by the spark exceeds a threshold value, the atomiser
and spark electrodes were positioned in close proximity to create a fuel-air mixture
containing minimal fuel vapour. The mixture velocity between these locations was
minimised, subject to ensuring negligible fuel evaporation.
Though mounted vertically, the test section was virtually identical to that de-
signed for earlier studies of homogeneous ignition (see Section 2.4.1). Glass flanges
were mounted on the electrodes to minimise variations in the spark breakdown volt-
age due to fuel wetting. A spinning cup atomiser was located a small distance
upstream of the test section, producing an essentially monodisperse, falling spray,
while allowing independent control of fuel flow rate and droplet size. The mixture
SMD was measured using a laser scattering technique described by Lorenzetto and
Lefebvre (1977). The equivalence ratio was calculated from the total flow rates of
air and fuel into the ignition zone. All mixtures tested were sub-stoichiometric, and
the significance of evaporation processes during ignition was investigated by testing
six fuels with different evaporation properties. Care was taken to ensure that val-
ues of minimum ignition energy were recorded following optimisation of both spark
duration and electrode separation.
In accordance with earlier findings for homogeneous mixtures, quenching distance
and minimum ignition energy were shown to increase with reductions in both pres-
sure and equivalence ratio. The minimum ignition energy was also found depend
strongly on droplet size, increasing with the cube of drop diameter. Predictably,
greater energies were required to ignite mixtures containing fuels with a compara-
2. Background 45
tively low volatility.
Adopting an approach similar to that advocated by Peters and Mellor (1980),
the authors proposed a model for the ignition of quiescent droplet mists. They as-
sumed that ignition success is governed solely by the fuel evaporation rate. This
model agreed well with experimental results, strengthening the assertion that fuel
evaporation is the rate-controlling step for a diverse range of conditions. This evap-
orative control was confirmed by supplementary experiments designed to produce
wide variations in flame temperature, thereby revealing any kinetic dependence. For
the heterogeneous mixtures examined, the critical diameter corresponded to approx-
imately 80% of the quenching distance; significantly larger than the value of 20%
reported for homogeneous mixtures.
Though offering pioneering insight into the ignition of fuel mists, the study of
Ballal and Lefebvre (1978a) was limited to lean mixtures containing relatively large
droplets. In addition, a single value of the minimum ignition energy was reported
for each test condition. Several researchers have noted that ignition is an innately
stochastic process. Consequently, at a specific test condition ignition is possible for
a range of spark energies, albeit with different probabilities of success. Singh and
Polymeropoulos (1986) have suggested that these probabilistic features result from
differences in the shape and position of the spark discharge channel, coupled with
spatial non-uniformities in the droplet distribution.
Working independently, Singh and Polymeropoulos (1986) and Danis et al. (1988)
investigated the ignition behaviour of both lean and rich monodisperse mixtures.
Though examining dissimilar fuels and using different means of atomisation, these
studies shared several common features. Both researchers characterised droplet size
by means of laser diffraction techniques. In addition, mixture equivalence ratio
was regulated by the addition of oxygen or air to the droplet suspension. Igni-
tion energies were evaluated at ambient conditions throughout, and standard spark
discharge apparatus was used, allowing independent control of spark energy and
2. Background 46
duration. Ignition energies were principally recorded following optimisation of both
spark duration and electrode separation.
Singh and Polymeropoulos (1986) investigated the influence of oxygen concen-
tration, equivalence ratio and droplet size on the spark energy required to ignite
Tetralin aerosols. Utilising the technique developed by Burgoyne and Cohen (1954),
droplet generation was achieved by controlled condensation of fuel vapour on sodium
chloride nuclei. Metered oxygen and air flows were introduced, inhibiting droplet
aggregation, whilst providing control of the overall equivalence ratio. Isokinetic
sampling at the spark gap confirmed these values.
Maintaining constant oxygen concentration and drop size, increasing spark en-
ergy predictably produced a higher probability of ignition success. However, this
improvement in performance approached an asymptotic limit that was often con-
strained to a comparatively low ignition probability. This was apparent for a 6.7 µm
mixture, where the success rate never exceeded 20% for spark energies less than
50 mJ. It was also evident that the maximum ignition probability for a given spark
energy occurred for a specific droplet diameter. The measurements indicated an
optimum droplet size of between 22 µm and 26 µm. In line with the change in
equivalence ratio, the increase in oxygen concentration produced improvements in
ignition performance due to the accompanying increase in flame velocity.
A model was also developed to estimate the non-reactive fuel vapour concentra-
tion and droplet motion generated following a spark discharge. This analysis was
based on earlier observations that rapid compression of a gaseous medium during the
early stages of an electrical discharge generates two primary features: a high-speed
shock wave and a constant-pressure hot gas kernel. Results suggested that the fuel
vapour concentration generated at the kernel centre was close to stoichiometric for
sprays containing droplets 24 µm in diameter. Larger drops produced a leaner mix-
ture while the vapour equivalence ratio exceeded 3 for a droplet diameter of 6.7 µm.
The model also indicated that smaller droplets are displaced radially by gas motion
2. Background 47
produced by the expanding shock wave, creating a fuel lean annular region around
the kernel. It was suggested that this fuel free region inhibits flame growth, resulting
in low ignition frequencies for droplet diameters of less than 10 µm.
Further evidence supporting the optimum droplet size hypothesis was provided
by Danis et al. (1988). Varying both equivalence ratio and droplet size, this study
compared ignition energy requirements for n-heptane and methanol sprays with
those for similar prevapourised mixtures. A vibrating orifice aerosol generator was
used to generate extremely uniform sprays, with the standard deviation in droplet
diameter typically less than 1% of the mean value.
Ignition events again appeared highly stochastic. For example, a lean n-heptane
spray containing 53 µm droplets required an order of magnitude increase in spark
energy to raise the ignition probability from 20% to 85%. The minimum ignition
energy was arbitrarily defined as the spark energy corresponding to an ignition
frequency of 50%. A plot of minimum ignition energy versus equivalence ratio
for prevapourised n-heptane corresponded closely to that reported by Lewis and
von Elbe (1961), suggesting an optimum equivalence ratio of between 1.5 and 2.
An optimum droplet size of between 10 and 30 µm was reported, and the lean
ignition limit was extended by the presence of droplets. This phenomenon was also
observed by Cohen et al. (1996) while examining Tetralin mixtures, and is believed
to result from regions of favourable gas-phase stoichiometry that form locally around
evaporating droplets.
2.4.3 Flowing heterogeneous mixtures
The minimum energy required to initiate a flame kernel in a gas turbine combustion
chamber depends on the flow field adjacent to the igniter. Recognising this, Ballal
and Lefebvre (1978b, 1979) extended the scope of their generic studies to assess
ignition energy variations with local flow velocity and turbulence intensity. The
influence of air pressure, droplet diameter and equivalence ratio on minimum ignition
2. Background 48
energy was again examined for flowing two-phase mixtures, with air temperature
effects also assessed.
The test facility remained largely unmodified throughout the experimental mea-
surements, incorporating a test section virtually identical to that designed for earlier
gaseous ignition studies (Ballal and Lefebvre 1977a). Fuel was admitted to the test
section through an interchangeable assortment of simplex, acoustic and spinning
cup atomisers mounted upstream of the test section. The principal fuel investigated
was aviation kerosene, though five hydrocarbon fuels of differing volatility were also
introduced during the later stages of the study. Droplet diameter was recorded for
the full range of fuel and air flow rates using the light-scattering technique developed
for quiescent aerosol measurements.
Prior to ignition testing, isokinetic sampling of the mixture in the spark gap
allowed calculation of the equivalence ratio for a range of flow conditions. To reduce
the risk of damage to the apparatus the prepared mixtures were all lean. Though
initial testing was conducted at ambient conditions, an altitude test facility was later
incorporated, generating air temperatures and pressures as low as 228 K and 0.2 bar
respectively. Physical conditions at the electrodes thus reflected those encountered
at altitudes in excess of 30000 ft (see Figure 2.3). A flow conditioner installed imme-
diately upstream of the fuel injector was later replaced by a gauze grid, generating
near isotropic turbulence in the ignition zone.
A preliminary investigation revealed those mixture parameters influencing spark
discharge energy and optimum duration. Maintaining a fixed spark current, a con-
siderable increase in discharge voltage and thus spark energy was observed for mix-
tures flowing at higher velocities. It was suggested that the associated increase
in discharge pathway resistance, resulted from ‘stretching’ of the spark during the
discharge period. This is plausible given that a 100 µs spark is displaced through
5 mm in a mixture flowing at 50 m/s. Though unsubstantiated, the authors also
proposed that this extension of the spark pathway reduced conductive heat losses
2. Background 49
to the electrodes. It was incidentally noted that the breakdown voltage required for
liquid fuel sprays was considerably higher than that necessary in gaseous mixtures.
This was reportedly due to the inevitable deposition of liquid fuel at each electrode
tip.
At the operating conditions considered, the optimum spark duration increased
with droplet diameter and decreased with flow velocity. These results are arguably
unsurprising given the finite time required for fuel evaporation together with the
need to ensure highly localised heat deposition. With the exception of the earliest
study, which was conducted with a fixed spark gap, values of minimum ignition
energy were recorded following optimisation of both spark duration and electrode
separation.
Results confirmed that higher equivalence ratios and fuel volatility promote suc-
cessful ignition. The dependence of minimum ignition energy on droplet size was
found to be stronger for heterogeneous flowing mixtures than for quiescent sprays:
Emin ∝ D4.5 compared to Emin ∝ D3. Furthermore, despite following consistent
trends, the measured droplet size exponent appeared to increase as the studies pro-
gressed. Ballal (2004) later suggested several possible reasons for these inconsisten-
cies. Early kerosene-air studies used an assortment of simplex, acoustic and spinning
cup atomisers to generate the required range of droplet diameters. Later research,
investigating the ignition behaviour of multiple fuels, employed 12 carefully matched
simplex atomisers producing consistent drop size distributions and minimising liquid
fuel deposition at the duct walls. Perhaps more critically, as research progressed,
the authors recognised the need to maintain constant turbulence intensity while
examining the influence of other variables. Only during the later studies could it
be argued that independent control of equivalence ratio, fuel volatility, droplet size,
flow velocity, pressure and turbulence intensity had been maintained.
In agreement with the behaviour reported for gaseous mixtures, the minimum
ignition energy was found to increase with turbulence intensity, and, as before, the
2. Background 50
influence of mean velocity on spark energy did not appear to extend beyond those
effects resulting from this increased turbulence intensity. The detrimental impact
of low air temperatures on ignition performance was also apparent. For a stoi-
chiometric mixture the decrease in temperature associated with high-altitude flight
required a three-fold increase in ignition energy, independent of pressure changes.
A similar deterioration in ignition performance with reductions in pressure was also
confirmed for a range of fuels. The measured decrease in minimum ignition energy
with pressure appeared more pronounced than predicted by theory. The nature of
this dependence appears consistent with the general form
Emin ∝ 1
P n
. (2.40)
Experimental ignition data acquired during the later studies suggested that the pres-
sure exponent, n, was between 1 and 1.3, varying with equivalence ratio. Theoretical
analysis suggested a much weaker dependence, with the exponent equal to 0.5. This
inconsistency derived from an inaccurate assumption associated with the physical
model employed.
Extending concepts developed for the analysis of quiescent fuel mists, the de-
vised model proposed that ignition of a heterogeneous mixture is governed solely by
the rate of fuel evaporation, remaining independent of chemical reaction rate for a
wide range of flow conditions. Good agreement was obtained between measured and
predicted values of minimum ignition energy for intermediate to large droplet diam-
eters at atmospheric pressure. As highlighted above however, theoretical projections
proved less accurate for conditions producing similar rates of evaporation and chem-
ical reaction; specifically low equivalence ratio mixtures of fine fuel droplets and
low-pressure air. Further limitations were apparent for moderate to high turbulence
levels. It was suggested that these conditions promoted fuel vapour formation prior
to ignition, leading to overestimation of the minimum ignition energy if Eq. (2.31)
is used.
Comparison of critical diameter expressions for flowing homogeneous and het-
2. Background 51
erogeneous mixtures produced expressions for a critical mean droplet size, below
which kinetic effects predominate. For a stoichiometric, monodisperse suspension
of kerosene droplets in ambient air, critical droplet sizes of 20 µm and 36 µm were
derived for quiescent and flowing mixtures respectively. This theoretical work con-
tributed significantly to the development of a general model for spark ignition, incor-
porating terms to account for both chemical kinetic effects and fuel vapour formation
prior to ignition. Including the chemical reaction time for conditions of low equiva-
lence ratio and pressure yielded a marked improvement in correspondence between
theory and experiment.
2.4.4 Laboratory-scale relight experiments
While many studies have investigated the creation of a viable spark kernel, relatively
little research has examined the migration and subsequent stabilisation of flame
during the second phase of the altitude restart process. These have been explored
by Mastorakos and co-workers in several studies using simplified flow configurations
that incorporate key aspects of the relight process. These studies are summarised
below.
In a simple experiment Ahmed and Mastorakos (2006) investigated spark ignition
of a turbulent methane jet in a concentric air-flow. The air-gap spark igniter used
to initiate ignition could be positioned at any point in the flow. The developing
flame was visualised using a high-speed camera and PLIF of the OH radical. The
flame kernel created by the spark was initially spherical, but quickly evolved into
a cylinder with a propagating edge that moved upstream. If successful, this flame
stabilised at an appropriate liftoff height. The propagation speed of this edge flame
relative to the flow was between three and six times the laminar burning velocity,
increasing with jet velocity.
Repeated ignition trials were conducted at a variety of spark locations, and the
ignitable region of the jet identified. The gas velocity between the electrodes at the
2. Background 52
instant of the spark was found to strongly influence ignition outcome, with higher
velocities leading to failure. The mean mixture fraction distribution, or flammability
map, was estimated using an empirical correlation. The ignitable region was found
to be somewhat wider than the flammable region, based on mean mixture fraction
contours, indicating that turbulent fluctuations of mixture fraction can cause a lean
region to become instantaneously enriched. The convection of developing kernels
away from the stabilisation point resulted in an ignition probability of zero in the
downstream portion of the jet, despite a favourable mean mixture fraction.
A difference in the areas covered by the flammable and ignitable regions was
also observed in a study of turbulent non-premixed counterflow flames (Ahmed
et al. 2007b). These results indicated that ignition was possible when the spark
was positioned at locations where the mixture fraction lay outside the rich and lean
flammability limits. In these cases the mechanism permitting ignition was strong
convection of the hot gas kernel from the spark gap to a region of flammable mixture
up to 10 mm away.
Ahmed et al. (2007a) designed and constructed a turbulent non-premixed bluff-
body burner to study the ignition process in a complex flow field more accurately
representing that produced in a gas turbine combustor. A conical bluff body was
mounted at the burner exit, and the flame area was enclosed by a quartz cylinder
with a diameter of 70 mm. Methane was injected radially into a concentric air-flow,
creating a well-defined central recirculation zone. The velocity and mixture fraction
fields were determined by means of laser Doppler anemometry (LDA) and acetone
PLIF respectively, while the ignition events were again visualised using a high-speed
camera and OH PLIF.
Without induced swirl, the mixture fraction in the CRZ was found to be uniform
and steady. When the spark occurred inside this zone, successful flame propagated
towards the bluff-body edges where it stabilised. When the spark electrodes were
located outside the CRZ, successful flame first propagated axially before expand-
2. Background 53
ing radially and tangentially to the opposite side of the quartz cylinder. Perhaps
unsurprisingly, the highest rate of successful ignition occurred when the spark was
located close to the stoichiometric mixture fraction contour at the edge of the CRZ.
However, ignition was still achieved inside the CRZ where the mixture was appar-
ently rich. In this case the discrepancy between the flammable and ignitable areas
was attributed to local and global fluid mechanics effects and to the comparatively
high turbulence levels. Three types of ignition failure were observed corresponding
to a spark generating no kernel, kernel initiation followed by subsequent blow-off,
and global extinction following successful stabilisation.
Introducing swirl to the air-flow changed the fuel distribution markedly, with
the mixture becoming too lean inside the CRZ while mixing improved around the
edges of the cylinder. The stability limits were expanded considerably by the swirl,
and the ignition and stability limits became closer. Under these conditions a flame
kernel generated by a spark at the edge of the quartz cylinder was seen to move
tangentially with the flow.
By adding an atomiser to the upstream face of the bluff body it was possible
to investigate ignition of a liquid fuel, namely n-heptane, in the same recirculating
flow (Marchione et al. 2007, July). As before, the highest rate of successful ignition
occurred when the spark was located close to the the edge of the CRZ. Phase Doppler
anemometry (PDA) measurements revealed that the mean droplet size decreased
considerably along the burner centre-line indicating that large droplets had impinged
on the quartz cylinder due to the swirling motion of the flow. Inside the CRZ, the
droplets moved away from the injector, against the direction of the mean air velocity.
Ignition tests revealed that flame can successfully propagate back towards the burner
face under these conditions.
In a separate study undertaken by Marchione et al. (2007, Sep), the influence of
multiple sparks on the ignition of recirculating sprays has been assessed. The sparks
generated were large and long compared to the length and time scales of the flow.
2. Background 54
Sparks with an energy of 30 mJ and a duration of approximately 8 ms were repeated
at a rate of 100 Hz for periods of between 1 and 5 s. A long spark sequence was
found to make ignition possible at a test condition where a single spark consistently
failed. This was attributed to the heat deposited during early sparks increasing the
ignition probability of later sparks. Furthermore, the likelihood of ignition was a
strong function of the igniter’s axial location at the edge of the quartz cylinder.
Flame kernels generated by the spark were stretched as far as 20 mm from the
electrodes by the turbulent motion of the local fluid. This work thus indicated that
the energy deposited in multiple sparks can have a spatially far-reaching effect that
can significantly increase the likelihood of ignition.
In summary, these laboratory-scale experiments have contributed valuable in-
sight into ignition and flame development processes in gaseous and liquid-fuelled
burners operating at atmospheric conditions. It remains to be demonstrated that
these phenomena occur inside gas turbine combustion chambers operating at the low
temperatures and pressures encountered during high-altitude relight. The current
work will clarify the nature of these processes in an actual combustor.
Chapter 3
Experimental Method
The objective of the experimental work performed in this study is to characterise
the fuel distribution and flame development inside a gas turbine combustor dur-
ing high-altitude relight. These have been examined using two imaging techniques
simultaneously: kerosene PLIF to map fuel placement (see Section 3.2), and high-
speed flame emission imaging to record flame activity (Section 3.3). This chapter
introduces the techniques and apparatus employed to investigate relight phenomena.
3.1 The altitude test facility
The Altitude Test Facility at Rolls-Royce, Derby, was used to simulate the operat-
ing conditions generated inside an aeronautical gas turbine combustor following an
altitude flameout. Significant modifications to the ATF were necessary to permit
high-resolution fuel and flame imaging, to be discussed later in Section 3.2. Firstly,
a general description of the layout and operating conditions are given.
3.1.1 Layout
The modified test section is illustrated in Figures 3.1 and 3.2. A two-sector com-
bustion chamber was mounted inside a pressure vessel. The combustion chamber
55
3. Experimental Method 56
was a LDI design housing a single fuel injector in the far-side sector. Though the
injector was fuel staged, only the pilot atomiser was fuelled during relight tests. The
Jet A kerosene fuel fed through the injector was stored at ambient conditions, and
supplied in controlled quantities via a mass flow controller.
Fig. 3.1: Photograph of experimental apparatus. Flow is from left to right.
A refrigeration unit upstream of the combustor chilled the inlet air. This cold
air was drawn through the combustion chamber by a vacuum pump. A drilled plate
at the outlet of the combustion chamber restricted the air mass flow rate through
the test section. The air pressure and mass flow rate in the combustion chamber
were independently regulated by adjusting the pressure drop generated by the pump
and the air flow through an upstream control valve. In this manner, it was possible
to replicate the conditions inside a lean-burn combustor prior to relight. The two
condition set points used during this study are discussed below in Section 3.1.2.
Optical access to the combustion chamber was provided by two quartz windows
3
.
E
x
p
erim
en
ta
l
M
eth
o
d
57
.........................................
..... flow
inside
outside
near sidefar side
Fig. 3.2: The modified test section. The combustion chamber is shown in red, the igniter in black, the periscope in blue,
and the pressure vessel in grey. The internal and external periscope covers are omitted.
3. Experimental Method 58
on the near side of the test section. The internal window, mounted in the side of
the combustor, was susceptible to liquid fuel contamination prior to and following
ignition. The near-side injector was therefore replaced by a vertical slot that created
an air curtain to prevent fuel deposition on the combustor window. The slot area
was adjusted to match the effective area of the fuel injector, thus ensuring an equal
air mass flow rate into the combustor at each injector location. Air-flow through the
injector and the slot together represented approximately 70% of the total combustor
air-flow (not including the additional purge air flows described in Section 3.2.5). In
addition, the external pressure-vessel window was warmed to prevent condensation
on its outside surface.
The sparks required for ignition were generated by a Champion SK00241 surface-
discharge igniter. Introduced from above, the igniter tip was located flush with the
outside combustor wall in the injector mid-plane. The energy required to generate
each spark was supplied by a Unison Trent exciter unit rated to deliver 10 J per
spark at a frequency of approximately 1.3 Hz.
3.1.2 Operating condition set points
Ignition testing was conducted at two high-altitude operating conditions referred
to as ‘safe’ and ‘unsafe’ in accordance with the likelihood of achieving ignition;
relight is possible at the safe condition but impossible at the unsafe set point. The
conditions generated inside the combustion chamber at these two set points are
listed in Table 3.1. The safe point corresponds to the conditions inside a lean-
burn combustor prior to relight, when the aircraft is flying at 30000 ft (9.1 km)
above sea level. These conditions were calculated using the procedure described in
Section 2.2.1 to predict the thermodynamic state of the cold combustor air as a
function of altitude and air speed.
The sole difference between the two operating conditions is the larger air mass
flow rate at the unsafe point. This influences the ignition process in two important
3. Experimental Method 59
Table 3.1: Safe and unsafe condition set points for relight testing.
safe unsafe
Tair (K) 265.0 265.0
Pair (kPa) 41.4 41.4
Tfuel (K) 288.0 288.0
AFR 33.0 60.5
respects. Firstly, as the air flow rate through an air-blast atomiser increases, a
finer fuel spray is generated (see Section 2.2.1). The minimum energy required to
ignite a droplet spray increases markedly with drop size (Rao and Lefebvre 1976;
Ballal and Lefebvre 1978a). A larger air-flow might therefore be expected to favour
ignition. However, flame stabilisation is impaired by the reduction in residence time
and increase in aerodynamic strain associated with high flow rates (Cohen et al.
1996). These combined effects enable ignition within fourteen sparks at the safe
condition, while preventing all successful flame development at the unsafe set point.
3.2 PLIF
3.2.1 Background
The distribution of fuel inside the ATF facility has been studied using planar
laser-induced fluorescence of kerosene fuel. This technique uses a planar sheet of
monochromatic light from a laser to excite fluorescent compounds present in the
fuel. When a molecule of one of these compounds absorbs an excitation photon,
its electronic energy is raised from a ground state to a higher level. In a process
known as Stokes fluorescence, some of these excited molecules relax to intermedi-
ate energy levels, emitting photons with a longer wavelength than the laser light.
By imaging these fluorescence emissions, a two-dimensional representation of the
fuel distribution in the sheet plane is recorded. This non-intrusive technique has
3. Experimental Method 60
been used previously to characterise fuel placement inside internal combustion en-
gines (Arnold et al. 1992; Nygren et al. 2002) and various gas turbine combustion
systems operating at simulated take-off or cruise conditions (McDonell et al. 1996;
Locke et al. 1998; Kohse-Ho¨inghaus and Jeffries 2002). Fuel PLIF has not, however,
been used previously to investigate the altitude relight process.
Numerous commercial fuels, including Jet A kerosene, contain a wide variety
of unsaturated and aromatic hydrocarbons that readily absorb ultraviolet light
and generate strong fluorescence. Examples of such compounds include the mono-
aromatics, benzene and toluene, the di-aromatic, naphthalene, and the polycyclic
aromatic hydrocarbon, fluoranthene. Each of these components generates a flu-
orescence signal that varies with the concentration of the fluorescing species, to-
gether with other factors such as the temperature, pressure and composition of
the surrounding gases. Furthermore, though aviation fuel standards in the United
Kingdom and the United States specify the maximum permissible aromatic content
(UK Ministry of Defence 2002; ASTM 2008), significant variation in the concen-
tration of these fluorescent compounds can occur from batch to batch (Goodger
1994). A well-characterised fluorescent compound is therefore frequently added to
a non-fluorescing fuel during quantitative studies of fuel concentration inside prac-
tical combustors (Schulz and Sick 2005). In the current study the goal is solely to
provide a qualitative comparison of the two test conditions, and such measures are
consequently unnecessary provided that the fuel used throughout testing is drawn
from a single batch.
The low air temperatures at relight conditions coupled with the low vapour pres-
sure of ambient Jet A kerosene (Coordinating Research Council 2004), suggested
that the fuel imaged inside the combustion chamber would be largely liquid. The
optical performance of the PLIF system was therefore specified according to the flu-
orescence behaviour of liquid Jet A kerosene. Previous spectroscopic studies indicate
that Jet A fuel fluoresces strongly when excited at 266 nm (Black 1995; Lo¨fstro¨m
3. Experimental Method 61
et al. 1996). This wavelength can be readily generated by using non-linear crystals
to quadruple the fundamental output of an Nd:YAG laser (1064 nm). Though slight
differences are evident, the emission spectra reported in the work of Lo¨fstro¨m et al.
(1996) and Black (1995) are remarkably consistent, with the signal extending from
approximately 300 to 450 nm and containing a single peak at 340 nm. Given the
possible batch-to-batch variations in composition noted above, a kerosene sample
was drawn from the ATF fuel tank prior to testing, and analysed using a spectroflu-
orometer. The emission spectrum generated by excitation at 266 nm is plotted in
Figure 3.3, and shows good agreement with the profiles reported previously.
×105
wavelength (nm)
in
te
n
si
ty
(a
rb
it
ra
ry
u
n
it
s)
..........................
..........
..............
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
......
...
.
...................................
............................................................................................................................................................................................
.............................................
....................................................................
250 300 350 400 450 500 550
0
1
2
3
4
5
6
Fig. 3.3: Fluorescence spectrum of Jet A fuel when excited at 266 nm.
While the emission spectrum of the fuel is clearly required for specification of
the imaging system, the absorption properties of Jet A are also critical to the suc-
cess of this work. Measurements reported in Coordinating Research Council (2004)
suggest that the fuel is virtually transparent to wavelengths exceeding 300 nm. Re-
absorption of emitted photons, a process known as fluorescence trapping, is therefore
3. Experimental Method 62
avoided. However, ultraviolet light at 266 nm is absorbed extremely strongly by the
fuel. This has two important implications for the present study. Firstly, it is crit-
ical that the laser sheet is introduced into the combustion chamber via a series of
clean optical surfaces, as fuel deposited on a mirror or lens can strongly attenuate
the sheet energy. Maintaining the cleanness of optics inside the ATF required the
development of several air purge protection systems, as described in Section 3.2.5.
Secondly, the laser energy may be significantly attenuated as it passes through a
dense spray. If this occurs, then the fluorescence signal does not represent the true
distribution of fuel. This phenomenon has been observed in previous studies (Mc-
Donell and Samuelsen 2000; Bazile and Stepowski 1995), and its magnitude in the
present work is assessed in Section 5.2.1.
With a knowledge of the absorption and fluorescence characteristics of undoped
Jet A kerosene, the PLIF imaging in this study was conducted as follows. The
fundamental output beam from a pulsed Nd:YAG laser was frequency quadrupled
to 266 nm and formed into a thin sheet. This ultraviolet light sheet was projected
into the region of the combustion chamber immediately below the igniter tip.
The induced fluorescence was imaged using a Princeton Instruments PI-MAX2
intensified charge-coupled device (ICCD) camera fitted with a Resolve Optics 60 mm
f/3.5 ultraviolet lens and a Hoya B-370 filter. This imaging system recorded the
broad-band fluorescence between 300 and 480 nm, registering a peak sensitivity at
370 nm. It was not possible for the camera to view the laser sheet at right angles
due to the limited optical access available. The correction required to compensate
for this viewing angle is discussed in Section 3.3.2.
3.2.2 Beam delivery configuration
The PLIF measurements of fuel distribution required the projection of a thin laser
sheet into the combustion chamber to illuminate the region immediately below the
igniter tip. As noted above, it was important to ensure that the projection optics
3. Experimental Method 63
were protected from fuel droplets and flame products. With this in mind, several
sheet projection and imaging configurations were considered.
An initial survey of the ATF suggested passing the laser sheet into the combus-
tor through a quartz window located downstream of the exhaust. This concept was
rejected, however, as the exhaust plate obstructed the line of sight to the igniter,
and the window was exposed to sooty combustion products. Horizontal projection
of the sheet through the near-side window was also discounted due to difficulties in
recording unobstructed fluorescence images from above. Furthermore, the incoming
laser sheet would have been significantly attenuated by fuel deposited on the inter-
nal, side window. This fuel contamination problem also prohibited introducing the
laser sheet from above. However, projecting the laser sheet from upstream provided
coverage below the igniter tip, while helping to maintain cool, uncontaminated op-
tical surfaces. For these reasons, the laser sheet was introduced into the combustion
chamber from upstream through a slit in the heatshield that surrounds the fuel
injector.
The path and profile of the laser beam and sheet are illustrated in Figures 3.4
and 3.5. The beam separation, attenuation, and elements of the sheet formation
were performed by optics mounted on a table positioned alongside the ATF. A de-
tailed description of this optical table is given below in Section 3.2.3. The partially-
formed sheet was projected into the combustor through a custom-built periscope,
and expanded by a planar lens mounted above the fuel injector. The design and
development of this periscope and sheet expansion assembly is discussed in Sec-
tion 3.2.4.
3.2.3 Optical table
The Nd:YAG laser required to generate the PLIF excitation wavelength was mounted
on an optical board together with the optics responsible for separating, attenuating
and geometrically modifying the beam. To reduce the required number of reflective
3. Experimental Method 64
optics and thus simplify the alignment process, the optical board was supported ap-
proximately 1.5 m above the ground by a steel frame. The optical table, comprising
this board and frame, is illustrated in Figure 3.4.
The optical table was stably supported by three adjustable feet at the base of the
frame, with the centre of mass lying well within the triangle formed by the ground
contact points. The length of each foot could be adjusted to set the table height and
to level the steel frame relative to the test facility. Two further degrees of freedom
were incorporated into the optical table to allow accurate alignment of the laser
sheet with the central plane of the far-side combustor sector. Firstly, the upper sub-
assembly of the support frame could translate parallel to the flow direction, allowing
accurate projection of the beam onto the top mirror of the periscope. Secondly, the
optical board could pivot about a horizontal axis located close to its downstream
edge. The slight inclination of the optical board apparent in Figure 3.4 (≈ 0.8◦),
was necessary to compensate for the pitch of the upstream combustor wall on which
the final sheet-formation lens was mounted.
A Continuum Surelite III-10 laser was used to generate the ultraviolet light
sheet required for kerosene PLIF. This pulsed Nd:YAG laser was operated with a
Q-switch delay of 300 µs, generating 250 mW at 10 Hz. The shot-to-shot variation
in beam energy produced by the laser is considered in Chapter 4. As illustrated in
Figure 3.4, the laser head was positioned along the far-side edge of the optical table.
The fundamental frequency of the laser (1064 nm) was quadrupled by two non-linear
optical crystals to produce light at 266 nm. As each crystal had a doubling efficiency
of approximately 50%, the laser beam at the exit aperture contained a mixture of
three wavelengths: 266, 532, and 1064 nm. It was therefore necessary to isolate the
266 nm wavelength using a pair of harmonic separators.
Having dumped the residual wavelengths, the ultraviolet beam was directed
through a variable power attenuator constructed from a half-wave plate and a thin-
film polariser. Using this arrangement, the power in the laser sheet was modulated
3. Experimental Method 65
Nd:YAG laser
harmonic separators power attenuator
mirror
sheet-formation optics (Figure B.1)
......................
...
..
...
.. to periscope (Figure 3.5)
Fig. 3.4: The optical table. The laser beam was projected from the table onto the
top mirror of the periscope (see Figure 3.5). The beam path is shown in green.
3. Experimental Method 66
by rotating the half-wave plate to direct a fraction of the highly-polarised input
beam to a dump (Kolner 1991; Kru¨ger et al. 2005). The period between ignition
tests could extend to several minutes, as large numbers of images were saved to
disk. The attenuator was therefore used to reduce the laser intensity to a minimum
during this period, thereby reducing the risk of damaging optics further along the
laser path without having to adjust the Q-switch delay.
The initial stages of forming the laser beam into a sheet were performed by three
planar lenses mounted on an optical rail. The first two lenses formed a Galilean tele-
scope that expanded the beam in the plane of the optical board. The third lens,
mounted at the end of the rail, had a comparatively long focal length (≈ 1.4 m
at 266 nm), and caused a gradual thinning of the sheet as it passed through the
periscope and into the combustion chamber. This progressive thinning approach
was used to reduce the risk of laser damage to the optical components mounted in
the periscope and final lens assembly. Dielectric-coated optics were used through-
out to maximise the proportion of transmitted or reflected light as appropriate.
The changes in beam cross-section produced by these three optics are calculated in
Appendix B, and summarised in Table 3.2.
Though elevating the optical board to the height of the periscope simplified the
beam alignment and reduced the number of mirrors required, projecting a class IV
laser beam at eye level presented a safety hazard. To avoid injury, the laser and
optics were fully enclosed throughout testing. During beam alignment, the compo-
nents mounted on the optical board required adjustment, and an elevated walkway
was provided for this purpose.
3.2.4 Periscope and sheet expansion assembly
A periscope was manufactured to direct the laser light from the optical table, through
the pressure vessel, and into the combustion chamber. In addition, a final lens assem-
bly mounted on the outside surface of the combustor provided the sheet expansion
3. Experimental Method 67
necessary to maximise the region of excitation below the igniter tip. The path of
the laser light through the modified test section is highlighted in Figure 3.5, while
details of the periscope and sheet expansion assembly are illustrated in Figure 3.6.
Fig. 3.5: Laser beam path through modified test section. Note that the unclipped
sheet extents are illustrated.
The incoming laser sheet struck the centre of the top periscope mirror, and
was reflected down along the periscope axis through a quartz window and into the
pressure vessel. The sheet continued down before being projected forwards by the
bottom periscope mirror into the sheet expansion assembly. This assembly, mounted
on the upstream outside surface of the combustor, contained a planar-concave lens
that created a diverging sheet in the centre plane of the far-side sector. This sheet
passed below the centre of the igniter tip.
The periscope and sheet expansion assembly were designed to fulfil several re-
quirements. Both assemblies were required to fit onto the existing test section,
without unduly disrupting the position of the combustor, fuel injector or igniter. As
the position of the combustion chamber relative to the pressure vessel was known
only approximately, it was necessary to incorporate means of adjusting the posi-
3. Experimental Method 68
gimbal mount
top mirror
window
rotation stage
goniometer
bottom mirror
sheet expansion assembly
igniter support block
support plate
igniter tip
laser sheet
imaged region
..........
...........
...........
...........
..........
...........
...........
......
.............
.............
.............
.............
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.
.........................
..............................................................................................................................................................
.....................
.....................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.
Fig. 3.6: The periscope assembly. The unclipped sheet extents are shown, and the
red square indicates the area of the PLIF image.
tion and orientation of the mirrors and lens to ensure accurate alignment of the
laser sheet. Notwithstanding these degrees of freedom, stable retention of the op-
tics following alignment was also necessary. An additional requirement was that
all components must either tolerate or be protected from condensation and liquid
kerosene. As the operating pressure is sub-atmospheric, the periscope as a whole
also needed to seal the pressure vessel effectively. A final design requirement was
that cooling the periscope and sheet expansion assembly from ambient temperatures
to the 265 K test condition must not adversely affect the adjustment mechanisms
or cause additional leakage of air into the pressure vessel.
As illustrated in Figure 3.2, the periscope support plate was mounted on the
far side of the pressure vessel, replacing the original igniter mounting plate. An
3. Experimental Method 69
adjustable igniter support block was manufactured to ensure that the igniter tip
could be accurately positioned relative to the internal surface of the combustion
chamber.
Adjustments to the upper and lower mirror positions and orientations were
achieved by several means. The periscope shaft consisted of three concentric tubes
that provided two degrees of freedom. The outside support tube remained station-
ary. The middle cylinder could translate ±15 mm axially by rotating a threaded
drive ring. The inside tube could rotate ±45◦ relative to the middle cylinder. The
quartz window was mounted at the top end of this inside tube. Despite the axial
load on this assembly that resulted from the sub-atmospheric operating conditions
inside the pressure vessel (≈ 36 N), free rotation was provided by incorporating a
needle roller thrust bearing. The bottom mirror assembly was clamped to the inside
tube, and could thus rotate and translate in a controlled manner.
To ensure free movement while minimising leakage, the concentric periscope
tubes were manufactured according to tight geometric tolerances, with dissimilar
materials at interfacing surfaces. The materials used were an austenitic stainless
steel (BS EN 1982:1999), and a phosphor bronze (BS EN 10088-2:2005). These
widely-available metals are corrosion-resistant and have an identical linear coefficient
of thermal expansion, preventing differential thermal expansion or contraction. Both
the inside and middle cylinders were clamped in position following laser alignment,
though little force was required to achieve this due to the low vibration levels and
aerodynamic loading generated by the air-flow.
The height and angle of the top periscope mirror were adjusted to ensure that
the sheet was projected along the axis of the periscope shaft. This was achieved
using the translation stage and gimbal mount illustrated in Figure 3.6. The pitch
and roll of the bottom mirror were controlled independently by a goniometer and a
rotation stage respectively. In addition to the positional and rotational adjustments
described above, the lateral position of the periscope assembly relative to its support
3. Experimental Method 70
plate could also be modified.
To prevent air leakage into the pressure vessel, glass and aramid fibre gaskets were
used to seal flat metal contact surfaces. The periscope window was supported be-
tween polytetrafluoroethylene (PTFE) seals that protected the anti-reflection coat-
ing, while spreading the load generated by the sub-atmospheric operating pressure.
The bottom mirror, goniometer and rotation stage of the periscope were enclosed
by a protective cover (see Figure 3.10(b)). As discussed in Section 3.2.5, ingress of
dust and fuel droplets through the laser exit slit was prevented by a purge of clean
air.
Condensation of water vapour on the periscope optics as a result of the low
operating temperatures would have compromised the quality of the PLIF results
and caused subsequent laser damage to the dielectric coatings. While driers removed
virtually all water vapour from the air flowing through the test facility, condensation
on the outside surface of the test section was likely due to the sub-zero operating
temperature. The severity of this condensation is determined by the atmospheric
conditions in the test cell, and the specific heat capacity of the test section. If the
relative humidity in the test cell is ψ at temperature Tamb, then the corresponding
dew-point temperature is given by
Tdp = Tsat(Pv) where Pv = ψPsat(Tamb) (3.1)
(C¸engel and Boles 1998). Thus, if ψ = 75% at Tamb = 20
◦C, then
Tdp = 15.3
◦C. (3.2)
As each test session was envisaged to last for several hours, there was a significant
risk that the surface temperature of the periscope would fall below this value. Two
mats containing heating elements were therefore installed to prevent condensation
forming on the periscope optics. A small mat generating 2 W of heat was wrapped
around the window housing. A larger, 5 W, mat was attached to the vertical plate
supporting the top mirror assembly. Thermocouples monitored the temperature of
3. Experimental Method 71
each heating mat and a thermostatic controller regulated the supply of energy to
maintain a local temperature of 70 ◦C. This system successfully prevented misting
of the top mirror and window.
The sheet expansion assembly was mounted above the fuel injector, on the far-
side, upstream face of the combustion chamber. The space required for the assembly
was provided by inverting the fuel injector so that the fuel lines that normally occupy
this space were moved to the bottom. The expansion lens was positioned as close as
possible to the upstream combustor face to maximise the excitation area. This lens
was mounted in a brass cylinder that rotated inside a housing to permit adjustment
of the inclination. The sheet passed into and out of the lens through a slit in the
housing. As described in Section 3.2.5 and illustrated in Figure 3.10(c), clean purge
air introduced into the brass cylinder flowed out through the slot, thus preventing
the admission of fuel droplets and dust.
Table 3.2: Beam dimensions at principal point of each lens.
lens number
width thickness
(mm) (mm)
1 8.0 8.0
2 12.0 8.0
3 12.0 8.0
4 12.0 1.8
Burn marks produced on photographic paper suggested that the diameter of the
beam was approximately 8 mm on exiting the laser head. A computer-aided design
(CAD) model of the optical table and periscope recorded the position and orientation
of each optical component. Using this information, the beam cross-section was
calculated at every point along the optical axis, as described in Appendix B. The
cross-sectional dimensions of the beam at the appropriate principal plane of each
lens are summarised in Table 3.2.
An investigation of the beam dimensions performed following testing revealed
3. Experimental Method 72
that the upper and lower sheet boundaries had been slightly clipped by the lens
mount and the fuel injector respectively. The boundary angles were therefore re-
calculated from geometric information provided by the CAD model. The unclipped
sheet expansion angle was recorded to provide a reference for the beam energy dis-
tribution correction. The position of the sheet relative to the combustion chamber
was also calculated from the model.
The clipped region of excitation is illustrated in Figure 3.7. Note that the sheet
thickness varies considerably inside the combustion chamber. Coupled with the 27.3◦
expansion, this causes significant point-to-point differences in the laser irradiance.
PLIF image corrections for these variations are described in Section 5.2.
............................................................................................................................................................................................................
.......................................................................................................................................................................................................................................................
..................................................................................................................................................................................................................................................................................................
..............
.........
...........
..............
...........
...............
........
...............
...........
..........
.............
..........
.............
.........
...............
...........
..........
..............
..........
..............
...............
..........
..........
..............
..........
..............
..........
..............
...........
...........
.............
...........
............
.........
.........................................................................................
................................................. ........................................................
...........................................................
..........................................................
.........................................................
....
.......
......
.......
.......
.......
.......
.......
.......
.......
.......
.......
.......
.......
.......
.....
.......
.......
.......
.......
.......
.......
.......
.......
.......
.......
.......
.......
.....
.......
.......
.......
.......
.......
.......
..............
...............
..............
......
...........................................
..........................................
........................
.........................
........
........
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.............................................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....................................................................................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..............................................................................................................................
.................................................................................................................
......
....
......
......
......
......
......
......
......
....
......
......
......
......
......
......
......
......
.....
•
..... .....
...... .....
...... ......
.... ....
..... .....
..... .....
igniter tip
•
lens focal point
injector centre-line
injector outside edge
sheet
centre-line
outside
clip
inside
clip
Fig. 3.7: The excitation volume inside the combustion chamber. Note that the sheet
thickness in each cross-section has been scaled by a factor of four.
3. Experimental Method 73
3.2.5 Air purges
Three air purge systems were developed to keep the optics clear of fuel. The internal
surface of the combustor window was protected by a curtain of filtered purge air
drawn into the combustion chamber from outside the pressure vessel. In addition,
comparatively small flow rates of cylinder air were introduced separately into the
enclosures surrounding the periscope bottom mirror and the sheet expansion lens.
Contamination of the optical surfaces was prevented by discharging this air through
apertures provided for the laser sheet.
As noted in Section 3.1.1, the near-side fuel injector mounted inside the ATF
was replaced by a vertical slot. Though the resulting air curtain reduced liquid fuel
impingement on the combustor window, an additional purge system was required to
minimise contamination and thereby improve the quality of all acquired images.
The combustor window purge system is illustrated in Figure 3.8. The sub-
atmospheric operating pressure inside the combustion chamber draws air in from
outside the pressure vessel and discharges it along the upstream edge of the com-
bustor window. Incoming air passes through a filter and a ball valve before flowing
into a manifold. Four pipes, each with an internal diameter of 19 mm, feed the
air through the pressure vessel wall and into a plenum 18 mm × 52 mm × 145 mm
situated immediately upstream of the combustor window. The flow then accelerates
through a slot 3 mm× 123 mm onto the internal surface of the window.
The ball valve provides a means of regulating the mass flow rate of purge air.
Though the ATF vacuum pump was capable of maintaining the set point pressure
with the valve completely open, introducing significant extra quantities of air in
this manner has the potential to modify the processes under investigation. It was
therefore necessary to assess the air mass flow rate entering the combustion chamber
through this system.
The mass flow rate of air passing through the slot has been estimated by per-
forming a one-dimensional compressible flow calculation. The air is assumed to have
3. Experimental Method 74
filter
ball valve
manifold
plenum window
plenum
window
........................................
.....
.........................................
.....
.......................................................
.....
top view
side view
Fig. 3.8: The combustor window purge system. The window itself is shown in red.
3. Experimental Method 75
negligible kinetic energy as it enters the plenum. In addition, if the flow through
the slot is adiabatic and isentropic then the stagnation temperature and pressure,
T0 and P0 respectively, remain constant throughout. Assuming the flow is steady,
the mass flow rate through a slot of area A is
m˙ = ρAV = AMP
√
γ
RT
, (3.3)
whereM is the local Mach number, P is the static pressure, T is the static tempera-
ture, and γ and R are the specific heat ratio and gas constant of air respectively. For
an ideal gas flowing isentropically, the static and stagnation values of temperature
and pressure are related as follows
T0
T
= 1 +
(
γ − 1
2
)
M2, (3.4)
P0
P
=
[
1 +
(
γ − 1
2
)
M2
]γ/(γ−1)
. (3.5)
Re-expressing Eqs. (3.4) and (3.5) in terms of T and P respectively, and substituting
into Eq. (3.3) gives
m˙ = AMP0
√
γ
RT0
[
1 +
(
γ − 1
2
)
M2
]
−(γ+1)/2(γ−1)
. (3.6)
The mass flow rate of air is therefore governed by the stagnation properties, the slot
area and the Mach number at the slot. If the plenum stagnation temperature and
combustor static pressure are assumed to remain constant, then the mass flow rate
is governed solely by the plenum stagnation pressure. This pressure increases as the
ball valve is opened.
The flow becomes choked when the following condition is satisfied
P0
P
≥
(
γ + 1
2
)γ/(γ−1)
= 1.89. (3.7)
When the valve is first opened, the stagnation pressure is not significantly higher
than the static pressure at the slot. In this unchoked case, the throat static pressure
is equal to the combustor pressure, and the Mach number is identified from Eq. (3.5).
3. Experimental Method 76
The mass flow rate then follows from Eq. (3.6). As the valve is opened further
however, the slot becomes choked and M = 1. In this case, Eq. (3.6) simplifies to
m˙ = AP0
√
γ
RT0
(
γ + 1
2
)
−(γ+1)/2(γ−1)
. (3.8)
In this manner the mass flow rate was calculated for a range of stagnation pressures,
and the characteristic curve is plotted in Figure 3.9(a). The plenum pressure was
monitored during testing using a digital manometer, and the valve adjusted to main-
tain a set point value of 56 kPa (cf. combustor pressure of 41.4 kPa). The purge air
discharged at this set point was sufficient to essentially eliminate fuel impingement
on the window. Importantly, the calculated purge mass flow rate does not constitute
a significant fraction of the total combustor air, so the combustor flow patterns are
not expected to be radically disturbed by the influx of purge air.
The air mass flow rate plotted in Figure 3.9(a) is almost certainly an overestimate
of the true flow rate. While the assumptions of negligible kinetic energy in the
plenum and adiabatic flow are broadly accurate, flow through the slot is likely to be
highly non-isentropic. Furthermore, the cross-sectional area of the vena contracta
associated with this flow is significantly smaller than the slot area. A discharge
coefficient has not been applied to account for these effects, and the mass flow rate
estimates are therefore conservative.
Relatively small flow rates of air were used to protect the periscope bottom
mirror and sheet expansion lens, each purge representing 0.5% of the air mass flow
rate flowing through a single injector sector at the safe condition. Polyurethane
tubing carried clean, dry air from high-pressure cylinders to enclosures surrounding
each optic. This air then flowed from the two enclosures through slots machined to
allow passage of the laser sheet, thus preventing the ingress of dust and fuel droplets.
Drawings of the protective enclosures are provided in Figures 3.10(b) and 3.10(c).
The mass flow rate of air introduced in this manner was regulated by inserting
a short metal pipe containing a 1 mm diameter orifice into each supply line. This
pipe is illustrated in Figure 3.10(a). As the combustor set point pressure was only
3. Experimental Method 77
(a)
stagnation pressure, P0 (kPa)
a
ir
m
a
ss
fl
ow
ra
te
,
m˙
(g
/
s)
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....
...
....
....
....
....
.....
.....
.....
......
......
......
......
......
......
......
......
......
......
......
......
......
....
unchoked choked
40 55 70 85 100
0
20
40
60
80
100
(b)
stagnation pressure, P0 (kPa)
a
ir
m
a
ss
fl
ow
ra
te
,
m˙
(g
/
s)
....
....
.....
....
....
....
.....
....
....
.....
....
....
....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
....
....
.....
....
....
....
.....
....
....
.....
....
....
....
.....
....
....
...
100 200 300 400 500
0.0
0.2
0.4
0.6
0.8
1.0
choked throughout
Fig. 3.9: Air purge characteristic curves: (a) combustor window; (b) periscope
bottom mirror and sheet expansion lens. The operating points are indicated by
dashed lines.
3. Experimental Method 78
(a)
(b)
(c)
from regulator
.........................................
.....
to enclosure
.........................................
.....
........
.........
........
.......
....
................................. .
..
........
........
.........
.......
....
................................. ..
..
........
.........
........
.... .......
Fig. 3.10: The periscope and sheet expansion purge systems: (a) choking pipe;
(b) periscope enclosure; (c) sheet expansion lens enclosure. Arrows indicate the
direction of air-flow.
3. Experimental Method 79
41.4 kPa, any upstream pressure supplied by the regulator was sufficient to choke
the orifice (see Eq. (3.7)), thus fixing the mass flow rate through each enclosure.
Using Eq. (3.8) it was therefore possible to derive a characteristic curve relating the
air mass flow rate to the upstream stagnation pressure set using the regulator. This
curve is plotted in Figure 3.9(b).
A set point pressure of 225 kPa was used throughout testing and proved effective
in preventing contamination. The additional mass of air introduced through the
periscope and sheet expansion lens enclosures is clearly insignificant compared to
that associated with the combustor window purge.
3.3 High-speed flame imaging
3.3.1 Background
Combustion of a hydrocarbon fuel causes the emission of electromagnetic radia-
tion. The spectrum of this emitted light provides information concerning the species
present in the flame. For small, Bunsen-type flames the maximum emission usually
occurs in the near infra-red part of the spectrum, where distinct bands reveal tran-
sitions in the vibrational energy levels of H2O, CO2, and CO molecules (Gaydon
1974). While the stable products of combustion do not emit strong visible or ultra-
violet radiation, many of the unstable free radicals produced in the flame reaction
zone generate characteristic bands in these spectral regions. This process, referred
to as chemiluminescence, results from electronic transitions in molecules such as OH,
CH, and C2. For example, excited OH and CH produce strong discrete emissions
at 306.4 and 431.5 nm respectively, while the Swan bands, generated by the excited
C2 radical, have outstanding heads at 473.7, 516.5, and 563.6 nm.
While point-wise and two-dimensional measurements of these narrow-band emis-
sions have provided insights into reaction zone equivalence ratio (Muruganandam
et al. 2005) and flame structure (Ikeda et al. 2002) respectively, small particles
3. Experimental Method 80
formed inside some flames profoundly affect the emission spectrum. Most impor-
tant amongst these is the soot that is often produced in fuel-rich flames (Gaydon
and Wolfhard 1970). The radiation emitted by these large organic compounds has
a broadly continuous spectrum resembling that of a black body at the flame tem-
perature.
The spectra produced by liquid-fuelled gas turbine flames of the type considered
in this study, depends on the composition and state of the fuel and oxidant, together
with the mixture equivalence ratio and droplet size. When the fuel is pre-vapourised
or very finely atomised, a premixed flame results that emits a Bunsen-type spectrum
featuring clear OH, CH, and C2 bands. However, most flames generated inside avi-
ation gas turbine combustors are only partially premixed and may therefore contain
large droplets that are surrounded by a local diffusion flame. In these flames the
continuum emission generated by hot particles is dominant, resulting in a yellow ap-
pearance. The low temperatures and poor atomisation generated during the present
study produced a highly nonpremixed, sooty, yellow flame.
The high-speed flame imaging performed in this study recorded broad-band emis-
sions generated during spark ignition and flame development. A Vision Research
Phantom high-speed camera situated on the near side of the ATF, and viewing
roughly perpendicular to the flow direction, recorded line-of-sight flame emissions
during both successful and unsuccessful ignition events. The spectral sensitivity of
this camera is illustrated in Figure 3.11, and extends from approximately 400 to
1000 nm, with a peak quantum efficiency of 40% at 660 nm. The high-speed camera
was fitted with a Nikon 28 mm f/2.8 lens that transmitted wavelengths between
approximately 350 and 900 nm.
As the flame emission due to soot is predominantly generated in areas of rich
combustion, regions of stoichiometric or lean combustion are more clearly identified
by imaging an individual excited radical (Smith and Sick 2005). However, the weaker
narrow-band emissions generated by CH, or C2 were not individually filtered and
3. Experimental Method 81
recorded due to the limited dynamic range of the camera at high recording rates.
Indeed, as explained in Section 4.3.1, even without a filter the camera could not
retain the broad-band signal at all times due to large changes in the measured flame
intensity during each ignition event. An intensifier was not fitted to the camera due
to the associated reduction in spatial resolution.
wavelength (nm)
se
n
si
ti
v
it
y
(m
A
/
W
)
...........
...
...
...
...
...
...
...
...
...
...
...
...
..................
..................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.....................
...
...
...
...
........
...
...
...
...
..............
...
...
...
...
...
...
...
...
...
...
..............................
...
...
...
...
...
...
...
...
..................
...
...
...
...
...
...
...
..............................................
...
...
...
.............................
...
...
...
...
...
................................................................
...
...
...
...
...
...
...
...
...
...
.......................................
...
...
...
...
...
...
...
...
..............................
...
...
...
...
...
........................................
...
...
...
...
.............................................................................
...
...
...
..............................................
...
...
.................................
..
400 550 700 850 1000
0
50
100
150
200
250
Fig. 3.11: Spectral sensitivity of high-speed camera. Data courtesy of Vision Re-
search Inc.
The horizontal and vertical pixel resolution was set to 384 and 256 respectively.
Images were recorded at a rate of 5.7 kHz, with an exposure of 165 µs. Though
each recording was limited to 1.96 s, the pre-trigger interval could be adjusted. As
described in Section 3.4, this feature was utilised to capture the successful ignition
event and preceding unsuccessful event at the safe set point.
3.3.2 Viewing angle
The position of the PLIF and flame cameras during testing is illustrated in Fig-
ure 3.12. Note that throughout this study neither camera could view the laser sheet
3. Experimental Method 82
plane at right angles due to the limited optical access available. In order to extract
geometric information from, and apply PLIF corrections to, the recorded images
(see Sections 6.2.3 and 5.2 respectively), a viewing angle correction was required to
map vectors in the sheet plane to the camera images and vice versa.
high-speed camera
PLIF camera
...
...
...
...
...
...
...
...
...
...
...
........
.....
flow
Fig. 3.12: Plan view of test section including cameras. See Figure 3.2 for description
of line colours.
The viewing angle correction is derived in Appendix A and is expressed in the
form of a single matrix transformation equation for each camera view. The general
form of these transformation equations is
u
v

 =

M11 M12
M21 M22



u∗
v∗

 , (3.9)
where u and v are sheet-plane translations parallel to the outside and upstream walls
of the combustor respectively, and u∗ and v∗ are horizontal and vertical translations
in the image. The elements of the transformation matrix Mij are calculated from
an image of a square calibration target aligned with the laser sheet.
3. Experimental Method 83
3.4 Timing and data acquisition
Data was acquired with the experimental apparatus operating in three different
modes. PLIF images of cold fuel were recorded in short sequences without operating
the high-speed camera. In order to capture images of fuel distribution and flame
activity during ignition at the safe and unsafe set points, two separate methods of
triggering were devised. A general description of the apparatus for timing and data
acquisition is described below, followed by detailed descriptions of each of the three
modes.
Images of fuel and flame together with data describing the performance of the
test facility were recorded concurrently to allow investigation of possible interactions
between the measured variables. As data from several sources was acquired simul-
taneously, a timing and data acquisition system was developed to control triggering
of the laser, cameras, and spark, while monitoring the igniter current, the flame
luminosity, and the performance of the ATF. The design of this system was heavily
influenced by timing considerations, specifically those relating to the spark and the
laser shot.
The proprietary surface-discharge ignition system used throughout testing pro-
vided no control of spark timing. Though rated to deliver sparks at a nominal
frequency of 1.3 Hz, the spark-to-spark interval generated by this system was ob-
served to vary between 700 and 850 ms. While this suggested triggering the laser
and PLIF camera from the spark, modifying the 10 Hz repeat rate of the laser can
adversely affect the shot-to-shot beam energy distribution by causing thermal lens-
ing variations in the Nd:YAG rod (Koechner 1970; Song et al. 2000). For this reason
the timing of the spark and laser were decoupled and PLIF images were acquired at
a random interval following each spark. While minimising laser energy fluctuations,
this timing strategy required a large number of tests to ensure that PLIF images of
the flame development phase were captured (see Sections 4.3.1 and 4.3.2).
3. Experimental Method 84
3.4.1 General setup
A schematic of the timing and data acquisition hardware is presented in Figure 3.13.
This figure illustrates the apparatus required for fuel and flame imaging at the safe
condition (see Section 3.4.4). As described in Sections 3.4.2 and 3.4.3, the control
system was simplified for cold fuel imaging and ignition testing at the unsafe set
point. The two control boxes labelled ‘PLIF’ and ‘spark and flame’, provided fast
control triggers to activate the laser and cameras at appropriate times. The signals
fed to and from these boxes during safe ignition testing are illustrated in Figure 3.14.
During all tests, the PLIF control box regulated the timing of the laser shot and
the intensifier gate. Pulses to trigger the flashlamp discharge and the Q-switch were
sent to the laser at a rate of 10 Hz (see Figures 3.14(e) and 3.14(f)). The separation of
these trigger pulses was set to 300 µs to regulate the laser energy (25 mJ per pulse).
In response to the input signals, the laser generated a sync out trigger approximately
700 ns before producing the 4 ns laser shot (see Figure 3.14(g)). Though every sync
out trigger was returned to the PLIF control box, only a minority of these pulses
were fed through to the intensified camera to trigger acquisition of a PLIF image.
As discussed in Sections 3.4.2 to 3.4.4, the selected pulse depended on the mode of
operation.
In addition to this hardware, the overall conduct of all tests was controlled
and monitored by a National Instruments SCXI data acquisition system running
purpose-written software in the LabVIEW development environment. This software
controlled the progress of a single test as follows. Having generated the desired test
condition inside the ATF, the program started, and reported the combustor oper-
ating conditions at 1 s intervals (i.e. total air mass flow rate, air temperature, air
pressure, fuel mass flow rate, and fuel temperature). Both cameras were then primed
to receive their trigger signals. After activating and stabilising the laser output, and
adjusting the beam attenuator to maximise transmission, the test conditions were
recorded once further before fuel was injected into the combustion chamber. After
3
.
E
x
p
erim
en
ta
l
M
eth
o
d
85
laser
test section
PLIF
spark and flame
operating
conditions
...
...
...
...
...
...
...
...
........
.....
exciter
CP
P
L
IF
H
S
PD
...
...
...
...
...
..
...
...
...
...
...
..
data acquisition system
...
...
...
...
...
..
...
...
...
...
...
..
imaging computer
• PLIF
• CT
• HS
• FT
• PO
•
•
AC
AC
•
•
ST
ST
•
•
PO
•
•
FT
• IG
•
•
ST
•
•
CT
•
SO
• •
FL QS
............................................................................
.................
key:
AC attenuated current
CP current probe
CT camera trigger
FL laser flashlamp
FT flame trigger
HS high-speed imaging
IG igniter exciter
PD photodiode
PLIF kerosene PLIF
PO photodiode output
QS laser Q-switch
SO laser sync output
ST spark trigger
Fig. 3.13: Timing and data acquisition hardware
3. Experimental Method 86
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..................................... ...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.....................................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..........................................................
......................................................
.....
....
....
...
...
...
.....
.............
...
...
...................
.......................
................................................................................................. .............................................................................................
.....
............................
..... ................................. ............................
..... ...............................
............................
..... .................................
............................
..... .................................
............................
..... .................................
............................
..... .................................
............................
..... .................................
............................
..... .................................
≈ 750 ms
80 µs 5 ms
15 ms
100 ms
100 ms
20 µs
300 µs
170 ns
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
.......
.......
.......
.......
.......
.......
.......
.......
.......
.......
.......
.......
.......
.......
.......
.......
Fig. 3.14: Timing of data acquisition (not to scale): (a) attenuated current; (b)
spark trigger; (c) photodiode output; (d) flame trigger; (e) laser flashlamp trigger;
(f) laser Q-switch trigger; (g) laser sync out trigger; (h) camera trigger.
3. Experimental Method 87
a two second delay for fuel flow stabilisation, data was acquired in a manner deter-
mined by the mode of operation (see detailed descriptions below). On completion
of the test, the fuel was switched off, all data were saved to disk, and the program
returned to monitoring the ATF operating conditions. A delay of 60 s was necessary
prior to the next test, both to provide sufficient time for data storage, and to allow
the dispersion of liquid fuel that had accumulated in the bottom of the combustion
chamber.
3.4.2 Cold fuel imaging
The simplest mode of operation was that used for PLIF imaging of fuel without
flame. Since no spark was produced, the fuel image could be freely acquired at any
instant without danger of damaging the camera. In addition, the high-speed camera
was not used, and consequently required no trigger signal. Though the acquisition
of a single PLIF image could occur at any time, the full-resolution frame rate of the
camera was limited to 4 Hz. For this reason, cold images of fuel distribution were
acquired at a rate of 2 Hz.
After the two second fuel flow stabilisation period, a series of 10 PLIF images
were recorded at 0.5 s intervals. This was achieved by using the PLIF control box
to pass every fifth laser sync out trigger to the intensified camera. A short period
after receiving the trigger signal, the camera intensifier was gated for a period of
100 ns, with the gate delay adjusted to coincide with the laser shot.
3.4.3 Fuel and flame imaging at unsafe condition
The following data acquisition procedure was employed during unsafe ignition test-
ing. After fuel flow stabilisation, the spark exciter unit was switched on, and pro-
ceeded to generate a spark approximately every 750 ms. By operating the two
cameras in tandem, PLIF images were acquired shortly following each spark, and
high-speed recordings of flame were captured during the early ignition attempts.
3. Experimental Method 88
Special care was required to protect the intensified camera from high-intensity emis-
sions generated by the spark. A test consisted of ten consecutive sparks, following
which the spark exciter and the fuel supply were turned off.
During ignition testing at the unsafe condition it was necessary to control the
timing of the PLIF image acquisition relative to the spark. A Tektronix TCP303
current probe and TCPA300 amplifier detected the spark by measuring the current
passing along the lead between the exciter unit and the igniter. The analogue signal
from the amplifier (see Figure 3.14(a)) was continuously monitored by the spark
and flame control box and compared to a preset threshold of 2.5 V. A negative-
going spark pulse was produced when the signal generated by the spark increased
to exceed this threshold (see Figure 3.14(b)). This spark trigger was directed to the
PLIF control box.
The sync out trigger corresponding to the laser shot following detection of a
spark was fed to the intensified camera provided that the shot occurred more than
5 ms following the spark (see Figure 3.14(h)). This buffer period ensured that the
initial luminosity of the spark did not cause damage to intensifier. If a laser shot
occurred during the 5 ms buffer period, then the camera trigger was postponed
until the following shot. Having received the camera trigger signal, the intensifier
was gated to coincide with the laser shot. A single kerosene PLIF image was thus
acquired at a random point between 5 and 105 ms following every spark.
In addition to fuel images, recordings of flame development were acquired during
ignition testing. At the unsafe condition the high-speed camera was triggered by
the spark and flame control box on detection of the first spark. The pre-trigger
interval of the camera was set to 8.8 ms. This period represented 0.5% of the total
recording period available during a single ignition test (1.90 s). As the sparks were
separated by roughly 750 ms, this pre-trigger interval was sufficient to ensure that
the high-speed imaging at the unsafe set point captured the first three ignition
attempts.
3. Experimental Method 89
Spark and flame luminosity were also monitored by an Agilent HSDL-5420 in-
frared photodiode mounted outside the ATF, adjacent to the external window. Using
the spark trigger produced by the spark and flame control box as a reference sig-
nal, the current probe and photodiode outputs were recorded throughout the period
from 10 ms before to 110 ms following each spark. In addition to these signals, the
spark, flame, and camera trigger pulses were also recorded during this interval.
3.4.4 Fuel and flame imaging at safe condition
The data acquisition process employed during safe ignition testing was broadly sim-
ilar to that used when operating at the unsafe set point. Indeed, the PLIF imaging
system functioned exactly as described in Section 3.4.3, generating a single fuel dis-
tribution image between 5 and 105 ms following every spark. The current probe,
photodiode and trigger signals were also recorded from 10 ms before to 110 ms fol-
lowing each spark. However, in contrast to the unsafe operation mode, where the
flame imaging began following the first spark, at the safe condition the high-speed
camera was triggered following the detection of a successful flame. Additional flame
detection hardware was necessary to achieve this, and the full functionality of the
system illustrated in Figure 3.13 was required.
The camera trigger was generated by the spark and flame control box, which
monitored the photodiode signal throughout each ignition test. If the amplified
signal from this photodiode (see Figure 3.14(c)) exceeded a threshold of 2.5 V at any
time between 15 and 100 ms following the detection of a spark, then a flame trigger
was generated (see Figure 3.14(d)), and directed to the high-speed flame camera.
The pre-trigger interval of the flame camera was set to 1.43 s during testing at the
safe condition. This period represented 75% of the total recording period available
during a single ignition test. This enabled every safe successful ignition event to
be recorded on the high-speed camera, together with each preceding unsuccessful
attempt. As discussed in Section 3.1.2, successful ignition was consistently achieved
3. Experimental Method 90
within fourteen sparks at the safe condition. Data acquisition therefore continued
until a flame was detected, whereupon the high-speed camera was triggered and the
test terminated.
3.4.5 Test numbers
Table 3.3 lists the number of successful and unsuccessful high-speed recordings of
flame development acquired at the two test conditions. One hundred safe tests
were conducted, recording the successful ignition event and any preceding failed
development. However, six tests identified as successful by the flame detection
hardware, failed to recover and survived for only a short period. These tests were
therefore reclassified as unsuccessful. A further five safe tests were successful on the
first spark. Thirty one unsafe tests were conducted, recording three unsuccessful
ignition events during each test.
Table 3.3: Number of tests recorded with high-speed flame camera.
safe unsafe
successful 94 –
unsuccessful 101 93
The PLIF image numbers recorded are listed in Table 3.4. The ignition categories
contain only those images that were acquired at the same time as a high-speed flame
recording. One and four tests at the safe and unsafe test points respectively were
discarded due to poor quality results.
Table 3.4: Number of tests recorded with PLIF camera.
safe unsafe
fuel only 200 200
successful ignition 93 –
unsuccessful ignition 100 81
Chapter 4
Results I: Preliminary Analysis
The previous chapter discussed the design requirements and development of appa-
ratus to perform specific measurements under a set of well-controlled conditions.
The success or failure of ignition is influenced not only by the random nature of the
relight process, but also by spark-to-spark variations in the experimental boundary
conditions. This chapter describes the measurement of these boundary conditions
and compares them to the expected values. For instance, the test facility operating
conditions, laser energy, and spark parameters all fluctuate during testing, and these
variations must be quantified to assess their impact on the results presented in later
chapters. In addition, the detailed analysis of PLIF images in Chapter 5 requires
the timescales of flame development to be identified, as also described here.
4.1 Operating conditions
The combustor operating conditions were recorded once during every test, prior to
the first spark. The measured parameters were air and fuel temperature and mass
flow rate, together with air pressure. The test-to-test variations in these conditions
are presented in Table 4.1. The mean temperatures and pressure correspond closely
to their set point values, and the standard deviation of each variable is comparatively
91
4. Results I: Preliminary Analysis 92
small. Though not listed, this is also true for the air mass flow rate. These param-
eters were therefore well controlled throughout testing. However, a malfunctioning
flow controller caused significant fluctuations in the fuel mass flow rate, leading to
imprecise control of the total AFR. This is demonstrated by the relatively large
standard deviation in the total AFR; approximately 4% of the mean at both condi-
tions. Despite this poor fuel metering, the air mass flow rate at the two set points
is sufficiently different to ensure that these fluctuations never transform the global
test conditions from safe to unsafe or vice versa.
Table 4.1: Experimental operating conditions during ignition testing.
safe unsafe
set point mean std. dev. set point mean std. dev.
Tair (K) 265.0 265.4 0.3 265.0 265.1 0.3
Pair (kPa) 41.4 39.8 0.2 41.4 43.9 0.4
Tfuel (K) 288.0 286.6 1.8 288.0 287.9 0.7
total AFR 33.0 33.8 1.5 60.5 60.5 2.6
Since the air and fuel mass flow rates were recorded only once at the start of
each ignition test, the sample rate was inadequate to reveal any possible correlation
between AFR variations and ignition outcome. Continuous monitoring of AFR
would therefore be desirable in future work.
4.2 Properties of laser beam and sheet
The following section describes two preliminary experiments to characterise the laser
beam and sheet, conducted as preparatory work prior to PLIF imaging during ig-
nition testing. Imaging of the beam passing through a fluorescent dye quantified
laser energy fluctuations, while also providing spatial profile information. A slit-
traverse experiment was conducted to corroborate the dye fluorescence results. The
4. Results I: Preliminary Analysis 93
revealed features of the beam energy distribution significantly influence the PLIF
image correction discussed in Section 5.2.
4.2.1 Dye fluorescence
The method described by Arnold et al. (1997) and Hartung et al. (2008) for
measuring the laser profile by projecting the beam through a dye has been followed
here. Images of the dye fluorescence provide information about the shot-to-shot
fluctuations and the spatial profile of the beam. Due to restricted access in and
around the ATF however, it was not possible to image the energy profile during
ignition testing. A supplementary laboratory experiment was therefore performed,
as described below.
A quartz cell with internal dimensions 10 mm×10 mm×44 mm, was filled with a
fluorescent dye solution, and positioned in front of the laser aperture. The intensified
camera used for kerosene PLIF imaging was positioned to view the dye cell from
beneath. The laser energy was adjusted to that used during ignition testing and
PLIF images of the fluorescing dye were recorded. A sample image is shown in
Figure 4.1(a).
While a smooth variation in intensity might be expected, the horizontal striations
apparent in Figure 4.1(a) reveal significant irregularities in the beam profile. In
addition, air bubbles in the dye solution have caused distinct spots in the image.
The signal along the indicated mid-plane of the cuvette has been extracted from 110
individual images, and these profiles are plotted in Figure 4.1(b), together with the
mean. The observed irregularities in the mean are partly attributable to these air
pockets.
Shot-to-shot variations in laser energy produced significant scatter in the fluores-
cence profiles. For a sample of 110 images, the standard deviation close to the centre
(i.e. l ≈ 5 mm) represents approximately 8% of the mean signal. Consequently, as
the energy evidently varies from shot to shot, individual PLIF images recorded dur-
4. Results I: Preliminary Analysis 94
(b)
×104
distance, l (mm)
P
L
IF
in
te
n
si
ty
co
u
n
t
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
.......................
..........
.........
....
...
...
....
...
.
...
...
...
....
...
...
...
.
...
...
.
.....
...
....
...
..
...
...
...
..
..
..
.........
...
.
...
...
...
...
.....
...
.
......
...
...
......
.
.....
...
...
...
...
..
........
...
...
...
.
........
...............
...
....
...
.....
....
...
.
.....
...
...
......
...
...
........
...
...
...
...
...
...
.
.............
...
...
...
.
...
...
...
...
.
......
.......
......
...
...
.
.........
...
.
..........
...
...
...
...
...
...
.
...
...
...
...
............
...
...
...................
.........
..........
...
...
...
.
........
...
.
..........
...
.......
...
...
..
.............................
............
..
...
...
........
...
..
.........
...
...
..........
...
...
....................
...
..
.........
...
..........
..................
.......
...
...
.
........
.......
.............
...........
..................
.................................
.
individual
mean
0 2 4 6 8 10
0
1
2
3
(× 104)
1 2
0 1 2 3
×104(a)
0 mm
10 mm
Fig. 4.1: Dye fluorescence measurements of beam energy profile: (a) typical single-
shot image; (b) 110 mid-plane profiles and their mean.
ing altitude testing cannot be compared directly. However, comparison of safe and
unsafe mean images is justified, as the laser fluctuations are not influenced by the
test conditions and the sample size of PLIF images is large.
The spatial energy profile deviates significantly from the expected Gaussian dis-
tribution. Though the signal is noisy, two distinct shoulders in the mean profile are
apparent at l ≈ 3.0 mm and l ≈ 7.0 mm. This observation has important conse-
quences for the fuel imaging experiment that are examined thoroughly in Section 5.2.
It is clear from Figure 4.1 that a correction for the beam energy distribution must
be based on a measured, rather than an assumed, energy profile.
4.2.2 Slit traverse
The spatial energy distribution generated by a laser can also be characterised by
measuring the power transmitted through a narrow slit as it scans across the beam
(Su and Clemens 1999; Hornak 2002). In the current study, a 0.55 mm vertical slit
4. Results I: Preliminary Analysis 95
was created using two razor blades. These blades were mounted in front of a laser
power sensor, and this assembly was traversed across the laser beam and sheet.
During the beam traverse, the slit was positioned 200 mm from the laser aperture.
With the laser generating the same energy as used during ignition testing, the slit
assembly was scanned across the beam incrementally, noting the transmitted power
after each positional adjustment. The slit assembly was similarly traversed across
the laser sheet. A laboratory mock-up of the optical arrangement used during rig
testing was constructed to recreate the sheet.
Figure 4.2 illustrates the rate of change of power with traverse distance through
the beam and sheet. These profiles were constructed as follows. The power sensor
readings noted during each traverse were plotted cumulatively as a function of dis-
tance. These profiles were then differentiated numerically, applying a least-squares
smoothing filter to remove low level fluctuations (Press et al. 2007). Note that it
is difficult to identify the centre of the measured energy distributions due to the
asymmetry of the profiles. For convenience, the central peaks have been aligned in
Figure 4.2.
The plot of beam power gradient confirms that the beam diameter is approx-
imately 8 mm. As observed in the dye fluorescence images, the beam profile de-
parts significantly from a Gaussian distribution, with two shoulders located at
l ≈ −2.5 mm and l ≈ 1.5 mm relative to the central maximum. Since the beam
exits the laser head and strikes the slit directly, these shoulders are not caused by
contaminated or damaged optics located beyond the harmonic separators. The ex-
tent of the sheet could not be identified due to the inadequate thermal sensitivity
of the power sensor. The beam and sheet profiles do, however, appear qualitatively
similar, confirming that the sheet energy distribution can be represented by linearly
scaling the beam profile.
The dye fluorescence and slit traverse profiles are in good agreement and the non-
uniformities in the beam profile are well characterised by both. However, air bubbles
4. Results I: Preliminary Analysis 96
distance, l (mm)
p
ow
er
g
ra
d
ie
n
t,
d
P
/
d
l
(m
W
/
m
m
)
••....•... •...••....•... •..
.•...•.
..•
....
•
...
...
...
..
•...................
•
...
...
...
...
...
...
•................
•
...
...
...
..
•.... •
...•
...
•
...
...
...
•...............
•
...
...
...
•....................
•
...
...
...
...
...
...
•..............
•
...
•...
•
..............
•..
...
...
...
...
...
...
...
...
...
.
•
.....................•...
•
...
...
...
..
•..
...
...
•
................................
•..
...
...
...
...
...
...
...
..
•
....................
•..
...
...
...
•
..........
•..
...
.
• beam
◦◦....... ◦...... ◦
....... ◦...... ◦
.......
◦
...
...
◦......
◦
...
...
◦....... ◦...... ◦
....... ◦.... ◦
....... ◦......
◦.......
◦
...
...
◦..........
◦
...
...
◦...
....
◦
............
◦..
...
...
..
◦......
◦......
◦
.......
◦.
...
...
.
◦
..........
◦..
...
..
◦
....... ◦...... ◦...... ◦....... ◦...... ◦....... ◦...... ◦.......
◦ sheet
−10 −5 0 5 10
0
30
60
90
120
150
180
Fig. 4.2: Slit traverse measurements of beam and sheet energy profiles.
in the dye solution adversely affect the dye fluorescence results (see Section 4.2.1),
and the slit traverse measurements of the beam are therefore considered to be more
accurate. A scaled form of this beam profile is therefore used to correct the kerosene
PLIF images for non-uniformities in the laser sheet (see Section 5.2.4).
4.3 Characteristic times
Preliminary analysis of the high-speed film was conducted to characterise the timescales
of ignition success and failure. These timescales are used to classify every PLIF im-
age according to the development stage of the flame at the instant of acquisition.
4.3.1 Recovery and failure times
The total intensity variation measured in each high-speed recording provides a sim-
ple assessment of combustion activity during ignition. This section describes the
results of a straightforward total-intensity analysis, and considers the distribution
4. Results I: Preliminary Analysis 97
of characteristic timescales identified from these intensity data. Further information
can be extracted from the high-speed images using the more sophisticated analysis
described in Chapter 6.
The total intensity traces plotted in Figure 4.3 are typical examples of ignition
success and failure. The high-intensity initial signal due to the spark decreases to a
low level within approximately 5 ms. This large reduction in flame emission has been
observed in previous gas turbine ignition studies (Naegeli and Dodge 1991; Pucher
et al. 2001). During the low-intensity delay period before recovery or failure, the
successful and unsuccessful plots are indistinguishable, and flames that later grow
and develop successfully emit very little light at this point. Indeed, during 6 of the
94 successful recordings, the flame emissions are so weak that the camera registers
no signal whatsoever. These ‘false failures’ typically occur for a period of less than
1 ms, and result from either a dramatic reduction in flame volume, or a shift in
the emission wavelength to beyond the detection range of the imaging system (see
Figure 3.11 for the spectral sensitivity of the high-speed camera).
If the total intensity plots of different successful events are compared, the height
of the recovery maximum can be seen to vary significantly from test to test, reflecting
differences in the initial mass of fuel in the combustor. After this excess fuel has
been consumed, the intensity falls to a much lower value that is broadly the same
for all successful tests.
The failure and recovery times used to characterise each test outcome, and fre-
quently referred to below, are also illustrated in Figure 4.3. The failure time is
defined as the interval between the spark and the last frame containing any flame
activity. The recovery time is the period following the spark when the signal first
increases to exceed a selected threshold value of 1 × 106 counts (see dashed line
in Figure 4.3). The intensity count never exceeds this value during a failure, but
always recovers strongly having exceeded it during a successful attempt.
The characteristic times of every ignition event are compiled in Figure 4.4. This
4. Results I: Preliminary Analysis 98
×106
interval following spark (ms)
fl
a
m
e
in
te
n
si
ty
co
u
n
t
..................................................................................................................................................................................
............
............................
....
...
...
....
....
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....
.........
................................................................................................................................................ ..................
•
......................................................................................................................................................................
.....
recovery time
•
............................................................................................................
.....
failure time
0 10 20 30 40
0
2
4
6
8
Fig. 4.3: Definition of recovery and failure times.
plot illustrates the proportion of unsuccessful tests that are yet to fail (red and blue
lines), and the fraction of successful tests that have already recovered (green line).
For a given condition and outcome, the range of recovery and failure times is large
compared to the mean. This is particularly true during successful events, when the
low-intensity phase before recovery can be as short as 10 ms, or can persist for a
period of up to 69 ms. This suggests that recovery is a random event not strictly
correlated with the operating conditions at the instant of the spark.
It is clear from Figure 4.4 that failures occur faster at the unsafe condition
than at the safe point. Indeed, 67% of unsafe failures occur before the first safe
death. Although the cumulative failure curves have similar profiles in Figure 4.4, the
flame development process is visually different for safe and unsafe failures. When
viewed directly, the sequence of high-speed images indicates that unsafe failures
result from disintegration of the flame kernel, whereas safe failures generally occur
following significant flame propagation. These two modes of failure have also been
4. Results I: Preliminary Analysis 99
interval following spark (ms)
p
ro
p
o
rt
io
n
o
f
te
st
s
(%
)
safe recovery
safe failure
unsafe failure
0 20 40 60 80
0
20
40
60
80
100
Fig. 4.4: Cumulative frequency plot of recovery and failure times.
observed during ignition studies of laboratory-scale flames (Ahmed et al. 2007a).
This important observation is explored further in Section 6.2.
4.3.2 Classification of PLIF images
Having extracted recovery and failure times from every high-speed flame recording,
the kerosene PLIF images can be categorised according to the state of flame devel-
opment at the instant of image acquisition. As the timing of the laser pulse relative
to the spark could not be controlled, every PLIF image was acquired at a random
interval between 5 and 105 ms following the spark (see Section 3.4). By recording
this spark-image interval, and comparing it to the corresponding recovery or failure
time, it is possible to determine the state of flame development when each PLIF
image was acquired, as shown in Table 4.2.
Images recorded before recovery or failure are categorised as ‘developing flame’.
These images were acquired when the flame was confined to small regions of the
4. Results I: Preliminary Analysis 100
Table 4.2: Classification of PLIF images.
safe unsafe
successful
{
developing flame 24 –
stabilised flame 69 –
unsuccessful
{
developing flame 13 0
post-failure 87 81
combustor, with unburnt fuel often present in the laser sheet. During ignition re-
covery, flame expands to occupy the majority of the combustor volume, consuming
most of the imaged fuel. The PLIF images acquired under these conditions were
grouped into a ‘stabilised flame’ category. The influence of flame on the fuel PLIF
signal is examined in more detail in Section 5.1.3.
Observe that in Table 4.2 there are relatively few PLIF images of developing
flame at the safe condition, and none at the unsafe condition. This is due to the
short timescales of success and failure compared to the 100 ms range of spark-image
intervals (see Figure 4.4). For future work, the number of developing flame PLIF
images could be increased either by providing control of the spark-image interval,
or by greatly increasing the total number of tests.
Although the acquisition timing limited the number of images acquired during
flame development, a comparatively large number of stabilised flame and post-failure
images were generated. By definition the post-failure shots were acquired in the
presence of fuel alone, and therefore increase the number of cold, fuel images from
200 (see Table 3.4), to 287 and 281 at the safe and unsafe conditions respectively.
These images have been included in the total fuel-only sample analysed in Chapter 5.
4.4 Spark effects
The ignition system used for relight testing is rated to deliver 10 J of energy per
spark. Minimum ignition energies at least ten times smaller than this have been
4. Results I: Preliminary Analysis 101
recorded during laboratory-scale ignition studies of kerosene-air sprays generated
under altitude conditions (see Table 2.3). Despite the comparatively large quantity
of energy deposited in the combustion chamber during the present study, only 17%
of sparks lead to a fully-stabilised flame. As highlighted in Section 4.3, there is
generally an extended period between the spark and the onset of recovery, provid-
ing ample time for variations in the flow field and fuel distribution to completely
extinguish the flame. This explains the low observed success rate.
A high-speed image of the spark acquired 0.35 ms following the discharge is
illustrated in Figure 4.5. The large amount of energy deposited by the spark is
demonstrated by the well-defined saturated region around the igniter tip. Further-
more, the light emitted by the spark is sufficiently intense to illuminate the injector
face, and to generate a strong reflection in the far-side combustor wall.
injector
spark
reflection
..............................................................
.....
.........................................................
...
..
...
..
............................................................................................................................................................... ..........
Fig. 4.5: Image of spark discharge. The colour map has been inverted so that black
corresponds to the highest intensity.
4. Results I: Preliminary Analysis 102
4.4.1 Measured spark parameters
As noted in Section 2.2.2, the extent to which relight outcome is determined by
the spark has not been previously investigated. Furthermore, fundamental studies
of droplet ignition suggest that in addition to energy, other characteristics of the
spark also affect the likelihood of success (see Section 2.4). This section therefore
considers the influence of three spark parameters on relight outcome: size, current,
and luminosity. The first of these is extracted from the high-speed images, while
the second and third are measured directly. The area of saturated signal below the
igniter tip in high-speed images of the spark (see Figure 4.5), provides a convenient
measure of the spark size.
Assuming that the discharge voltage remains approximately constant, the spark
current provides an indication of the energy introduced into the combustion cham-
ber. Throughout testing, the current was recorded by an inductive probe mounted
on the igniter lead. Electromagnetic shielding around the cable massively attenu-
ated the measured signal however, generating data that is quantitatively inaccurate.
Despite this, the results provide a useful comparative measure of spark current, and
in turn, energy.
A further source of spark information was provided by the photodiode. Though
introduced for control purposes, the peak voltage generated by the photodiode re-
veals variations in the spark luminosity. In contrast to the high-speed camera, the
photodiode did not saturate when exposed to the intense spark emissions, and thus
provides a reliable measure of spark brightness.
Figure 4.6 illustrates the distributions of measured spark size, peak current,
and peak luminosity, grouped according to set point and ignition outcome. The
measured spark size (see Figure 4.6(a)) has been extracted from every high-speed
sequence, yielding an approximately equal number of samples for each histogram,
as indicated previously in Table 3.3. In contrast, Figures. 4.6(b) and 4.6(c) present
measurements recorded during a much larger number of sparks. In these figures,
4. Results I: Preliminary Analysis 103
the relatively small number of safe successes (green line), reflects the low ignition
probability.
Accounting for the differences in sample size, the distributions do not differ
significantly with set point or ignition outcome. This observation is reinforced by
the closely matched mean and standard deviation of each spark parameter, presented
in Tables 4.3 and 4.4 respectively. These results therefore imply that the measured
variations in spark size, current, and luminosity do not influence ignition outcome,
and are not significantly affected by the test conditions.
Table 4.3: Mean measured spark parameters.
kernel area peak attenuated peak photodiode
(×103 pixels) current (A) output (V)
safe success 5.77 21.4 4.51
safe failure 6.19 21.4 4.52
unsafe failure 6.25 21.6 4.50
Table 4.4: Standard deviation of measured spark parameters. The standard devia-
tion as a percentage of the mean is included in parentheses.
kernel area peak attenuated peak photodiode
(×103 pixels) current (A) output (V)
safe success 0.87 (15%) 1.63 (8%) 0.04 (1%)
safe failure 1.16 (19%) 1.74 (8%) 0.05 (1%)
unsafe failure 1.85 (30%) 1.63 (8%) 0.04 (1%)
The spark variability can be assessed from the standard deviation of each data set
(see Table 4.4). While the peak attenuated current and photodiode output are both
tightly grouped around their mean values, the scatter in the kernel area is relatively
large. The spread in this data is partly associated with the limited recording rate
of the high-speed camera. The finite exposure period causes the time between the
spark discharge and the analysed frame to fluctuate by ±88 µs from spark to spark.
4. Results I: Preliminary Analysis 104
(a)
×104
kernel area (pixels)
fr
eq
u
en
cy
•••••••••
•
...
...
...
...
...
...
...
...
...
..
•....................................................................................................
•
......................
•..
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..
•
............................•
....
•
....•
•
....
•..
.. •...
•..
..
•••••••••••••••••••
•
...
...
...
•..............................................
•
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.
...
...
...
...
...
...
...
•.
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.
•
................................
•..
...
...
...
...
...
.
..... •....
•
...
•
•
........•
... ... •...
.
....
•............
•
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.
•..........
.....
•
...
.....................
•.
...
...
...
...
...
...
...
...
...
...
...
...
..
•
..
...
•..
...
..
...
...
...
...
.
•
...
...
...
..
...
...
.
•..
.. .
•.............
•
.......
•..
..
•....
safe success
safe failure
unsafe failure
0.0 0.4 0.8 1.2 1.6 2.0
0
10
20
30
40
50
(b)
peak attenuated current (A)
fr
eq
u
en
cy
••••••••••••••...
..••...
..•
.... •
....•..... ••
•
...
...
...
...
•........ •
....
•...........
•
..............•
....
•
.............
•.
...
•
....••••••••••
•....
.•
•
...
...
...
...
...
•..
...
...
.
•
...
...
...
...
..
•..
..
•
...
...
...
..•
....
•
........
•.............
•
...
...
...
...
...
...
...
...
...
...
...
...
...
...
•..................
•...
•..................................................
•
..................................
•..
...
...
...
...
...
...
...
...
...
•
............................................................
•..
...
...
...
..
•
........... ....•
....
•..
•..............
•.
..
•..
.. ... •
.........
... •...
•.........................................................
•
...
...
...
...
...
...
...
...
...
...
...
•
...
...
...
...
...
...
...
...
...
...
.
•
......
•.
...
...
...
...
..
10 14 18 22 26 30
0
20
40
60
80
100
(c)
peak photodiode output (V)
fr
eq
u
en
cy
•••••••••••••••••
.....
•...
.•.....
•
...
..
•......... •
...
•..
...
...
..
•
.........
••
.....•••••••••••••••...
..
•.....
•
...
...
...
...
...
...
•........
•
...
...
...
...
...
...
...
...
...
...
...
...
...
..•
.... •
•.
...
.
•
.....................................
•..
...
...
...
...
...
...
...
.
........•...
•...........
•
...
...
...
...
...
...
...
...
...
...
...
...
.
•........................
•
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..
•.................
•
.....................................
•..
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.
•
............................
•.
...
...
...
...
4.0 4.2 4.4 4.6 4.8 5.0
0
20
40
60
80
100
Fig. 4.6: Histograms of measured spark parameters: (a) size; (b) current; (c) lumi-
nosity.
4. Results I: Preliminary Analysis 105
This generates scatter in the spark size measurement, as the initial kernel growth is
rapid and not temporally resolved by the camera. As the standard deviation in the
kernel area is exaggerated by the frame timing, the spark appears to be relatively
consistent.
4.4.2 Interaction between successive sparks
A single ignition test at the safe condition consisted of repeated sparking at 750 ms
intervals until a successful developing flame was detected by the photodiode. The
spark-to-spark interval significantly exceeds both the combustion chamber residence
time (24 ms and 13 ms at the safe and unsafe set points respectively), and the longest
recorded recovery time (69 ms). The hot gases and chemical radicals produced
during an unsuccessful ignition attempt have therefore been evacuated from the
combustor well before the next spark. While this might lead us to expect every
spark to have a constant probability of success, the distribution of safe, successful
spark numbers illustrated in Figure 4.7, provides evidence to the contrary.
If sparks are independent, then the probability of success on the k-th spark
is analogous to the probability of tossing a coin and obtaining k − 1 tails before
achieving a head. Therefore, the probability is given by
P (k) = (1− p)k−1p where k = 1, 2, 3, . . . , (4.1)
where p is the probability of any spark being successful. This is known as a geometric
distribution (Papoulis and Pillai 2002). The expected number of successes on the
k-th spark for a sample of Ntot ignition tests is then NtotP (k). Unlike the coin
flipping example, where p = 0.5, the probability of individual spark success is less
than that of failure.
Assuming that the entire sample of ignition tests consists of independent sparks,
p is estimated by the total number of successful sparks divided by the total spark
4. Results I: Preliminary Analysis 106
spark number, k
fr
eq
u
en
cy
•
•
...................
•.
...
...
...
...
...
•
..............
•..
....
...
...
..
•
................
•...
......
......
...
•
..................
•...
........
......
•
.................
•.....
...........
.
•
................
•......
...........
•
................
•..........
.......
• all sparks (Eq. (4.2))
◦
◦
◦
........................................................
◦..
...
...
...
...
...
...
...
...
...
...
...
...
..
◦
........................
◦.
...
...
...
...
...
◦
...............
◦...
......
.
◦
.................
◦.....
............
◦
................
◦...........
......
spark five onwards (Eq. (4.3))
1 3 5 7 9 11 13 15
0
5
10
15
20
25
30
Fig. 4.7: Measured and expected distributions of successful spark number. The
expected distribution is calculated using Eq. (4.1).
number
p =
Nsuc
Ntot
=
94
565
= 0.17. (4.2)
Substituting this value of p into Eq. (4.1) generates the distribution represented by
filled circles in Figure 4.7. This distribution over-predicts the number of successes
for low values of k. Furthermore, the increase in recorded successes from k = 1
to k = 5 indicates that, contrary to the predicted trend, the ignition probability is
growing during these early sparks.
If the sample is instead restricted to tests where success occurs later than spark
four, then p is given by
p =
Nsuc
Ntot
=
66
212
= 0.31. (4.3)
As illustrated by the open circles plotted in Figure 4.7, the distribution generated
using this ignition probability, corresponds more closely to the measured data. This
indicates that the ignition events are independent from spark five onwards, but not
4. Results I: Preliminary Analysis 107
before.
The early spark-number dependence may result from a change in the spark de-
livery system over time. As noted in Section 4.4.1, none of the measured spark
parameters exhibit any correlation with outcome. However, the possibility remains
that sparks could change in some way that has not been measured here. For ex-
ample, discharge duration has previously been shown to affect the ignition outcome
independently of spark energy (Ballal and Lefebvre 1975). In addition, variations
in the breakdown voltage may cause spark losses to vary, producing fluctuations in
the heat supplied.
Alternatively, repeated sparking may produce a change in operating conditions
that favours later sparks. Such behaviour has been observed by Marchione et al.
(2007, Sep). In their laboratory-scale experiment however, the interval between
sparks was considerably shorter than the evacuation time of the combustion cham-
ber. As highlighted above, the spark interval in the present study is more than
adequate to ensure that air and fuel flows are unaffected by previous sparks. Nev-
ertheless, a change in the boundary conditions could result from an accumulation
of heat in the combustor walls caused by early sparks. While this possibility was
further investigated by mounting a thermocouple immediately downstream of the
igniter, no evidence of warming was found.
Despite this, the wall temperature measurements provide a key insight into the
distribution of liquid fuel inside the combustor, and suggest an alternative explana-
tion for weak early sparks. At the safe condition, the wall temperature was observed
to increase when fuel injection commenced, indicating the impingement of fuel on
the thermocouple. This suggests that a film of fuel may cover the igniter tip initially,
before being dislodged or vapourised by early sparks. The presence of excessive fuel
on a surface-discharge igniter tip is known to adversely affect spark performance
(see Section 2.2.2).
For a quantitative study of the factors influencing ignition events per se, analysis
4. Results I: Preliminary Analysis 108
should be confined to the later, independent sparks. In the present study, however,
a limited number of ignition tests were performed, and discarding data would re-
duce the sample size significantly. Furthermore, the increase in ignition probability
during early sparks may occur in-flight, when attempting high-altitude restart. As
a consequence, the modes of success and failure during the opening sparks are of
considerable practical interest, and will be considered together with those of later
sparks.
Chapter 5
Results II: Fuel Imaging
In this chapter individual and mean PLIF images are studied to identify differences
in fuel placement at the two set points. Significant corrections must be applied to
the raw images to account for the beam energy distribution and sheet geometry.
To allow comparison of uncorrected PLIF images, analysis is initially restricted to
a small region in the centre of the sheet. Entire images are later corrected and
compared in Section 5.2. CFD flow predictions are considered, and the influence of
the flow field on the fuel distribution is discussed.
5.1 Analysis of uncorrected PLIF images
5.1.1 Individual cold images
Instantaneous images of fuel distribution acquired at the safe and unsafe test condi-
tions are presented in Figures 5.1 and 5.2 respectively. Also shown in these figures
is a detail from the centre of the laser sheet, denoted region A. Both images are
representative, and have been selected so that the count inside region A is close to
the mean at their respective set points.
Distinct droplets of various sizes are apparent in both Figures 5.1 and 5.2, to-
gether with lower intensity regions of cloudy, diffuse signal. These areas correspond
109
5. Results II: Fuel Imaging 110
(× 104)
1 2 3 4 5
0 1 2 3 4 5 6
×104
.......................
......................
......................
......................
......................
......................
......................
...
......................................................................................................................................................................
A
Fig. 5.1: Uncorrected, individual, cold PLIF image acquired at safe condition.
to fuel vapour and small droplets with diameters below the spatial resolution of
the camera. The finite sheet thickness also makes it difficult to resolve clusters of
sub-pixel mist.
From the detail images of typical ‘safe’ and ‘unsafe’ signal, it can be seen that
though there are more droplets at the unsafe condition, the droplet sizes are similar
in both cases. These observations are confirmed by Figure 5.3, which shows the five
detail images with region A counts closest to the mean at each set point. Figure 5.4
has been produced by applying a logarithmic colour map to the detail images shown
in Figure 5.3. The low-level signal revealed in Figure 5.4 suggests that the proportion
of fuel contained in very small droplets and vapour is larger at the unsafe condition
(particularly pronounced in sub-figures (g) and (h)), but that this signal represents
only a small fraction of the total region A count. The range of droplet sizes evident
in the detail images has been selected by the CRZ, which is expected to occupy the
region below the igniter tip (see Section 2.1). Recirculating flows have the ability to
select particle size as the inertia of large liquid droplets often causes them to deviate
5. Results II: Fuel Imaging 111
(× 104)
1 2 3 4 5
0 1 2 3 4 5 6
×104
.......................
......................
......................
......................
......................
......................
......................
...
......................................................................................................................................................................
A
Fig. 5.2: Uncorrected, individual, cold PLIF image acquired at unsafe condition
from the streamline of the surrounding gas (Hardalupas et al. 1994; Becker and
Hassa 2003). The observed uniformity of droplet sizes in region A suggests that the
centrifugal effect of the recirculation vortices is similar at the two test conditions.
Figure 5.4 also confirms that the number density of droplets is larger at the unsafe
condition. This is perhaps unexpected as the fuel flow rate is identical at both set
points. However, the global distribution of droplet size is significantly different at
the two set points as demonstrated by the back-lit images of cold atomised fuel
presented in Figure 5.5. At the safe condition, where the air flow rate is relatively
low (Figure 5.5(a)), a cone of large droplets can be seen emerging from the pilot
injection point, whereas no large droplets are apparent at the unsafe set point where
the air flow rate is relatively high (Figure 5.5(b)). The droplets generated at the
unsafe condition are smaller due to the improved performance of air-blast atomisers
at higher air flow rates (see Section 2.2.1). As noted above, the PLIF signal in
region A is due to small droplets only. The momentum of the larger droplets carries
them onto the internal surfaces of the combustor. Consequently, there is a high
5. Results II: Fuel Imaging 112
1
2
3
4
5
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
(i)
(j)
0
6
×104
Fig. 5.3: Cold PLIF details: (a)-(e) safe condition; (f)-(j) unsafe condition
5. Results II: Fuel Imaging 113
10
00
10
00
0
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
(i)
(j)
102
103
104
105
Fig. 5.4: Cold PLIF details with logarithmic scale: (a)-(e) safe condition; (f)-(j)
unsafe condition
5. Results II: Fuel Imaging 114
level of signal on the outside wall at the safe condition (as demonstrated in the next
section), and a relatively low level of signal in region A. The opposite is true at the
unsafe set point.
(a) (b)
Fig. 5.5: High speed images of cold fuel droplets: (a) safe; (b) unsafe. The colour
map has been inverted so that black corresponds to the highest intensity.
As the relative amounts of fuel deposited on the combustor walls and fuel in sus-
pension are unknown, the local equivalence ratio in region A is not well represented
by the global equivalence ratio. On the basis of the images presented above, the
equivalence ratio below the igniter tip is higher at the unsafe condition than at the
safe set point, despite the global equivalence ratio being much leaner at the unsafe
condition (unsafe: 0.24, safe: 0.45). Relight is impossible at the unsafe condition,
so the relatively high fuel signal can be interpreted in two ways: either the mixture
strength is more favourable at the unsafe condition and failure is entirely due to
turbulent strain, or the mixture is locally too rich for ignition to occur.
The possibility of a locally rich mixture can be evaluated from the PLIF images
in Figure 5.4. These indicate that a rich mixture would be produced if the fuel
contained in region A was fully vapourised by the spark. This could occur at both
5. Results II: Fuel Imaging 115
test conditions, but would be more extreme at the unsafe set point since more fuel
is present in the detail images at this condition. The lack of flame growth at the
unsafe condition could therefore be caused by an equivalence ratio that is beyond
the rich flammability limit. The combined effects of an unfavourably rich mixture in
the CRZ and excessive turbulent strain are the most likely causes of ignition failure
at the unsafe condition.
5.1.2 Mean cold images
Mean, cold images illustrating the average fluorescence signal recorded at the two
test conditions are presented in Figure 5.6. With the exception of a CCD background
subtraction, these images are uncorrected
〈Iu〉 = 1
N
N∑
j=1
(Ij − B) , (5.1)
where Ij are the individual uncorrected images and B is the CCD background.
Figures 5.6(a) and 5.6(b) represent the average signal from 287 and 281 images
acquired at the safe and unsafe test points respectively (see Section 4.3.2). The
upper and lower extents of the sheet are identified by solid white lines. The fuel
injector centre-line is also included. These features were mapped onto the image
using information from a CAD model of the apparatus, together with the camera
viewing angle correction described in Appendix A. Note that the imaged region
extends from the fuel-injector face on the left to immediately downstream of the
igniter on the right, as identified previously in Figure 3.6.
Though exaggerated by the colour map, the signal level is higher in Figure 5.6(b)
than in Figure 5.6(a). This is consistent with the observations about the detailed
images noted in Section 5.1.1 where it was shown that the number density of fuel
droplets is higher at the unsafe condition. It is this higher number density of droplets
that leads to overall higher signal levels in the mean unsafe image compared to the
mean safe image.
5. Results II: Fuel Imaging 116
(× 103)
2 4 6 8
(× 103)
2 4 6 8
0 2 4 6 8 10
×103
0 2 4 6 8 10
×103
10 mm
(a) (b)
Fig. 5.6: Uncorrected, mean, cold PLIF images: (a) safe; (b) unsafe. The igniter tip
is identified by a filled red circle. A black box is shown around region A.
The mean images illustrated in Figure 5.6 are qualitatively similar in several
respects. The spatial distribution of fuel is broadly comparable at the two test
conditions. In both cases, the recorded fluorescence originates principally from the
bottom half of the excitation volume. The signal is relatively weak immediately
beneath the combustor outside wall. Indeed, the intensity count is negligible in the
upstream, outside corner of the chamber. A region of comparatively high intensity is
apparent in both images, centred approximately 50 mm downstream from the fuel
injector and 25 mm below the outside wall. The measured signal decreases both
upstream and downstream of this region.
Fuel is introduced close to the injector centre-line. It is therefore credible that
the comparatively low levels of fluorescence observed upstream accurately reflect
an absence of fuel. This is, however, less likely downstream of the high-intensity
region. The lower fluorescence intensity recorded at this location may have resulted
from either attenuation of the excitation energy or a change in sheet cross-section.
5. Results II: Fuel Imaging 117
Corrections for these factors are discussed below in Section 5.2.
In addition to these qualitative similarities, a further feature is also apparent in
both images. A narrow band of low-intensity signal extends across the top half of
the excitation volume. This dark stripe, particularly evident in the unsafe image,
coincides with the path of an optical ray originating from the final lens, and was
therefore initially attributed to lens contamination. As explained in Section 5.2.5
however, the stripe is an artefact of the non-Gaussian laser energy profile.
1000
102 103 104
...
...
...
...
.......
..................................................
...
...
...
..
...
...
...
...
........
...................................................
...
...
...
..
signal on
outside wall
reflection
............................................................
... ...
...
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
....
.....
....
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.......
...
scattering effects
Fig. 5.7: Uncorrected, mean, cold PLIF image acquired at safe condition. Note the
logarithmic intensity scale.
At the safe condition (see Figure 5.6(a)), a localised region of strong fluorescence
is observed on the outside wall at the downstream edge of the igniter port. To high-
light this low-level signal, the mean, safe image is reproduced in Figure 5.7 with
the intensity colour map scaled logarithmically. The droplet signal downstream of
the igniter is clearly visible. In addition, the reflection of this localised fluorescence
can also be seen below the excitation volume. Though the wall signal and its re-
flection are also present at the unsafe condition, they are considerably weaker (see
Figure 5.6(b)), indicating a difference in the quantity of fuel on the outside wall at
5. Results II: Fuel Imaging 118
the two test conditions.
The mean, safe, cold image illustrated in Figure 5.7 contains clear evidence of
scattering, as internal features of the combustion chamber are visible. This effect
is largely associated with fuel contamination of the combustor walls. Liquid fuel
residing on the internal surfaces not only fluoresces when excited by scattered laser
light, but also reflects the fluorescent wavelengths.
The fluorescence count was extracted from a small region of every image to char-
acterise the shot-to-shot distribution of fuel signal. The area of interest, identified
as region A in Figure 5.1 and also outlined in Figure 5.6, is situated in the high-
intensity region identified previously. As this area is comparatively small, the beam
energy profile and sheet geometry corrections can be neglected. Histograms of the
total region A intensity are plotted in Figure 5.8, and the sample mean and standard
deviation at each test condition are listed in Table 5.1.
Table 5.1: Statistics of cold, total PLIF intensity in region A.
mean std. dev.
(×108) (×108)
safe 0.50 0.23
unsafe 0.71 0.40
Significant image-to-image variation in the fuel signal is evident at the two test
points. The large data spread (s/x ≈ 50%), in part results from variations in the
bulk AFR (s/x = 4%), and shot-to-shot fluctuations in the laser energy (s/x = 4%).
Since the combination of these factors accounts for only a small proportion of the
variation seen in Table 5.1, the flow field is evidently fluctuating strongly.
As noted previously, the mean intensity is higher at the unsafe condition. The
sample standard deviation is also larger at the unsafe condition. Since the AFR
and laser energy fluctuations are the same in both cases, the difference between the
standard deviations is due to higher rms turbulence levels at the unsafe condition.
5. Results II: Fuel Imaging 119
×108
(a)
PLIF intensity count
fr
eq
u
en
cy
0.0 0.6 1.2 1.8 2.4 3.0
0
15
30
45
60
×108
(b)
PLIF intensity count
fr
eq
u
en
cy
0.0 0.6 1.2 1.8 2.4 3.0
0
15
30
45
60
Fig. 5.8: Histograms of cold PLIF signal in region A: (a) safe; (b) unsafe. Note that
region A is identified in Figure 5.1.
5. Results II: Fuel Imaging 120
5.1.3 Influence of flame
In addition to the numerous images of cold fuel, 106 kerosene fluorescence images
were acquired in the presence of developing and stabilised flame. The flame emission
image acquired at the same instant as each fuel fluorescence shot has been extracted
from the high-speed flame recordings. The bounds of region A can be transformed
from the PLIF image to the corresponding high-speed image using the method
described in Appendix A. The effect of flame on the fuel signal is illustrated in
Figure 5.9, where the corresponding PLIF and flame intensities contained in region
A are plotted.
×104
×108
flame intensity count
P
L
IF
in
te
n
si
ty
co
u
n
t
•
•
•
•
•
•
•
•
•
•
•
• •
•
•
•
•
•
•
•
•
•
•
•
•
•
•
• developing flame
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦
◦◦ ◦
◦
◦
◦
◦
◦
◦
◦ ◦
◦
◦ ◦ ◦◦◦
◦
◦ ◦
◦
◦
◦◦
◦
◦ ◦
◦◦
◦ ◦◦
◦
◦
◦
◦
◦ stablised flame
...
...
...............................
...
...
...
..............................
...
Fig. 5.10
Fig. 5.11
0 2 4 6 8
0.0
0.3
0.6
0.9
1.2
Fig. 5.9: Scatter plot of PLIF intensity versus flame intensity.
When no flame is apparent in region A, the fluorescence intensity count has a
distribution broadly corresponding to that observed for the cold images. The mean
intensity count is 0.41× 108, compared to 0.50× 108 for fuel alone. When flame is
detected in region A, the fluorescence count is significantly lower, with a mean of
0.08× 108. Once flame is detected, the fuel signal is also largely independent of the
5. Results II: Fuel Imaging 121
recorded flame intensity.
For similar levels of detected flame, the fluorescence signal recorded during flame
development (filled circles in Figure 5.9), is generally larger than that measured
following stabilisation (open circles). The reason for this is that stabilised flame
occupies the entire combustor, whereas developing flame apparent in region A may
lie outside the plane of the laser sheet. In this case the PLIF signal is elevated due
to the presence of unburnt fuel.
Two pairs of simultaneous shots have been selected for closer examination. These
are circled in Figure 5.9, and are displayed in Figures 5.10 and 5.11. Note that it
was not possible to acquire perfectly simultaneous images due to uncertainties in
the spark discharge timing. However, the maximum offset of the camera exposure
periods never exceeded 88 µs.
In Figure 5.10, flame has consumed the regions of fuel vapour and small droplets
evident in the cold images (cf. Figure 5.1), significantly reducing the global signal.
In the enlarged details of region A, significant flame activity is apparent, causing
the consumption of all but three large fuel droplets.
Each of these droplets is surrounded by a low-intensity halo which may result
from fluorescence of intermediate combustion products, such as polycyclic aromatic
hydrocarbons (PAH). Such compounds are known to fluoresce in the near ultraviolet
and blue regions of the electromagnetic spectrum (Ciajolo et al. 2001; Ossler et al.
2001). Alternatively, this signal may derive from flame interference. As discussed
in Section 3.3.1, excited radicals produce emissions within the sensitivity range of
the intensified camera (300 to 480 nm). However, the highly non-premixed nature
of the flame suggests that radiative emissions from soot may interfere with the fuel
fluorescence signal most strongly, even at these comparatively short wavelengths
(Seyfried et al. 2007).
Despite the intense flame signal in Figure 5.10, little interference is apparent in
the fuel fluorescence image. This is also true in the example of exceptionally high
5
.
R
esu
lts
II:
F
u
el
Im
a
g
in
g
122
(× 104)
1 2 3 4 5 32 64 96 128 160 192 224
(a) (b)
0 1 2 3 4 5 6
×104
0 32 64 96 128 160 192 224 256
.........................................................................................................................................................................................
...........................................................................................................................................................................................................
...........................................................................................................................................................................................................
.....................................................................................................................................................................................................
Fig. 5.10: Simultaneous images of fuel and flame: (a) uncorrected hot PLIF image; (b) flame emission image.
5
.
R
esu
lts
II:
F
u
el
Im
a
g
in
g
123
1000 10000 32 64 96 128 160 192 224
(a) (b)
102 103 104 105 0 32 64 96 128 160 192 224 256
.........................................................................................................................................................................................
...........................................................................................................................................................................................................
...........................................................................................................................................................................................................
.....................................................................................................................................................................................................
Fig. 5.11: Simultaneous images of fuel and high-intensity flame: (a) uncorrected hot PLIF image. Note the logarithmic
intensity scale; (b) flame image containing highest intensity emission in region A.
5. Results II: Fuel Imaging 124
flame intensity illustrated in Figure 5.11. The flame image illustrated in this figure
contains the largest intensity count recorded in region A. Note that the fluorescence
image and detail are presented with a logarithmic intensity colour map to highlight
the low-level signal. It is clear from a comparison of Figures 5.11(a) and 5.11(b), that
the shape of the saturated flame activity broadly corresponds to that of low-level
signal in the PLIF image.
Whatever its origin, the flame interference signal is evidently comparatively
small. While large droplets can generate pixel counts approaching 100% of the cam-
era’s dynamic range, the brightest recorded flame contributes signal representing no
more than 3% of this detection range. This kerosene PLIF technique is therefore
suitable for the detection of fuel in flames operating under altitude conditions.
(a) (b)
Fig. 5.12: Fuel droplets following flame stabilisation: (a) individual image; (b)
composite of all acquired images. A threshold value of 104 was used to binarise the
PLIF images.
The droplets apparent in Figures 5.10 and 5.11 are sufficiently small to remain
suspended in the CRZ without impinging on the combustor walls, yet large enough
to avoid being totally consumed by flame. Since only a small fraction of the total fuel
5. Results II: Fuel Imaging 125
mass survives in this form, the individual droplets in each hot PLIF image appear
as well-spaced, similarly-sized, regions of high-intensity signal. By applying a fixed
threshold to each fluorescence image it is possible to construct a scatter plot of fuel
droplet placement that requires no correction. This approach was not possible in
Section 5.1.1 due to the continuous signal variation caused by the presence of small
droplets and fuel vapour. The regions of the PLIF image presented in Figure 5.11
with a signal count exceeding 104 were identified and used to construct the binary
image illustrated in Figure 5.12(a). A composite image formed by overlaying all of
the binary hot PLIF images of stabilised flame is shown in Figure 5.12(b). Droplets
most frequently occupy the bottom half of the excitation volume, with relatively
little fuel found immediately below the combustor outside wall. Note that compared
to the uncorrected mean images, the concentration of signal has moved downstream
and inwards towards the centre-line of the combustion chamber.
5.2 PLIF image correction
Though the preliminary evaluation described in Section 5.1 provided useful insight,
the global fuel distribution is not well represented by the uncorrected images. Point-
to-point differences in measured signal arise from absorption and scattering of the ex-
citation wavelength, the varying cross-section of the sheet, and the non-uniform dis-
tribution of laser energy. Corrections for these factors are described in Sections 5.2.1
to 5.2.4.
Fluctuations in laser energy are also responsible for shot-to-shot differences in
signal level as described previously in Section 4.2.1. No correction has been applied
for these variations, and individual images are not therefore directly comparable.
However, the influence of the fluctuations is removed by averaging, and the mean
safe and unsafe images are compared in Section 5.2.5.
5. Results II: Fuel Imaging 126
5.2.1 Scattering and reflections
When PLIF is used to investigate the fuel distribution in a spray, the magnitude
and placement of the fluorescence signal are influenced by optical absorption and
scattering. These effects have proved to be significant during previous studies, most
notably those investigating the dense, hollow-cone sprays produced by automotive
fuel injectors (Sick and Stojkovic 2001; Stojkovic and Sick 2001; Koh et al. 2003).
In the absence of flame, absorption of the excitation wavelength is governed by
the mass of fuel contained in the excited region of the spray. The degree of optical
scattering is determined by the spray density, usually defined as the liquid surface
area per unit volume. These processes modify the apparent fuel concentration in
three ways. Firstly, a dense spray can significantly attenuate the laser sheet. Sec-
ondly scattered light may induce secondary fluorescence of fuel droplets outside the
sheet. Finally, the fuel lying between the sheet and the camera can attenuate the
primary fluorescence signal.
The cold PLIF images acquired during this study generally contain a region of
diffuse signal below the excitation volume. Liquid droplets in this region fluoresce
when excited by scattered laser light and reflect fluorescence radiated from the sheet.
The structure of this signal is illustrated in Figure 5.13 (cf. Figure 5.2), where a
logarithmic scale has been used to emphasise low-intensity regions. Two portions of
the image are shown in detail: region A inside the sheet as previously identified in
Section 5.1, and region B at a point outside the sheet where scattering was suspected
to be significant. The intensity legend indicates that the maximum signal in region
B (light blue) is approximately two orders of magnitude less than that recorded in
region A (red). Scattering effects are nonetheless evident and should be accounted
for.
The shot-to-shot distribution of scattered signal was characterised by calculating
the total intensity contained in region B for each individual cold image. Histograms
of the region B intensity are plotted in Figure 5.14, and the sample mean and
5. Results II: Fuel Imaging 127
1000 10000
102 103 104 105
A
B
.....
....
....
.....
....
.....
....
....
.....
....
....
....
.....
.....
....
....
....
....
....
....
....
.....
....
....
.....
....
....
.....
....
.....
....
....
.....
....
..
.............................................................................................................................................................
....................................................................................................................................
.........................
.....................................................................................................................................................
Fig. 5.13: Uncorrected, individual, cold PLIF image acquired at unsafe condition.
Note the logarithmic intensity scale.
standard deviation at each test condition are listed in Table 5.2. The fluorescence
intensity distributions extracted from region B, reflect trends observed in region A.
Significant image-to-image variation in the fuel signal is evident, and the mean and
standard deviation are both larger at the unsafe condition than at the safe set point.
Table 5.2: Statistics of cold, total PLIF intensity in region B.
mean mean std. dev.
per pixel (×107) (×107)
safe 746 0.78 0.25
unsafe 1186 1.24 0.63
To minimise the sheet signal resulting from these effects, scattering constants
were subtracted from both uncorrected mean images. The safe scattering constant,
extracted from the safe mean image, corresponded to the mean count per pixel
recorded in region B. This value of 746 counts was subtracted from every element
5. Results II: Fuel Imaging 128
×107
(a)
PLIF intensity count
fr
eq
u
en
cy
0 2 4 6 8
0
25
50
75
100
125
×107
(b)
PLIF intensity count
fr
eq
u
en
cy
0 2 4 6 8
0
25
50
75
100
125
Fig. 5.14: Histograms of cold PLIF signal in region B: (a) safe; (b) unsafe. Note
that region B is identified in Figure 5.13
5. Results II: Fuel Imaging 129
of the safe mean image. The unsafe scattering constant, equal to 1186 counts,
was calculated in an identical manner, and similarly subtracted from every pixel of
the unsafe mean image. These scattering corrections were applied prior to further
processing.
The assumption justifying this approach is that scattering raises the fluorescence
signal radiating from every point in the laser sheet by an equal amount. Though
the situation is clearly more complex than this, assessing the exact spatial variation
in these effects is challenging without droplet size information, and perhaps unnec-
essary under these conditions, where the effects are not particularly significant.
Having subtracted the scattering constants, the resulting mean images were mul-
tiplied by the sheet thickness, frontal area and beam energy distribution correction
matrices. The operations performed during corrective processing can therefore be
represented as follows
〈Ic〉 =
(〈Iu〉 − S) 3∏
k=1
ψk , (5.2)
where 〈Ic〉 and 〈Iu〉 are the mean corrected and mean uncorrected images (see
Eq. (5.1)), S is the scattering constant, and ψk (k = 1, 2, 3) are the three correction
factors. The derivation of these correction matrices is discussed in Sections 5.2.2
to 5.2.4.
5.2.2 Sheet thickness
The thickness of the laser sheet varies considerably within the imaged area. Conver-
gence of the beam caused by the third sheet-formation lens causes a reduction in the
centre-line thickness from 1.7 mm at the left edge of the PLIF image to 1.2 mm at
the right edge. More significant changes in sheet thickness throughout the imaged
area result from the elliptical cross-section of the sheet. Since the edges of the sheet
are particularly thin (minimum thickness = 0.6 mm), a correction factor based on
the sheet thickness is required to correct the mean images.
The correction factor at a point is inversely proportional to the sheet thickness.
5. Results II: Fuel Imaging 130
This correction factor therefore varies with both the radial distance from the focal
point, and the angle to the sheet centre-line. The sheet thickness correction factor,
ψ1, is therefore given by
ψ1 =
tref
t
, (5.3)
where tref and t are the sheet thicknesses at a fixed reference point and the point
requiring correction respectively. The three correction factors are, by definition,
equal to 1 at the fixed reference point. This is located 100 mm from the focal point of
the final lens, on an optical ray bisecting the lower half of the sheet (white, filled circle
in Figure 5.18(a)). The sheet thickness at any point can be calculated by substituting
its radial and angular position into Eq. (B.14). Note that the correction factor was
not calculated beyond the clipped boundaries of the sheet, and singularities at the
unclipped edges were thus avoided.
Discrete droplets are resolved in individual PLIF images, generating localised
regions of high intensity count. Corrective amplification or attenuation of an indi-
vidual image therefore requires the addition or removal of signal representing discrete
droplets. Since the location of these droplets is unknown, it is not possible to apply
the sheet thickness correction to individual PLIF images. However, element-wise
averaging eliminates the localised high-intensity regions, allowing the correction to
be applied to mean images.
5.2.3 Frontal area
The PLIF excitation volume is illustrated in Figure 3.7. During propagation through
the combustion chamber, the inside and outside edges of the sheet diverge, and the
sheet thickness decreases. A single correction factor has been derived to account
for the point-to-point irradiance variations resulting from these changes in the sheet
cross-section.
On entering the combustion chamber, the propagating front of the laser sheet
forms a series of curved surfaces centred at the focal point of the final lens. The
5. Results II: Fuel Imaging 131
correction factor required at a point is proportional to the area of the curved surface
containing the point, and thus varies solely with the radial distance from the focal
point (see Appendix B). The frontal area correction factor, ψ2, is therefore given by
ψ2 =
A
Aref
, (5.4)
where A and Aref are the areas of the surfaces containing the point requiring cor-
rection, and the fixed reference point respectively. As described in Appendix B, a
general expression was derived for the area of the surface passing through every im-
aged point. Each area was evaluated numerically, and the corresponding correction
factor calculated.
Both the sheet thickness and frontal area correction factors vary with the radial
distance from the focal point of the final lens. These correction factors are plotted as
a function of radius in Figure 5.15(a). A black, filled circle identifies the correction
factor reference point. Both factors increase monotonically with distance from the
focal point. The combined effect of the two radial corrections varies between a factor
of 0.5 and 1.5. This is relatively small compared to the angular corrections described
in the next section.
The effect of the combined radial corrections is also illustrated in Figure 5.15(b).
The uncorrected fluorescence profile was extracted from the mean, safe image along
the optical ray passing through the fixed reference point (radial, white dashed line in
Figure 5.18(a)). At larger radial distances, the combined correction factor flattens
the fluorescence profile. This indicates that the downstream region of low-intensity
signal observed in the uncorrected, mean images, is predominantly due to sheet
geometry effects. Excitation wavelength attenuation is therefore less significant than
might be suspected based on the uncorrected images alone.
5.2.4 Beam energy distribution
The principal cause of spatial irradiance variations is the non-uniform energy profile
of the laser beam. The slit traverse results described in Section 4.2.2 were used to
5. Results II: Fuel Imaging 132
(a)
radial distance from focus (mm)
co
rr
ec
ti
o
n
fa
ct
o
r
.
.
.
.
.
.
.
. .
. .
. .
. .
. .
. .
. .
frontal area
.... ....
.... ....
.... ..
.. ....
...
.... .
... .
....
....
..
sheet thickness
.....
.....
.....
....
.....
.....
.....
.....
.....
.....
.....
.....
....
.....
.....
....
.....
....
.....
.....
.....
....
.....
total
•
0 50 100 150
0.0
0.5
1.0
1.5
2.0
×104
(b)
radial distance from focus (mm)
P
L
IF
in
te
n
si
ty
co
u
n
t
....
.......
....
...
...
...
...
....
...
...
...
.
...
...
.
....
...
...
...
...
...
...
...
...
...
...
....
...
...
...
.
..
...
...
...
...
..
....
...
...
...
.........
.
...
...
...
..
.....
...
...
..
......
...
...
...........
...
..
....
.
..
....
...
...
...
...
..
........
...
.
..
.....
...
...
...
...
...
...
...
...
.
..........
...
...
...
...
....
...
...
...
...
...........
...
...
...
....
...
...
...
.
..
....
..
..
.....
..
...
.
...
..
..
..
...
...
...
.
...
.
...
..
...
.
....
...
.............
..
.....
...........
..
..
...
...
...
.
....
...
...
..
...
...
...
.....
....
...
..
...
..
..
...
...
..
..........
.
..
...
...
.......
.
.
.
...
.
......
..
.
...
...
...
..............
..
...
...
.
...
.
.
..
...
..
.
.
...................
.............
...
...
...
...
...
...
...
.........
..
.........
..............
..
........
...
..
.
..
..
.
...
..
..
....
...
...
...
...
...............
...
.
....
...
...
.
...
..
..........
...
...
..
...
...
..
...
.
......
...
..
...
...
...
...
...
..
..
............
.....
...
...
...
..
...
....
..
...
..
...
................
.
....
...
.
..
...
.
.
...
...
...
...
...
.
..............
.......
..
........
..
...
...
..
...
.
..
.
..
...
....
.
..
.
.
..
...
................
.
..
...
...
...
...
.
....
..
...
.
..........
.....
.
....
...
...
...
...
...
.
.
...
...
..
........
..
...
..
...
..........
...
.
..
...
...
.......
...
.
...
...
..
.
.
.............................................
...
...
...
..
.....................
...
...
...
...
...
..
....
...
....
...
.
.....
.....
...
...
...
..
..................
.
....
...
...
...
...
..
.....
......
...
.
..
...
...
.
...
...
..
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..
................
...
...
..
.....
..
.
...
...
...
...
..
...
....
...
.
...
.
.....
.
..
..
...
...
...
...
.
.....
.
...
...
...
...
...
..
......
...
...
...
...
...
..
....
...
...
.
......
...
.
.
.
.........
...
..
...
...
..
.
.
.
...
...
...
...
.
.......
...
...
..
...
...
...
.
.
.....
...
...
...
.
..........
...
...
.
....
.....
..........
...
...........
...
...
...
.
.............
uncorrected
..
...
...
...
...
.
.
.
...
...
..
.
...
...
...
..
...
...
...
..
...
...
..
...
..
...
......
..
...
..
...
...
..
.....
...
.
.......
...
.
..
.....
...
...
..
.....
....
...
...
.
.....
...
...
..
...
...
...........
...
.
..
...
...
...
...
.
.....
...
.
...
...
.
....
...
...
...
..
........
...
..
.......
..
.......
...
...........
......
...
...
.
.....
...
...
.
...
..
.
.
.
.
...
..
.......
.
...
...
...
....
.........
........
......
.............
....
.
.
..
......
....
......
..
..
.
.
.
..
.
...
...
.
...
.
...
....
........
...
.
..
...
...
...
..
.
.
...
...
...
..
...
...
...
.
...
..
...
...
...
...
.
..
...
...
.
...
...
...
...
...
...
..
.....
...
...
...
.
.............
....
...
........
..
...
...
...
...
...
...
...
..
...
...
..
......
...
.......
...
...
.
.....
.
...
.
...
...
...
...
.
....
..
......
...
..
...
.
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..
...
...
...
...
...
...
...
...
...
...
...
.
...........
..
...
.
...
...
...
...
...
...
.
..........
.
...
...
...
.
.
...
...
.
...
...
...
...
.
...
.
...
...
.
..
...
...
.
...........
...
...
...
.....
.......
...........
...
.
.
.........
..
...
...
...
...
...
.
......
corrected
0 50 100 150
0.0
0.2
0.4
0.6
0.8
1.0
Fig. 5.15: Radial correction effects: (a) variation in radial correction factors with
distance from focal point; (b) effect of combined corrections on mean, safe cold
profiles along radial line (see Figure 5.18(a)).
5. Results II: Fuel Imaging 133
correct for these non-uniformities in the incoming beam.
The beam energy profile was characterised as the measured change in power gra-
dient with distance through the beam. On leaving the final lens, the propagating
front of laser light forms a series of concentric arced surfaces. The irradiance varia-
tions in the sheet can therefore be quantified by mapping the measured beam profile
along the frontal arcs so that the edges of the beam lie on the inside and outside
boundaries of the unclipped sheet. The correction factor required at a point is then
inversely proportional to the measured power gradient mapped onto the point. The
beam energy distribution correction factor, ψ3, is therefore given by
ψ3 =
(dP/dl)ref
(dP/dl)
, (5.5)
where (dP/dl)ref and (dP/dl) are the power gradients mapped onto the correction
factor reference point and the point requiring correction respectively.
The measured beam power gradient is plotted in Figure 5.16 as a function of
normalised angle to the optical axis. The fluorescence profile along a frontal arc
was extracted from the mean, unsafe image, and is also presented in Figure 5.16.
Normalised angles of -1, 0 and 1 correspond to the unclipped inside edge, centre-line
and unclipped outside edge of the sheet respectively. The clipped sheet boundaries
are represented by dashed lines. The non-Gaussian energy shoulders first noted
in Section 4.2, clearly influence the fluorescence profile. Indeed, the upper feature
(normalised angle ≈ 0.5), has been used to locate the beam energy profile relative
to the sheet. This was necessary due to difficulties in identifying the centre of the
measured energy distribution (see Section 4.2.2).
These observations confirmed the importance of calculating the beam energy
correction based on a measured profile. The assumption of a Gaussian distribution
would not have removed image features associated with the laser output.
The beam energy distribution correction factor varies solely with the angle to
the optical axis. In addition to the radial dependence discussed in Section 5.2.3, the
sheet thickness factor also changes with angle. These correction factors are plotted
5. Results II: Fuel Imaging 134
×103
normalised angle to optical aixs
P
L
IF
in
te
n
si
ty
co
u
n
t
p
ow
er
g
ra
d
ie
n
t
(m
W
/
m
m
)
•• • •....... •...... •
....... •.....
.. •......
•.......
•............
•
...
...
...
...
...
...
..
•....................
•
...
...
...
...
...
..
•...........
•....
..
•...
.... •.......
•.........
•
...
...
...
...
...
.
•.........
•
...
...
...
...
...
...
...
•....................
•
...
...
...
...
...
•...
....
•
......
•..
...
...
...
...
..
•
...............................
•..
...
...
...
...
...
...
...
•......
•........... •
.......
•.
...
...
...
...
...
...
...
...
...
...
...
•
..........................
•..
...
...
...
...
...
...
.
•
...........
•..
...
...
..
•
......
• sheet energy
........
.....
.....
......
.
...
...
....
...
....
..
...
.
....
.........
...
...
...
.
.......
...
.
....
...
...
........
...
...
......
...
...
.
...
.......
...
...
...
...
...
...
...
...
........
...
...
...
.
..
...
.
.......
...
.
..
...........
...
...
...
...
...
...
...
.........................
...
...
...
...
.
...............
...
...
...
.
.......................
......
...
...
.
..
...
.
......
......................
...
...
...
.
...
...
...
...
..
...............
...
...
...
.
..............
......
......
...
...
...
............
...
...
.
....
...
...
...
.
........
...
...
...
...
...
...
...
..
..............
...
...
...
...
...
...
...
...
...
...
...
...
...
.
........................
...
...
...
...
...
...
...
...
...
..
.................
...
...
...
...
...
...
...
..
....
...
...
...
...
..
.......
...
...
...
.....................
...
...
...
...
...
...
.
.......
...
...
.
....
...
...
...
....
...
...
...
...
...
.....
...
...
.
.......
.
...
...
...
...
.......
...
...
...
...
.
................
...
...
...
.
...
...
...
...
...
...
...
...
...
.
..................
...
....................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.........
.
.
.....................
...
...
...
...
...
...
...
...
...
...
.
...........
...
......
....
...
.
....
.......
...........................
.
...
...
...
...
...
...
...
...
...
...
...
...
...
.
....................................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....
...
...
...
...
..
..............
....
...
...
...
...
...
.
.......
...
...
...
...
...
...
...
...
.
.......
...
.......
...
...
...
..
.....
..
...
...
...
..................
...
...
...
...
.
....
..
.......
...
...
...
...
.
...............
...
...
...
...
...
...
...
.......
................
...
..
......
...........
.............
...
.
...............
...
...
.
...
.
..................................
...
...
...
.
.................
...
...
...........
...
.
.
...
..
...
...
..
.
.....
.......
...
...
...
...
...
...
...
...
...
...
..
.................
...
.......
..
..
.............
...
...
...
...
...
...
...
...........
...
......
...
...
...
...
................
...
...
...
......................
........
...
...
...
...
...
...
...
.....
...
...
...
..
............
.....
...
...
...
...
...
..
.........
...
...
...
...
..
.............
.................
...
...
...
...
...
...
...
...
..
...........
...
...
...
...
..............................................
...
...
...
...
...
...
...
..
.......................................
...
...
...
...
...
.
.............
...
...
...
.
................
.
....
......
...
...
....
.
................
......................................
..................................
...
.
....
...........
..
........
..
...
..
................
.
.........
...
...
.....
.
..............
............................. .........
fluorescence
................................
.....
−1.5 −1.0 −0.5 0.0 0.5 1.0
0
1
2
3
4
5
6
7
0
25
50
75
100
125
150
175
Fig. 5.16: Comparison of beam energy profile and fluorescence signal.
as a function of normalised angle in Figure 5.17(a). Note that the plots extend
between the clipped boundaries of the sheet. The correction factor reference point
is again identified by a black, filled circle.
The beam energy distribution correction factor is similar in magnitude to the
radial correction factors in the central two quadrants of the sheet (i.e. normalised
angle between -0.5 and 0.5). However, at the outside edges of the sheet, the beam
energy correction is dominant and large. This effect is particularly pronounced at
the outside wall, where a correction factor of 8.8 is required. At both boundaries,
the magnitude of the correction factor is extremely sensitive to the position of the
beam energy profile relative to the sheet. As the shoulder feature used to locate
the energy profile is rather broad, and the extracted fluorescence plot is relatively
noisy (see Figure 5.16), the exact placement of the energy profile is uncertain. This
error is less important towards the centre of the sheet, where the correction factor
is much less sensitive to misalignment of the energy profile.
5. Results II: Fuel Imaging 135
(a)
normalised angle
co
rr
ec
ti
o
n
fa
ct
o
r
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...........
.....
.
.
.
.
.
.
.
.
beam energy
............................................................
......
.
sheet thickness
............................................................................................................................................................................................
........
......
...
...
...
...
..
...
...
...
...
...
...
...
...
...
...
...
...
...
...
total
•
−1.5 −1.0 −0.5 0.0 0.5 1.0
0
6
12
18
×104
(b)
normalised angle
P
L
IF
in
te
n
si
ty
co
u
n
t
....
....
..
...
..........
......
.........
....
.....
...
.....
.....
......
.....
......
...
.
...
..
...
...
...
........
...
..
....
...
......
.
...
..
.....
........
......
...
...
........
..
...
...
...
...
...
...
............
...
...
.....
...
.........
...
...
...
.
.....
...
..
...
...
...
..
.........
...
...
...
.
.....
.
...
..
......
...
...........
...
...
...
...
...
.........
...
..
...
...
...
...
...
...
.
..........
.....
.
..
..
.
.
............
.
...
..
............
....
...
.
..........
...
...
.
...
...
....
...
...
.
......
...
...
.
........
...
...
...
...
..
.......
.
.......
.....................
...
...
...
...
...
.
...
................
........
...
...
...
...
...
.
..................
...
...
...
.
...............
......
...
.
...........
...
...
.
...........
..
...
...
...
...
...
...
.
......
...
...
....
.....
...
...
........
...
..
..
.....
.
.....
....
...
...
...
...
...
...
..
....
..................
.............
..............
.........................
....................................................
uncorrected
...
.
...........................
...
...
...
...
..
......................................
...
...
.........
...
..
...
...
...
...
...
...
...
......
...
.
............................................
...
...
...
.
.........
...
...
...
................
.......
......
...
...
.....
....
...
.....
.
.....
..
......
...
.
.......
.
......
.......
.....
...
.......
...
.
...
..........
...........
...
..
....
....
.
......
.....
......
.......
...
.......
.....
...
......
....
...
......
........
...
.
......
...
....
.
..
...
...
...
...
.
.....
...
..
...
.......
......
.............
...
...
...
...
..
...
.
........
......
...
..
.....
...
.
...
.
..
..
.
.
.
...
....
.
...
...
......
...
...
.
..
..
.....
..
...
...
..
..
...
...
...
...
...
...
...
...
...
.
...
...
...
..
..
...
...
...
...
.
.........
...
..
............
.
...
...
...
...
.
..........
...
...
...
...
.
.................
...
..
......
...
...
.
......
...
...
...
.
....
...
...
...
...
...
...
...
...
...
...
....................
...
...
...
..
corrected
−1.5 −1.0 −0.5 0.0 0.5 1.0
0.0
0.5
1.0
1.5
Fig. 5.17: Angular correction effects: (a) variation in angular correction factors with
normalised angle to optical axis; (b) effect of combined corrections on mean, safe
cold profiles along frontal arc (see Figure 5.18(a)).
5. Results II: Fuel Imaging 136
The sheet thickness correction factor also increases towards the edges of the sheet
to account for thinning due to its elliptical cross-section. Though the sheet thickness
correction varies by a factor of only 2.1, it combines with the beam energy factor
to generate comparatively large corrections at the sheet edges. The largest total
angular correction factor is applied at the outside boundary and is equal to 16.3.
The effects of the combined angular corrections on the mean, safe fluorescence
signal are illustrated in Figure 5.17(b). The uncorrected fluorescence profile was ex-
tracted from the mean, safe image along the arc passing through the correction factor
reference point. This curve is identified as a white dashed line in Figure 5.18(a).
The uncorrected profile is radically modified at the edges by applying the angular
corrections. A correction factor greater than about 3 is regarded as questionable
since the noise associated with low-level signal is amplified along with the signal
itself. Therefore subsequent analysis of the mean images in the following section
considers only those regions where a total correction factor of less than 3 is applied.
5.2.5 Interpretation of corrected images
The mean, corrected, cold PLIF images are presented in Figure 5.18. Comparison
of the corrected and uncorrected images (Figure 5.6), reveals important differences.
Firstly, the size of the downstream, low-intensity region has been reduced by the
radial corrections. This indicates that attenuation of the laser energy is not as severe
as the uncorrected images would suggest. Secondly, the dark stripe extending across
the top half of the images in Figure 5.6 has been virtually eliminated by the beam
energy distribution correction. This confirms that the stripe was not caused by
lens contamination as originally suspected, but resulted from non-uniformities in
the beam profile, as measured in Section 4.2.
A further key difference between the corrected and uncorrected images is that the
high-intensity signal close to region A in Figure 5.6 has moved downwards towards
the fuel injector centre-line in Figure 5.18. This results in a wider, low-intensity
5. Results II: Fuel Imaging 137
(× 103)
2 4 6 8
(× 103)
2 4 6 8
0 2 4 6 8 10
×103
0 2 4 6 8 10
×103(a) (b)
Fig. 5.18: Corrected, mean, cold PLIF images: (a) safe; (b) unsafe. Note that areas
with a correction factor greater than 3 have been set to zero.
corridor close to the outside wall of the combustor.
The signal on the outside wall itself was amplified to such an extent by the
corrections that it was considered unreliable and has therefore been removed from
the images. Nonetheless, the uncorrected images analysed previously in Section 5.1
showed that more fuel is deposited on the outside wall at the safe condition than at
the unsafe condition.
With the exception of the scattering constant, identical corrections have been ap-
plied to images acquired at the two test conditions. Though the scattering constant
subtracted from the images is smaller at the safe condition, the mean fluorescence
intensity remains higher at the unsafe set point, reproducing the trend identified in
the uncorrected images (see Section 5.1.2). The elevated signal at the unsafe condi-
tion therefore indicates the presence of more fuel in suspension below the igniter.
The distribution of fuel apparent in the safe corrected image corresponds closely
to that observed in the hot PLIF scatter plot of mid-sized droplets presented in
5. Results II: Fuel Imaging 138
Figure 5.12. The placement of each fuel droplet in this plot is unaffected by the PLIF
signal strength, and no correction is therefore required. Figures 5.12 and 5.18 thus
provide evidence that fuel is present in greater concentrations towards the inside,
downstream boundary of the excitation region. This suggests that flow recirculation
is trapping fuel in this region whether or not a flame is present.
5.3 CFD comparison
Computational modelling of the combustor air-flow has been performed by Rolls-
Royce plc to develop an understanding of the cold flow fields at the safe and unsafe
set points (Goddard and Zedda 2007). A non-reacting large-eddy simulation was
conducted using the Rolls-Royce corporate combustion program, PRECISE. The
modelling implemented in this software is described by James et al. (2006). The ge-
ometry and boundary conditions defined in the model were not identical to those of
the test facility, but instead corresponded to a single sector with periodic boundary
conditions (i.e. representing a full combustor annulus). The results of the simula-
tion nonetheless provide a useful qualitative comparison with the experiment. The
CFD results presented below are the time-averaged components of the velocity field
extracted from the injector mid-plane.
The LES method provides a more accurate and complete description of the flow
field than alternative Reynolds-averaged Navier-Stokes (RANS) techniques (Pope
2000; Boudier et al. 2007). There is, however, a relatively high computational cost
associated with LES, particularly for high Reynolds number, wall-bounded flows.
Furthermore, incorporating finite-rate chemistry into an LES method to model tur-
bulent reacting flows is prohibitively expensive, ruling out combustion LES as a
design tool at present (Cant and Mastorakos 2008).
Time-averaged velocity vector maps generated by LES were extracted from the
plane of the laser sheet and transformed into the PLIF imaging plane according to
the process described in Appendix A. In this way the flow field in Figure 5.19 appears
5. Results II: Fuel Imaging 139
as if viewed from the PLIF camera. Thus the corrected PLIF images (Figure 5.18)
can be directly compared with the transformed CFD velocity fields.
(a) (b)
20
m/s
Fig. 5.19: Time-averaged velocity in sheet plane predicted by CFD: (a) safe; (b)
unsafe. The igniter tip is identified by a filled red circle.
The following features are common to both the safe and unsafe flow fields (Fig-
ures 5.19(a) and 5.19(b) respectively). A high-velocity stream of air enters the
combustion chamber towards the edge of the fuel injector at an angle of approxi-
mately 45◦ to the injector centre-line. The flow is deflected by the combustor outside
wall, creating a corridor of high-velocity air along the underside of the wall. A lo-
cal recirculation region is evident roughly 25 mm below the igniter tip. As noted
in Section 5.1.1, the flow patterns are broadly similar at the safe and unsafe set
points, reflecting the extremely high Reynolds number in both cases. At any given
point, the ratio of the vector length in Figure 5.19(b) to that in Figure 5.19(a) is
approximately equal to the ratio of air mass flow rates (cf. Table 3.1).
The flow velocities and patterns at the two test conditions support the fuel
placement observations reported earlier in this chapter. The high-velocity air stream
entering at 45◦ behaves as an air curtain preventing fuel injected close to the centre-
5. Results II: Fuel Imaging 140
line from reaching the outer boundary. As argued in Section 5.1.1, large fuel droplets
can penetrate this curtain by virtue of their momentum, resulting in the deposition
of fuel on the outside combustor wall and the associated PLIF signal observed in
Figure 5.7. As indicated in Figure 5.19(a), the air curtain is also weaker at the safe
condition, enabling further droplets to impinge on the wall.
The comparatively high fuel concentration in the lower half of the sheet observed
in Figure 5.18 coincides with an area of low-velocity, recirculated flow. Smaller
droplets tend to accumulate in this recirculation region as evidenced by the PLIF
details in Figures 5.3 and 5.4. As discussed in Section 5.1.1, the recirculating flows
generated under altitude conditions therefore control the point-to-point drop size
distribution inside the combustor. It is this feature that explains the disparity in
the magnitude and distribution of the PLIF fuel signals at the two test conditions.
Thus the CFD results are consistent with the PLIF findings, namely that there is
more fuel on the wall and less in the flow at the safe condition, and less fuel on the
wall and more in the flow at the unsafe condition.
Chapter 6
Results III: Flame Imaging
While the PLIF imaging described in Chapter 5 provides useful insight into fuel
placement prior to the spark event, it is the high-speed recordings of flame radiation
that provide the key information concerning relight ignition and flame development.
The analysis and interpretation of these results is described in this chapter.
6.1 Flame tracking software
Given the large distance separating the igniter from the fuel injector, the spatial
development of the flame is critical in determining the success or failure of a relight
attempt. Image processing software has therefore been developed to monitor the
motion and breakup of flame in each high-speed ignition sequence (Read et al.
2008). The design and implementation of an algorithm to generate a flame centroid
trajectory map corresponding to each ignition event is described in detail below. As
explained later in Sections 6.2 and 6.3, the output from this program was analysed
extensively to investigate the differences between successful and unsuccessful relight
attempts.
141
6. Results III: Flame Imaging 142
6.1.1 Search principle
A schematic illustration of flame development and breakup is presented in Figure 6.1.
The intense light emitted by each spark produces a spatially well-defined saturated
region immediately below the igniter tip (see Figure 4.5). The flame tracking process
is initialised by searching a small square contained in this region (dashed line in
Figure 6.1(a)) for pixels exceeding a high-level threshold. The extent of the spark
kernel is determined by searching outwards from this square, thus excluding the
reflections of the spark apparent in Figure 4.5.
The following frame, represented by Figure 6.1(b), is then searched to identify
‘child’ flame kernels in the area formerly occupied by the ‘parent’ spark. Consecutive
frames are processed in the same manner, on each occasion noting the parents of
any detected flame kernel. By recording the centroid location of every kernel, it is
possible to construct a unique trajectory map illustrating the motion of visible flame
during each ignition event.
(a) igniter
...
......
.............................................................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....
.....
.....
......
.......
........
.......
......................................................................................................................................................................................
.........................................................................
.....
...
...
...
...
...
...
...
...
...
...
.
(b)
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....
.....
.....
......
.......
........
..........
......................................................................................................................................................................................
...........................................................................
.....
...
...
...
...
...
...
...
...
...
...
.
..
..
...
...
...
...
...
...
...
...
...
...
....
......
................................................................................................................
.......
......
....
...
...
...
...
...
...
...
...
..
.....
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....
...
....
....
.....
......
.......
........
........................................................................................................................................................................................
....
parent
child 1
child 2
...............................
...
..
.....
.......
..................
..
............................... .
..
............................
•
•
•
Fig. 6.1: The flame tracking process: (a) first frame; (b) second frame.
The program records the location of the intensity centroid associated with every
flame kernel captured in the high-speed recordings. The kernels are constructed by
searching the pixels that contain flame in the preceding image. Figure 6.2 shows two
illustrative examples of flame kernel construction. The pixels constituting a single
6. Results III: Flame Imaging 143
flame kernel are assigned a unique number. The grey pixels without a number
in Figure 6.2 register a count that is above the threshold, but have not yet been
assigned to a specific kernel.
Let us consider kernel number ‘2’ shown in Figure 6.2(a). Here, the black square
represents the first discovered member of the group. The full extent of this flame
body is identified by searching the area around the black pixel for other high-
intensity pixels. The path of this search follows a series of expanding, concentric
circuits. The 8 elements bordering the black square are searched first, followed by
the 16 pixels surrounding this square, and so on. To join kernel 2, a pixel must have
a count that exceeds the threshold, and must border a pixel that already belongs
to the group. The expanding search continues until a circuit is completed without
assigning a new pixel to the group. For the case illustrated in Figure 6.2(a), the
search continues to the edge of the image, revealing an existing kernel (number 1),
and several unassigned pixels that are not connected to kernel 2.
Though a single expanding search is often sufficient to reveal the entire flame
kernel, this is not always the case. The grey pixels at the top of Figure 6.2(b) exceed
the threshold but remain unassigned to a kernel as the ‘linking’ pixel lies outside the
circuit during the first search. If unassigned cells are encountered during the first
expansion, then a further search is required, this time proceeding from the perimeter
to the centre of the search area. The search oscillates inwards and outwards in this
manner until no new cells are allocated, whereupon kernel 2 may be regarded as
complete.
6.1.2 Algorithm features
Though simple in principle, the implementation of the search algorithm described
in Section 6.1.1 is complicated by several factors. Firstly, the measured flame in-
tensity varies significantly during each ignition event. As noted in Section 4.4, and
illustrated in Figure 4.5, the spark itself is extremely bright and the initial high
6. Results III: Flame Imaging 144
(a)
2
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
2
2
(b)
2
1
1
1
1
1
1
1
2
2 2
2
2
2
2
2
2
Fig. 6.2: The kernel search process: (a) single expansion required; (b) multiple
expansions required.
speed images contain strong reflections on the combustor walls. In contrast, during
later development, small, faint, kernels are observed separating from larger volumes
of flame. These low-intensity fragments often grow to occupy large areas, and it
is therefore important for the tracking program to retain them in addition to the
bright kernels. An adaptive threshold is therefore required to ensure that bright
spark reflections are not classified as flame, while faint flame fragments are success-
fully identified. This is given by
Ithr = pImax where p = 0.5. (6.1)
Imax is the maximum intensity in the parent kernel. Any pixel with a count exceed-
ing this value is assigned to a new child kernel that is expanded to its full extent
according to the procedure described in Section 6.1.1.
However, the small regions of flame described above produce an additional dif-
ficulty. Fragments of flame are often ejected from a parent kernel at comparatively
high velocity. The child kernels depicted in Figure 6.1(b) can therefore lie outside
the boundary of the parent. To ensure detection of these child kernels, a second
low-threshold search is conducted of the parent flame area and an expanded region
6. Results III: Flame Imaging 145
around it (corresponding to ≈ 5% of the image height).
The construction of a child kernel during this expanded search is illustrated in
Figure 6.3. A relatively bright flame body lying within the bounds of its parent is
identified as kernel C1. A faint flame body lying outside the boundary of the original
search must also be identified. An expanded region is searched for pixels exceeding a
low threshold given by Ithr = 1. The filled circle in Figure 6.3(a) represents the first
identified point where I (xi) > Ithr. On discovering this pixel at xi, the threshold is
immediately reset
Ithr = max {Ithr, pI (xi)} , (6.2)
and a new child is constructed as described in Section 6.1.1. If the existing child C1
is encountered while building the new kernel, the threshold is again reset to remove
the pixels in the current child with the lowest count
Ithr = Imin . (6.3)
The dashed line in Figure 6.3(a) indicates the threshold intensity that will separate
the new child from C1. This is achieved by repeatedly resetting the threshold ac-
cording to Eq. (6.3). The separated kernel is then examined to determine whether
Imin > pImax, where Imax is the maximum pixel count contained in the new child. If
this condition is satisfied, then the child build is complete; otherwise the threshold
is again reset
Ithr = pImax , (6.4)
and the final build commences at the pixel registering the maximum intensity. This
process generates the new kernel C2, illustrated in Figure 6.3(b).
The adaptive threshold and expanded search features improve the program’s
ability to retain faint flame, and produce satisfactory tracks of flame activity. How-
ever, these features can also lead to misrepresentation of the flame break-up that
must be addressed. For example, the expanded search tends to create child kernels
with a large number of parents. While this is valid in the case where several flame
6. Results III: Flame Imaging 146
(a)
x
I
(x
)
C1
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....
...............................................................................................................................
...
...
...
...
...
...
...
..
..
..
...
...
...
...
...
..............................................................
•
....................................
...
.. ...
.
xi
(b)
x
I
(x
)
C1
C2
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....
................................................................................................................................
....
...
...
...
...
...
...
..
..
..
...
...
..
...
...
...............................................................
Fig. 6.3: The expanded search process: (a) kernel isolation; (b) final build.
kernels have recombined, the program should also permit independent flame bodies
to pass close to each other without being identified as a single entity. To address
this, the relative intensity of the parents and their proximity to the child is exam-
ined. An apparent parent is regarded as valid if the distance between the parent
and the child is no more than twice that from the closest parent to that child. In
addition, the parent of a given child must not be less than half the size of the largest
parent.
As noted in Section 4.3.1, complete loss of signal occurs during 6 of the 94
successful events recorded at the safe set point. However, a much higher number
of the recordings, 56 in total, include images containing no pixels with a count
6. Results III: Flame Imaging 147
64 128 192
(a)
(b)
(c)
(d)
0 64 128 192 256
Fig. 6.4: Early high-speed images (left) and corresponding kernel outlines (right).
The timing of frames (a) to (d) is indicated in Figure 6.5.
6. Results III: Flame Imaging 148
exceeding the low threshold. These low-intensity episodes typically occur for a
period of less than 1 ms, between 10 and 20 ms following the spark discharge. The
program can continue to track the flame following each of these episodes as follows.
Having discovered an image registering no flame activity, a fixed-threshold search
of the subsequent frames is performed until a new kernel is identified, whereupon
the tracking process continues as described above. Consequently, though breaks in
the trajectory plot occasionally occur, the program is able to track successful events
through to recovery.
5
10
15
20
0
5
10
15
20
25
time (ms)
20 mm
................................
................................
...............................
................................
(a)
(b)
(c)
(d)
Fig. 6.5: Example of a trajectory map. The end of the igniter is represented by
the white filled oval. The labels to the right of the map indicate the timing of each
frame in Figure 6.4.
The primary results obtained from the flame tracking software are a series of tra-
jectory plots, illustrating the motion and breakup of flame during a single ignition
attempt (see Figure 6.4). Figure 6.5 shows an example of a complete trajectory plot
for an unsuccessful ignition attempt at the safe condition. Flame centroid trajecto-
ries are colour-coded with respect to time, progressing from dark red immediately
6. Results III: Flame Imaging 149
following the spark to dark blue 25 ms later. The centre of the spark kernel is
identified by an unfilled red circle. Each line represents the path of an individual
flame kernel. Fragmentation and recombination of these kernels is represented by
branching and merging of the lines respectively. The disappearance of a flame kernel
following extinction is indicated by a cross.
The broken blue trajectories in the top right hand corner of Figure 6.5 indicate
very faint flame exiting the combustion chamber around 20 to 25 ms following the
spark. The periodic loss and recovery of this weak signal illustrates the program’s
ability to continue the flame track following low-intensity episodes.
6.2 Developing flame
Trajectory plots of flame motion have been examined qualitatively to identify cat-
egories of ignition development. The variations in flame size and spark velocity
provide valuable quantitative information concerning the process, and permit inves-
tigation of the systematic differences between successful and unsuccessful ignition
events.
6.2.1 Flame trajectories
The flame tracking software was used to generate a flame trajectory plot for every
high-speed recording of an ignition attempt. A trajectory plot corresponding to a
safe, successful relight event is presented in Figure 6.6(a). As in Figure 6.5, the plot
is colour-coded with respect to time, progressively changing from dark red to dark
blue during a period of 25 ms.
Figure 6.6(a) resembles the flame development maps associated with many suc-
cessful ignition events. Following initial activity below the igniter tip, flame kernels
move upstream, before developing strongly in the lower, upstream quadrant of the
combustion chamber. When recovery occurs, 24.6 ms following the spark discharge,
6. Results III: Flame Imaging 150
5
10
15
20
0
5
10
15
20
25
time (ms)
(a)
(b)
I
II
III
Fig. 6.6: Trajectory plots during 25 ms following spark: (a) safe success,
recovery time = 24.6 ms; (b) unsafe failure, failure time = 5.4 ms.
6. Results III: Flame Imaging 151
flame has spread to occupy a significant proportion of the imaged area.
This result is in stark contrast to the typical unsafe trajectory plot displayed
in Figure 6.6(b). Disintegration of the spark kernel occurs rapidly with little op-
portunity for the flame to spread upstream. This fast breakup results from the
increased flow velocities and turbulence intensities generated at the unsafe test con-
dition, and is reflected in the comparatively short unsafe failure times highlighted
in Section 4.3.1.
Whereas all unsafe ignition failures occur in a similar manner, safe unsuccessful
events fail in a variety of ways. Figure 6.7 consists of four trajectory plots each
representing a different mode of safe failure. Although upstream flame motion is
clearly required for successful relight, flame is also observed leaving the combustion
chamber downstream. This convective loss of flame occurred during many safe tests,
and was sometimes directly responsible for a failure. The four safe failure modes
are in contrast to the unsafe failure in Figure 6.6(b) which involves disintegration
and minimal propagation. The flame tracking program thus confirms the visual
observations of failure categories reported earlier in Section 4.3.1.
The experiments of Ahmed et al. (2007a) identified three types of failures.
Firstly, a spark followed by no indication of kernel initiation, which most resem-
bles an unsafe failure in the current study, but could also broadly describe the safe
failure mode in Figure 6.7(a). Secondly, flame initiation followed by downstream
blow-off that is analogous to Figure 6.7(b). Thirdly, flame propagation and momen-
tary stabilisation before extinction exemplified by Figure 6.7(d). The safe failure
in Figure 6.7(c) represents a failure mode that has not been previously reported,
namely a split in the initial kernel followed by downstream blow-off with simultane-
ous upstream propagation.
The relationship between flame placement and ignition success has been inves-
tigated by calculating the proportion of tests registering flame activity in three
overlapping regions, labelled I, II, and III in Figure 6.6(b). Region I is an upstream
6. Results III: Flame Imaging 152
(a) (b)
(c) (d)
Fig. 6.7: Trajectory plots of safe failures: (a) rapid disintegration; (b) flame exit;
(c) flame split; (d) no stabilisation at fuel injector.
portion of the image close to the fuel injector, region II contains all points below
the injector centre-line, and region III is a downstream area close to the combustor
exit. The statistics obtained from the tracking program are listed in Table 6.1.
The vast majority of safe, successful ignition events register flame in all three
regions of the combustion chamber during the 25 ms following the spark. Most
safe failures also exhibit activity below the centre-line and at the combustor exit.
However, approximately half of these failed attempts display no evidence of flame
at the fuel injector during the first 25 ms. This result indicates that although flame
movement onto the injector face is a prerequisite for successful ignition, it does not
guarantee success – at least 55% of failures occur having achieved this. Furthermore,
as illustrated in Figure 6.8, the distribution of times required for flame to enter the
fuel-injector region is similar for successful and unsuccessful ignition events. This
6. Results III: Flame Imaging 153
Table 6.1: Ignition activity in three regions of the combustor (see Figure 6.6(b))
during the 25 ms following the spark.
safe safe unsafe
% successes failures failures
I: fuel injector 100 55 3
II: below centre-line 99 89 44
III: combustor exit 86 85 74
implies that those unsuccessful flames that enter the injector region, are transported
or propagate at a similar rate to flame kernels that are ultimately successful. Note
the extremely broad range of these detection times, again indicating the highly
stochastic nature of ignition.
At the unsafe condition, flame is observed below the fuel injector centre-line in
less than half of the ignition attempts, and signal is recorded at the injector face in
virtually no cases. Disintegration of the initial flame kernel occurs so rapidly at the
unsafe set point that failure occurs before flame can be transported into regions I
and II. However, the downstream convection of the early flame is sufficiently fast to
generate signal at the exit during the majority of unsafe failures.
Insight into flame propagation is also provided by identifying the positions of last
recorded flame activity during unsuccessful ignition events at the safe and unsafe
conditions, as shown in Figure 6.9. This figure broadly reflects the results observed in
the region analysis. In general, the safe failure occurs at a considerable distance from
the igniter tip, indicating significant migration of the flame prior to extinction. At
the unsafe condition, the failure points are closer to the igniter tip, and concentrated
downstream of this point. At both conditions, a significant number of failures result
from convective loss of flame out of the combustor exit, in the manner evident in
Figure 6.7(b).
6. Results III: Flame Imaging 154
(a)
interval following spark (ms)
fr
eq
u
en
cy
0 5 10 15 20 25
0
5
10
15
20
25
(b)
interval following spark (ms)
fr
eq
u
en
cy
0 5 10 15 20 25
0
5
10
15
20
25
Fig. 6.8: Flame detection times in fuel-injector region at safe condition: (a) success;
(b) failure.
6. Results III: Flame Imaging 155
•
•
safe
unsafe
Fig. 6.9: Points of last activity during failed ignition events.
6.2.2 Flame size distributions
The flame tracking program records the size of all identified flame kernels. By sum-
ming the size of every kernel contained in a single image, the variation in recorded
flame area during each ignition event can be identified. Note that the flame area
extracted from each image does not correspond to the flame surface area, and also
differs from the number of active pixels in the frame as the tracking program excludes
all reflections. Scatter plots of flame area versus time are presented in Figure 6.10.
The mean areas corresponding to safe successful, safe unsuccessful, and unsafe un-
successful ignition trials, are plotted as green, red, and blue lines respectively. Note
that the horizontal axis is logarithmic.
A large degree of scatter is apparent in the data, again highlighting significant
variability in the ignition process. Despite this data spread, the mean plots illustrate
several important features of relight flame development. Firstly, the pixel counts
in the earliest recorded frames indicate that the three spark size distributions are
indistinguishable. As discussed in Section 4.4, this implies that the spark kernel size
6. Results III: Flame Imaging 156
×104
interval following spark (ms)
fl
a
m
e
a
re
a
(p
ix
el
s)
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
••
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
••
•
•
•
•
•
•
•
•
•
•
••
•
•
•
•
•
•
•
•
•
•
•
•
••
•
•
•
•
•
•
•
•
•
•
•
•
•
•
••
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
••
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
••
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
••
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
••
•
•
•
•
•
•
•
•
•
••
•
••
•
•
••
•
••
•
•
•
•
•
•
•
•
•
•
•
••
•
•
•
•
•
•
•
•
•
• •
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•••
........
........
......
....
....
....
...
...
....
....
..........................................................................................................................................................................................
......
...
...
...
...
...
...
..
..................
....
...
....
...
....
...
....
...
.
..............................................................................................................................................................................................
...
...
....
..........................................................................................................................................................................................
safe success
safe failure
unsafe failure
......................
...
..
...
..
......................
...
..
...
..
......................
...
.
...
..
......................
...
..
...
..
......................
...
..
...
..
......................
...
..
...
..
(a) (b) (c) (d) (e) (f)
0.1 1 10 25
0.0
0.3
0.6
0.9
1.2
1.5
Fig. 6.10: Variation in imaged flame area with time. The labels above the plot
indicate the timing of each sub-plot in Figure 6.11.
plays no role in determining ignition outcome, and is uninfluenced by the change in
test conditions.
Immediately following the spark, the area occupied by flame increases at the
safe condition. This is due to the hot kernel expanding and breaking apart. Rapid
quenching of the spark at the unsafe set point causes the pixel count to decrease
monotonically with time. After approximately 2 ms, the flame area at the safe con-
dition begins to decrease rapidly, suggesting that the kernels fail to liberate sufficient
heat to survive. At every point during this phase, flames that develop successfully
demonstrate a larger mean flame area than their unsuccessful counterparts. This
difference, however, is small compared to the sample standard deviation. The in-
crease in mean, successful, flame area after approximately 20 ms indicates the onset
of flame recovery. However, only 9% of recoveries occur during the 25 ms following
the spark. The rapid flame growth generating these early recoveries therefore skews
6. Results III: Flame Imaging 157
the mean between 20 and 25 ms.
These observations are confirmed in Figure 6.11, where the distributions of indi-
vidual kernel area are plotted at six intervals following the spark. These times were
selected to ensure that the sub-figures are separated regularly in logarithmic time,
as shown in Figure 6.10. At all intervals following the spark, very small kernels are
present in significantly greater numbers than large kernels, requiring the use of a
logarithmic vertical axis in Figure 6.11.
Figure 6.11(a) shows plots of the kernel size distributions extracted from the sec-
ond frame following the spark. At both the safe and unsafe conditions, the histogram
of kernel sizes created shortly following the spark features a Gaussian distribution
of large kernels and a collection of small flame fragments. The evident disparity
between the initial kernel size at the two set points can be attributed to early flame
growth at the safe condition. This initial growth is also apparent in Figure 6.11(b)
where the safe success and failure histograms are again indistinguishable. Breakup
of the initial kernel into smaller flame bodies is clearly evident in the unsafe his-
togram. In Figure 6.11(c), the number of unsafe tests is starting to decrease due
to ignition failures. At this point safe kernels are also exhibiting the breakup be-
haviour first observed at the unsafe condition (cf. Figure 6.11(b)). Figures 6.11(d)
and 6.11(e) indicate that during the low intensity period before recovery (see Sec-
tion 4.3.1), flame at the safe condition separates into a collection of small fragments.
The kernel growth apparent in Figure 6.11(f) is associated with the early stages of
recovery.
The fact that safe successes and failures are indistinguishable during the first
25 ms following the spark (corresponding to Figure 6.11(f)), suggests that ignition
outcome is not determined by kernel size, but by another factor such as the flame
path. This possibility is addressed below in Section 6.3.2.
6. Results III: Flame Imaging 158
×104
(a)
kernel area (pixels)
fr
eq
u
en
cy
•
•..
...
...
...
...
...
..•.
...
...
..•......
•
...
...
...
.•...............
•.............•............•
•
•..
...
...
...
..••...•...•....•.....................
............
•
...
...
...
...
...
.•................
•
•
•.
...
...
...
...
...
...
...
...
.•.
...
...
...
...
.•.
..•.
...
.•....•...
...........................•
...
...
...
.•.....................• •
safe success
safe failure
unsafe failure
0.0 0.3 0.6 0.9 1.2
0.1
1
10
102
103
104
×104
(b)
kernel area (pixels)
fr
eq
u
en
cy
•...............................................................................................•• •..
...
...
...
.••..
...
...
...
...
...
...
...
.•....••..
...
...
.•....•.
....•....•..........•.
...
...•...............•.................
•
...
...
...
..•
•
•
••......
...
...
.•.
...
...
...
..
...
..•......
..•..
.•. ..
•......•.
..•.
...•
•
•.....
.....................................•......•.....••................•
...
...
...•.........••...............•
...
...
...
..•....................
•
...
...
...
...
...
.•.
...
...
.•
••............
•
0.0 0.3 0.6 0.9 1.2
0.1
1
10
102
103
104
×104
(c)
kernel area (pixels)
fr
eq
u
en
cy
•...................................................................
•....•.......................
•
...
...
...
...•....•.....•.
...•.............•....•
...
.•.
..•..
...•.
...•...............................
•
...
...
...
..•.
...
..•.
...
...
...
...•.............•......•.
...
..•
...
.••......................
•
...
...
...
...
...
...
..•....................
•
•
•..
•..............•..............
.•.......
.............
•
...
...
..
...
...
...
.•........................
•
...
..•......•.
...
...
...
...
.•....•..
...•....
.....
••..
...
...
...
...
...
.• •......••
•
•
•
0.0 0.3 0.6 0.9 1.2
0.1
1
10
102
103
104
×104
(d)
kernel area (pixels)
fr
eq
u
en
cy
•..................................................................•................•
•
•
..........•............
•
•
•
0.0 0.3 0.6 0.9 1.2
0.1
1
10
102
103
104
×104
(e)
kernel area (pixels)
fr
eq
u
en
cy
•................................................................
•...............................
••
•.....................................................................................
•
•
0.0 0.3 0.6 0.9 1.2
0.1
1
10
102
103
104
×104
(f)
kernel area (pixels)
fr
eq
u
en
cy
•..........................................................•..........................
•..........•........
•
• • • ••
•
0.0 0.3 0.6 0.9 1.2
0.1
1
10
102
103
104
Fig. 6.11: Distributions of imaged kernel area. The timing of each sub-plot is
indicated in Figure 6.10.
6. Results III: Flame Imaging 159
6.2.3 Spark kernel velocities
As noted earlier, the tracking software records the centroid locations of every flame
body. It is therefore possible to estimate the spark kernel velocity by dividing the
apparent displacement of the spark between the first and second frames by the time
between them. Tables 6.2 and 6.3 present a comparison of the measured spark
velocity and the CFD-predicted gas-phase velocity at the average spark-centroid
location.
The spark velocity calculation accounts for the oblique viewing angle of the high-
speed camera using the correction described in Appendix A. The measured axial
velocities therefore describe spark motion in the central plane of the combustor
sector (i.e. in the plane of the laser sheet). Though the camera cannot detect
lateral motion of the flame, the CFD predictions of circumferential velocity have
been included. These values demonstrate that while there is significant swirl, the
axial component of velocity is dominant.
Table 6.2: Flame-centroid and CFD velocities at safe condition.
m/s u v w
flame 21.1 -8.7 –
CFD 17.5 -2.8 -7.6
Table 6.3: Flame-centroid and CFD velocities at unsafe condition.
m/s u v w
flame 32.8 -2.8 –
CFD 33.2 -4.3 -11.8
Notwithstanding the differences in geometry and boundary conditions (see Sec-
tion 5.3), the stream-wise components of velocity show good agreement at both test
conditions. However, the radial components of the flame centroid and CFD veloc-
ities are significantly different, with spark kernels at the safe condition descending
6. Results III: Flame Imaging 160
significantly faster than predicted by the CFD (−8.7 m/s compared to −2.8 m/s for
the CFD). This disparity is caused by a difference in the mode of failure at the two
set points. As previously observed, unsafe failure is caused by rapid disintegration of
the spark body, with no evidence of kernel growth (see Section 6.2.2). At the unsafe
set point the initial kernel therefore follows the flow, resulting in broad agreement
with the CFD (see Table 6.3). In contrast, significant flame growth occurs at the
safe condition. While upward growth of the kernel is prevented by the combustor
outside wall, the flame is unconstrained from below. The centroid of flame activ-
ity therefore descends significantly faster than the flow field, as demonstrated in
Table 6.2.
Ahmed and Mastorakos (2006) have demonstrated a correlation between the igni-
tion probability of a turbulent non-premixed methane jet and the gas-phase velocity
at the instant of the spark. Histograms of instantaneous axial velocity measured
using LDA revealed that successful events generally occur at lower velocities than
ignition failures. This result raises the question of whether ignition outcome in the
altitude test facility depends on the local air velocity at the igniter tip.
From the results quoted above in Tables 6.2 and 6.3, it is reasonable to assume
that the axial velocity of a spark kernel is equal to that of the local air-flow. Spark
velocity information was extracted from recordings of successful and unsuccessful
ignition events at the safe condition. Histograms of the axial component of velocity
are shown in Figures 6.12(a) and 6.12(b). The mean axial velocity immediately
following an unsuccessful spark is approximately 2 m/s greater than that measured
during a successful ignition event. Furthermore, the distribution of failures is skewed
towards higher velocities. This is illustrated in Figure 6.12(c), where the difference
between the number of successes and failures associated with each velocity bin has
been calculated, normalised to 100 samples, and plotted as a histogram. How-
ever, the difference between the mean velocities is less than both sample standard
deviations, implying that the dependence of ignition outcome on axial velocity sug-
6. Results III: Flame Imaging 161
(a)
velocity (m/s)
fr
eq
u
en
cy
mean = 20.2
std.dev. = 5.8
0 15 30 45 60
0
5
10
15
20
(b)
velocity (m/s)
fr
eq
u
en
cy
mean = 21.8
std.dev. = 6.9
0 15 30 45 60
0
5
10
15
20
(c)
velocity (m/s)
su
cc
es
se
s
−
fa
il
u
re
s
0 15 30 45 60
−10
−5
0
5
10
Fig. 6.12: Spark stream-wise velocity at safe condition: (a) success; (b) failure; (c)
normalised difference between number of successes and failures.
6. Results III: Flame Imaging 162
gested by Figure 6.12(c) is relatively weak compared to that observed by Ahmed and
Mastorakos (2006) for a methane jet flame. A more significant difference between
successful and unsuccessful spark velocities may emerge for a sample size greater
than the 100 or so tests analysed here.
6.3 Successful ignition
Employing the same method used to determine the spark kernel velocity, the veloc-
ities of flame kernels that emerge during the later stages of successful high-speed
recordings have also been calculated. A key question to address is whether flame re-
covery occurs from a single location in the combustion chamber. The stabilised flame
that is generated following successful ignition also fluctuates in a regular manner,
and this frequency content is analysed in some detail. Measures to enhance ignition
probability based on these observations are considered.
6.3.1 Flame velocities
Velocity data derived from the flame tracking program were spatially and tempo-
rally averaged, and constructed into a series of vector maps illustrating the general
motion of the flame. Vector maps constructed from recordings of safe, successful
ignition events, are presented in Figure 6.13. It is important to note that these
vectors represent the three-dimensional motion of flame that may occupy any point
across the width of the combustor. Ascribing a magnitude to these velocity vectors
therefore requires the assumption that all flame motion is two-dimensional and oc-
curs in the central plane of the combustor sector. Each vector represents the average
velocity of all flame kernels in an imaged area 3.8 mm× 3.8 mm, during an interval
of 0.88 ms. A velocity vector is displayed at locations where at least 10 flame kernels
are contained in the averaging square. These vectors are colour-coded according to
direction (see legend in Figure 6.14). The flame motion illustrated in these figures
6. Results III: Flame Imaging 163
is due to both convection of each kernel, and flame propagation relative to the cold
gas phase.
(a) (b)
(c) (d)
Fig. 6.13: Safe, successful, flame centroid velocity averaged over t ± 0.4 ms: (a)
t = 0.8; (b) t = 2.5; (c) t = 4.3; (d) t = 6.1. Vectors are coloured according to the
direction legend illustrated in Figure 6.14.
Figure 6.13(a) illustrates the initial flame centroid velocities of all safe successes,
averaged over the interval from 0.4 to 1.2 ms following the spark. The pocket of
hot gas generated by the spark discharge moves downstream and towards the fuel-
injector centre-line, at relatively high velocities. Shortly following the spark, success-
ful ignition is characterised by upstream flame motion back towards the fuel-injector
face, as shown in Figs. 6.13(b) to 6.13(d). Separate regions of upstream and down-
stream flame movement are clearly evident.
Figure 6.14 illustrates the flame centroid velocities extracted from all safe suc-
cessful tests and averaged over the 25 ms following the spark. The plot is broadly
6. Results III: Flame Imaging 164
representative of flame motion in the combustion chamber as distinct regions of the
chamber tend to register flame activity at different times (see Figure 6.13). The red
vectors in Figure 6.14 demonstrate that flame moves predominantly upstream in
the region to the left of the igniter, confirming the presence of a recirculation zone.
Flame kernels situated downstream of the igniter move towards the combustor exit,
as indicated by the light blue arrows.
direction
20 m/s
Fig. 6.14: Safe, successful, flame-centroid velocity averaged over 25 ms following
spark.
A significant feature of the velocity vector map is the clockwise recirculation
region that is observed below and downstream of the igniter tip. Recall from Sec-
tion 5.3 that the cold LES predicts a similar recirculation zone, in this case located
slightly upstream of the igniter. The fact that the same structure is observed in both
the flame centroid velocity map (Figure 6.14) and the CFD flow field (Figure 5.19),
suggests that flame motion is dominated by convection rather than propagation.
6. Results III: Flame Imaging 165
6.3.2 Recovery origin
The possibility that all safe successful ignition events share a common development
path has been investigated by examining the spatial distribution of flame kernels
during the lead up to recovery. In Figure 6.15, the total image intensity recorded
during each successful ignition event is plotted versus time, with every data set
shifted so that the recovery time coincides with t = 0. The mean intensity is plotted
as a solid line, and increases as expected during recovery.
×106
interval before recovery (ms)
fl
a
m
e
in
te
n
si
ty
co
u
n
t
•
•
•
•
•
••
•
•
•
•
•
•
•
•
•
•••••••
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
...........................................................................................................................................................................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.......................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
......................................................................................
...
...
...
...
...
...
...
...
...
...
...
.............................................
...
...
...
...
...
...
...
...
..............................
...
...
...
...
...
...
...
...
...
...
...
....
...
...
.........................................................
...
...
...
...
...
...
...
...
...
...
...
........
...
..............
...
...
...
.
.......
...
..........
.......
...
...
...
.......
.......................
...
...
...
...
...
...
.....
........
...............
...
...
.........
........
...
....................................
....................
....................................................................................................
...........
......
.....
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....
.............................
......................
...
....
.
......................
...
.
...
.
......................
...
.
...
.
......................
...
.
...
.
......................
...
.
...
.
Figure 6.16Figure 6.17
(a)(b)(c)(d)
−72 −60 −48 −36 −24 −12 0 12
0
1
2
3
4
Fig. 6.15: Flame intensity variations during successful ignition referenced to recovery
time. The labels above the plot indicate the timing of Figures 6.16 and 6.17.
As discussed in Chapter 4, the range of safe recovery times is large, extending
from 11 to 69 ms. Consequently, when the interval before recovery is 69 ms, only one
intensity trace contributes to the mean signal. Conversely, when the interval before
recovery is 11 ms or less, intensity data from all of the safe successful recordings
is used to calculate the mean. Between these times, an additional ignition event
contributes to the mean intensity count whenever a spark occurs. Every new spark
contaminates observations of the flame recovery process, and the analysis is therefore
6. Results III: Flame Imaging 166
limited to the first 20 ms before recovery, during which few sparks influence the mean
intensity trace and kernel distribution.
The kernel number density map at the moment of recovery appears as shown in
Figure 6.16. This contour map of average kernel number per unit area represents
the probability of discovering flame activity at any point inside the combustor.
Flame activity extends across virtually the entire chamber, with a concentration
of kernels upstream of the igniter. From this density map it is not possible to
determine whether the upstream concentration of flame bodies represents the source
of recovery, or simply results from steady-state flame anchoring to the injector face
following recovery. This uncertainty is due to the fact that a proportion of the
recovery process occurs before the attributed recovery time (see Section 4.3.1).
30
25
20
15
10
5
35
30
25
20
15
10
5
0
number density
Fig. 6.16: Example of a kernel number density map averaged over t± 0.4 ms, where
t = −0.6 ms (relative to recovery time).
It is therefore necessary to examine the kernel number density maps at earlier
intervals as shown in Figure 6.17. Progressing backwards in time and presenting the
density maps corresponding to the intervals indicated in Figure 6.15, it is apparent
6. Results III: Flame Imaging 167
that flame disappears from the majority of the combustor, but remains active in
the region close to the injector face (see Figure 6.17(d)). This region can now be
confidently identified as the origin of recovery.
(a) (b)
(c) (d)
Fig. 6.17: Safe, successful kernel number density averaged over t ± 0.4 ms: (a)
t = −4.1; (b) t = −8.5; (c) t = −12.9; (d) t = −17.3. Contours are coloured
according to the legend illustrated in Figure 6.16.
Having located a region that is favourable to recovery, it would be desirable to
deliver developing kernels to this area of the combustor. At present this requires
the migration of flame upstream from the igniter. Relocating the igniter tip slightly
upstream of its current position may improve the likelihood of ignition success by re-
ducing the distance that developing flame must move. As described in Section 2.4.4,
this is supported by the work of Marchione et al. (2007, Sep), who observed that
the likelihood of ignition was a strong function of the igniter’s axial location.
6. Results III: Flame Imaging 168
6.3.3 Frequency content
Visual examination of the high-speed images acquired following flame recovery sug-
gests that the stabilised flame fluctuates in a highly regular manner. This observa-
tion is confirmed in Figure 6.18, where the variation in total flame intensity during
a single ignition event is plotted in blue. Following the recovery peak approximately
30 ms after the spark, the flame intensity count decreases, and begins to oscillate at
approximately 50 Hz. These fluctuations are predominantly associated with cyclic
changes in the total kernel area (red line in Figure 6.18), rather than variations in
kernel luminosity. Note that the flame-tracking software was not used to perform
this analysis, and the flame area is simply defined as the number of pixels in each
frame with a count exceeding 10.
Fourier analysis of the total flame intensity signal indicates a strong peak in
power at a frequency of 50.5 Hz, as shown in Figure 6.19. To minimise spectral
leakage, a single Hamming window was applied to the time series before performing
a discrete Fourier transform (DFT) (Press et al. 2007). The length of the DFT was
set to 11400 points, thereby providing a spectral resolution of 0.5 Hz (fs = 5.7 kHz).
Virtually all of the power is contained in frequencies less than 100 Hz. This is also
true for many of the stabilised flame signals recorded following recovery. The Fourier
analysis was repeated for the 88 successful ignition events that generate a continuous
signal between 150 and 540 ms. All of these signals demonstrate maximum power
spectral densities at frequencies less than or equal to 60 Hz. Indeed, half have their
maximum at between 50 and 55 Hz. This fluctuation of steady-state flame may
either be combustion-induced, or could result from a 50 Hz instability generated in
the cold flow prior to the spark.
Thermoacoustic oscillations result from the positive feedback between pressure
and heat release that can occur inside a gas turbine combustion chamber. Aircraft
engine spray flames are susceptible to low-frequency ‘rumble’ instabilities at idle and
sub-idle conditions (Zhu et al. 2001). Previous work by Eckstein et al. (2006) has
6. Results III: Flame Imaging 169
×106 ×104
interval following spark (ms)
fl
a
m
e
in
te
n
si
ty
co
u
n
t
fl
a
m
e
a
re
a
(p
ix
el
s)
.....................................................................................................
.......................
...
..........
..
...
...
...
...
...
...
...
....
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
....................................................................................................................
........
...
...
...
...
...
...
...
.....
.....
...
......................................
....
......................
...
...
.................................
.........
...
...
......................................
....
............
....................................
......
.....
.......................
.................
...
...
...
.....
........................................
...
...
...
.....
...
..
.............
.......................................
........................
............ ....
...
........
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..
...
...
..........
........................................................................................
...
...
...
...
........
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
................................................................
...
...
..
...
...
...
...
...
............................
..............
...
...
.......................................................
...
...............
.....
...
...
...
...
....................................
...
...............
...
......
...
...
..
.......................................
.............
...
...
.....
....
...
...
...
...........................
.....
...........
...
...
...
...
...
...
...
...
...
...
...
......
...
...
...
...............................................
..........
..
...
..
...
...
...
...
...
...
.....
...
...
...
...
.
... ....
.....
.... ....
...
..
...
.
......
....
.....
..... .
...
...
..
...
....
.....
....
......
...
..
...
..
...
. ...
....
....
.....
...
..
...
....
....
.....
... ..
.
...
..
...
.
.....
.....
....
....
.
...
..
...
. ....
.....
....
...
.
..... ....
.....
..........
...
...
...
...
...
....
..................................
.
...
...
...
...
....
.........................................
.....
sample
period
Fig. 6.19
flame intensity
flame area
extracted sinusoid
0 30 60 90 120 150 180
0
1
2
3
4
5
0
2
4
6
8
×106 ×104
interval following spark (ms)
fl
a
m
e
in
te
n
si
ty
co
u
n
t
fl
a
m
e
a
re
a
(p
ix
el
s)
..............................
...
...
....
...
...
...
...............
...
.....................................
....
...
...
...
...
...
...
...
....
..................................................
...
...
...
...
...
....
...
..........................................
...
...
...
...
...
...
...
...
.......................................
...
...
...
...
...
...
...
...
...
..
...
...
..........................................................
...
...
...
....
...
...
..
....
.............
................................
....
...
...
...
...
....
...
...
....
...
..
..........................................
...
...
...
...
...
...
...
...
...
...
.............................................
....
.....
......
...
...
...
...
...
......
...
...
...
.....................................................
...
...
....
...
.....
...
...
...
...
...
...
...
....
...
.........................
.......................................................
....
....
.....
...
...
....
...
...........
....................................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.................................................................
...
...
...
...
...
...
...
...
...
...
...
....
...
...
........................
...
..................................
........
...
...
....
...
...
...
...
...
...
...
...
...
...
...
...
.................................................................
....
...
...
...
...
...
...
...
...
...
...
...
...
........................................................................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...............................................................
....
...
...
...
....
...
...
...
...
...
...
...
...
...
...
....................................................................
.......
...
...
...
...
....
...
...
...
...
...
...
....
...
...............
..................
...................................
...
...
...
...
...
...
........................
...........
...
...
...
...
...
.....................
.... ...
...
...
...
...
...
...
...................
.
............
...
...
...
...
...
...
.....................................
...
...
...
...
...
...
.
.........................
........
...
...
...
...
...
...
.....................................
...
...
...
...
...
...
....
..................................
...
...
...
...
...
...
...
.........
.....................
...
...
...
...
...
...
.........
180 210 240 270 300 330 360
0
1
2
3
4
5
0
2
4
6
8
×106 ×104
interval following spark (ms)
fl
a
m
e
in
te
n
si
ty
co
u
n
t
fl
a
m
e
a
re
a
(p
ix
el
s)
...
...
...
..
...
...
...
.......................................
........
..
...
...
...
....
.....
...
...
...
...
.................................................
.....
...
..........
...
...................
...
...
...
...
...
...
...
...
...
.............................................
....
...
...
..
...
.........................................
...
...
..
...
...
...
...
...
...........................................
...
.....
...
...
...
...
...
...
...........................................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.............................................................
...
...
...
...
.......................................
....
....
...
..
.....
....
.........
...
...
...
...
......
...
...
...
......................................................
...
...
....
...
....
...
...
...
...
...
...
...
....
...
.........................
.......................................................
...
....
....
...
...
....
...
.............
....................................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
................................................................
...
...
...
...
...
...
...
...
...
...
...
....
...
...
........................
...
...................................
........
...
...
....
...
...
...
...
...
...
...
...
...
...
...
...
..................................................................
....
...
...
...
...
...
...
...
...
...
...
...
...
...........................................................................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..................................................................................
...
...
....
...
...
...
...
...
...
...
...
...
...
......
................................................................
.......
...
...
...
...
...
....
...
...
...
...
...
....
...
...
.............
..................
.............................
..
...
...
...
...
...
...
....................................
...
..
...
...
...
...
................
...................
..
...
...
...
...
...
....
...............................
...
...
...
...
...
...
...
............
....................
...
..
...
...
...
...
....................................
...
...
...
...
...
...
................................. ..
...
...
...
...
...
..
......................................
...
.
.....................
............
...
...
...
...
...
...
...
...........
360 390 420 450 480 510 540
0
1
2
3
4
5
0
2
4
6
8
Fig. 6.18: Stabilised flame oscillations.
6. Results III: Flame Imaging 170
×1010
frequency (Hz)
p
ow
er
sp
ec
tr
a
l
d
en
si
ty
(W
/
H
z)
.....................
...
...
..................
...............
.............
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..
................................................................................................................................................................................
.........................
.......
.......
101 102 103 2850
0
1
2
3
4
Fig. 6.19: Spectral analysis of measured flame intensity signal.
shown that these frequencies typically extend from 50 to 120 Hz, encompassing a
large proportion of the frequencies observed in the present study.
The observed fluctuations may alternatively be caused by a cold flow instability.
A previous study of isothermal flow through a fuel injector has demonstrated that
induced swirl can generate instabilities at low frequencies (Carrotte and Batchelor-
Wylam 2003). If such a cold instability is causing the flame oscillations in the
present study then the spark timing assumes great importance, as earlier results
suggest that the local axial velocity below the igniter tip is skewed towards lower
values for successful ignition (see Section 6.2.3).
The existence of a cold flow instability has been investigated by extracting the
most powerful frequency component contained in the flame intensity signal, and
extrapolating back to find its phase relative to the spark (at t = 0). This process
is illustrated in Figure 6.18, and the resulting phase information extracted from all
test data is collated in Figure 6.20. There appears to be no consistent relationship
6. Results III: Flame Imaging 171
phase shift (rad)
fr
eq
u
en
cy
0 pi
2
pi 3pi
2
2pi
0
2
4
6
8
10
Fig. 6.20: Phase shift of stabilised flame oscillations relative to spark.
between the phase of the spark relative to the most powerful intensity sinusoid, and
the number of successful events. However, many more samples would be required
to draw a definitive conclusion.
6.4 CFD comparison
Several features of the CFD simulation can be compared with the results produced by
the flame tracking program. As already noted in Section 6.2.3, the CFD predictions
of axial velocity below the igniter tip are in good agreement with the observed kernel
motion. Three streamlines generated by the cold LES and passing through this
region are illustrated in Figure 6.21. The mean spark centroid is indicated by an open
white circle. These streamlines terminate on the side of the computational domain,
indicating that the circumferential component of velocity is significant compared
to the axial velocity. Swirl produced by the fuel injector may therefore heavily
influence the relight process, possibly generating the yellow and green downward
6. Results III: Flame Imaging 172
vectors evident in Figure 6.14. Though not measured during this study, the out-
of-plane velocity could be identified using a stereoscopic imaging approach such as
that described by Cheung and Zhang (2006).
A well-developed recirculation zone is apparent in the upstream portion of the
combustion chamber, bounded by the red line in Figure 6.21. This provides an
explanation for the observed upstream motion of flame kernels at the safe set point.
Furthermore, the recovery origin identified in Section 6.3.2 is fully contained inside
this recirculation zone. It is reasonable to conclude that the combustor flow patterns
tend to promote stabilisation of flame at the injector face provided that the air mass
flow rate does not prohibit kernel survival.
(a) (b)
u < 0
u > 0
Fig. 6.21: CFD streamlines passing below the igniter tip at the safe test condition:
(a) side view; (b) rear view. The igniter tip and mean spark centroids are represented
by red and white filled circles respectively. The zero contours of axial velocity are
also shown in red.
Chapter 7
Conclusion
In this study experiments have been performed in a high-altitude test facility to
increase understanding of ignition and flame development during relight of a LDI
combustor. The distribution of kerosene fuel inside the combustion chamber was
measured using PLIF, and high-speed images of flame emission were recorded during
numerous ignition tests. The analysis of the experimental results has provided
important insights into the relight process, and these are summarised below.
7.1 Operation of test facility
Testing was conducted at two conditions, designated ‘safe’ and ‘unsafe’ to indicate
the relative probability of ignition. The sole difference between these conditions
was the larger air mass flow rate at the unsafe set point (see Table 3.1). Roughly
one in five sparks produced successful relight at the safe condition whereas ignition
was never achieved at the unsafe condition. During an ignition test, sparks were
generated at 1.3 Hz and were halted following successful relight at the safe condition,
and after 10 sparks at the unsafe condition. Several hundred PLIF images of cold
fuel distribution were recorded, and around 100 ignition tests were conducted at
each set point, during which PLIF and flame images were recorded simultaneously.
173
7. Conclusion 174
7.1.1 Apparatus modifications
Fuel and flame imaging at altitude restart conditions are made extremely difficult
by the large quantities of fuel deposited on the internal surfaces of the combustion
chamber. Contamination of the laser sheet optics would cause significant attenuation
of the excitation wavelength, while fuel on the combustor side windows would reduce
the resolution of both PLIF and flame images. In addition, simultaneous fuel and
flame imaging requires careful timing control, that must also handle large variations
in the spark-to-spark interval. The following modifications were implemented to
overcome these difficulties:
• Design and installation of a beam delivery apparatus. Laser light was directed
from an optical table onto a laser periscope and sheet expansion assembly
mounted on the ATF, which projected the sheet into the combustion chamber
from behind the fuel injector.
• Addition of several protective air purges. These purges were introduced at
the bottom periscope mirror, sheet expansion lens, and combustor window,
to prevent contamination of the optical surfaces by dust, condensation, and
liquid fuel.
• Development of a timing and data acquisition system. Signals from the laser,
igniter current probe, and a photodiode registering successful ignition events,
were used to control the acquisition of images with the PLIF and high-speed
cameras.
7. Conclusion 175
7.1.2 Apparatus characterisation
The success or failure of ignition is influenced not only by the random nature of the
relight process, but also by bulk variations in the experimental boundary conditions.
Fluctuations in the ATF operating conditions, laser energy, and spark parameters
were quantified in Chapter 4, together with the timescales of flame development.
These preliminary investigations into the performance of the apparatus led to the
following findings:
• The spatial energy profile of the laser sheet did not have the expected Gaus-
sian distribution. Asymmetric features of the profile were identified that later
influenced correction of the PLIF images.
• Flame activity during an ignition test was characterised by a low-intensity
delay period prior to recovery. The length of this delay period varied signifi-
cantly, as did the failure time of unsuccessful sparks, indicating that recovery
is a random event not governed by the conditions at the instant of the spark
discharge.
• The spark was relatively consistent. Measured values of spark size, current,
and luminosity showed no correlation with ignition outcome and were not
significantly affected by the test conditions.
• At the safe condition, ignition probability increased until the fifth spark and
remained constant thereafter, indicating that the early sparks were not inde-
pendent events. It is likely that the observed variation in spark performance
is associated with large amounts of fuel deposited on the igniter tip, that may
be displaced by the action of early sparks.
7. Conclusion 176
7.2 Main findings of fuel and flame imaging
It was shown in Chapter 5 that the laser sheet was not significantly attenuated by the
kerosene droplets, and the PLIF signal remained strong even along the downstream
edge of the image. Nevertheless, the PLIF images required significant corrections,
the largest of which accounted for the non-uniform laser energy profile. Initial
analysis avoided the need for correction by considering a small area close to the
middle of the sheet. The full mean images were later corrected and analysed.
• In the small area of interest, the mean PLIF signal was higher at the unsafe
condition than at the safe set point. This is due to improved atomisation at
the unsafe condition leading to high concentrations of fuel in suspension. The
larger droplets produced at the safe condition tend to be deposited on the
combustor walls rather than remaining in the CRZ.
• The local AFR was rich at the safe condition and richer still at the unsafe
condition. This may contribute to consistent ignition failure at the unsafe
condition, though excessive turbulent strain is also important.
• PLIF images were successfully acquired in the presence of flame and showed
negligible interference from PAH fluorescence and flame emission. The effect
of flame was to significantly reduce the overall PLIF signal, but a base level
of signal remained due to the presence of large burning droplets.
• Corrected mean images of cold fuel distribution were compared to CFD sim-
ulations of the gas-phase flow, confirming that small kerosene droplets are
trapped in the CRZ. Virtually no fuel is found in a channel of fast-moving air
between the CRZ and the igniter tip.
The analysis of the high-speed flame recordings was discussed in Chapter 6.
An image processing algorithm was developed to analyse the motion and breakup
of flame during relight, allowing the construction of a single flame trajectory plot
7. Conclusion 177
representing each ignition attempt. This program recorded the size and centroid
velocity of every flame kernel, and led to the following conclusions:
• Trajectory plots revealed several modes of relight failure including the rapid
disintegration, blow-off, and unsuccessful stabilisation modes identified by
Ahmed et al. (2007a) for a laboratory scale burner. An additional failure
mode was also identified that consisted of a split in the initial kernel followed
by simultaneous downstream blow-off and upstream propagation towards the
fuel injector.
• Initial kernel velocities were found to be in broad agreement with CFD predic-
tions of the cold flow. Flow patterns apparent from measurements of average
flame velocity were also observed in the CFD results, demonstrating that flame
motion is dominated by convection rather than propagation.
• A recovery origin was identified close to the fuel injector face. Since the size
of flame kernels decreased continuously from shortly following the spark until
recovery, successful ignition may be characterised by the ability of flame kernels
to survive long enough to reach the fuel injector face.
• Stabilised flame fluctuated sinusoidally at a rate of approximately 50 Hz. This
may be a combustion-induced instability, or the fluctuation may already be
present in the cold flow.
7. Conclusion 178
7.3 Recommendations for future work
A major finding of this study is that kerosene PLIF can be used to map the fuel
distribution inside a LDI combustor operating under altitude relight conditions.
As described above, several insights into the relight process have been provided
by the combined imaging of fuel and flame. Further useful information would be
obtained by simply repeating the experiment to increase the sample size of data
available. For example, supplementary measurements would confirm whether there
is indeed a correlation between low air velocity at the igniter tip and successful
ignition outcome, as suggested in Section 6.2.3. In addition, the possibility of a
periodic cold flow instability has important consequences for relight performance,
and this could again be confirmed by analysing a larger data set.
7.3.1 Measures to improve LDI relight performance
The results of the current work suggest the following methods of enhancing relight
performance:
• Increase the spark energy. The flame tracking results suggest that increasing
the initial kernel size may improve the probability of successful flame propa-
gation.
• Move the igniter upstream. Ignition probability may be increased by reducing
the distance that a developing flame kernel must travel to reach the recovery
origin.
• Improve atomisation at the safe condition. This would assist survival of the
flame kernels as finer droplets evaporate more easily and feed the flame.
• Adjust the spark timing to take advantage of periodic velocity fluctuations
that may exist in the cold flow.
7. Conclusion 179
7.3.2 Further investigations
The apparatus developed during this study can be used to apply other laser di-
agnostic techniques to examine relight processes inside the ATF. In addition, the
experiment techniques employed could be improved in several respects. The follow-
ing are promising avenues for future research:
• Use the double-pulse functionality of the laser to perform droplet or gas-phase
particle image velocimetry (PIV).
• Mie-scattering measurements of liquid fuel distribution could be made using
the second laser harmonic at 532 nm.
• Improve the beam energy distribution correction by diverting a small fraction
of the laser light to a reference dye cell during testing. This would also allow
each PLIF image to be corrected for shot-to-shot energy fluctuations.
• Provide improved timing control of the spark to increase the probability of
obtaining fuel distribution images during flame development.
• Conduct high-speed measurements of emissions from an specific radical pro-
duced in the flame reaction zone e.g. OH, CH, or C2. This would be achieved
by placing an appropriate optical filter in front of an intensified CCD camera,
and may also eliminate the loss of signal experienced during some successful
ignition events.
Appendices
180
Appendix A
Viewing Angle Corrections
The imaging of a calibration square is illustrated schematically in Figure A.1. The
blue square, PQRS, represents a calibration target and lies in the plane of the
laser sheet. The centre of this square is at the origin. The red parallelogram is the
projection of square PQRS onto the x-y plane, and the camera observes this plane
at right angles.
.................................................................................................................................................. .
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
.
.....
.....
.............................................................................................................................................
.......
•
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
•
••
•
•
•
•
•
................................................................
......................................................................
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...........................................................................
O
x
y
z
P Q
RS
(a, b, 0)
(c, d, 0)
(a, b, z1) (c, d, z2)
Fig. A.1: Rotated square.
181
A. Viewing Angle Corrections 182
The coordinates (a,b) and (c,d) are measured from the image of the calibration
target. The unknowns are z1 and z2. As the square is centred at the origin, points
P and Q are equidistant from O and thus
a2 + b2 + z21 = c
2 + d2 + z22 . (A.1)
Also, the vectors
−→
OP and
−→
OQ are perpendicular, so
z1z2 = −(ac + bd) . (A.2)
Substituting (A.2) into (A.1) gives
z41 + (a
2 + b2 − c2 − d2)z21 − (ac+ bd)2 = 0. (A.3)
Eq. (A.3) is a quadratic in z21 that is easily solved. The positive root is chosen, since
z21 ≥ 0 is required. The choice of sign for z1 is arbitrary as the viewing correction
is the same for the two orientations of the square that generate the same image.
Suppose, without loss of generality, that point P is projected forward in the positive
z-direction, as depicted in Figure A.1. Then,
z1 =
1√
2
[
− (a2 + b2 − c2 − d2) +
√
(a2 + b2 − c2 − d2)2 + 4(ac+ bd)2
]1/2
, (A.4)
and the value of z2 follows from Eq. (A.2).
Having identified the coordinates of P , Q, R, and S as described above, let us
define a normal vector to the square
n =
−→
OP ×−→OQ . (A.5)
Note that the plane of the square contains all points r such that n · r = 0.
Suppose that a vector (u∗,v∗) is measured in an image captured by either camera.
Then from Eq. (A.5) the z-component, w∗, of this vector is given by
w∗ =
u∗(dz1 − bz2) + v∗(az2 − cz1)
(ad− bc) . (A.6)
A. Viewing Angle Corrections 183
It remains to find the components of (u∗,v∗,w∗) in the plane of the square. Let us
call this vector (u,v). A suitable coordinate system for the square is obtained from
the midpoints of QR and PQ respectively
is =
−→
OQ+
−→
OR∣∣−→OQ+−→OR∣∣ , js =
−→
OP +
−→
OQ∣∣−→OP +−→OQ∣∣ . (A.7)
Taking the dot product of (u∗,v∗,w∗) with each of the unit vectors is and js gives
the following result 
u
v

 = M

u∗
v∗

 =

M11 M12
M21 M22



u∗
v∗

 , (A.8)
where the elements of the transformation matrix are
M11 =
[
c− a + (dz1 − bz2)(z2 − z1)
(ad− bc)
]
[
(c− a)2 + (d− b)2 + (z2 − z1)2
]1/2 , (A.9)
M12 =
[
d− b+ (az2 − cz1)(z2 − z1)
(ad− bc)
]
[
(c− a)2 + (d− b)2 + (z2 − z1)2
]1/2 , (A.10)
M21 =
[
a+ c+
(dz1 − bz2)(z1 + z2)
(ad− bc)
]
[
(a+ c)2 + (b+ d)2 + (z1 + z2)2
]1/2 , (A.11)
M22 =
[
b+ d+
(az2 − cz1)(z1 + z2)
(ad− bc)
]
[
(a+ c)2 + (b+ d)2 + (z1 + z2)2
]1/2 . (A.12)
The coordinates required to evaluate Eqs. (A.9) to (A.12) were extracted from an
image of the calibration target recorded by each camera. The image of this square
was not a true parallelogram due to perspective effects, optical distortion by the
lens, and uncertainties in the coordinates caused by resolution limits. These effects
A. Viewing Angle Corrections 184
were removed by averaging the distance between the origin and opposite vertices.
The coordinates corresponding to the viewing positions of the two cameras are listed
in Table A.1. The matrix elements calculated from these coordinates are presented
in Table A.2. Having identified these elements, Eq. (A.8) can be used to map vectors
in the sheet plane to the camera images, and vice versa.
Table A.1: Calibration target vertex coordinates (see Figure A.1)
a b c d
PLIF camera -427.5 409.0 427.0 412.0
high-speed camera -86.0 83.0 80.5 82.0
Table A.2: Viewing angle transformation matrices and scaling factors.
M11 M12 M21 M22 mm/pixel
PLIF camera 1.000 0.001 -0.004 1.041 0.08
high-speed camera 1.016 0.034 0.006 1.025 0.45
Table A.2 indicates that the required corrections are comparatively small for
both camera viewing angles i.e. Mij is approximately equal to the identity matrix.
A scaling factor was required to convert between units of pixels and millimetres.
This was obtained by applying Eq. (A.8) to a vector in the reference image joining
the centre of the calibration square to one of its vertices. As the true length of this
line is known, the scale factor (mm/pixel) follows. The scaling factors corresponding
to the two viewing angles are also listed in Table A.2.
Appendix B
Geometry of Laser Sheet
B.1 Lens calculations
The 8 mm diameter beam produced by the laser is formed into a sheet by four planar
lenses, arranged as illustrated in Figure B.1. If the dimensions and refractive index
of each lens are known, then the dimensions of the resulting elliptical beam cross-
section can be evaluated at every point along the optical axis using a first-order
analysis (Hecht 2002).
If it is assumed that the convergence and divergence angles are small (≤ 10◦),
then the paraxial approximation is valid. In this case, the refraction of incoming
light rays can be considered to occur in one of two principal planes, the choice
of which depends on the direction of beam propagation through the optic. For a
general singlet lens, the distance from these principal planes to their respective focal
points is known as the effective focal length
f =
[
(n− 1)
(
1
R1
− 1
R2
+
d (n− 1)
nR1R2
)]
−1
, (B.1)
where n is the refractive index of the lens material, R1 and R2 are the radii of
curvature of the first and second lens surfaces respectively, and d is the lens thickness
at the optical axis. It is generally more useful to state the position of the focal point
relative to a lens vertex. The front focal distance (FFD) is defined as the distance
185
B
.
G
eo
m
etry
o
f
L
a
ser
S
h
eet
186
lens 1 lens 2 lens 3
p
erisco
p
e
lens 4
..............................................................
....
propagation
...
...
...
...
...
...
...
...
...
.......
.....
.............................
...
..
...
..
tf
........................
. ...
.
...
...
...
...
...
...
...
......
...
2θ
Fig. B.1: Sheet-formation optics. Lenses 1, 2, and 3 are mounted on the optical table (see Figure 3.4). Lens 4 is installed
in the sheet expansion assembly (see Figure 3.6).
B. Geometry of Laser Sheet 187
from the front focal point to the vertex of the first lens surface
FFD = f
[
1 +
d (n− 1)
nR2
]
. (B.2)
The back focal distance (BFD) is defined as the distance from the vertex of the
second lens surface to the rear focal point
BFD = f
[
1− d (n− 1)
nR1
]
. (B.3)
Eqs. (B.1), (B.2), and (B.3) can be used to calculate the effective focal length, FFD,
and BFD of each sheet-formation lens, and these results are listed in Table B.1.
Note that all four lenses are manufactured from ultraviolet-grade fused silica with
a refractive index of 1.500 at 266 nm. With these focal distances and a knowledge
Table B.1: Dimensions and focal distances of sheet-formation lenses. All dimensions
are given in mm.
lens number R1 R2 d f FFD BFD
1 -50.9 ∞ 3.1 -101.9 -101.9 -103.9
2 ∞ -76.3 5.0 152.7 149.4 152.7
3 704 ∞ 2.1 1408.9 1408.9 1407.5
4 ∞ 12.7 4.0 -25.4 -28.1 -25.4
of the separation between optics, simple geometry can be used to establish the
width and thickness of the beam at critical points along the optical axis. The cross-
sectional dimensions of the beam at the appropriate principal plane of each lens are
listed in Table 3.2.
Referring to Figure B.1, lenses 1 and 2 expand the beam by a fixed amount
horizontally relative to the surface of the optical table. This finite expansion is
achieved by positioning the optics so that the front focal point of lens 2 coincides
with the back focal point of lens 1. Following this alignment, the width of the beam
after expansion is simply the ratio of the effective focal lengths multiplied by the
B. Geometry of Laser Sheet 188
original beam width
w2 =
f2
f1
w1 =
152.7
101.9
× 8 = 12.0 mm. (B.4)
Lens 3 has a comparatively long focal length that generates a gradual thinning of
the beam as it passes through the periscope and into the combustion chamber. The
change in thickness per unit distance along the optical axis generated by lens 3 is
m =
t1
f3
=
8
1408.9
= 5.68× 10−3. (B.5)
The distance from the principal plane of lens 3 to the focal point of lens 4 can be
extracted from a CAD model of the experimental apparatus. The sheet thickness
at the focal point of lens 4 is then
tf = t1 −mx = 8−
[(
5.68× 10−3) (1064.8)] = 2.0 mm. (B.6)
Lens 4 expands the laser beam into a sheet. The beam height at the principal plane
is relatively large compared to the effective focal length, producing a divergence
at the upper and lower beam extent of 27.3◦. This comparatively large expansion
invalidates the paraxial assumption, and the effective focal length must be modified
to account for curvature of the principal plane. Then
|f ′4| =
√
f 24 −
(w4
2
)2
=
√
(−25.4)2 −
(
12.0
2
)2
= 24.7 mm. (B.7)
Based on this modified effective focal length, the divergence half angle of the sheet
is
θ = tan−1
(
w4/2
|f ′4|
)
= tan−1
(
12.0/2
24.7
)
= 0.24 rad. (B.8)
Note that this angle corresponds to the unclipped sheet extent (see Section 3.2.4).
B.2 Frontal area calculation
A correction must be applied to the PLIF images to account for changes in the
cross-section of the sheet as it passes through the combustor. As illustrated in
B. Geometry of Laser Sheet 189
Figure B.2, the propagating laser front forms a series of curved surfaces, composed
of points equidistant from the focal axis of the lens. Neglecting non-uniformities
in the beam energy profile, the irradiance is constant throughout a frontal surface,
and proportional to the surface area. The correction factor required at a general
point, P , is therefore proportional to the area of the frontal surface containing P .
A general expression for this frontal area is therefore required.
......
......
.......
......
......
......
.
.....................................
.......
......
......
......
......
.......
......
......
.......
......
......
......
......
......
......
......
......
......
......
.....
......
......
......
......
......
......
......
......
.......
......
......
.......
......
......
......
......
......
......
......
......
................. ...................................................................................................................................................................
.....
..................................................................................................................................
............
............
............
............
............
............
............
............
............
............
............
.........
....
•
• .....
..................... .........................................................
.....
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
........
.....
.................................................................................................................................................................
......................................................................................................................................
...........................
............................................................................................................................................
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
..................
...
.
...
..
...
...
...
...
...
.......
.....
..................
......
.
...
...
...
...
...
.......
.....
x
dA
z
y
tf t = tf −mr cosα
P
r
α θ
lens focal axis
Fig. B.2: Laser sheet geometry.
Point P is situated distance r from the focal axis of the lens. An ellipse is
formed by taking a cross-section of the sheet perpendicular to the horizontal axis.
The elliptical cross-section passing through the end of a radial line, length r, inclined
at angle α, has the outline
y2
(r cosα tan θ)2
+
4z2
t2
= 1. (B.9)
If y = r sinα then
tan2 α
tan2 θ
+
4z2
t2
= 1. (B.10)
B. Geometry of Laser Sheet 190
The maximum thickness of the elliptical cross-section is
t = tf −mr cosα. (B.11)
The area of the surface element dA is
dA = 2zr dα. (B.12)
Combining Eqs. (B.10), (B.11), and (B.12), and integrating yields
A = 2r
∫ θ
0
(tf −mr cosα)
√
1− tan
2 α
tan2 θ
dα . (B.13)
Numerical evaluation of Eq. (B.13) provides the frontal area at every imaged point
in the sheet. The values of m, tf , and θ, have been identified in Eqs. (B.5), (B.6),
and (B.8) respectively.
From Eqs. (B.10) and (B.11), the sheet thickness at a specific radial and angular
location is
z =
tf −mr cosα
2
√
1− tan
2 α
tan2 θ
. (B.14)
This equation is used in Section 5.2.2 to identify the sheet thickness correction factor
at every point in the sheet.
References
Advisory Council for Aeronautics Research in Europe (2002). Strategic Research
Agenda 1, Volume 2.
Aggarwal, S. K. (1998). A review of spray ignition phenomena: present status
and future research. Prog. Energy Combust. Sci. 24, 565–600.
Ahmed, S. F., R. Balachandran, T. Marchione, and E. Mastorakos (2007a). Spark
ignition of turbulent nonpremixed bluff-body flames. Combust. Flame 151,
366–385.
Ahmed, S. F., R. Balachandran, and E. Mastorakos (2007b). Measurements of
ignition probability in turbulent non-premixed counterflow flames. Proc. Com-
bust. Inst. 31, 1507–1513.
Ahmed, S. F. and E. Mastorakos (2006). Spark ignition of lifted turbulent jet
flames. Combust. Flame 146, 215–231.
Allouis, C., F. Beretta, and A. Amoresano (2008). Experimental study of lean pre-
mixed prevaporized combustion fluctuations in a gas turbine burner. Combust.
Sci. Technol. 180, 900–909.
Anacleto, P., M. V. Heitor, and A. L. N. Moreira (1996). Mean and turbulent
flowfields in a model RQL gas-turbine combustor. Exp. Fluids 22, 153–164.
Arnold, A., R. Bombach, B. Kappeli, and A. Schlegel (1997). Quantitative mea-
surements of OH concentration fields by two-dimensional laser-induced fluo-
rescence. Appl. Phys. B 64, 579–583.
191
References 192
Arnold, A., F. Dinkelacker, T. Heitzmann, P. Monkhouse, M. Scha¨fer, V. Sick,
J. Wolfrum, W. Hentschel, and K.-P. Schindler (1992). DI diesel engine com-
bustion visualised by combined laser techniques. Proc. Combust. Inst. 24,
1605–1612.
ASTM (2008). Standard ASTM D1655-08.
Atkins, D. L., J. S. Ervin, and L. Shafer (2005). Experimental studies of jet fuel
viscosity at low temperatures, using a rotational viscometer and an optical
cell. Energy Fuels 19, 1935–1947.
Ballal, D. R. (2004, Jan). private communication, Kettering Laboratories, Room
462, 300 College Park, Dayton, Ohio, 45469-0140, United States.
Ballal, D. R. and A. H. Lefebvre (1975). The influence of spark discharge char-
acteristics on minimum ignition energy in flowing gases. Combust. Flame 24,
99–108.
Ballal, D. R. and A. H. Lefebvre (1977a). Flame quenching in turbulent flowing
gaseous mixtures. Proc. Combust. Inst. 16, 1689–1698.
Ballal, D. R. and A. H. Lefebvre (1977b). Ignition and flame quenching in flowing
gaseous mixtures. Proc. R. Soc. Lond. A 357, 163–181.
Ballal, D. R. and A. H. Lefebvre (1978a). Ignition and flame quenching of quies-
cent fuel mists. Proc. R. Soc. Lond. A 364, 277–294.
Ballal, D. R. and A. H. Lefebvre (1978b). Ignition of liquid fuel sprays at subat-
mospheric pressures. Combust. Flame 31, 115–126.
Ballal, D. R. and A. H. Lefebvre (1979). Ignition and flame quenching of flowing
heterogeneous fuel-air mixtures. Combust. Flame 35, 155–168.
Ballal, D. R. and A. H. Lefebvre (1981). A general model of spark ignition for
gaseous and liquid fuel-air mixtures. Proc. Combust. Inst. 18, 1737–1746.
Bazile, R. and D. Stepowski (1995). Measurements of vaporized and liquid fuel
References 193
concentration fields in a burning spray jet of acetone using planar laser induced
fluorescence. Exp. Fluids 20, 1–9.
Beck, J. E., A. H. Lefebvre, and T. R. Koblish (1991). Airblast atomization at
conditions of low air velocity. J. Propul. Power 7, 207–212.
Becker, J. and C. Hassa (2003). Liquid fuel placement and mixing of generic aero-
engine premix module at different operating conditions. J. Eng. Gas Turbines
Power 125, 901–908.
Black, J. D. (1995). Investigation of fluorescence from bulk liquid fuel excited
by Nd:YAG laser harmonics. Technical Report RR(OH)1443, Rolls-Royce Ap-
plied Science Laboratory, Derby, UK.
Blanc, M. V., P. G. Guest, G. von Elbe, and B. Lewis (1947). Ignition of explosive
gas mixtures by electric sparks. I. Minimum ignition energies and quenching
distances of mixtures of methane, oxygen, and inert gases. J. Chem. Phys. 15,
798–802.
Blanc, M. V., P. G. Guest, G. von Elbe, and B. Lewis (1949). Ignition of explosive
gas mixtures by electric sparks. III. Minimum ignition energies and quenching
distances of mixtures of hydrocarbons and ether with oxygen and inert gases.
Proc. Combust. Inst. 3, 363–367.
Blomeyer, M., B. Krautkremer, D. K. Hennecke, and T. Doerr (1999). Mixing
zone optimization of a rich-burn/quick-mix/lean-burn combustor. J. Propul.
Power 15, 288–295.
Boileau, M., S. Pascaud, E. Riber, B. Cuenot, L. Y. M. Gicquel, T. Poinsot,
and M. Cazalens (2008a). Investigation of two-fluid methods for large eddy
simulation of spray combustion in gas turbines. Flow Turbul. Combust. 80,
291–321.
Boileau, M., G. Staffelbach, B. Cuenot, T. Poinsot, and C. Be´rat (2008b). LES
of an ignition sequence in a gas turbine engine. Combust. Flame 154, 2–22.
References 194
Boudier, G., L. Y. M. Gicquel, T. Poinsot, D. Bissie`res, and C. Be´rat (2007).
Comparison of LES, RANS and experiments in an aeronautical gas turbine
combustion chamber. Proc. Combust. Inst. 31, 3075–3082.
Braig, W., H. Schulte, and C. Riegler (1999). Comparative analysis of the wind-
milling performance of turbojet and turbofan engines. J. Propul. Power 15,
326–333.
Brandt, M., K. O. Gugel, and C. Hassa (1997). Experimental investigation of the
liquid fuel evaporation in a premix duct for lean premixed and prevaporized
combustion. J. Eng. Gas Turbines Power 119, 815–821.
Burgoyne, J. H. and L. Cohen (1954). The effect of drop size on flame propagation
in liquid aerosols. Proc. R. Soc. Lond. A 225, 375–392.
Burland, G. N., K. A. Goreham, and R. J. Taunt (1984). The development of the
high energy surface discharge spark igniter. In 29th International Gas Turbine
Conference Exhibit. ASME.
Caines, B. N., R. A. Hicks, and C. W. Wilson (2001, Jul). Influence of sub-
atmospheric conditions on the performance of an airblast atomiser. In 37th
AIAA/ASME/SAE/ASEE Joint Propulsion Conference, Salt Lake City, Utah.
Calcote, H. F., C. A. Gregory Jr., C. M. Barnett, and R. B. Gilmer (1952). Spark
ignition, effect of molecular structure. Indus. Eng. Chem. 44, 2656–2662.
Canepa, E., P. D. Martino, P. Formosa, M. Ubaldi, and P. Zunino (2006). Un-
steady aerodynamics of an aeroengine double swirler lean premixing prevap-
orizing burner. J. Eng. Gas Turbines Power 128, 29–39.
Cant, R. S. and E. Mastorakos (2008). An Introduction to Turbulent Reacting
Flows. London: Imperial College Press.
Carrotte, J. F. and C. Batchelor-Wylam (2003). Characterisation of the instanta-
neous velocity and mixture field issuing from a lean-premixed module (LPM).
In Proc. ASME Turbo Expo, Volume 2, New York, pp. 489–498. ASME.
References 195
C¸engel, Y. A. and M. A. Boles (1998). Thermodynamics: An Engineering Ap-
proach (3rd ed.). New York: McGraw-Hill.
Chakraborty, N., E. Mastorakos, and R. S. Cant (2007). Effects of turbulence on
spark ignition in homogeneous mixtures: A direct numerical simulation (DNS)
study. Combust. Sci. Technol. 179, 293–317.
Cheung, K. Y. and Y. Zhang (2006). Stereo imaging and analysis of combustion
process in a gas turbine combustor. Meas. Sci. Technol. 17, 3221–3228.
Ciajolo, A., A. Tregrossi, R. Barbella, R. Ragucci, B. Apicella, and M. De Joannon
(2001). The relation between ultraviolet-excited fluorescence spectroscopy and
aromatic species formed in rich laminar ethylene flames. Combust. Flame 125,
1225–1229.
Cohen, H., G. F. C. Rogers, and H. I. H. Saravanamuttoo (1996). Gas Turbine
Theory (4th ed.). Harlow, Essex: Longman.
Coordinating Research Council (2004). Handbook of aviation fuel properties.
Technical report.
Danis, A. M., I. Namer, and N. P. Cernansky (1988). Droplet size and equivalence
ratio effects on spark ignition of monodisperse n-heptane and methanol sprays.
Combust. Flame 74, 285–294.
Dietrich, D. L., N. P. Cernansky, M. B. Somashekara, and I. Namer (1990). Spark
ignition of a bidisperse, n-decane fuel spray. Proc. Combust. Inst. 23, 1383–
1389.
Dowling, A. P. and S. Hubbard (2000). Instability in lean premixed combustors.
Proc. Inst. Mech. Eng. Part A 214, 317–332.
Eckstein, J., E. Freitag, C. Hirsch, and T. Sattelmayer (2006). Experimental study
on the role of entropy waves in low-frequency oscillations in a RQL combustor.
J. Eng. Gas Turbines Power 128, 264–270.
European Aviation Safety Agency (2003). Certification Specification CS-E 910.
References 196
Federal Aviation Adminisatration (2000). Federal Aviation Regulation, Section
25.903.
Gaydon, A. G. (1974). The Spectroscopy of Flames (2nd ed.). London: Chapman
and Hall.
Gaydon, A. G. and H. G. Wolfhard (1970). Flames - Their Structure Radiation
and Temperature (3rd ed.). London: Chapman and Hall.
Glassman, I. (1996). Combustion (3rd ed.). San Diego: Academic Press.
Goddard, C. and M. Zedda (2007, Jul). private communication, Combustion Sys-
tems Department, Rolls-Royce plc., PO Box 31, Derby, DE24 8BJ, UK.
Goodger, E. M. (1994). An Overview of Jet Fuel Supply and Quality. Norwich:
Landfall Press.
Gupta, A. K., M. S. Ramavajjala, J. Chomiak, and N. Marchionna (1991). Burner
geometry effects on combustion and NOx emission chracteristics using a vari-
able geometry swirl combustor. J. Propul. Power 7, 473–480.
Hardalupas, Y., C. H. Liu, and J. H. Whitelaw (1994). Experiments with disk
stabilised kerosene-fuelled flames. Combust. Sci. Technol. 97, 157–191.
Hartung, G., J. Hult, C. F. Kaminski, J. W. Rogerson, and N. Swaminathan
(2008). Effect of heat release on turbulence and scalar-turbulence interaction
in premixed combustion. Phys. Fluids 20, 035110.
Hecht, E. (2002). Optics (4th ed.). Harlow, Essex: Addison-Wesley.
Hill, P. G. and C. R. Peterson (1992).Mechanics and Thermodynamics of Propul-
sion (2nd ed.). Addison-Wesley.
Hornak, J. P. (Ed.) (2002). Encyclopedia of Imaging Science and Technology,
Volume 1. New York: John Wiley & Sons.
Ikeda, Y., N. Yamada, T. Nakajima, M. Ohta, M. Inada, and S. Mandai (2002).
References 197
Spray combustion characteristics in a highly pressurized swirl-stabilized com-
bustor. Proc. Combust. Inst. 29, 853–859.
International Civil Aviation Organisation (1993). Manual of the ICAO Standard
Atmosphere (3rd ed.). ICAO (CD-ROM).
International Civil Aviation Organisation (2006). Annexes to the Convention on
International Civil Aviation, Annex 16 – Environmental Protection, Volume
II – Aircraft Engine Emissions.
James, S., J. Zhu, and M. S. Anand (2006). Large-eddy simulations as a design
tool for gas turbine combustion systems. AIAA J. 44, 674–686.
Kaminski, C. F., J. Hult, M. Alde´n, S. Lindenmaier, A. Dreizler, U. Maas, and
M. Baum (2000). Spark ignition of turbulent methane/air mixtures revealed
by time-resolved planar laser-induced fluorescence and direct numerical simu-
lations. Proc. Combust. Inst. 28, 399–405.
Kampa, M. and E. Castanas (2008). Human health effects of air pollution. Envi-
ron. Pollut. 151, 362–367.
Kerrebrock, J. L. (1977). Aircraft Engines and Gas Turbines. Cambridge, Mas-
sachusetts: The MIT Press.
Klein, M., N. Chakraborty, K. W. Jenkins, and R. S. Cant (2006). Effects of
initial radius on the propagation of premixed flame kernels in a turbulent
environment. Phys. Fluids 18, 055102.
Koechner, W. (1970). Thermal lensing in a Nd:YAG laser rod. Appl. Opt. 9,
2548–2553.
Koh, H., J. Jeon, D. Kim, Y. Yoon, and J.-Y. Koo (2003). Analysis of signal
attenuation for quantification of a planar imaging technique. Meas. Sci. Tech-
nol. 14, 1829–1838.
Kohse-Ho¨inghaus, K. and J. B. Jeffries (Eds.) (2002). Applied Combustion Diag-
nostics. New York: Taylor and Francis.
References 198
Kolner, B. H. (1991). Intensity noise enhancement in the half-wave plate/polarizer
attenautor. Opt. Lett. 16, 784–786.
Kru¨ger, V., C. Wahl, R. Hadef, K. P. Geigle, W. Stricker, and M. Aigner (2005).
Comparison of laser-induced incandescence method with scanning mobility
particle sizer technique: The influence of probe sampling and laser heating on
soot particle size distribution. Meas. Sci. Technol. 16, 1477–1486.
Lefebvre, A. H. (1989). Atomization and Sprays. New York: Hemisphere.
Lefebvre, A. H. (1995). The role of fuel preparation in low-emission combustion.
J. Eng. Gas Turbines Power 117, 617–654.
Lefebvre, A. H. (1999). Gas Turbine Combustion (2nd ed.). Philadelphia: Taylor
and Francis.
Lewis, B. and G. von Elbe (1961). Combustion, Flames, and Explosions of Gases
(2nd ed.). New York: Academic Press.
Li, Y. G. and R. L. Hales (2003). Steady and dynamic performance and emissions
of a variable geometry combustor in a gas turbine engine. J. Eng. Gas Turbines
Power 125, 961–971.
Locke, R. J., Y. R. Hicks, R. C. Anderson, and M. M. Zaller (1998). Optical fuel
injector patternation measurements in advanced liquid-fueled, high pressure,
gas turbine combustors. Combust. Sci. Technol. 138, 297–311.
Lo¨fstro¨m, C., H. Kaaling, and M. Alde´n (1996). Visualization of fuel distribu-
tions in premixed ducts in a low-emission gas turbine combustor using laser
techniques. Proc. Combust. Inst. 26, 2787–2793.
Lorenzetto, J. E. and A. H. Lefebvre (1977). Measurements of drop size on a
plain-jet airblast atomiser. AIAA J. 15, 1006–1010.
Marchione, T., S. F. Ahmed, R. Balachandran, and E. Mastorakos (2007, July).
Effectiveness of localised spark ignition in recirculating n-heptane spray flames.
References 199
In 21st International Colloquium on the Dynamics of Explosions and Reactive
Systems, Poitiers, France.
Marchione, T., S. F. Ahmed, and E. Mastorakos (2007, Sep). Ignition behaviour
of recirculating spray flames using multiple sparks. In 5th Mediterranean Com-
bustion Symposium, Monastir, Tunisia.
McDonell, V. G., L. Arellano, S. W. Lee, and G. S. Samuelsen (1996). Effect
of hardware alignment on fuel distribution and combustion performance for a
production engine fuel-injection assembly. Proc. Combust. Inst. 26, 2725–2732.
McDonell, V. G. and G. S. Samuelsen (2000). Measurement of fuel mixing and
transport processes in gas turbine combustion. Meas. Sci. Technol. 11, 870–
886.
Muruganandam, T. M., B. -H. Kim, M. R. Morrell, V. Nori, M. Patel, B. W.
Romig, and J. M. Seitzman (2005). Optical equivalence ratio sensors for gas
turbine combustors. Proc. Combust. Inst. 30, 1601–1609.
Naegeli, D. W. and L. G. Dodge (1991). Ignition study in a gas turbine combustor.
Combust. Sci. Technol. 80, 165–184.
Nicholas, J. (Ed.) (2005). The Jet Engine (6th ed.). Rolls-Royce plc.
Nygren, J., J. Hult, M. Richter, M. Alde´n, M. Christensen, A. Hultqvist, and
B. Johansson (2002). Three-dimensional laser-induced fluorescence of fuel dis-
tributions in an HCCI engine. Proc. Combust. Inst. 29, 679–685.
Odgers, J. and A. Coban (1977). Energy release to static gas from a 12-Joule high
energy ignition system. ASME Pap (77-GT-18).
Ossler, F., T. Metz, and M. Alde´n (2001). Picosecond laser-induced fluorescence
from gas-phase polycyclic aromatic hydrocarbons at elevated temperatures. I.
Cell measurements. Appl. Phys. B 72, 465–478.
Papoulis, A. and S. U. Pillai (2002). Probability, Random Variables and Stochastic
Processes (4th ed.). New York: McGraw-Hill.
References 200
Penner, J. E., D. H. Lister, D. J. Griggs, D. J. Dokken, and M. McFarland
(Eds.) (1999). Aviation and the Global Atmosphere. Cambridge: Cambridge
University Press.
Peters, J. E. and A. M. Mellor (1980). An ignition model for quiescent fuel sprays.
Combust. Flame 38, 65–74.
Polymeropoulos, C. E. and S. Das (1975). The effect of droplet size on the burning
velocity of kerosene-air sprays. Combust. Flame 25, 247–257.
Pope, S. B. (2000). Turbulent Flows. Cambridge: Cambridge University Press.
Press, W. H., S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery (2007).
Numerical Recipes - The Art of Scientific Computing (3rd ed.). New York:
Cambridge University Press.
Pucher, G. and W. D. Allan (2004). Turbine fuel ignition and combustion fa-
cility for extremely low temperature conditions. In Proc. ASME Turbo Expo,
Volume 1, pp. 385–392. ASME.
Pucher, G., G. Wang, M. F. Bardon, D. P. Gardiner, A. M. Namet-Allah, and
A. Benaissa (2001). Enhanced ignition systems for aircraft altitude relight.
In 48th Annual Conference of the Canadian Aeronautics and Space Institute,
Volume 48, pp. 611–620.
Rah, S.-C., A. F. Sarofim, and J. M. Bee´r (1986). Ignition and combustion of
liquid fuel droplets. Part I: Impact on pollutant formation. Combust. Sci.
Technol. 48, 273–284.
Rao, K. V. L. and A. H. Lefebvre (1976). Minimum ignition energies in flowing
kerosine-air mixtures. Combust. Flame 27, 1–20.
Read, R. W., J. W. Rogerson, and S. Hochgreb (2008, Jan). Relight imaging at
low temperature, low pressure conditions. In 46th AIAA Aerospace Sciences
Meeting and Exhibit. Reno, Nevada.
References 201
Richards, G. A. and A. H. Lefebvre (1989). Turbulent flame speeds of hydrocarbon
fuel droplets in air. Combust. Flame 78, 299–307.
Rose, H. E. and T. Priede (1959). An investigation of the characteristics of spark
discharges as employed in ignition experiments. Proc. Combust. Inst. 7, 454–
463.
Rosin, P. and E. Rammler (1933). The laws governing the fineness of powdered
coal. J. Inst. Fuel 7, 29–36.
Rowe, A. L. (2005, July). private communication, Performance Department,
Rolls-Royce plc., PO Box 31, Derby, DE24 8BJ, UK.
Schulz, C. and V. Sick (2005). Tracer-LIF diagnostics: quantitative measurement
of fuel concentration, temperature and fuel/air ratio in practical combustion
systems. Prog. Energy Combust. Sci. 31, 75–121.
Seyfried, H., C. Brackmann, A. Lindholm, M. Linne, F. Barreras, and R. v. d.
Bank (2007). Optical diagnostics applied to a gas turbine pilot burner. AIAA
J. 45, 2702–2709.
Sick, V. and B. D. Stojkovic (2001). Attenuation effects on imaging diagnostics
of hollow-cone sprays. Appl. Opt. 40, 2435–2442.
Singh, A. K. and C. E. Polymeropoulos (1986). Spark ignition of aerosols. Proc.
Combust. Inst. 21, 513–519.
Smith, J. D. and V. Sick (2005). High-speed fuel tracer fluorescence and OH
radical chemiluminescence imaging in a spark-ignition direct-injection engine.
Appl. Opt. 44, 6682–6691.
Song, J., C. Li, N. S. Kim, and K. Ueda (2000). Passively Q-switched diode-
pumped continuous-wave Nd:YAG-Cr4+:YAG laser with high peak power and
high pulse energy. Appl. Opt. 39, 4954 – 4958.
Stojkovic, B. D. and V. Sick (2001). Evolution and impingement of an automotive
References 202
fuel spray investigated with simultaneous Mie/LIF techniques. Appl. Phys.
B 73, 75–83.
Su, L. K. and N. T. Clemens (1999). Planar measurements of the full three-
dimensional scalar dissipation rate in gas-phase turbulent flows. Exp. Flu-
ids 27, 507–521.
Turns, S. R. (1996). An Introduction to Combustion - Concepts and Applications.
New York: McGraw-Hill.
UK Ministry of Defence (2002). Defence Standard 91-91.
Warnatz, J., U. Maas, and R. W. Dibble (2006). Combustion (4th ed.). Berlin:
Springer-Verlag.
Xiong, T., Z. Huang, and Y. Wang (1981). Some factors affecting altitude relight-
ing performance of turbojet engine combustor. ASME Pap (81-GT-49).
Zhu, M., A. P. Dowling, and K. N. C. Bray (2001). Self-excited oscillations in
combustors with spray atomizers. J. Eng. Gas Turbines Power 123, 779–786.