Document [original]

Shockless Explosion Combustion

Controlled Autoignition in Stratified Mixtures

for

Pressure Gain Combustion

vorgelegt von

Fatma Cansu Yücel, M.Sc.

ORCID: 0000-0003-1252-7013

von der Fakultät V – Verkehrs- und Maschinensysteme

der Technischen Universität Berlin

zur Erlangung des akademischen Grades

Doktorin der Ingenieurwissenschaften

– Dr.-Ing. –

genehmigte Dissertation

Promotionsausschuss:

Vorsitzender: Prof. Dr.-Ing. Ennes Sarradj

Gutachter: Prof. Dr.-Ing. Christian Oliver Paschereit

Prof. Dr.-Ing. Rupert Klein

Prof. Ephraim Gutmark, Ph.D., D.Sc.

Prof. Dr. sc. Panagiotis Stathopoulos

Tag der wissenschaftlichen Aussprache: 20.07.2021

Berlin 2021

Preface

I would like to start by expressing my gratitude to the German Science Foundation for

financing my project as a part of the “Collaborative Research Center 1029”. It has been

a great privilege to work under these conditions.

For giving me the great opportunity to be a part of his team and to be allowed to

work on such interesting projects, I would like to thank Prof. Oliver Paschereit. As

my supervisor he has been supportive throughout my graduation, which helped me

complete this work. Furthermore, I would like thank Prof. Rupert Klein as my second

supervisor, who always had an open ear for my questions and helped out in countless

discussions on research questions. Thanks also to Prof. Ephraim Gutmark and Prof.

Panagiotis Stathopoulos who have agreed on reviewing my work and also have been

providing valuable assistance to my research projects in the past years with their great

expertise on the topic of pressure gain combustion.

Next, I would like to thank my colleague Fabian Habicht with whom I have been working

very closely. I am very grateful for the successful and enjoyable collaboration and for

all his support that I received over the years. Special thanks to Joshua Gray who has

been supervising me as a graduate student and who got me interested in the topic of

pressure gain combustion. My greatest gratitude go to Prof. Neda Djordjevic and Prof.

Myles Bohon from whom I have been receiving incredible mentorship professionally as

well as personally. They kept me motivated through rough times and working with

them left a lasting effect on me which I value greatly.

I want to thank everyone from the Collaborative Research Center 1029, especially

Prof. Rudibert King and Florian Arnold. Moreover, I would like to mention the

“Integrated Research Training Group” led by Prof. Panagiotis Stathopoulos and Prof.

Neda Djordjevic and coordinated by Sonja Hoßbach for enabling me an exchange which

allowed me to spend a month at Prof. Kasahara’s Laboratory at the Nagoya University.

Thanks also to Prof. Kasahara, Prof. Matsuoka and Prof. Kawasaki for the great

insights on the topic of pressure gain combustion I have been receiving throughout my

stay in Japan.

I can not thank enough my colleagues from the Hermann-Föttinger-Institute for the

great times we shared during numerous activities; collaboration in research, traveling to

conferences, bouldering or fishing. Thanks a lot to Niclas Hanraths, Tom Tanneberger,

Richard Blümner, Eric Bach, Lisa Zander, Johann Vinkeloe, Finn Lückoff, Tim Rähse

and Mohammad Rezay Haghdoost. I also want to thank all student research assistants

Iwo Gotthans, Viktor Morell, Robert Kanisch, Friederike Großmann, Adrian Stobbe,

iii

Oliver Klement and Alexander Jaeschke who contributed to this work with great

enthusiasm. Thanks also to Andy Göhrs who has been supporting me regarding the

technical implementation of my projects. He always managed to find creative solutions

to bring forward ideas that seems unfeasible at first whether it comes to the SEC,

the PDC or even my UAZ Buchanka. Also many thanks to Thorsten Dessin, Robert

Bahnweg, Andrea Maneck, Heiko Stolpe, Colin Muxlhanga, Christian Menzel, Martin

Franke and Sebastian Schimek. Further, I highly appreciate all who helped me coping

with the german bureaucracy; Sandy Meinecke, Maria Lück and Steffi Stehr. Moreover,

I would like to thank Alexandra Elbakyan for disseminating knowledge without barriers

and providing invaluable access to scientific research.

And finally I would like to thank my family and friends Beyazlale Yücel, Orkun Yücel,

Sila Yücel, Jacob Kuntzsch and Valeria Netesova and of course my partner and best

friend Dirk van den Biggelaar. I am infinitely grateful to have you by my side and I

love you all from the bottom of my heart!

Berlin, 2021 Fatma Cansu Yücel

dedicated to Carolin Gries

Zusammenfassung

Im Rahmen dieser Arbeit wird das zuverlässige und wiederholbare Erzeugen einer homo-

genen Selbstzündung innerhalb einer kontinuierlichen Strömung als neuartiges Konzept

zur druckerhöhenden Verbrennung untersucht. Die druckerhöhende Verbrennung ist

ein vielversprechender Ansatz zur Steigerung des Wirkungsgrades von Gasturbinen

verglichen zur herkömmlichen Gleichdruckverbrennung.

Das Konzept basiert auf einem zyklischen Betrieb bei dem ein geschichtetes Brennstoff-

profil bei hohen Frequenzen in ein kontinuierlich durchströmtes Rohr eingedüst wird und

anschließend homogen zündet. Das eingedüste Brennstoffprofil ist präzise geschichtet,

um so den Gradienten in der Verweilzeit zu kompensieren der während des Eindüsvor-

ganges zustande kommt. Somit wird das gleichzeitige Zünden des gesamten Brennkam-

mervolumens erreicht. Die daraus resultierende aerodynamische Begrenzung führt zu

einem moderaten Drucksanstieg innerhalb der Brennkammer. Dieser Druckanstieg

ist eine Funktion der Homogenität der Selbstzündung, welche durch die Anzahl der

simultan auftretenden Zündfronten innerhalb des Brennkammervolumens charakterisiert

werden kann. Eine höhere Homogenität der Selbstzündung ist einhergehend mit einer

gesteigerten Druckerhöhung bei gleichzeiter Minimierung des Auftretens von Stößen.

Diese Art von Zündung wird als stoßfreie Explosionsverbrennung (englisch: shockless

explosion combustion, kurz: SEC) bezeichnet.

Das wesentliche Ziel dieser Arbeit ist es den SEC Prozess experimentell zu untersuchen

und einen zuverlässigen und wiederholbaren Betrieb zu ermöglichen. Dies wurde im

Rahmen von vier Veröffentlichungen realisiert. Hierfür wurde zunächst ein Prüfstand zur

präzisen Eindüsung eines geschichteten Brennstoffprofils ausgelegt. In einem nächsten

Schritt wurde, unter Anwendung optischer Messverfahren zur Konzentrationsmessung,

die Kontrollierbarkeit der Brennstoffverteilung innerhalb des Brennkammervolumens

durch gezielte Eindüsung gezeigt. Im Rahmen von reaktiven Verbrennungsversuchen

wurden drei ausgewählte Brennstoffprofile hinsichtlich ihres Zündverhaltens untersucht.

Ein signifikanter und wiederholbarer Einfluss der Befüllungskurve auf die Homogen-

ität der Selbstzündung konnte gezeigt werden. Basierend auf diesen Beobachtungen

wurde ein Regelalgorithmus zur zyklus-gemittelten Generierung verschiedener Selbstzün-

dungsmoden durch die Optimierung der Endüsung entwickelt und angewandt. Optische

Messungen in Kombination mit Druckmessungen zeigen ein komplexes Zusammenspiel

zwischen Wärmefreisetzung und Druckerhöhung beeinflusst durch das Zündverhalten

des verwendeten Brennstoffs. Vier verschiedene Moden der Selbstzündung wurden

identifiziert: turbulente Deflagration, subsonische Selbstzündung, supersonische Selb-

stzündung und das simultane Enstehen mehrerer Selbstzündungen.

Die Ergebnisse dieser Arbeit zeigen, dass eine stoßfreie Explosionsverbrennung experi-

mentell realisierbar ist. Die Homogenität von Selbstzündungsprozessen wurde erfolgreich

durch das gezielte Eindüsen eines geeigneten Brennstoffprofils beeinflusst. Es wurde

gezeigt, dass die erzeugten Druckamplituden mit der Homogenität der Zündfront korre-

lieren. Weiterhin wurde beobachtet, dass die Brennstoffeingenschaften eine entschei-

dende Rolle bei der Entstehung bestimmter Flammmenausbreitungsmoden spielen.

Diese Brennstoffeigenschaften können durch das gezielte Eindüsen eines Brennstoffpro-

fils ausgenutzt werden, um somit bestimmte Moden zu Generieren. Dies wurde im

Rahmen dieser Arbeit durch die erfolgreiche Anwendung eines Reglers gezeigt.

vii

Abstract

This work investigates the reliable generation of a homogeneous autoignition in a

reactive mixture flow as a pressure gain combustion approach. Pressure gain combustion

represents a promising concept to achieve an increase in the thermal efficiency of gas

turbine applications compared to conventional constant pressure combustion. The

concept is based on a pulsating operation with high frequency injection of a defined

mixture profile into a continuous air flow that undergoes homogeneous autoignition.

The injected fuel profile has been tailored to compensate for the gradient in residence

time and hence, enable the simultaneous ignition of the entire combustor volume leading

to an aerodynamic confinement which ultimately results in an increase in pressure. This

pressure rise is highly dependent on the homogeneity of the autoignition which can

be characterized by the number of quasi-simultaneous ignitions occurring inside the

combustor volume. An increased homogeneity results in a greater pressure rise and

simultaneously minimized the occurrence of shock waves. This type of combustion is

termed as shockless explosion combustion (SEC).

The main objective of this work is investigate the SEC process experimentally and

enable a repeatable and reliable operation which has been realized within the frame of

four publications. First, an initial test rig was modified to allow for the precise injection

of a desired fuel profile into a continuous air flow and subsequently observe autoignition

within a desired combustor section. Secondly, optical measurement techniques were

applied to quantify the successful injection of the desired fuel profile which stays

largely preserved during convection in the combustor. As a next step, the correlation

between three model injection profiles and the resulting autoignition was investigated.

A significant and reproducible influence of the fuel injection on the ignition distribution

is observed. These observations are subsequently used to apply an extremum seeking

control algorithm which controls the cycle–averaged formation of different autoignition

modes by optimizing the fuel injection profile. Optical and pressure measurements reveal

a complex interaction between heat release and pressure waves influenced by low and high

temperature chemistry of the applied fuel. Four different modes of ignition have been

identified which are classified, namely: turbulent deflagration, subsonic autoignition,

supersonic autoignition, and aerodynamic confinement by multiple simultaneous ignition

fronts.

The results presented in this work, demonstrate the experimental feasibility of a

shockless explosion combustion. It was shown that autoignition, which is primarily

driven by chemical kinetics, and thus, highly sensitive to perturbations, can be greatly

affected by the injected fuel profile. The amplitude of the pressure rise was found to

viii

strongly correlate with the autoignition homogeneity. Moreover, it was found that the

autoignition modes observed, are impacted by the applied fuel which exhibits multi-stage

ignition behavior. These fuel characteristics can be exploited to trigger different modes

of autoignition by the proper adjustment of the injected fuel trajectory. By this, the

probability of the occurrence of distinct autoignition modes can be greatly impacted,

which was shown by the successful application of a closed-loop control algorithm.

Contents

List of Figures xii

Nomenclature xv

1 Introduction 1

1.1 Pressure Gain Combustion . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.2 ScopeofthisThesis............................. 5

1.3 Shockless Explosion Combustion . . . . . . . . . . . . . . . . . . . . . . 6

2 Experimental Setup 19

2.1 SECwithBypass.............................. 19

2.2 Evolution of the SEC Test Rig . . . . . . . . . . . . . . . . . . . . . . . 21

2.3 SingleTubeSEC .............................. 24

3 Methodology 29

3.1 Measurements without Chemical Reaction . . . . . . . . . . . . . . . . 29

3.2 Measurements with Chemical Reaction . . . . . . . . . . . . . . . . . . 36

4 Post Processing 45

4.1 Proper Orthogonal Decomposition . . . . . . . . . . . . . . . . . . . . . 45

4.2 Cycle Averaging and Standard Deviation . . . . . . . . . . . . . . . . . 46

4.3 ImageProcessing .............................. 46

5 Publications 49

5.1

Publication I: Effect of the Switching Times on the Operating Behaviour...

5.2

Publication II: Investigation of the Fuel Distribution in a Shockless

ExplosionCombustor............................ 66

5.3

Publication III: Autoignition in Stratified Mixtures for Pressure Gain

Combustion................................. 77

5.4 Publication IV: Controlled Autoignition in Stratified Mixtures . . . . . 88

6 Discussion 107

6.1 Fuel Stratification Approach . . . . . . . . . . . . . . . . . . . . . . . . 107

6.2 Correlation of Fuel Stratification and Autoignition . . . . . . . . . . . . 111

6.3 SingleCycleAnalysis............................ 112

6.4 Autoignition Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

7 Conclusion 117

8 Outlook 119

8.1 Operation under Elevated Pressure Conditions . . . . . . . . . . . . . . 119

8.2 Further Improvements of the SEC Operation . . . . . . . . . . . . . . . 121

Bibliography 123

A Publications Associated with this Thesis 133

xii

List of Figures

1.1 Total energy supply by source. Adapted from [36]. . . . . . . . . . . . . 2

1.2

Comparison of the ideal constant volume (Humphrey) and the constant

pressure cycle (Joule). Adapted from [74]. . . . . . . . . . . . . . . . . 4

1.3 Sketch of the SEC cycle. Adapted from [89]. . . . . . . . . . . . . . . . 7

1.4

a) ignition time

τiri

(black), ignition delay time

τidt

(green) and injec-

tion time

tinj

(red) for a constant fuel injection and b) for a stratified

fuel injection (with varying equivalence ratio

designed to deliver a

simultaneous autoignition of the entire mixture volume). . . . . . . . . 8

1.5 Thermal criticality. Adapted from [68]. . . . . . . . . . . . . . . . . . . 10

1.6 Flammability limits of hydrocarbons. Adapted from [71]. . . . . . . . . 12

1.7

Low (A, B and C), intermediate (A and B) and high temperature (A)

oxidation of dimethyl ether. Adapted from Curran et al. [16]. . . . . . . 13

2.1 Sketch of the atmospheric SEC test rig with bypass. . . . . . . . . . . . 20

2.2

Valve array with mounting block and five high-speed solenoid valves

connected in parallel and connectors at inlet (red) and outlet (blue) a).

Sketch of the cross section of the mounting block with one example valve

mountedb). ................................ 22

2.3

Detailed view of the injection station with 10 circumferentially distributed

ports each equipped with one high-speed solenoid valve. . . . . . . . . . 23

2.4

Sketch of the atmospheric test rig equipped with ionization probes (I1–

I7), high frequency pressure transducers (P1–P4), thermocouples (T1–

T2) and low frequency pressure transducers (FF and FA). A second

exchangeable combustor with optical access is sketched in b). The

connection of convection and ignition delay time is visualized in c). . . 24

2.5

Calculated ignition delay times for a DME–air mixture at atmospheric

pressure, a constant fuel temperature of 330 K, a varying air temperature

and varying equivalence ratios using Cantera [23]. . . . . . . . . . . . . 26

3.1

Sketch of the experimental setup for laser Doppler anemometry. Adapted

from[75]. .................................. 30

xiii

3.2 Beer-Lambertlaw.............................. 32

3.3 Schematic energy level diagram. Adapted from [28]. . . . . . . . . . . . 34

3.4

The connection of ignition time measured by the ionization probes

τiri

the position xof the reactive mixture and the ignition delay time τidt. 37

3.5

Correlation between standard deviation of the autoignition front obtained

from ionization probes and optical OH* emissions for ignitions originating

within the combustor section(green) and outside of the combustor section

(black). ................................... 38

3.6 Example data of pressure and ionization probe. . . . . . . . . . . . . . 40

3.7

Example spectrometer data representing the characteristic wavelengths

of OH*, CH* and C2*. ........................... 42

4.1

Visualization of the image calibration using Matlab ’fitgeotrans’ func-

tion [

] applied to target image using an image doubler (

publication IV

6.1 Overview and context of the publications. . . . . . . . . . . . . . . . . 108

6.2

Model injection trajectories with a) valve control trajectory and respective

modeled injection curve b) measurement and simulation data. Adapted

from Publication IV and included for convenience. . . . . . . . . . . 109

6.3

Maximum pressure amplitude over delay time ∆

τip

for three different

injection trajectories. Adapted from

publication III

and included for

convenience.................................. 111

6.4

Different modes of autoignition. Adapted from

publication IV

and

included for convenience. . . . . . . . . . . . . . . . . . . . . . . . . . . 113

6.5

Probability of the occurrence of autoignition modes for different control

steps...................................... 113

6.6

Controller performance for optimizing cost function

. Adapted from

publication IV and included for convenience. . . . . . . . . . . . . . . 115

8.1

Sketch of the high pressure SEC test rig equipped with ionization probes

–I

5), high-speed pressure transducer (P

–P

) and low-speed pressure

transducers (FUS and FDS)......................... 119

xiv

Nomenclature

Roman Symbols

Symbol Unit Name

T[K] temperature

s[J/K] entropy

t[s] time

p[bar] pressure

x[m] spatial coordinate

l[m] combustor length

r[J/s] heat generation rate

l[J/s] heat loss rate

∆U[J/s] change internal energy

vm3vessel volume

Q0[J] total heat generation

w[mol/m3s] reaction velocity

N[-] Avogadro number

q[J] specific heat generation

ka[mol/s] reaction rate

E[J/mol] activation energy

R[kg m2/s2mol K] universal gas constant

Z[-] pre-exponential factor

A[m2] surface area

u[m/s] propagation velocity

a[m/s] speed of sound

d[m] distance

df[m] fringe spacing

f[Hz] frequency

I[W] laser intensity

G[cm−2/atm] line strength

i[-] modulation amplitude

a[cm−1] modulation depth

H[-] Fourier component

c[m/s] speed of light

h[Js] Planck constant

S[V] detection signal

V[m3] volume

R[Ω] resistance

z[-] mathematical field

a[-] time coefficient

U,V[-] orthogonal matrices

P[-] pressure matrix

x[-] data point

N[-] number of probes

Greek Symbols

Symbol Unit Name

η[-] efficiency

Π[-] pressure ratio

τ[s] delay time

ϕ[-] equivalence ratio

χ[W/m2K] heat transfer coefficient

ξ[-]

parameter for autoignition mode clas-

sification

θ[◦] intersection angle

λ[m] wavelength

ν[Hz] laser frequency

γ[-] transmission coefficient

α[-] absorption coefficient

χ[-] species

Φ[-] line shape

ω[Hz] modulation frequency

ψ[rad] phase shift

φ[-] absorption yield

σ[m2] absorption cross section

Ω[◦] solid angle

ρ[kg/m3] density

xvi

Φ[-] spatial function

Σ[-] singular values

Subscripts

Index Meaning

th thermal

C compressor

ign ignition

inj injection

0 initial

cr critical

ai autoignition

fl fluorescence

Acronyms

Acronym Meaning

IEA International Energy Agency

CVC constant volume combustion

SEC shockless explosion combustion

PGC pressure gain combustion

PDC pulse detonation combustion

RDC rotating detonation combustion

RCM rapid compression machine

DME dimethyl ether

NTC negative temperature coefficient

WG wastegate

TDLAS tunable diode laser absorption spectroscopy

PLIF planar laser induced fluorescence

LDA laser Doppler anemometry

WMS wavelength modulation spectroscopy

PMT photomultiplier

CWL center wavelength

FWHM full width at half maximum value

POD proper orthogonal decomposition

HCCI homogeneous charge compression ignition

xvii

BFGS Broyden-Fletcher-Goldfarb-Shanno

LTC low temperature chemistry

ITC intermediate temperature chemistry

HTC high temperature chemistry

xviii

1 Introduction

“Real generosity towards the future lies in giving all to the present.”

Albert Camus

Humanity has faced challenges for as long as its existence. New technologies were

developed that steadily increased our living standards but also required ever more

resources. Nowadays, we have come to the realization that our progress as a civilization

is accompanied by a destructive and severe habit, namely the generation of energy from

unsustainable resources. The challenge humanity is facing today is man-made. If a

global effort to make a transition to sustainability will not deliver on the promises made,

global warming will continue and will destabilize the environment and societies all over

the world, inevitably accounting for countless environmental and human tragedies. The

century is the moment in which our situation is rapidly intensifying but where

also our awareness, ability and willingness to combat against the current problems

are increasing. Limiting the total energy consumption, increasing the exploitation of

renewable resources and making the remaining energy consumption from unsustainable

resources more efficient is the main technical objective for combating global warming.

This three-pronged approach is highlighted by the International Energy Agency (IEA)

[

], which confirms that tremendous efforts towards the increase of exploitation of

renewable energy sources still leave a significant need for the combustion of (fossil) fuels

in conventional gas turbines for energy generation.

According to Masnadi et al. [

], the world population will rise from around 7 to 9

billion by 2050, leading to an increase in energy consumption by approximately 56 %.

As illustrated in Fig. 1.1, the energy demand by 2018 was mainly covered by fossil

fuels (i.e., coal, oil and natural gas). Nuclear energy provided about 5 % and renewable

resources such as wind, solar as well as biofuels and hydro power reach a share of about

13.8 % according to the IEA. Roughly 67% of the worlds renewable energy resources

involves biomass energy. Biomass energy is referred to as energy derived from renewable

organic material such wood or crop residues [

]. Sunlight which is being absorbed

and stored as chemical energy during growth by flora can be converted into combustible

liquids or gases such as methane, ethanol or bio diesel which are referred to as biofuels.

1 Introduction

Figure 1.1: Total energy supply by source. Adapted from [36].

Renewable energy from biomass is a controversial subject since its advantages come

along with certain disadvantages. Compared to fossil fuels the energy density of a

biofuel is less. Additionally, quick cultivation of biomass is disadvantageous since it can

harm fertile land, leading to deforestation and decreases natural diversity. Nevertheless,

biofuels can be produced widely and are considered carbon-neutral since the released

during combustion has been captured by the organic material from the atmosphere

during its photosynthetic growth [1].

Energy from solar and wind are considered the most environmentally friendly way of

energy generation. The main challenge arising from a high dependency on wind and

solar in the energy system is the unpredictability and, thus, volatility of this kind

of energy supply. This is because of the inherent dependency on the weather, which

naturally exhibits strong volatility. In periods with high energy generation and low

consumer demand, excess energy remains, while periods with high energy demand and

low supply cause an energy deficiency. To transform our energy system such that it

is capable incorporating a high percentage of its energy from wind and solar, it is

important to have strong supply-side infrastructure in place to cope with moments

of low supply. This fall-back generation should be able to rapidly respond to the

fluctuating demand and supply and should, therefore, be able to ramp-up to significant

energy output in a short period of time to fill the gaps.

One of the prime candidates for this role are gas turbines. A gas turbine has good

throttle behaviour and is able to provide a short reaction time compared to conventional

1.1 Pressure Gain Combustion

power plants. Additionally, a major advantage of gas turbines is that they can be fuelled

with carbon-neutral biofuels or hydrogen [

], which are both sustainable energy sources.

Furthermore, both fuels can be produced with power-to-gas/power-to-liquid technology.

This provides a more convenient way of energy storage in the form of liquid and gaseous

fuels such as methanol, dimethyl ether, methane and other hydrocarbons [

]. Placing

a gas turbine in a power-to-gas/power-to-liquid cycle justifies its application within a

sustainable energy system. Additionally, these synthetic fuels can be easily fed into the

existing supply line. Energy generation from sustainable resources, in combination with

small-scale gas turbines to fill in the gaps, may be the key when it comes to enabling

the transition towards a green energy system. Therefore, research on increasing the

efficiency of gas turbines is of significance.

In this work, pressure gain combustion (PGC) is presented as an alternative to conven-

tional constant pressure combustion for the implementation into a gas turbine cycle.

The presented approach is based on a cyclic operation in which a homogeneous autoigni-

tion of a reactive mixture profile is achieved, inducing an aerodynamic confinement

throughout the combustor volume. Thus, an increase in the pressure is achieved. The

primary motivation is to achieve a significant increase in the thermal efficiency of a gas

turbine once implemented.

1.1 Pressure Gain Combustion

For many years research has focused on the topic of gas turbines and efficiency increase.

Optimizations of compressor and turbine components reached technological boundaries

and, thus, a stagnation in efficiency improvement. A promising approach to achieve a

break through is to overcome losses during the combustion process by shifting towards

new technologies. Hence, PGC has been in the scope of research in recent years as an

alternative to conventional constant pressure combustion (CPC). Implementing PGC

into a gas turbine cycle promises an increase in thermal efficiency [

]. The PGC

concept is associated with a (quasi-)constant volume combustion (CVC) of a fuel–air

mixture resulting in an increase in pressure.

The ideal cycle for the constant volume and constant pressure combustion are compared

in Fig. 1.2. As visualized in the temperature-entropy diagram (Fig. 1.2a), the entropy

generation is reduced in the Humphrey cycle (CVC) compared to the Joule cycle (CPC),

leading to an elevated work extraction due to an increased available energy output.

Figure 1.2b visualizes the thermal efficiency

ηth

as a function of the compressor ratio

simulated for a constant turbine inlet temperature. Clearly, the CVC cycle results

1 Introduction





Humphrey cycle

Joule cycle

a) b)

Figure 1.2:

Comparison of the ideal constant volume (Humphrey) and the constant

pressure cycle (Joule). Adapted from [74].

in an increased thermal efficiency. However, for high compressor ratios this advantage

diminishes. At lower operating pressures, a 20 % gain in efficiency can be achieved

according to Stathopoulos et al. [74].

Different approaches have been investigated in the past decades. Among the most known

concepts are detonation-based approaches such as pulse detonation combustors (PDC)

and rotation detonation combustors (RDC). PDCs are cyclic operating tubular combus-

tors which are filled with a flammable mixture. This mixture is subsequently ignited

using an external ignition source to initiate a propagating flame front. After initiation,

the flame front propagates through a section composed of different obstacles which

promote flame acceleration, such that a fast transition into a propagating detonation

wave is achieved. This detonation wave burns the mixture quasi-instantaneously with a

propagation speed of about 2000 m/s (for stoichiometric hydrogen–air mixtures). Thus,

the gases have no time to expand resulting in an quasi-CVC of the entire mixture. The

combustor is subsequently purged with air and refilled to start the next cycle [

By achieving a high frequency operation and assembling multiple tubes into a multi-tube

configuration, a quasi-steady operation is enabled approximating steady state inlet

conditions for a downstream attached turbine.

RDCs, also called continuous detonation combustors, consist of an annular chamber

which is fed by a continuous detonative mixture. The mixture is ignited using a

spark plug initiating a detonation wave which travels around a closed-loop combustion

chamber [

]. Compared to PDCs, RDCs operate at much higher frequencies on

the order of kHz, resulting in steadier conditions downstream of the combustor which

is advantageous for a downstream attached turbine. However, major drawbacks of

detonation-based concepts are sharp pressure peaks, which impede the refilling process at

high frequencies and additionally harm mechanical components. Moreover, detonation

1.2 Scope of this Thesis

waves and the DDT process generally are associated with considerable losses [94, 74].

An alternative for achieving a quasi-CVC based on the Humphrey cycle (see Fig. 1.2) was

proposed by Bobusch et al. [

]. This concepts aims for a quasi-simultaneous autoignition

of a reactive mixture without the presence of detonation waves. The so-called shockless

explosion combustion (SEC), similar to the PDC, is based on a cyclic operation using

a tubular combustor, which will be further discussed in Sec. 1.3. At the beginning

of each cycle, a stratified fuel–air mixture is injected into a continuous air mass flow.

After a certain delay time, this mixture undergoes a quasi-simultaneous autoignition

resulting in an aerodynamic confinement. Thus, a rather gradual rise in pressure is

achieved compared to sharp pressure peaks associated with detonation waves. A pressure

wave is then formed traveling downstream in the combustor. At the acoustically open

combustor outlet, the pressure wave is reflected as an expansion wave propagating

upstream. Once the expansion wave reaches the combustor inlet, the pressure drops

below the supply pressure supporting the refilling of the combustor to restart the cycle.

The SEC provides two main advantages: i) shock waves, which are associated with losses,

can be avoided and ii) the refilling process is greatly facilitated. Both features highly

contribute to an overall increase in efficiency along with an improved practicability in

terms of implementation.

1.2 Scope of this Thesis

As PGC provides a good basis for increasing the thermal efficiency in gas turbines,

there is still a fundamental understanding to acquire. While detonation-based concepts

have been under research for many years under different aspects, the fundamentals of

an SEC have not been yet fully elucidated.

This work investigates the SEC phenomenon experimentally. A broad range of mea-

surement techniques are used to understand the interplay of physics, chemical kinetics,

flow turbulence and autoignition modes. From the starting point of this work until its

final completion many small steps had to be taken demanding deep understanding of

different fields and disciplines as well as creativity and endurance.

This thesis is divided into eight chapters which are shortly described in the following:

Chapter 1 gives a brief introduction and motivation for this work followed by the

theoretical aspects of the SEC in terms of autoignition, modes of autoignition as well as

hydrocarbon oxidation. The experimental setup and its evolution during the research

period are further discussed in Ch. 2. A short introduction and theoretical aspects to

each applied measurement technique is provided in Ch. 3 followed by an overview of

1 Introduction

the data processing tools used for post processing in Ch. 4. Chapter 5 contains four

publications as a part of the results of this work. The final discussion is provided in

Ch. 6. Here, the contributions of each publication are set into context and the reader is

guided through the progress of this work. A conclusion and an outlook are provided in

Ch. 7 and Ch. 8, which emphasize and summarize the key findings and provide a basis

for future work on the topic of the SEC.

1.3 Shockless Explosion Combustion

The theoretical aspects discussed in this chapter provide the framework for the inter-

pretation of experimental results and help to gain a profound understanding of the

SEC process. In this section first, the SEC cycle is introduced and the concept of fuel

stratification is described. Next, an analytical model of (spontaneous) autoignition

based on the theoretical considerations of Semenov is described in Sec. 1.3. An overview

of hydrocarbon oxidation is provided in Sec. 1.3 followed by the description of dimethyl

ether oxidation as the fuel of choice in this work in Sec. 1.3. Subsequently, the theory

of autoignition modes is introduced in Sec. 1.3. Finally, a brief overview of chemical-

kinetic models and experimental facilities for the determination of fuel characteristics is

provided in Sec. 1.3.

In order to allow for a suitable description of the SEC process two different time scales

with regard to the autoignition process will be introduced in the following: i) the

ignition delay time

τidt

and ii) the ignition time relative to the start of the fuel injection

τiri

. The ignition delay time

τidt

is a characteristic time scale, which describes the

time delay between the initial radical formation of a discrete reactive mixture particle

until the onset of autoignition. The ignition time relative to the start of the fuel

injection

τiri

is referred to as the time delay between the start of the fuel injection and

the autoignition of a certain mixture particle. As previously introduced, the SEC is

based on a periodic combustion process aiming for a quasi-simultaneous ignition of

the entire combustor volume, meaning each injected reactive mixture particle ignites

simultaneously (

τiri

const.

). This quasi-simultaneous autoignition is realized through

the precise stratification of the injected fuel profile to achieve a gradient in ignition delay

time

τidt

to compensate for the variation in residence time throughout the injection

duration

tinj

. The ignition delay time is a function of the local temperature

, pressure

and equivalence ratio

. Thus, the ignition time

τiri

can be well controlled by the

proper adjustment of the local equivalence ratio when assuming

and

to remain

constant.

1.3 Shockless Explosion Combustion

restarting cycle

purging process

f lling the tube with a stratif ed mixture

stratif ed fuel-air mixture

burned gas

air-buffer

τidt(φ)



const., τiri=const.

p>p0

quasi-simultaneous autoignition

quasi-simultaneous

autoignition

Figure 1.3: Sketch of the SEC cycle. Adapted from [89].

Figure 1.3 sketches the schematic SEC cycle. Each cycle starts by adding a defined

mixture profile into the air flow (Fig. 1.3a). After a well-defined delay time, the reactive

mixture undergoes a quasi-simultaneous autoignition in a geometrically constraint

volume and induces an increase in pressure in the combustion products. Subsequently,

a pressure wave, which is created at the contact surface of the burned gas and the

previously injected air buffer, travels downstream in the combustor (Fig. 1.3b). At the

acoustically open combustor outlet, the pressure wave is reflected as an expansion wave

and travels upstream in the combustor (Fig. 1.3c). Once the expansion wave reaches

the combustor inlet, the pressure drops below the supply pressure restarting the cycle

(Fig. 1.3d).

When injecting a fuel profile into a convecting air flow at a spatially fixed point (

= 0)

the resulting fuel–air package has locally differing residence times as a function of the

injection time

tinj

(

)throughout the combustor length

(as shown in Fig. 1.4). To clarify,

an example will be given in the following: injecting a constant fuel profile (while ensuring

and

to remain constant), meaning every mixture volume having the same ignition

delay time

τidt

, leads to an ignition at the downstream part of the combustor (

τiri,min

)

inducing an upstream propagating ignition front (Fig. 1.4a). As a result, the combustor

volume does not undergo quasi-simultaneous ignition, which ultimately impedes the

desired aerodynamic confinement. Thus, to achieve a simultaneous ignition of the entire

mixture volume, the equivalence ratio

is varied to allow for a variation in the ignition

delay time

τidt

along the combustor length (Fig. 1.4b). By this, a constant distribution of

the ignition time relative to the start of the fuel injection

τiri

τidt

(

) +

tinj

(

) =

const.

and thus, a simultaneous autoignition is achieved. In other words: each infinitesimal

1 Introduction

τiri=const.



inj

τidt(φ=const.)

τiri

tinj

τidt(φ=const.)

x x

τiri,min

Figure 1.4:

a) ignition time

τiri

(black), ignition delay time

τidt

(green) and injection

time

tinj

(red) for a constant fuel injection and b) for a stratified fuel injection (with

varying equivalence ratio

designed to deliver a simultaneous autoignition of the entire

mixture volume).

fuel–air package undergoes autoignition simultaneously before being affected by the

heat release of a neighboring particle resulting in a quasi-CVC.

Spontaneous Autoignition

Ignition processes, occur in terms of multi-stage or single-stage ignitions. Multi-stage

ignitions are characterized by the significance of radical formation and will be discussed

separately in the context of hydrocarbon oxidation in Sec. 1.3. Generally, flammable

mixtures undergo ignition according to their flammability limits which are dependent

on the spark energy, temperature, pressure and fuel concentration [

]. An ignition

can be achieved by applying an external energy source that provides a localized high

density energy release sufficient enough to increase the temperature of the nearby

gases, and thus, initiate a self-sustaining flame. In the absence of an external source a

reactive mixture becomes flammable at sufficient temperatures. Once the autoignition

temperature is reached, different reaction mechanism are initiated which eventually are

strong enough to ignite the mixture.

In this section, the description of the single-stage ignition phenomenon, referred to as

thermal explosion, is introduced providing a qualitative understanding based on the

analytical consideration by Semenov [

]. In his theory Semenov describes a thermal

explosion as a fast transition between a slow reaction into an explosion governed by the

competing effects of heat release by chemical reactions and heat conduction through

the vessel walls [

]. This instantaneous transition from slow to fast reaction is

observed at high temperatures and is associated with low ignition delay times. The

theory of Semenov provides a good basis for the description of the ideal SEC process.

However, in reality the SEC process is impacted by the occurrence of multi-stage ignition

phenomena which induce premature heat release by slow reactions impeding a perfectly

1.3 Shockless Explosion Combustion

homogeneous autoignition. This will be addressed in Sec. 1.3.

According to Semenov the ignition process in a closed vessel can be modeled by the

energy equation when approximating the chemistry by a one step reaction and assuming

the density of the reactants to remain constant during the initial stage, as follows [

∆U=Q=r−l. (1.1)

Here, ∆

is the change in internal energy (accumulation rate) and

displays the sum

of heat generation (r) and heat loss rate (l). The rate of heat generation

during an

exothermic reaction is defined as

r=vQ0w=vQ0Ze−E

RT (1.2)

with

q/N

representing the heat generation by a single molecule,

being the heat

release by one gram of the reactants and

= 6

being the Avogadro number

representing the number of particles per mole [

]. The reaction rate

−E

is expressed by the Arrhenius law with the activation energy

, the pre-exponential

factor

and the gas constant

. The volume of the vessel is represented by

. The

heat loss rate can be expressed by the heat which conducts through the vessel walls

l=χA(T−Ta)(1.3)

with the Newtonian heat transfer coefficient

, the surface area

, the temperature

of the reacting gas

and the vessel wall temperature

. According to the energy

conservation law a stationary state arises for ∆U= 0 and thus

vQ0Ze−E

|{z }

rate of heat generation r

=χA(T−Ta)

|{z }

rate of heat loss l

.(1.4)

This relation is also known as thermal criticality, describing the critical value once the

heat released by chemical reactions exceeds the heat conduction through the vessel walls

leading to an explosion [

]. Figure 1.5 visualizes the heat generation rate

and heat

loss rate

as a function of the gas temperature

. From Eq.

(1.4)

it becomes apparent

that the rate of heat generation

increases exponentially with the gas temperature

At the same time

is a function of the gas pressure since the reaction rate is pressure

dependent. The heat loss rate

has a linear dependency (see Eq.

(1.3)

) on the gas

temperature. In Fig. 1.5a the heat release is visualized for a constant wall temperature

and three different gas pressures, resulting in different heat release rates

and

1 Introduction

r, l

T0,0 T0,1 T0,cr T0,2

T0,0 T1,0 T2,0

T1,cr

a) b)

r1r2r3

r, l

l1l2l3

Figure 1.5: Thermal criticality. Adapted from [68].

. Figure 1.5b visualizes the scenario for a constant gas pressure and different wall

temperatures (T0,T1and T2) resulting in different heat loss rates l1,l2and l3.

In Fig. 1.5a

and

intersect at

T0,1

. This temperature is considered a stationary

state point, thus

. Here, first the gain in heat is greater compared to the heat

loss rate and the gas is heated until

is reached, meaning the gas temperature

increases until

T0,1

. Beyond this point the released heat will be conducted away

such that the gas temperature does not increase further hindering the mixture from

autoignition. A second intersection is visible at

T0,2

. This state is not significant

in the theory of autoignition and is associated with an external heating of the gas until

ignition occurs [

]. For

and

a single intersection point at

T0,cr

exists. This

tangential intersection point represents the respective vessel wall temperature

T0,0

and heat release rate

(

)which is considered the minimum required temperature and

pressure to achieve autoignition. Hence, for a greater gas pressure (

) and a constant

wall temperature

T0,0

no intersection is observed, meaning the gas is continuously

heated until finally autoignition occurs.

Figure 1.5b illustrates the same mechanism for differing heat loss terms

and

as a

function of the vessel wall temperature. The observations are equivalent to Fig. 1.5a. A

spontaneous autoignition occurs for

and

represents the critical wall temperature

and no ignition occurs in the case of

, since the heat generated by the reaction is

being conducted through the vessel walls. Hence, the condition for tangency

dT=dL

dT(1.5)

is the condition for the onset of ignition [

]. The analytical consideration of Semenov

provides a good model of the (spontaneous) autoignition or thermal explosion. The

1.3 Shockless Explosion Combustion

onset of autoignition, as a result of exponential heat generation from fast reactions,

causes a “thermal feedback mechanism”, leading to a quasi-instantaneous transition

from slow reactions into an explosion. In hydrocarbon oxidation, however, slow reactions

are predominant and, hence, play a key role in the SEC application. Therefore, in the

following section the ignition behavior of hydrocarbons followed by the DME oxidation

will be introduced.

Hydrocarbon Oxidation

As described in Sec. 1.3, single-stage ignition or thermal explosion is based on a self-

acceleration process in which the heat released by chemical reactions competes with the

heat being conducted through the vessel walls. For purely thermal explosion processes

the temperature increase is quasi-instantaneous. For hydrogen or hydrocarbon oxidation,

thermal explosion takes place after a certain induction time in which radical formation

and chain-branching mechanism are promoted. This induction time or ignition delay

time is exploited in the fuel stratification process. During this induction time different

stages of ignition are observed [

]. These ignition stages are dominated by the

presence of slow reactions resulting in rates of heat release prior to the main ignition,

which have a notable impact on the SEC process.

For hydrocarbons, the onset of single-stage ignition generally occurs under very high

temperature conditions (above 1100 K). The flammability limits of hydrocarbons is

visualized in Fig. 1.6a. At low temperatures chemical reactions requiring high activation

energies do not take place. Thus, the fuel molecules do not disintegrate into reactive

species immediately, but rather undergo a slow auto-oxidation process in which peroxide

molecules and aldehydes are formed first. The energy released during this “cool flame”

formation makes up about 10 to 15 % of the total energy release. A detailed overview

about competing mechanism in the formation of peroxide and aldehyde molecules is

provided in [

]. The peroxide radicals subsequently decay into formaldehyde, which is

characterized by a bluish radiation and thus referred to as “blue flame”. Blue flames are

considered the onset of a multi-stage ignition and are associated with an increased level

of energy release. At sufficiently high temperatures, carbon-monoxide molecules that

are formed in the blue flame which react with the remaining oxygen leading resulting

in a thermal explosion.

Figure 1.6b illustrates the schematic pressure history of a multi-stage ignition as the

consecutive appearance of the described ignition stages. Each ignition stage converts

the reactants up to a certain level. In the first stage, a cool flame is formed which

lasts until the onset of formaldehyde formation, associated with a blue flame. The blue

1 Introduction

thermal explosion

multi-stage

ignition

cool flame

blue flame

thermal explosion

τ1τ2τ3

Figure 1.6: Flammability limits of hydrocarbons. Adapted from [71].

flame, however, lasts until a sufficient temperature in combination with a sufficient

concentration of carbon-monoxide is reached. Thermal explosion occurs as the final

step where the reactants are converted into the final product species.

Three stages during hydrocarbon combustion are observed, cool, blue and hot flames

which depend on the temperature and pressure conditions. In the following section

DME oxidation will be introduced and the underlying kinetics will be discussed with

regard to the occurrence of each ignition stage.

Dimethyl Ether Oxidation

In this section, first, the advantages of DME as fuel will be highlighted, followed by a

brief introduction of DME oxidation mechanisms. DME, which can be produced carbon-

neutral, has the largest potential for conventional use as an alternative to fossil fuels

[

]. Conventionally, DME is produced in a two step process in which first methanol

is generated from syngas followed by a second dehydration step resulting in DME.

Syngas, however, is composed of primarily hydrogen and carbon-monoxide and can

be generated environmentally friendly using biomass as primary source. A single step

production technology was proposed in the past based on bi-functional catalysts [

The efficient synthesis of DME is still remaining in the scope of research [

DME offers advantages in terms of distribution and transportation compared to other

biofuels since it can be fed into the existing supply line.

In this work, DME was used as fuel due to its relatively low ignition delay times

(35 ms<

τidt

<80 ms) at high temperature (

= 1023 K) and atmospheric pressure con-

ditions, which allowed for the observation of autoignition using the “single tube SEC”

1.3 Shockless Explosion Combustion

CH3O+CH3

. .

CH3OCH3-H

CH2O + CH3



scission



scission

CH3OCH2

O2CH2OCH2O2H

.HO2CH2OCHOOH

OCH2OCHO + OH

OH + HO2CH2OCHO



scission



scission

CH2O+CH2O+OH

CH3OCH2O2

CH2OCH2O2H

unimolecular

decomposition



scission

CH2O + HCO2

HOCHO + H

HOCH2O + CO

HOCH2OCO

CH2OH +CO2



scission



scission

LTC

ITC

HTC

Figure 1.7:

Low (A, B and C), intermediate (A and B) and high temperature (A)

oxidation of dimethyl ether. Adapted from Curran et al. [16].

setup which will be introduced in Ch. 2. Generally, like all hydrocarbons, DME exhibits

different ignition stages dependent on the temperature range as previously discussed in

Sec. 1.3. In the literature, these are broadly classified as low, intermediate and high

temperature oxidation. Each temperature range promotes different chemical paths that

can lead to either a single-stage or multi-stage ignition. In Fig. 1.7 the overall reaction

scheme of the DME oxidation is visualized. A detailed analysis of DME kinetics and

oxidation mechanisms can be found in Curran et al. [

] and Fisher et al. [

Here, red (Fig. 1.7A) arrows mark the chemical paths dominant by high temperature

chemistry (above 1100 K), red and green arrows combined represent oxidation taking

place at intermediate temperature ranges (between 800 and 1100 K) (Fig. 1.7A–B)

and during low temperature combustion (below 800 K) each paths (red, green and

black, Fig. 1.7A–C) is being promoted. As visualized in the reaction scheme, the

unimolecular decomposition (dotted red line in Fig. 1.7A) of the DME molecule into

1 Introduction

two radicals CH

3˙

and

(methoxy and methyl) is a reaction taking place solely at

high temperature conditions. This is directly related to the fact that the unimolecular

decomposition is a highly endothermic reaction and thus, requires a very high activation

energy which is provided at sufficiently high temperatures. The high temperature

chemistry of DME is quite clear. Beside the unimolecular decomposition the reaction

is dominated by a

–scission, which is the thermal cracking of hydrocarbon bonds,

resulting in the formation of two radicals, namely

CH2O

and

CH3

(formaldehyde and

methyl radical) [

]. In general, high temperature oxidation of DME is associated with

a rapid heat release rate through the thermal cracking of chemical bonds resulting in a

single-stage ignition. Hence, an immediate onset of autoignition or thermal explosion

without the presence of cool and/or blue flames is being observed.

DME oxidation in the low and intermediate temperature range is dominated by complex

multi-step reactions (Fig. 1.7A, B and C). At low temperatures, a direct cracking of

chemical bonds is not possible due to the high activation energy required for this process.

Instead, first the molecules are broken down into radicals subsequent to the addition

of molecular oxygen and intramolecular H-isomerization. While chemical reactions

proceed, the temperature rises due to heat release and once a sufficient temperature

level is reached

–scission leads to the formation of two stable molecules of formaldehyde

and a hydroxyl radical (OH-). These stable molecules cause a stagnation in the overall

reactivity in the temperature range of 600 to 725 K. This range is known as the negative

temperature coefficient (NTC) area, where an increase in temperature leads to decrease

in reactivity and increased ignition delay times. This NTC behavior is characteristic

for hydrocarbons. Generally, at the intermediate and low temperature ranges DME

exhibits two-stage ignition behavior characterized by an initial ignition dominated

by low temperature chemistry promoting the formation of formaldehyde molecules

as discussed in Sec. 1.3. Shortly after a second ignition follows primarily driven by

high temperature chemistry. The time delay, between both stages, decreases at low

temperature ranges [

]. Besides, single-stage and multi-stage ignition behavior, a

third mixed state has been observed in recent studies [

]. This mixed state is

characterized by the simultaneous existence of hot and low temperature chemistry,

initiating a “hot” and a “cool” flame simultaneous, called “double flame”. The “double

flame” evolves in the initiation of a hot flame within the reactive products (formaldehyde)

which are remaining from an initial propagating cool flame. Both flames eventually

merge since the hot flame propagates at a much higher propagation speed compared to

the cool flame.

The reactivity of DME is predominated by the chemical paths promoted in the different

temperature ranges. Within the NTC range, an increase in temperature leads to a

1.3 Shockless Explosion Combustion

decrease in reactivity. Gradients in reactivity originating from local deviations in the

initial state of a mixture greatly impact the occurrence of different modes of autoignition.

The theory of autoignition modes will be discussed in the following section.

Modes of Autoignition

In this section, different modes of autoignition are described based on the theory

of Zel’dovich [

]. As discussed in Sec. 1.3, the SEC aims for a quasi-simultaneous

autoignition of a reactive mixture profile. This is achieved by the injection of a defined

fuel profile creating a “global” gradient in reactivity throughout the combustor. A

perfectly homogeneous autoignition as considered for the “ideal” SEC process, is a

theoretical consideration. In practical devices, however, infinitesimal perturbations

are unavoidable causing undesired deviations in “local” reactivity, which prevent the

exact adjustment of the ignition delay time. In the scope of this work, the theory

of Zel’dovich is important in two ways: i) to understand the impact of the desired

fuel stratification on the autoignition mode and ii) to understand the impact of local

perturbations causing undesired deviations in reactivity leading to stochasticity of the

autoignition process.

In his work Zel’dovich [

] describes different propagation modes that evolve in the

presence of fluctuations in reactivity resulting in deviations in the local ignition delay

time

τidt

. Based on his calculations Zel’dovich defined different regimes of autoignition

modes as a function of the propagation velocity of the autoignition front

uai

originating

from an exothermic center (region with increased reactivity) with a local gradient in

temperature ∂T/∂x:

uai = ∂τidt

∂x !−1

= ∂τidt

∂T

∂x !−1

.(1.6)

He classifies these modes into four types: deflagration or subsonic autoignition, formation

of a detonation front, supersonic autoignition and thermal explosion. According to

Zel’dovich the critical temperature gradient for the formation of a detonation can be

determined once the propagation velocity of the autoignition is equal to the local speed

of sound ∂T

∂x !cr

=a ∂τidt

∂T !.(1.7)

For fuels exhibiting an NTC behavior, and a complex low-temperature chemistry, as

discussed in Sec. 1.3, Eq.

(1.7)

has two solutions donating the upper and lower bounds

of the detonation limits. The impact of the initial temperature on the development

of a detonation for DME is studied in more detail by Dai et al. [

]. Gu et al. [

]

1 Introduction

introduced two dimensionless parameters to quantify different modes of autoignition.

The normalized temperature ξis defined as

ξ=a

uai

=a∂τidt

∂T

∂x (1.8)

with the speed of sound

and the temperature

. The non-dimensional time

introduced as a second parameter:

ε=x

aτe

,(1.9)

representing the ratio between the acoustic time

x/a

and the excitation time

τe

. The

excitation time is defined as the time delay between 5 % and maximum heat release

rate [

]. Different studies [

] found that detonations occur at certain values for

and

, which allow for the determination of upper and lower values for the detonation

development regime

ξu

and

ξl

. Based on these limits different modes of autoignition

can be formulated: for

ξ≥ξu

a deflagration or a subsonic autoignition is observed that

is characterized by a propagation velocity lower than the speed of sound

. A coupling

between the reaction front and the pressure wave, potentially leading to a detonation

through the process of deflagration-to-detonation transition occurs within the bounds of

ξu> ξ ≥ξl

. A supersonic autoignitive flame is initiated for values between 0

< ξ < ξl

And finally a thermal explosion which is a theoretical consideration, corresponds to a

perfectly simultaneous autoignition (

= 0). Outside of the NTC region the gradient in

reactivity can be expressed as a function of the equivalence ratio

. Since the SEC is

based on the gradient in reactivity achieved by the variation of the equivalence ratio,

Eq. (1.6) can be written as:

uai = ∂τidt

∂x !−1

= ∂τidt

∂ϕ

∂x!−1

.(1.10)

A perfectly simultaneous autoignition in which the entire mixture ignites simultaneously

(

= 0) is not feasible when it comes to the technical implementation of the SEC. In

practice rather a quasi-simultaneous autoignition can be achieved instead. Thus, it is

aimed for multiple consecutively or chaotically quasi-simultaneous autoignition kernels

that are spatially distributed causing an aerodynamic confinement inside the combustor.

Hence, combustion products are prevented from expansion resulting in an increase

in pressure. The theory of Zel’dovich provides a good basis for understanding the

importance of fluctuations in reactive systems. Also, it shows that a proper adjustment

of the equivalence ratio provides a solid tool for the control of autoignition processes

as implemented in this work. However, these modes of autoignition are very much

1.3 Shockless Explosion Combustion

dependent on the chemical structure of the utilized fuel [18, 27, 12, 62].

Chemical-Kinetics Models

Chemical-kinetic models are routinely used to model the chemical decomposition of

reactants into products at an elementary level. In the recent years these models greatly

contributed to the prediction of the combustion behavior of different fuels. A review of

chemical-kinetic models for the oxidation of hydrocarbons is provided in [

]. In this

work, a detailed chemical-kinetic model has been used for ignition delay time calculations

of DME with Cantera [

] for a zero-dimensional constant volume reactor using the

AramchoMech2.0 mechanism [

]. The applied mechanism covers the chemistry of C

to C

compounds as well as oxygenated fuels including DME. This mechanism was

further developed by Zhou et al. [

] by including a series of new sub-mechanisms

into AramchoMech3.0. Different chemical-kinetic models, including both AramcoMech

mechanism, were compared to validate experimental data obtained from shock tube

measurements with DME in [

]. Both mechanism show generally good agreement

with the experimental data. Hence, to reduce simulation time, for ignition delay time

calculations, AramcoMech2.0 was used in the scope of this work.

These chemical-kinetic models are validated by experimental data obtained in different

facilities for conducting kinetic studies. In the past several facilities have been developed,

optimized and implemented including shock tubes [

], continuous stirred (flow) tank

reactors (CSTR) [

] and rapid compression machines (RCMs) [

], which will be

introduced in the following.

CSTRs chemical reactors which are commonly used to investigate kinetics in the gas-

phase during thermal decomposition of a fuel. Here, gas chromatography and mass

spectroscopy is used to monitor the species evolution during combustion [35].

RCMs represent a single stroke of a diesel engine and are used to investigate fuels under

engine-like operating conditions. A homogeneous mixture is injected and subsequently

compressed leading to a rapid increase in temperature until ignition occurs [

RCMs are mainly used to investigate the ignition delay time of mixtures under differing

temperature and pressure conditions. They provide geometric requirements for the ap-

plication of additional diagnostics to measure flow fields and concentration/temperature

distribution inside the combustion chamber.

Shock tubes are facilities enabling very precise measurements of ignition delay times

at high temperature and pressure conditions. A simple shock tube consists of a long

tube with two sections namely driver and driven section separated by a diaphragm [

1 Introduction

A pressure difference is created between both sections until the diaphragm comes to

a burst. Hence, a shock wave is initiated compressing the test gas provided in the

driven section. The precise control of initial conditions allows for reproducing quasi

one-dimensionality in addition to a very accurate measurement of the shock velocity.

Once the shock wave is reflected at the end wall, the pressure and temperature behind

the reflected shock wave can be determined by one-dimensional calculations.

In this chapter different theoretical aspects have been intensively discussed to provide a

solid basis for the investigations conducted in this work. In the following chapter the

experimental setup will be introduced.

2 Experimental Setup

In this chapter, the experimental setup for investigating the SEC process is introduced

and an overview of its step-by-step evolution is provided. Two different test rig designs

were investigated in this work. The first version of the SEC test rig was designed

by Bobusch et al. [

] and subsequently modified by Yücel et al. [

] to increase the

reproducibility of the autoignition process. This first setup was used for investigations

in the context of

publication I

. Subsequently, further modification were implemented,

resulting in a second setup used in the context of

publications II–IV

. In the following,

the first setup will be referred to as the “SEC with bypass” and the second setup

will be referred to as the “single tube SEC”. All experiments described in this work

were conducted at the experimental facilities of the Technische Universität Berlin,

Hermann-Föttinger Institut.

2.1 SEC with Bypass

In the following, the first setup referred to as “SEC with bypass” will be introduced.

As sketched in Fig. 2.1, the test rig consists of a preheater, a Y-pipe, a bypass, a fluidic

diode, a combustor and an exhaust tube. Two wastegates, WG1 and WG2, are attached

to the outlets of the combustor and the bypass path, respectively. A continuous air flow

is led through the preheater in order to increase the air temperature up to 923 K, as

measured at the combustor inlet. This ensures the ignition delay time for the applied

DME–air mixture to be approximately

τidt

= 100 ms. Downstream of the preheater,

a Y-pipe is mounted, dividing the air path into a combustor and a bypass path. The

need for a bypass results from the preheater requirement of a continuous air flow with a

minimum mass flow rate of 30 kg/h, which is equivalent to an average air flow velocity

of 18 m/s. A Coriolis mass flow meter in combination with a proportional valve is used

to control the air mass flow rate to a steady-state value during operation. To control

the position of the reactive mixture along the combustor path, the continuous air flow

is switched actively between combustor and bypass by closing/opening the respective

wastegate, ensuring autoignition to take place within the desired section. A quartz tube

2 Experimental Setup

preheater

Y-pipe

combustor exhaust tube

wastegate

(WG1)

wastegate (WG2)

bypass

fluidic diode and injection

Figure 2.1: Sketch of the atmospheric SEC test rig with bypass.

which serves as an optically accessible combustor is mounted between the injection

station and the exhaust tube. When the air flow is directed through the combustor,

fuel is added to the continuous air flow at the position of the fluidic diode, using two

valve arrays. Each array consists of five parallel-connected highspeed solenoid valves

which will be further described in Sec. 2.2. Both arrays are connected to an annulus

through four tubes feeding ten fluidic oscillators which guide the injected fuel into

the combustor. These fluidic oscillators were designed in a previous study [

] to

enhance the mixing quality of the fuel–air mixture.

At the beginning of each cycle, a predefined fuel profile is added to the air flow. This

fuel profile is defined by adjusting the number of open valves between 0 and 10 to

achieve the desired stratification throughout the injection duration. Subsequent to

injection, WG1 is closed while WG2 is opened simultaneously, allowing for the air flow

to switch from the combustor into the bypass path. This position is held until ignition

occurs. Once the flammable mixture undergoes autoignition, the ignition times along

the combustor are measured by five axially distributed photomultiplier sensors. After

the ignition process, the air flow is switched to the combustor path to purge the exhaust

gases and restart the cycle.

This setup allows for the investigation of the autoignition processes under atmospheric

pressure and elevated temperature conditions. Due to the switching feature, this test

rig design allows for the observation of autoignition at increased ignition delay times by

simply adjusting the switching times of the wastegates. However, increased ignition

delay times impede the repeatability of the ignition process since the reactive mixture

is exposed to turbulence promoting stochasticity. Furthermore, the test rig provides

a limited number of applicable diagnostics due to the absence of sensor ports in the

combustor section. Thus, PMTs are used to evaluate the autoignition process in terms

2.2 Evolution of the SEC Test Rig

of ignition homogeneity. To fully understand the SEC process, further experimental

data is required. This includes pressure or chemiluminescence data to allow for the

correlation of heat release, pressure rise and kinetics, which play a key role in the

SEC process (see Sec. 1.3). Therefore, a second test rig was designed as a result of a

step-by-step modification of the setup presented in this section.

2.2 Evolution of the SEC Test Rig

Autoignition processes are very sensitive to any source of disturbances. Therefore, one

major challenge is to provide reliable and repeatable operation. In the course of this

study, the SEC with bypass has been associated with several disadvantages leading

to high stochasticity in the autoignition process, thus hindering a sufficient operation.

Major disadvantages were found to be originating from the low frequency operation of

the wastegates (0.5 Hz), the heat loss caused by the Y-pipe, and the preservation of the

injected mixture profile until the onset of ignition. In the following, these points will be

further discussed, followed by the modifications implemented.

•

Towards the implementation of the SEC in practical approaches, the precise control

of the fuel distribution inside the combustor poses a central challenge. A defined

mixture profile is created by the injection of a prescribed fuel trajectory into a

continuous air flow. Subsequent to injection, this mixture profile is exposed to

turbulent fluctuations until ignition occurs. During this time frame, a smoothing of

the gradients along the injected concentration profile is promoted. This smoothing

effect increases with an increasing ignition delay time which is linked to the time

duration in which the mixture is exposed to turbulent fluctuations. Therefore,

low ignition delay times are favorable to reduce diffusion and enable a sufficient

control of the fuel distribution inside the combustor. Since the ignition delay time

for DME, outside its NTC region, is decreasing with increasing temperature, high

preheat temperatures inside the combustor are necessary. One major drawback of

the SEC with bypass is the heat loss caused by the number of components (e.g.

Y-pipe, fluidic diode, partially bypass) mounted between the preheater and the

combustor section impeding high temperature conditions at the combustor, thus,

promoting increased ignition delay times.

•

Another disadvantage with regard to the fuel injection was found to originate from

the annulus design. For short injection durations, i.e. high operating frequencies,

the annulus volume acts as a damper, impeding the precise injection of a stratified

mixture. Instead, the fuel accumulates in the gap before entering the combustor,

2 Experimental Setup

inlet valve

valves

inlet

outlet

d1d2

a) b)

Figure 2.2:

Valve array with mounting block and five high-speed solenoid valves con-

nected in parallel and connectors at inlet (red) and outlet (blue) a). Sketch of the cross

section of the mounting block with one example valve mounted b).

causing a blurring of the injection profile.

•

Fuel concentration measurements have shown that the design of the valve arrays,

which are used in order to connect multiple high-speed valves in parallel (as shown

in Fig. 2.2), cause variations in the convection time depending on the position

of the operating valve (

and

). Hence, the convection time of the fuel is a

function of the respective valve position leading to a variation in the order of 1 ms

to 2 ms depending on the fuel mass flow rate. These variations lead to notable

deviations of the fuel concentration distribution from the injected fuel trajectory.

•

As discussed earlier, the applied wastegates provide a suitable tool for controlling

the air flow path between the combustor and the bypass. However, they become

a limiting factor when it comes to high frequency operation of the test rig. The

operation frequency is limited to a maximum of 0.5 Hz. Furthermore, increased

opening and closing times between 17 and 31 ms lead to a great variation in the

cross section area of the combustor and bypass outlet, respectively. Thus, during

wastegate opening/closing the flow inside the combustor is accelerated/decelerated.

Consequently, this uncontrolled mixing of the combustor volume affects the in-

jected fuel profile and impedes a precise control of the equivalence ratio throughout

the combustor.

•

The quartz combustor applied between the injection station does not allow

for sensor instrumentation and, thus, impedes the simultaneous monitoring of

additional parameters by applying pressure or ionization sensors.

To overcome the challenges listed above, a new “single tube SEC” test rig, without

a bypass has been designed, leaving a “single tube” as the SEC setup. This design

significantly reduces heat loss between the preheater and the combustor. Furthermore,

the distance and number of components is reduced by avoiding the Y-pipe and bypass,

2.2 Evolution of the SEC Test Rig

DME

preheated

air

high-speed

solenoid valve

mounting

block

convection

tube

injection port

Figure 2.3:

Detailed view of the injection station with 10 circumferentially distributed

ports each equipped with one high-speed solenoid valve.

and replacing the fluidic diode with a simple restriction. As a result, the fuel–air

mixture is exposed to higher inlet temperatures, causing decreased ignition delay times.

Consequently, the advection time between the fuel injection and the combustor section

is sufficient to ensure autoignition in the desired section. Hence, no stalling is required

and the bypass path becomes redundant. Also, the decrease in ignition delay time

ultimately leads to a reduced time duration in which the fuel profile is exposed to

turbulent diffusion. Without the application of wastegates, the firing frequency has been

increased up to 5 Hz. A higher firing frequency enables a more reliable and repeatable

operation since the combustor wall temperature reaches a steady-state condition within

a short time duration, providing identical initial conditions for each operating cycle.

In a next step, the injection geometry was modified in order to improve the fuel

stratification process. Three design aspects were improved: i) avoiding an annular gap

between the injection ports and the combustor, ii) ensuring constant convection times

for each operating valve and iii) implementing a dome-loaded pressure regulator to

reduce the pressure drop during injection. Based on this points, a new injection station

2 Experimental Setup

pressure

regulator vaporizer

fuel valve

optical accessible combustor

1234

preheater restriction convection

tube

injection station

P1 P2 P3 P4

I2 I3 I4 I5 I6 I7

combustor

exhaust tube

I1 I2 I3 I4 I5

0.7

01.2

x (m)

(ms)

039 67

uair=18 m/s

Figure 2.4:

Sketch of the atmospheric test rig equipped with ionization probes (I1–I7),

high frequency pressure transducers (P1–P4), thermocouples (T1–T2) and low frequency

pressure transducers (FF and FA). A second exchangeable combustor with optical access

is sketched in b). The connection of convection and ignition delay time is visualized in

c).

was designed, as shown in Fig. 2.3, which will be further describes in the following

section.

Furthermore, the previously applied quartz combustor was replaced by two exchangeable

combustor designs. The first design consists of a stainless-steel tube, which is used to

further decrease the heat loss along the combustor section. Multiple sensor ports enable

simultaneous measurements with pressure and ionization probes to obtain pressure rise

and ignition homogeneity.

A second combustor was designed for the integration into the single tube SEC test

rig to provide optical access in addition to the existing sensor ports. This design was

used for the detailed analysis of single ignition cycles. All these modifications highly

contributed towards an improved performance of the SEC test rig paving the way to a

systematic investigation of the SEC process.

2.3 Single Tube SEC

In the following, the “single tube SEC” setup used in the context of

publications II–

will be discussed. The setup consists of a preheater, an injection station and three

sections, namely, convection, combustion and exhaust section, as sketched in Fig. 2.4.

Each section has an inner diameter of 40 mm. The convection section consisting of

injection tubes and the convection tube has a total length of 700 mm. The combustor

section and the exhaust section have a length of 500 mm and 1200 mm, respectively.

A restriction of the cross section area is applied downstream of the preheater in order

2.3 Single Tube SEC

to choke the air flow. Hence, hot exhaust gases and pressure waves subsequent to

ignition are hindered from reaching the preheater. The injection station is designed

to deliver a prescribed charge of ignitable mixture within a given injection time. For

this, 10 highspeed solenoid valves (Staiger VA 204-716) with a maximum operating

frequency of 250 Hz are mounted on 10 circumferentially distributed ports. Each port is

connected to an individual injection tube with an inner diameter of 1 mm and a length

of 0.2 m. A high-speed dome-loaded pressure regulator (Swagelok RD6) is mounted

into the fuel line to minimize pressure fluctuations during the fuel injection. To ensure

gaseous injection, the fuel is vaporized and guided through a heated line (

Tfuel

=330 K)

before injection. A Coriolis mass flow meter is used to determine the average fuel

mass flow under steady-state conditions and subsequently correlate with the number

of simultaneously operated valves for calibration. Two type K thermocouples enable

temperature measurements upstream of the injection station (T1) and in the exhaust

tube (T2). The modular combustor setup allows for exchanging the stainless steel

design for an optical accessible design (see Fig. 2.4b), which is assembled from a series

of four quartz tubes, each 120 mm long, supported by stainless steel flanges fitted with

one pressure sensor and ionization probe each.

This single tube SEC setup enables the observation of autoignition of a prescribed fuel

profile, injected into a continuous air flow, during convection. For this, the ignition delay

times and the length of the convection and combustor section need to be matched. It is

important to remember that low ignition delay times are favorable in order to minimize

diffusion during convection, and thus, enable a proper control of the fuel distribution

inside the combustor. For this, a preheater is applied to rise the air temperature

up to 1023 K, which corresponds to the maximum allowable preheater temperature,

measured at T1. When conducting combustion experiments, the air mass flow rate is

set to a constant value of

˙mair

=30 kg/h, resulting in a bulk velocity of

¯uair

= 18 m/s.

Non-reactive fuel concentration measurements, which will be described in Sec. 6.1,

were conducted at low preheat temperatures at varying mass flow rates between 30 to

100 kg/h to ensure a constant bulk velocity at different temperature and, thus, different

density ranges. As previously applied, a Coriolis mass flow meter in combination with

a proportional valve was used to control the air mass flow rate during operation.

Numerical calculations were conducted to assess the ignition delay times

τidt

for the

given boundary conditions (

Tair

= 1023 K and atmospheric pressure,

ϕ∈

8]) in

order to determine the required length of the convection tube. Figure 2.5 shows the

operation map of the test rig. The simulations were conducted with Cantera [

] for a

zero-dimensional constant volume reactor using the AramchoMech2.0 mechanism that

has been validated for DME kinetics as discussed in Sec. 1.3. The air temperature is

2 Experimental Setup

τ (s)

mix (K)

800 850 900 950 1000 1050 1100

750

φ=0.6

φ=0.8

φ=1.0

φ=1.2

φ=1.4

φ=1.6

φ=1.8

10-2

10-1

39ms

67ms

fuel=330K

air=1000K

air=950K

air=1050K

air=1100K

air=1150K

air=900K

air=1023K

Figure 2.5:

Calculated ignition delay times for a DME–air mixture at atmospheric

pressure, a constant fuel temperature of 330 K, a varying air temperature and varying

equivalence ratios using Cantera [23].

varied from

Tair

= 850 K up to 1300 K and the equivalence ratio is varied from

= 0

up to 1.8. The mixture temperature

Tmix

is calculated for a DME temperature of 330 K

and varying equivalence ratios, according to:

Tmix =cp,air ˙mairTair +cp,DME ˙mDMETDME

cp,air ˙mair +cp,DME ˙mDME

(2.1)

with

donating the specific heat capacity and

˙m

representing the respective mass flow

rate. Adding the cool fuel to the preheated air flow results in a decrease in the overall

mixture temperature. With increasing equivalence ratio the added fuel mass flow rate

increases causing a lower mixture temperature.

The mixture temperature calculated using Eq.

(2.1)

for the given air and fuel tempera-

2.3 Single Tube SEC

tures of

Tair

= 1023 K and

Tfuel

= 330 K ranges from 810 K up to 930 K depending on the

equivalence ratio (see Fig. 2.5). The simulations reveal ignition delay times in the order

of 90 to 140 ms at atmospheric pressure and the respective mixture temperature (see

Fig. 2.5 point

). For a given bulk velocity (

¯uair

= 18 m/s), the convection distance

xconv of each injected particle can be calculated according to

xconv =τidt ¯uair.(2.2)

To keep the length of the convection tube reasonably short, the lower limit (90 to

100 ms) of the ignition delay time is considered, meaning operation at high equivalence

ratios is required (see Fig. 2.5). The convection distance for each injected particle within

100 ms, determined using Eq.

(2.2)

, is 1.8 m. Hence, in order to observe autoignition

within the measurement section, a convection section with a length of approximately

1.1 m is required, when accounting for the combustor length of 0.5 m.

However, when conducting combustion experiments, the length of the convection section

has been varied from 0.7 to 1.2 m. In the experiments a convection length of 0.7 m

turned out to be sufficient. This was verified by measurements using ionization probes

and optical measurements, which indicated actual ignition times to be in the order

of 40 ms to 75 ms. Therefore, it can be assumed that the actual mixing temperature

Tmix

is higher than expected from the calculations. This can be explained as follows:

after the mixing of fuel and air at the injection station, the reactive mixture convects

downstream in the combustor until the onset of ignition. During this convection time,

the mixture is being heated continuously due to heat conduction from the hot combustor

walls (

Twall

= 1023

). This leads to an overall higher mixture temperature, assumingly

between 940 to 1023 K depending on the local equivalence ratio, as indicated in Fig. 2.5.

Hence, lower ignition delay times are observed in the experiments than expected from

the calculations. Meaning, the operation range of the test rig is shifted as indicated

by the demarcation

in Fig. 2.5. It is important to note that the location of the

autoignition strongly correlates with the local equivalence ratio, rather than the global

equivalence ratio.

The simulation results further reveal that low mixture temperatures (

Tmix

< 850 K)

lead to a decrease in ignition delay time. As discussed in Sec. 1.3 this is caused by the

NTC behavior of the applied fuel. This effect is more prominent for high equivalence

ratios ϕ > 1.2.

Four operation areas are marked in Fig. 2.5 of which two have been described above

(

and

). The operation range

is favorable in terms of low preheat temperatures

along with a lower temperature sensitivity of the ignition delay times. On the other hand,

2 Experimental Setup

the resulting ignition delay times are high, which would require a longer convection tube

in order to observe autoignition in the combustor at atmospheric pressure conditions.

Furthermore, an increase in convection time, as a result of extended ignition delay times,

is undesirable since turbulent diffusion highly affects the preservation of the injected

mixture profile.

Operating point

is characterized by very low ignition delay times, enabling high

frequency operation. By this quasi-steady inlet conditions are provided for a downstream

turbine, once integrated into a gas turbine cycle. Additionally, low ignition delay times

offer an improved control of the injection profile due to very short convection times.

However, in order to operate at

, very high mixing temperatures are required. An

alternative is to operate at higher pressure conditions, which is outside the scope of

this work but will be further discussed in the outlook provided in Ch. 8.

3 Methodology

In this chapter, measurement techniques applied in the scope of this work will be

introduced. Different methods have been applied for measurements with and without

the presence of chemical reaction. First, measurement techniques are introduced which

have been applied for fuel concentration and flow velocity measurements without

the presence of a flame. Next, measurement tools are described which are used for

autoignition diagnostics.

3.1 Measurements without Chemical Reaction

Concentration measurements are a vital part of this work since controlling the fuel

distribution inside the combustor is of great importance for achieving a reliable and

repeatable quasi-simultaneous autoignition. Hence, different methods have been applied

and optimized in the course of this study. Two measurement techniques namely i) tunable

diode laser absorption spectroscopy (TDLAS) and ii) acetone planar laser induced

fluorescence (acetone-PLIF) have proven well-suited for concentration measurements

in the “single tube SEC” and were applied in the scope of

publications II-IV

. Laser

Doppler Anemometry (LDA) was applied in

publication I

for measuring the flow

velocity inside the combustor using the “SEC with bypass”. All three methods provide a

non-intrusive way of measuring the fuel concentration/flow velocity inside the combustor

and were applied under non-reacting conditions.

Laser Doppler Anemometry

In the scope of

publication I

, an LDA system from Dantec Dynamics was applied for

measuring the flow velocity inside the combustor. The main incentive was to determine

the opening and closing times of the applied wastegates (see Ch. 2). The measured

parameters were subsequently used under reacting conditions enabling a sufficient

control of the test rig.

3 Methodology

Laser

Beamsplitter module Front lens

Measuring volume

Forward

scatter

optics Photo-detector

Measuring volume

with interference fringe pattern

Expample signal

Figure 3.1:

Sketch of the experimental setup for laser Doppler anemometry. Adapted

from [75].

LDA is a non-intrusive measurement technique for velocity measurements based on the

principles of the Doppler shift. The Doppler shift describes the change in frequency of

an interfering laser signal scattered by an object moving relative to the source and is a

function of the moving object’s velocity and its geometry [2].

Figure 3.1 shows a basic arrangement of the LDA system. The laser beam of a continuous

wave laser is splitted into two equal laser beams using multiple beam splitters. Both

beam parts are directed through a front lens with a diameter

converging to a focus

with a focal distance

within the measurement volume. In this work, an aerosol of

liquid seeding (di-ethyl-hexyl-sebacate) is added to the air flow which is previously

atomized using an aerosol-generator. The scattered light is focused by the scatter optics

and received by the photo-detector. The light interference between the two laser beams

form a fringe pattern with a distinct fringe spacing

which can be calculated according

to:

df=λ

2 sin(θ/2) (3.1)

where

is the wavelength of the incident laser light and

is the intersection angle

between the two laser beams [

]. Once a particle passes through the measurement

volume, the light is scattered with a Doppler shift resulting in a modulation at a

frequency fproportional to the velocity uof the moving particle

f=u2 sin(θ/2)

λ.(3.2)

The photo-diode receives a fluctuating light intensity which is generated due to light

3.1 Measurements without Chemical Reaction

absorption by the particles. This light emission is converted into an electrical signal

which is subsequently correlated with the Doppler frequency. With the known Doppler

shift (temporal parameter) and the fringe spacing (spatial parameter) the velocity of

the moving particle can be calculated [19].

Tunable Diode Laser Absorption Spectroscopy

In this work, TDLAS is applied as proposed by Li et al. [

] and Blümner et al. [

The TDLAS setup available at the Hermann-Föttinger-Institut and applied in this work,

was developed for measurements of fuel flow fluctuations in gas turbine combustors [

This technique enables a precise real-time measurement of the fuel concentration inside

the combustor and was used in

publications II-IV

. These measurements serve as

preliminary investigations that were conducted to quantify the capability of the injection

strategy to inject a desired fuel profile into a continuous air flow within a defined time

frame. Furthermore, the measured concentration profiles were used to quantify the

effect of turbulent diffusion on the gradients in fuel concentration to determine the

grade of preservation of the injection profile.

Tunable diode lasers are available at various wavelengths and modulation frequencies of

several hundred kilohertz. Therefore, they serve as an ideal tool for rapid, non-intrusive

concentration measurements. At near-infrared wavelengths, they can be used for the

detection of different species such as methane or water vapor [

]. The concept of

absorption spectroscopy is based on the Beer-Lambert law which describes the change

in intensity occurring when a monochromatic light passes through a volume containing

a medium with absorbing characteristic as visualized in Fig. 3.2. The light source is a

fixed-wavelength laser with an output intensity

(

)which is a function of the laser

frequency

(

)is the transmitted laser signal after passing through an absorbing

volume with a length of

(assuming the quartz to have an absorption coefficient of

zero). After the emitted laser signal passes through the measurement volume, it is

absorbed by a fixed-wavelength absorption sensor. The transmission coefficient

(

)

and the absorption coefficient α(ν)can thus be calculated according to [44]:

γ(ν) = It

I0ν

= e−α(ν)L.(3.3)

The absorption coefficient of a medium is a function of the total gas pressure

, the

mole fraction of the absorbing species

χi

, the line strength

, and line shape Φ

of the

absorption feature determined from an appropriate line shape mechanism (depending

3 Methodology

I0()It()

quartz glass

combustor

Figure 3.2: Beer-Lambert law

on the operating conditions) such as Gaussian, Lorentzian or Voigt profile [

]. For

optically thin samples (absorbance <0.05) [59], the spectral absorption coefficient can

be expressed as

α(ν) = pχiLX

GjΦj.(3.4)

The line shape for various species can be found in the HITRAN data base [61].

The laser used in this work has a wavelength of 1,65

µm

which matches well with the

absorption features of methane. Thus, all concentration measurements were conducted

using methane as the tracer fuel. Wavelength modulation spectroscopy (WMS) was

applied to increase the signal-to-noise ratio of the measurements. This technique allows

for the modulation of the laser signal within the absorption spectrum of the tracer gas.

The experiments presented in this work used a rapid sinusoid at a modulation frequency

of ω=10 kHz. The modulation frequency can be expressed as:

ν(t) = ¯ν+acos(ωt)(3.5)

where

¯ν

is the center laser frequency and

is the modulation depth. Frequency modu-

lation goes along with a simultaneous intensity modulation. This intensity modulation,

however, is linear for small modulation amplitudes. Therefore the resulting laser

intensity I0can be expressed as

I0(ν) = ¯

I0[1 + i0cos(ωt +ψ)] (3.6)

with the average laser intensity

, the normalized linear intensity modulation ampli-

tude

and the phase shift

. The non-linear absorption features of the tracer gas

3.1 Measurements without Chemical Reaction

methane and the modulation frequency result in a detector signal containing higher

harmonics Hk[60], yielding

γ(¯ν+acos(ωt)) =

∞

k=0

Hk(¯ν, a) cos(kωt).(3.7)

The detector signal is processed by a lock-in amplifier, which allows for the extraction

of the first and second harmonic in noisy environments [

]. The second harmonic is

a function of the spectral parameters and gas properties and sensitive to the absorption

line shape curvature [

]. For diode lasers, the first harmonic of the measurement

signal is mainly dependent on the laser intensity modulation and can thus be used to

normalize the second harmonic signal in order to account for perturbation in the laser

signal, e.g. laser drift [

]. The measured species concentration scales linearly with the

normalized second harmonic extracted from the detector signal.

Calibration measurements were conducted prior to each measurement to determine

the proportionality coefficient by measuring the detector output signal for different

well-known steady state fuel mass flow rates using a Coriolis mass flow meter. The

calibration data was subsequently used to quantify the fuel mass flow rate obtained at

high frequency operation.

Acetone Planar Laser Induced Fluorescence

The acetone-planar laser induced fluorescence (acetone-PLIF) setup available at the

Hermann-Föttinger-Institut and applied in this work was developed by Alexander

Jaeschke in the scope of his master thesis [

]. Planar laser induced fluorescence is

a widely used non-intrusive laser diagnostic technique for flow visualization and is

generally applied for spatially-resolved measurements of velocity [

], temperature [

]

and/or concentration [47, 82] distribution.

In the scope of this work, acetone-PLIF was applied to the single tube SEC (see

Ch. 2) to obtain a two-dimensional visualization of the fuel concentration distribution

inside the combustor. Compared to the previously introduced one-dimensional TDLAS

measurements, additional features of the stratification process were identified, which

can only be accounted for as an integrated value from TDLAS measurements. To

assure comparability of the results obtained applying TDLAS and acetone-PLIF, all

measurements were conducted under the same boundary conditions using methane as

tracer fuel.

Acetone-PLIF is a spectroscopic method which utilizes a laser for the excitation of

3 Methodology

acetone molecules to higher energetic levels. Once excited, a spontaneous de-excitation

of the molecules follows along with the emission of light. The emitted light can be

characterized according to its electromagnetic features by its wavelength

and frequency

νwhich are linked through the speed of light constant cas follows [28]:

ν=c

λ.(3.8)

Once light is being absorbed by matter, energy is transferred to the molecule. This

energy

is called quantum and can be expressed by the Planck’s constant

and is

inversely proportional to the wavelength:

E=hν =hc

λ.(3.9)

Figure 3.3 illustrates a schematic energy level diagram for a dielectric atom. Every

molecule is characterized by multiple energy levels, but only specific molecules are

capable of changing their energy level by the absorption of light. Thus, once de-exciting,

these molecules emit light in the form of fluorescence or phosphorescence. Each main

energy level consists of various vibrational levels (e.g. in Fig. 3.3 as 0, 1, 2, 3, 4). While

a molecule is at a higher energetic state, it dissipates vibrational energy by collision

resulting in a lower vibrational level.

Potential energy

Interatomic distance along

critical coordinate



Absorption

Fluorescence

Phosphorescence



Loss of

vibrational

energy by

collision

intersystem

crossing

Ground state

singlet

state

triplet

state

Figure 3.3: Schematic energy level diagram. Adapted from [28].

Phosphorescence occurs as a result of intersystem crossings in case molecules are de-

excited from a different energetic state, called triplet state (see Fig. 3.3). This state

3.1 Measurements without Chemical Reaction

is reached solely in case the spin of the electrons is reversed and does not involve the

process of radiation. The relaxation into ground state from the triplet state is again

associated with an electronic spin. Thus, the return into ground state takes more time

and is delayed which typically results in an afterglow. Additionally, phosphorescence is

characterized by a more pronounced shift in wavelength due the higher dissipation rate

compared to fluorescence.

One source of error when applying fluorescence-based measurement techniques is quench-

ing. Quenching is an intra-molecular deactivation process of molecules causing a decrease

in the obtained fluorescence signal. According to [

] four common sources promoting

quenching are observed in technical approaches, namely temperature, concentration,

oxygen and impurity. Despite, there are certain species known to create greater quench-

ing like oxygen, iodide and nitrogen oxide, which exhibit increased molecular motion,

and thus, promoting molecule collision at higher temperatures.

Beside quenching, a high concentration of the fluourescent species may lead to a

significant decrease in fluorescence as well. This effect is caused by the increased

absorption from an elevated number of molecules hindering the excitation source from

passing through. Hence, during calibration of the fluorescence signal a linear range of

fluorescence signal and tracer concentration is followed by saturation. In applications it is

essential to operate within the linear range in order to assure quantitative measurements.

Hence, the fluorescence signal

Sfl

obtained in measurements can be calculated according

to [65]

Sfl= (E/hν)V ρflσabsφflηΩ/4π, (3.10)

where

E/hν

is the photon flux,

ρfl

density of the fluorescence tracer in the observed

volume

that have the absorption cross-section

σabs

, the efficiency of the detection

system

, the observed solid angle Ω

and the absorption yield

φfl

. The equation

Eq.

(3.10)

is governed by the absorption cross-section and the fluorescence quantum

yield

Sfl(λ, T)∝φfl(λ, T)σabs(λ, T).(3.11)

The quantum yield is defined as the ratio of photons being absorbed to those being

emitted by a system. Both quantities dependent on temperature conditions and the

excitation laser wavelength which corresponds to the maximum absorption wavelength of

the applied tracer. The quantum yield of acetone was determined by Heicklen et al. [

]

to be around 0.0021 for a temperature of 40

C and a excitation wavelength of 313 nm,

and by Halpern and Ware [

] to be 0.0012 at room temperature for the same excitation

3 Methodology

wavelength. Furthermore, only little dependence on the pressure and relatively little

dependence on the wavelength was found by Heicklen et al. [

]. These values are

relatively small compared to those of aromatic compounds. However, since oxygen is

considered a main source of quenching causing a degeneration of the fluorescence signal,

and as research showed that acetone is only minor impacted by oxygen quenching [

in realistic setups in which fuel is mixed with air, the obtained signal is comparable to

tracers with higher quantum yields. The second value governing Eq. 3.10, the absorption

cross-section, is a measure for the probability of an absorption process to occur. For

acetone at ambient temperature conditions this value is approximately 4.7x10

−20

for a wavelength of 270 nm [65].

For the acetone measurements conducted in this work, a fourth-harmonic pulsed high

energy Nd:YAG laser (dual cavity, heated BBO crystal) operating at a wavelength of

266 nm was applied, resulting in a total absorption cross section of 4.4x10

−20

to the maximum value of acetone. Due to the strong temperature dependence of the

absorption cross section the temperature was kept constant and monitored during the

each measurement. Thus, when assuming pressure, temperature and wavelength to

remain constant, the fluorescence signal is a function of the laser energy and acetone

concentration only. The acetone was used to saturate a defined methane mass flow

rate. For this, a specific seeder was used equipped with a bypass in order to adjust the

acetone/methane ratio to ensure an optimal signal. Different well-known steady state

methane mass flow rates were led through the seeder resulting in different fluorescence

signal intensities. By this, the detected intensity levels can be related to the known

methane concentrations allowing for a semi-quantitative measurement ensuring operation

within the linear range of the acetone fluorescence.

3.2 Measurements with Chemical Reaction

In the scope of this work, autoignition characteristics in terms of ignition homogeneity

and pressure rise subsequent to ignition were quantified by applying pressure, ionization

and optical measurements, which will be introduced in the following sections. Mea-

surements with chemical reaction imply harsh conditions in terms of temperature and

pressure, thus limiting the options of measurement techniques available.

3.2 Measurements with Chemical Reaction

convection section

0.7 1.2

combustor exhaust section

x (m)



iri (ms)

ionization probes

{



idt=39



idt=67



idt=53

start of the

injection

stratified fu

mixture

Figure 3.4:

The connection of ignition time measured by the ionization probes

τiri

, the

position xof the reactive mixture and the ignition delay time τidt.

Ionization Probe Measurements

Ionization probes are commonly used in combustion research to detect the reaction

zone of an autoignition without optical access. Spark igniters have been extensively

used for diagnostic purposes due to low purchasing costs and availability [

]. In this

work, ionization probes developed at the TU Berlin were used for the application in the

SEC combustor. The ignition times

τiri

were measured at multiple axial positions to

obtain the homogeneity of the autoignition event.

The principle is based on two electrodes on which a potential is applied. Once autoigni-

tion takes place, the ions created in the reaction zone are attracted by the electric

field generated by the electrode. After a sufficient concentration of ionized gas near

the electrode is reached, a short-circuit between both electrodes occurs, resulting in

a voltage drop. The current of the probe is a function of the bias voltage, the ion

concentration and the applied potential.

Figure 3.4 visualizes the ignition times measured by the ionization probes

τiri

with

respect to the position

of the autoignition inside the SEC setup. The fuel is injected

= 0 and

τiri

= 0 and subsequently convected downstream, resulting in a variation

in residence time

τres

throughout the injected fuel profile. In order to achieve quasi-

simultaneous autoignition, the injected fuel profile is stratified (

ϕ3> ϕ2> ϕ1

), as

3 Methodology

std(Io) (ms)

01234

std(OH*) (ms)

Outside

Inside

Figure 3.5:

Correlation between standard deviation of the autoignition front obtained

from ionization probes and optical OH* emissions for ignitions originating within the

combustor section(green) and outside of the combustor section (black).

described in Sec. 1.3. For a bulk velocity of

¯uair

= 18 m/s, the mixture reaches the

combustor inlet/outlet after 39 ms/67 ms, respectively. Consequently, the equivalence

ratio needs to be adjusted according to the desired ignition delay times

τidt

to enable

quasi-simultaneous ignition of the combustor volume within the combustor section.

Multiple ionization probes are flush mounted into the combustor walls, as indicated in

Fig. 3.4. The ignition time measured by ionization probes is relative to the start of the

injection process

τiri

(see Ch. 2). Ignition times in the range of 67 ms indicate successful

ignition of the entire mixture in the combustor section. Values below indicate an early

ignition in the convection section, while higher values imply late ignition inside the

exhaust section.

To validate the accuracy of the obtained ignition times when applying ionization probes,

two-dimensional measurements using a highspeed camera were conducted simultaneously.

This allows for a qualitative comparison of the obtained ignition times applying two

independent measurement techniques. Detailed results of this study can be found in

Yücel et al. [

]. However, a summary is provided by Fig. 3.5, which visualizes the

correlation between the standard deviation of the autoignition front, as a measure

of the autoignition homogeneity, obtained by applying i) three ionization probes and

ii) optical measurements of the OH* intensity. The ionization probes were used to

measure the ignition times at different axial positions in the combustor which were

used to calculate the standard deviation of the autoignition as measure of the ignition

homogeneity. The optical measurements were used to detect the continuous spatial

distribution of the ignition times along the measurement section (not distinct positions),

which were subsequently used to calculate the ignition homogeneity. Measurement

points marked in green indicate autoignition within the measurement section, whereas

3.2 Measurements with Chemical Reaction

black markers represent autoignition outside of the measurement section. The obtained

data reveal a stronger correspondence between both methods for autoignitions detected

within the measurement section. Nevertheless, the ignition times obtained by each

method slightly deviate. Since the ionization probes are flush mounted into the tube

walls, a signal is only detected once a sufficient ion concentration is present at the

respective electrode. Thus, after the onset of ignition, the ions created in the reaction

zone first need to accumulate before a signal is measured by the ionization probe.

When measuring the OH* intensity, the entire tube diameter is considered, meaning

that any autoignition kernel appearing within the measurement section is sufficient for

determining the ignition time. Thus, depending on the ion concentration, the signals

are shifted.

In this work, ionization probes were applied to obtain cycle-averaged ignition character-

istics. The measured ignition times were evaluated in terms of the standard deviation

as a measure of autoignition homogeneity. A low standard deviation in ignition time is

associated with a higher number of simultaneous autoignitions, while a high standard

deviation is less homogeneous. Even though the precision of the ionization probes is

assumed to be lower compared to optical measurement methods, reliable data can be

extracted on a cycle-averaged basis. However, to sufficiently quantify the degree of

homogeneity of an autoignition solely by using ionization probes, a large number of

sensors is required.

Pressure Measurements

Pressure measurements are of utmost importance for the characterization of the SEC as

a pressure gain combustion device. In this work, pressure measurements were applied

in combination with ionization probes and/or optical chemiluminescence measurements.

High preheat and combustion temperatures require the applied sensors to withstand

extreme operation conditions. All pressure measurements inside the combustor section

of the SEC setup are conducted using fast response pressure sensors equipped with

a miniature water–cooled jacket (Kulite Sensor Type EWCTV-312), which allows for

pressure measurements in extreme temperature environments. The water circuit is

controlled to a constant water temperature of 323 K using a closed-circuit temperature

controller.

The measurement principle is based on a piezoresistive silicon sensor which measures

the change in electrical resistance of its piezoresistor element as a function of the applied

pressure and subsequently transforms it into an electric output signal. This phenomenon

was discovered first by Lord Kelvin in 1856 [

] who observed that the resistance of a

3 Methodology

-0.2

0.2

0.4

pmax

40 80 100 120

-0.2

0.2

0.4

0.6

p (bar)

t (ms)

Uion (V)

τpτi



τip

pthresh Uion,thresh

reflected

pressure wave

pressure rise

through autoignition

expansion

wave

expansion

wave

reflected

pressure wave

pressure rise

through autoignition

-0.2

0.2

0.6

40 80 100 120

-0.2

0.2

0.4

0.6

p (bar)

t (ms)

Uion (V)

τpτi



τip

pmax

Uion,thresh

pthresh

b)a)

Figure 3.6: Example data of pressure and ionization probe.

conductor increases under tension due to a change in length and thickness of conductor

through which the current is forced. The resulting resistance

is thus a function of

the density ρ, length Land cross section area Aof the conductor:

R=ρL/A. (3.12)

Figure 3.6 shows pressure and ionization probe data for two example cycles (a and b).

The illustrated pressure data is averaged from pressure measurements obtained by four

sensors flush mounted into the combustor walls. The ionization probe signal qualitatively

represents the ion concentration measured by a single ion probe at the respective

combustor position as described in Sec. 3.2. Two parameters are extracted from

pressure measurements in combination with ionization probes to evaluate autoignition

characteristics: i) maximum averaged pressure rise

¯pmax

and ii) the difference in time

delays ∆

τip

τi−τp

(see Fig. 3.6). A high

¯pmax

and a low ∆

τip

implies a more

homogeneous ignition due to the quasi-simultaneous detection of ignition and pressure

rise indicating an aerodynamic confinement during heat release. A low rise in pressure in

combination with a high ∆

τip

is considered a less homogeneous autoignition/deflagration.

The data visualized in Fig. 3.6a indicates the onset of ignition approximately 65 ms after

the start of the fuel injection. Thus, the reactive mixture ignites within the combustor

section. The pressure rises up to a maximum value

¯pmax

, followed by a decrease due

to the expansion of the burned gas in upstream and downstream directions. Pressure

waves that travel upstream are reflected at the acoustically closed inlet (choked inlet

conditions) and result in a second rise in pressure, as shown in Fig. 3.6. Pressure

waves propagating downstream are reflected at the tube outlet resulting in an upstream

propagating expansion wave, as indicated. Due to the pronounced pressure peak and the

low ∆

τip

visible in the data, this type of autoignition is considered a quasi-homogeneous

autoignition. The second cycle, visualized in Fig. 3.6b, indicates a low rise in pressure

3.2 Measurements with Chemical Reaction

and an increased value in ∆

τip

. Here, no aerodynamic confinement is achieved due to

the extended time delay which prevents the coupling of heat release and pressure.

Throughout this work, the determination of ignition characteristics by combining

pressure transducers and ionization probes has proven to deliver good results and thus

allows the classification of autoignition modes based on the identified parameters.

Chemiluminescence Measurements

In the scope of this work, chemiluminescence measurements were applied for a qualitative

evaluation of autoignition characteristics in terms of ignition homogeneity as well as

for the observation of multi-stage ignition behavior of the applied fuel as discussed in

Sec. 1.3.

Chemiluminescence is the spontaneous emission of light as a result of energy release due

to chemical reaction. Compared to fluorescence, as discussed in context with acetone-

PLIF (see Sec. 3.1), chemiluminescence is triggered by a chemical reaction instead of

light absorption. Chemiluminescence measurements provide a tool for a non-intrusive

combustion diagnostic utilizing the light emission by naturally occurring excited species.

In the literature, these type of measurements are employed to determine the onset of

ignition (e.g. in shock-tube measurements), the heat release rate or equivalence ratio.

Most common excited species found in hydrocarbon oxidation are CH*, OH*, C

* and

*. The formation of chemiluminescent species during hydrocarbon combustion is

studied in more detail by [39, 40].

In the scope of this work, the light emission during autoignition was captured by a

number of photomultiplier sensors (PMTs) in combination with a spectrometer allowing

for a one-dimensional line-of-sight measurement. PMTs were used to obtain the temporal

evolution of different species at a distinct axial position inside the combustor. When

applying an optical filter to a PMT, all light emission within the chosen range is

captured. Generally the emission spectra of multiple species overlap, meaning that

by a single filter the emission of multiple species are overlayed in the detection signal.

Therefore, this techniques requires the application of multiple filters to allow for a

precise allocation of the identified species. This specifically applies to CO

* which is

generally known to cause background emission by its broad-band emission character.

The spectrometer (Ocean Optics QE6500) was applied to determine the spectrum of

the light emission during combustion showing distinct peaks at characteristic species

wavelengths that occur in the process of autoignition. An example signal is shown in

Fig. 3.7 which visualizes the characteristic emission wavelengths of OH*, CH* and C

3 Methodology

OH*

CH*

C2*

intensity

wavelength (nm)

200 250 300 350 400 450 500

307 431 470 516

Figure 3.7:

Example spectrometer data representing the characteristic wavelengths of

OH*, CH* and C2*.

In the emission spectrum of formaldehyde, CH

O* (350-505 nm), which typically occurs

in cool flames as discussed in Sec. 1.3, and CO

* (300-600 nm) are characterized

by broad-band emission, while OH*, CH*, and C

* show distinct single or multiple

peaks (called Swan band) [

]. Optical band-pass filters centered around a defined

wavelength, called center wavelength (CWL), were applied to selectively transmit light

allowing for the observation of different species. The transitive window width is defined

by the full width at half maximum value (FWHM). This allows for precise measurement

of species that emit light at distinct wavelengths. However, the measured signal is

generally overlayed by background emission such as broad-band emitting CO

*. Thus,

it is reasonable to apply filters that measure the chemiluminescence emission slightly

off the main wavelength in order to allow for background corrections (e.g. using a

multi-connector cable).

The spectrometer data was primarily used for further interpretation of the PMT

signals. While the PMT data is limited to a number of species for the applied filter,

the spectrometer enabled the detection of further species such as C

*. Moreover, two-

dimensional chemiluminescence measurements were conducted using a highspeed camera

equipped with an intensifier. The data was subsequently used to create

diagrams

which visualize the temporal and spatial evolution of the autoignition front inside the

combustion chamber.

The methodology described in this chapter, helped to manifest the interplay of pressure

waves and heat release with regard to fuel stratification and the ignition characteristics

of the fuel. The fuel stratification strategy and its impairment by turbulent fluctuations

was investigated applying acetone-PLIF and TDLAS. Autoignition homogeneity was

analyzed based on ionization probes as well as two-dimensional chemiluminescence

3.2 Measurements with Chemical Reaction

data. Pressure transducers were applied to measure the pressure rise subsequent to

autoignition. Furthermore, PMTs and spectrometer data was used to investigate the

impact of fuel properties on the autoignition characteristics. In the following chapter,

post processing of the obtained data will be introduced.

4 Post Processing

In the scope of this work, a variety of measurement techniques were applied, as discussed

in Ch. 3. In this chapter the post processing of the obtained data will be described.

4.1 Proper Orthogonal Decomposition

Proper orthogonal decomposition (POD) as described by Berkooz et al. [

] and Chatter-

jee [

] was applied to a large amount of pressure data to extract dominant structures

allowing for the identification of characteristic pressure responses inside the combustor.

These responses were subsequently correlated to the obtained autoignition homogeneity.

POD is a well understood method for low-dimensional modeling of high-dimensional

data. Pressure data obtained by the application of four pressure transducer was

evaluated.

A field

(

x, t

)can be decomposed into deterministic spatial functions Φ

(

)and the

respective time coefficients ak(t)[87, 14]:

z(x, t)≈

k=1

ak(t)Φk(x).(4.1)

Here, Φ

(

)are referred to as POD modes which are chosen to be orthonormal, meaning

the time coefficients are solely dependent on the spatial mode Φ

. To achieve a sufficient

approximation, the basis functions are chosen to represent the most dominant features by

the first

spatial modes. These orthonormal basis functions can be used to approximate

the original data. The discrete version of the POD can be calculated by singular value

decomposition. For this, the pressure data of 50 cycles are assembled resulting in a

m×nmatrix Pwhich is subsequently factorized into:

P=





pcycle01(t)

pcycle50(t)







=UΣVT,(4.2)

4 Post Processing

where

is an orthogonal

m×m

matrix, Σis a

m×n

diagonal matrix and

is an

orthogonal matrix with the size

n×n

. The diagonal entries of Σare called singular

values and represent the “dominance” of each calculated mode. Therefore, the singular

values are arranged in a decreasing order. Thus, the first POD mode corresponding to

the highest singular value is the most dominant one. To understand the link between

Eq. 4.1 and Eq. 4.2, the term

Σis substituted by

. Thus,

can be expressed as

QV T

. Obviously, the time coefficients

correspond to

and the POD modes Φ

represent the respective columns of the matrices

. Subsequently, the first POD modes

are examined representing the dominant pressure profile extracted from a set of 50

cycles.

4.2 Cycle Averaging and Standard Deviation

The SEC is exposed to fluctuations causing cycle-to-cycle variations. Therefore, the

operation was controlled on a cycle-averaged basis. Cycle-averaged values

¯x

were

calculated according to

¯x=1

i=1

xi(4.3)

with

being the time-resolved measurement signal and

donating the number of

probes.

For the evaluation of the autoignition homogeneity the standard deviation of the ignition

times measured by ionization probes was calculated according to

std(x) = v

N−1

i=1

|xi−¯x|2.(4.4)

4.3 Image Processing

Image processing techniques were applied for two-dimensional optical measurements

with chemical reaction.

–

diagrams were generated to visualize the temporal evolution

of the autoignition inside the combustor. For this, two-dimensional image data recorded

with a high-speed camera in combination with an intensifier was used. An image doubler

was attached to the intensifier to capture the light emission at two different wavelengths.

The distortion resulting from the curvature and the thickness of the quartz combustor

towards the bounds is obtained. For spatial calibration purposes, images are taken

initially from a target placed into the center plane of the combustor visualizing uniform

4.3 Image Processing

raw

1. transformation

2. transformation

final

cropped

raw

1. transformation

2. transformation

final

OH*CH*

polynomial

piecewise linear

Figure 4.1:

Visualization of the image calibration using Matlab ’fitgeotrans’ function [

]

applied to target image using an image doubler (publication IV).

squares (as shown in Fig. 4.1). A Matlab function ’fitgeotrans’ is applied to the

calibration snapshots, which uses pairs of ’control points’ from the ’non-distorted target

image, called fixed points’, and from the distorted image, called ’moving points’ to infer

a transformation [

]. These points are selected manually and are chosen to be

the grid points due to convenience. As shown in Fig. 4.1 the transformation matrix

was obtain in two steps. First, a second degree polynomial fit was used as a global

transformation in order to correct for the curvature along the combustor radius (see

Fig. 4.1a). Subsequently, a line-wise linear fit was applied allowing for the correction of

the grid distances. The obtained transformation matrix was applied to each snapshot.

The resulting images are cropped in order to reduce the amount of data to be processed.

Subsequently, the mean image of the first ten snapshots prior to the onset of ignition was

subtracted from the transformed data in order to background-correct each measurement.

The resulting image data was used to create

–

diagrams illustrating the temporal

4 Post Processing

evolution of the autoignition inside the combustor. For this, each individual image is

condensed into a one-dimensional array by averaging the obtained intensity over the

tube diameter. The resulting one-dimensional arrays are subsequently aligned into a

two-dimensional matrix representing the intensity for each time step at the respective

axial position inside the combustor. This data was used to illustrate the spatial and

temporal evolution of the autoignition front and determine propagation speed and/or

ignition homogeneity.

5 Publications

All publications included in this work target the main objective of understanding

the fundamentals of the SEC process and enabling a reproducible and reliable SEC

operation. Each publication delivers a key insight towards this goal. The main

outcomes for each paper resembling the evolution of the research topic are summarized

in the following: I) designing a test rig to increase reproducibility of the operation, II)

verifying the capability of the designed injection strategy to allow for a precise control

of the fuel stratification, III) investigating the correlation between fuel stratification and

autoignition characteristics and finally IV) applying an optimization approach to control

the formation of different autoignition modes. Upfront information is provided for each

publication summarizing the contribution, the applied methods and the outcomes.

5 Publications

5.1 Publication I: Effect of the Switching Times on the

Operating Behaviour of a Shockless Explosion

Combustor

Contribution:

This work serves as a starting point to evaluate operational param-

eters that impact the flow motion inside the “SEC with bypass”.

Methods:

Laser Doppler Anemometry was applied to obtain valve operating

parameters. PMTs were used to measure ignition timings.

Results:

The results reveal the controllability of the ignition position by

a proper adjustment of the valve timings. Moreover, the repro-

ducibility of the autoignition was greatly improved. However, the

results also reveal that further improvements are necessary, includ-

ing the decrease of ignition delay times, which ultimately results

in a higher grade of preservation of the injected fuel profile. This

goal is achievable by drastically simplifying the test rig to overcome

heat losses and thus ensure a higher combustor inlet temperature

resulting in lower ignition delay times.

Reference:

Yücel, F. C., Völzke, F., & Paschereit, C. O. (2019). Effect of the

switching times on the operating behavior of a shockless explosion

combustor. In Active Flow and Combustion Control 2018 (pp.

121-134). Springer, Cham. URL:

https://doi.org/10.1007/978-

3-319-98177-2_8

Effect of the Switching Times

on the Operating Behavior of a Shockless

Explosion Combustor

Fatma C. Yücel, Fabian Völzke and Christian O. Paschereit

Abstract In the past, a wide range of investigations are made in order to increase the

efficiency gain in gas turbines by using constant volume combustion. In comparison to

detonation-based concepts, such as pulse detonation engine and rotation detonation

engine, a new promising way was proposed by Klein and Paschereit and firstly

assessed by Bobusch et al. (Combust Sci Technol 186(10–11):1680–1689 (2014),

[1]), the so-called shockless explosion combustion (SEC). The principle is based

on a quasi-homogeneous auto-ignition process that leads to an approximate constant

volume combustion (aCVC). In order to achieve a quasi-homogeneous auto-ignition,

it is necessary to achieve constant ignition delay times along the combustor. The

combustion process in the SEC is similar to the one in internal combustion engines,

namely Homogeneous Charge Compression Ignition (HCCI). This paper focuses on

the use of wastegates to actively control filling and flow motion in the combustor

dedicated to perform quasi-homogeneous auto-ignition. The results clearly show the

ability to actively control the fuel distribution and purging time in the combustor

which is an important step in the evolution of the SEC.

Keywords Constant volume combustion ·Shockless explosion combustion

Ignition delay time ·Purging time

F. C. Yücel (B)·F. Völzke ·C. O. Paschereit

Technische Universität Berlin, Institut für Strömungsmechanik und Technische Akustik,

Müller-Breslau-Str. 8, 10623 Berlin, Germany

e-mail: [email protected]

F. Völzke

e-mail: fabian.v[email protected]

C. O. Paschereit

e-mail: oliver[email protected]

R. King (ed.), Active Flow and Combustion Control 2018,

Notes on Numerical Fluid Mechanics and Multidisciplinary Design 141,

https://doi.org/10.1007/978-3-319-98177-2_8

121

5.1 Publication I: Effect of the Switching Times on the Operating Behaviour...

122 F. C. Yücel et al.

1 Introduction

Today, the potential of achieving a notable gain in gas turbine efficiency by improv-

ing components such as compressor or turbine is only incremental. A leapfrogging

step could be achieved by replacing the conventionally utilized Brayton cycle (ideal

constant-pressure cycle) by the Humprey cycle (ideal constant-volume cycle) [2].

In the past decades, different concepts on this topic were investigated, e.g., pulsed

detonation combustors (PDC) [3] and rotation detonation combustors (RDC) [4].

The PDC and RDC are both based on periodic combustion processes that utilize a

detonation wave to achieve an approximate constant volume combustion (aCVC).

By igniting a flammable mixture a detonation wave is initiated that propagates with

a high velocity into the unburned mixture, the gas has no time to expand and burns

quasi-instantaneously. However, detonation waves are inefficient due to sharp pres-

sure peaks that are associated with strong losses. The shockless explosion combustor

suggested by Bobusch et al. [1] is a new promising way to implement an aCVC to

achieve an increase in the efficiency of a gas turbine cycle. The combustion process

itself is similar to the HCCI process used in internal combustion engines [5,6]. In

the HCCI a homogeneous ignitable mixture is compressed until auto-ignition occurs

while the SEC achieves a quasi-homogeneous auto-ignition by stratifying the fuel-air

mixture along the combustor close to auto-ignition conditions. The fuel stratification

is needed in order to compensate the residence time of the fuel such that a constant

ignition delay time along the combustor is achieved. One challenge in the HCCI

process is the ignition timing. The occurrence of too early or late combustion turned

out to be disadvantageously for the HCCI process. For this, closed-loop control is

applied to actively control the ignition timing. One similar challenge in the evolution

of the SEC process is to control the ignition timing along the combustor to increase

homogeneity of the auto-ignition process.

The SEC is based on a periodic combustion cycle that achieves an aCVC by a

quasi-homogeneous auto-ignition process as shown in (Fig. 1). At the beginning of

the process (Fig.1, top), the combustor is filled with a well-defined stratified fuel-

air mixture. This axial stratification leads to a quasi-homogeneous auto-ignition of

the entire gas volume after a certain ignition delay time. Due to the simultaneous

combustion, a pressure wave is induced that propagates in downstream direction.

At the open end of the combustor, the pressure wave is reflected as a suction wave

that travels upstream (Fig.1, bottom). As the suction wave reaches the tube inlet,

the pressure drops below supply pressure and the recharge cycle can start. First, the

combustor is purged with pure air creating a buffer to the hot combustion products.

After a certain purging time, fuel is injected and the combustor is filled with the

axially stratified fuel-air mixture (Fig.1, left) while the suction wave is reflected and

propagates downstream the tube. The suction wave is then reflected at the tube outlet

as a pressure wave traveling upstream again. The process is restarted by another

simultaneous combustion of the fuel-air mixture.

The ignition delay time of a mixture depends on temperature, pressure, equiva-

lence ratio and the type of fuel and oxidizer. When assuming all other parameters to

5 Publications

Effect of the Switching Times on the Operating … 123

Fig. 1 SEC cycle

be constant, the spatial distribution of the ignition delay time τ(x)can be controlled

by a stratification of the equivalence ratio. The ignition delay time decreases with

increasing temperature, pressure and with decreasing deviation from stoichiometric

conditions. Neglecting small temperature differences along the tube, a constant fuel

injection would lead to an early ignition at the rear end of the combustor which would

impede a quasi-homogeneous auto-ignition. In order to achieve a quasi-simultaneous

auto-ignition, the ignition delay time must increase from the injection position to the

outlet of the combustor. Further investigations have shown that a criteria for a quasi-

simultaneous ignition along the combustor is that the difference in ignition delay

time Δτign between two neighboring infinitesimal small volumes should not exceed

the excitation time τet, which is a value for the rate of chemical energy release [7,8]:

Δτign

<τet.(1)

One challenging task is the realization of resonant operation. The frequency of

the SEC cycle is determined by the acoustic frequency of the combustor—as it relies

on the suction wave to initiate the refill process—which is in the order of 100Hz. For

this, depending on tube length and speed of sound, very short ignition delay times

are required. For atmospheric pressure conditions this can not be achieved, as the

ignition delay times of all known fuels are not small enough. This crucial question

gives room for future investigations and is not in the scope of this paper.

Another challenge is maximizing the gain in efficiency by achieving a quasi-

homogeneous auto-ignition. Therefore, an increase in homogeneity of the combus-

tion process is aimed since it goes along with an increase in efficiency. This requires

efficient control of the axial fuel stratification to realize an operational SEC process.

5.1 Publication I: Effect of the Switching Times on the Operating Behaviour...

124 F. C. Yücel et al.

Investigations showed that the application of an iterative learning controller into a

closed-loop control is a promising way to minimize the variance of the ignition time

by controlling the spatial fuel distribution [9].

For further improvements of the homogeneity of the auto-ignition, it is important

to gain a deeper knowledge of this procedure and to analyze the impact of different

input parameters. Two parameters are investigated in this work: the ability to con-

trol the spatial distribution of the fuel concentration and the impact of temperature

fluctuations due to the purging time of the combustor.

2 Experimental Setup

An atmospheric test rig (Fig.2) is designed to investigate the behavior of a shock-

less explosion combustor at non-resonant conditions. The test rig has been used in

earlier investigations [9] and was modified in order to increase the experimental

reproducibility.

The air is preheated to a maximum temperature of 700◦C using an electrical air

heater to realize lowest ignition delay times. Dimethylether (DME) is used as fuel

which exhibits an ignition delay time of around 100ms at the given atmospheric

pressure conditions and high temperatures [1,10]. Additionally, DME has a charac-

teristic small ignition delay time variation with temperature. Fuel lines are preheated

to 90◦C preventing liquification of the DME.

The application of the preheater requires a minimal mass flow of 30 kg/h. This

results in a combustor flow velocity of about 22m/s and a residence time in the range

of 30ms. With an ignition delay of 100ms the ignitable mixture would have already

left the combustor before ignition. Enabling also tests under atmospheric pressure

conditions a bypass is installed allowing the combustor flow to rest after charging.

Currently, valves for turbocharged engine systems that can be controlled actively

are used, called wastegates, replacing the previously used fluidic switch and valve

Fig. 2 Sketch of the modified atmospheric test rig of a shockless explosion combustor equipped

with wastegates (WG)

5 Publications

Effect of the Switching Times on the Operating … 125

Fig. 3 Detailed view of the combustor with fluidic diode (blue) with oscillators (yellow), quartz

tube and PMTs

heads. They have been installed respectively at the combustor (WG1) and at the

bypass (WG2). These valves are equipped with several ports where boost pressure

can be supplied such that they can be opened and closed actively. The advantage

of these wastegates when compared to the fluidic switch in combination with valve

heads is that the wastegates prevent backflow more reliably and the opening and

closing timings can be adjusted actively. This has not been the case for the valve

heads since they closed and opened passively depending on the pressure at the tube

outlets. Furthermore, replacing the fluidic switch by a Y-pipe decreased the heat loss

upstream of the combustor significantly.

A fluidic diode prevents the backflow of hot gases due to pressure increase after

the combustion process and was already successfully used in earlier investigations

[11]. The fuel is injected using two identical fuel injector arrays each equipped with

four parallel-connected high-speed solenoid valves. The fuel inlet section is equipped

with eight radial injection ports and fluidic oscillators to increase the mixing quality

in the radial direction.

The combustor has two sections with an inner diameter of d=40mm. The first

section is made from a 0.5m long quartz tube in order to enable the detection of

the ignition event via photomultipliers (PMTs). These PMTs are used to evaluate

the ignition delay times at five axial positions in the combustor (Fig.3). The second

part is a steel tube equipped with two water-cooled pressure sensors to detect the

combustion-induced pressure wave.

3 Investigation in the Wastegate Behavior

The wastegates used in this work are composed of a main body and an actuator body

(Fig.4). The actuator body contains a piston that is connected to the valve head in the

main body. There are two different ways of controlling control the wastegates: passive

and active. Equipping the actuator body with springs of different spring constants, the

5.1 Publication I: Effect of the Switching Times on the Operating Behaviour...

126 F. C. Yücel et al.

Fig. 4 Sectional view of a

wastegate; 1valve head,

2boost pressure port for

opening, 3membrane, 4

boost pressure port for

closing, 5spring, 6

diaphragm

required input pressure for opening the valve can be controlled passively by setting

the reset force of the piston. In this work, the wastegates were controlled actively by

regulating the supplied boost pressure. A diaphragm separates the two different areas

inside the wastegate for the boost pressure supply. By setting the pressure in the upper

area to higher levels, the wastegate closes. When the pressure in the upper region

is decreased and the pressure in the lower area is increased, the wastegate opens.

3/2-way valves have been applied to allow for controlling the boost pressure. It is

important to note that the wastegate behavior and especially its response times are

dependent on the operating speed of the 3/2-way valves. One voltage signal is needed

to operate both wastegates simultaneously since the 3/2-way valves are installed on

each wastegate at inverted positions.

The switching process and its effect on the combustor flow was investigated using

a laser Doppler anemometry system (LDA). This non-intrusive optical measure-

ment technique enables a high spatial and temporal resolution. An aerosol (Bis(2-

ethylhexyl) sebacate) was used as seeding which was atomized using the aerosol-

generator.

The LDA measurements were done under non-reactive conditions using air only.

The air flow has been set to 30 kg/h, which matches the conditions for reacting

tests. The laser was positioned 50mm downstream the injection geometry. Figure 5

displays the voltage signal for controlling the 3/2-way valves and the measured

combustor flow velocity as a function of time for the first three periods at a switching

frequency of 0.5Hz. When the air flow is guided to the combustor, the mean velocity

measured was umean =6.3m/s with an RMS-value of urms =0.68m/s implying a

turbulence intensity of 10 %. When the voltage signal for controlling the 3/2-way

valves is set to UC=5 V, WG1 closes and WG2 opens simultaneously. Subsequently,

5 Publications

Effect of the Switching Times on the Operating … 127

Fig. 5 Measured particle

velocity uwith Laser

Doppler Anemometry and

switching signal UCover

time

Fig. 6 Mean velocity umean,

RMS-value urms and

switching signal UCover

time

after a certain delay time the velocity decreases fast until the flow comes to rest. The

main velocity is overlaid by velocity oscillations due to high turbulence during the

opening and closing procedure. When the voltage signal is set back to UC=0V,the

flow velocity increases again until the mean velocity of ¯v=6.3m/s is reached. Both

switching events include two time delays respectively. To specify these time delay

times, the velocity was averaged over 10 cycles and the delay times were assessed

(Fig.6).

The time delay for closing the WG1 is determined to be ΔτD1 =215ms, while the

flow needs a time span of ΔτST1 =17 ms to be decelerated to u=0 m/s. The values

for opening WG1 have been determined to be ΔτD2 =175ms and ΔτST2 =31ms.

All values are listed in Table 1.

The RMS-value shown in Fig. 6displays the stochastic fluctuations of the flow

velocity when the air is guided through the combustor. The fluctuations between the

cycles that appear when stopping the air flow in the combustor are very low which

5.1 Publication I: Effect of the Switching Times on the Operating Behaviour...

128 F. C. Yücel et al.

Table 1 Wastegate timings

WG1 closing (ms) WG1 opening (ms)

Wastegate delay τD215 175

Switching time τST 17 31

emphasizes the reproducibility of the switching events. Even though the flow cannot

be stopped instantly, the reproducibility of the process enables the consistency of the

stratification of the fuel-air mixture for reacting tests.

4 Ignition Timing Measurements

Preliminary investigations showed that the purging time can be actively controlled

by triggering the wastegates. For this, the delay times discussed in the previous

section (see Table 1) have to be taken into account. In this section, the impact of the

actual wastegate timings and the purging time as an input control parameter on the

combustion process of the SEC are evaluated under reacting conditions. The success

rates for auto-ignition and the spatial distribution of the ignition delay time have

been investigated as a function of the switching times. To investigate this behavior,

a periodic non-resonant combustion process is applied.

One operational period of the SEC process can be divided into three different parts:

charging, homogeneous ignition and purging. During the charging process, the air

mass flow is guided through the combustor and a constant fuel mass flow is added for

30ms. Since this work focuses on the control aspect, the combustible mixture is not

stratified as no homogeneous combustion is aimed here. After a well defined time

span τWT1, that is varied between 40 and 120ms, the air flow is switched into the

bypass by closing the wastegate WG1 and opening wastegate WG2. To specify the

appropriate timing of the voltage signal, the time delays τD1 and τD2 are taken into

account. While the mass flow is directed through the bypass, the ignition event takes

place in the combustor. After a second defined time span τWT2 the purging process

starts by guiding the air flow through the combustor again. Thus, an increasing

residence of the air mass flow in the bypass is equal to a decreasing purging time.

Figure7shows an exemplary period of a 1 Hz SEC cycle for τWT1 =40ms and

τWT2 =300ms. During the interval from 0 to 40 ms the air mass flow is guided

through the combustor while fuel is injected at the same time from 0 to 30 ms, such

that the quartz tube is entirely filled with a flammable mixture. After 40ms the closing

procedure starts such that after the required switching time of 17ms the wastegate

is entirely closed and the air mass flow is guided through the bypass for a duration

of 260ms. The air flow is then switched back into the combustor at 300 ms with

an additional time delay of 31ms until the end of the cycle. In total this induces a

purging time of 669 ms.

5 Publications

Effect of the Switching Times on the Operating … 129

Fig. 7 Simplified representation of the different timings, 1fuel injection time τinj,2wastegate

timing τWT1,3wastegate timing τWT2,3to 4purging time τpt,Iswitching time τST1,II

switching time τST2

Fig. 8 Output-signal of PMTs

As shown in Fig.3, PMTs are used to detect the ignition event at five different

axial positions. Figure8shows two exemplary output signals of the PMTs. The PMTs

detect the light emission of the flame due to auto-ignition. The time of ignition is

assumed to be the time when the voltage signal of a certain PMT reaches the threshold

of 0.45V. In Fig.8a the PMT signal displays an output, where a combustion event is

detected by all PMTs, while in Fig.8b combustion event was only detected by PMT4

and PMT5. Although there is a peak in the signal of PMT3, this is not interpreted as

an ignition event at this position, since it does not reach the threshold of 0.45V. It

rather can be interpreted as an ignition event close to the position of PMT3 causing

some light emission entering the PMT3. As a criteria for auto-ignition only the first

PMT signals are considered as actual auto-ignition events since in this paper, it is

not aimed for a fuel stratification. Starting from the first ignition it is considered that

the remaining PMTs only detect a deflagrative propagating flame front.

Figure9illustrates the ignition events at different positions in the combustor for

different wastegate timings τWT1 and τWT2. Every node at the x-axis is representative

5.1 Publication I: Effect of the Switching Times on the Operating Behaviour...

130 F. C. Yücel et al.

Fig. 9 Investigation on the impact of the purging time on the success rate on every single PMT by

varying τWT1 and τWT2

for one PMT, while the y-axis represents different timings for switching from bypass

into the combustor. The colorbar represents the success rate of auto-ignition whereby

40 cycles where evaluated in total for every combination of tested wastegate timings.

A success rate of 1 means that the PMT detected an ignition event in every single

cycle. The blue areas, representing a value of 0, show that no ignition could be

detected, respectively.

It is apparent that for relatively high purging times (τWT2 ≤540 ms) and 40 ms ≤

τWT1 ≤80ms, every single PMT has a success rate of 1. It can be assumed that the

entire quartz tube is filled with a flammable mixture as visualized in Fig.10 case B.

Thus, this regime contains a periodic auto-ignition event that induces a combustion

process at all observed axial positions in the combustor.

Increasing τWT1 results in the injected fuel package traveling further downstream

before the flow is stopped due to the closing of WG1 (see Fig. 10). Thus, the ignition

event is shifted more and more into the exhaust tube. Therefore, an increase to τWT1 =

100ms reduces the detected ignition events at the first three PMTs drastically (Fig.10

case D). For τWT1 =120ms, no PMT detects a stable combustion for τWT2 <690 ms

as the success rate is zero at every position since the entire burnable mixture has been

moved outside of the quartz tube, where the combustion event cannot be detected by

the PMTs (Fig.10 case E). This observation demonstrates the ability to control the

spatial fuel distribution by using actively controlled wastegates.

5 Publications

Effect of the Switching Times on the Operating … 131

Fig. 10 Position of the fuel-air package in the combustor for an increasing τWT1 value

For an increased τWT2 and a constant τWT1, no ignition event can be detected at

PMTs 3–5. A presumption for this effect is that due to small periods of purging,

hot combustion products remain in the combustor due to recirculation areas near the

injection geometry and cause contact burning when the next fuel package is injected.

Thus, a auto-ignition fails and the combustion is only detected by the first PMTs

since there is no time for the fuel-air mixture to travel downstream before ignition.

This phenomenon can be observed when τWT2 exceeds a certain threshold that equals

the time that is needed to purge all the exhaust gases that have entered the injection

geometry and the supply line due to the increased pressure in the combustor after the

combustion event.

This threshold for τWT2, where the success rate changes from 1 to 0 for a constant

τWT1, is shifted upwards by increasing τWT1. As mentioned earlier, increasing τWT1

causes the burnable mixture to travel further downstream before the auto-ignition

takes place. This leads to a smaller amount of hot exhaust gases in the upstream part

of the combustor. Thus, a shorter time is needed to purge the line.

By further increasing τWT2, a second characteristic of the SEC test rig is visible.

When the preheated air flow is guided through the bypass, the temperature in the

combustor decreases. The more time the air flow is guided through the combustor, the

higher the temperature gets. Since the ignition delay time of DME is still sensitive

to the mixing temperature as shown in [1,10], a small value of τWT1 and a large

5.1 Publication I: Effect of the Switching Times on the Operating Behaviour...

132 F. C. Yücel et al.

Fig. 11 Investigation of the impact of the purging time on the ignition delay time on every single

PMT by varying τWT1 and τWT2

value of τWT2 causes the ignition delay time to exceed the cycle duration. Therefore,

no combustion event is detected by the PMTs. Depending on the purging time the

temperature in the combustor varies significantly. A higher purging time causes an

increase in temperature but is also the limiting factor for higher frequencies. Low

purging times cause burnt gas to remain in the combustor causing an uncontrolled

ignition. This proves that the purging time is a decisive parameter in order to achieve

an increased homogeneity of the auto-ignition process.

The mean ignition delay time is measured as the time span from the start of

injection to the detection until combustion event. The local ignition delay time varies

from 100 to 250ms. When no combustion event is detected, the ignition delay time

exceeds the value of 250 ms (see Fig. 11).

The velocity over time shown in Fig.5implies that an increase in τWT1 leads to an

increased time span in which the mixture is exposed to high turbulence. Turbulent

diffusion causes mixing of the flammable mixture with the surrounding air at the

bounds of the fuel-air package (see Fig. 10 case C). This leads to a decrease in the

gradient of the equivalence ratio with increasing τWT1. Thus, the observed ignition

delay time increases with increasing τWT1, which is contradicting the wish for a

specifically shaped profile of fuel stratification. Increasing τWT2 induces a slight

decrease in the ignition delay time.

5 Publications

Effect of the Switching Times on the Operating … 133

5 Conclusion and Outlook

The SEC process was investigated under atmospheric pressure conditions using a

modified test rig. Wastegates, which replaced the former used fluidic switch and

valve heads, were used to obtain the influence of purging time and to find improved

conditions for successfully reducing the ignition delay time at atmospheric condi-

tions. It has been shown that actively controlled wategates enable precise control of

the fuel distribution. It has been proved that the purging time has a decisive impact

on the performance of a shockless explsoion combustor. New parameters were found

that can be utilized as input data for closed-loop control in order to optimize the spatial

fuel distribution which is essential for achieving a homogeneous auto-ignition.

The obtained data showed that the auto-ignition itself is a complex process with

a high number of variables. This work gives a detailed look to the sensitivity of

the SEC process to the switching timings of the wastegates as an important control

parameter.

The next step is to significantly decrease the ignition delay times. Therefore, exper-

imental investigations in an intermediate pressure test rig are planned. A decrease in

ignition delay time would lead to higher firing frequencies that are needed to realize

resonant operation. A new stainless steel combustion chamber has been designed,

where ionization probes will be used for flame detection at elevated pressures.

Acknowledgements The authors gratefully acknowledge support by the Deutsche Forschungs-

gemeinschaft (DFG) as part of Collaborative Research Center SFB 1029 “Substantial efficiency

increase in gas turbines through direct use of coupled unsteady combustion and flow dynamics” on

project A01.

References

1. Bobusch, B.C., Berndt, P., Paschereit, C.O., Klein, R.: Shockless explosion combustion: an

innovative way of efficient constant volume combustion in gas turbines. Combust. Sci. Technol.

186(10–11), 1680–1689 (2014)

2. Stathopoulus, P., Vinkeloe, P., Paschereit, C.O.: Thermodynamic evaluation of constant volume

combustion for gas turbine power cycles. In: 11th International Gas Turbine Congress, Tokyo,

Japan, pp. 15–20 (2015)

3. Roy, G.D., Frolov, S.M., Borisov, A.A., Netzer, D.W.: Pulse detonation propulsion: challenges,

current status, and future perspective. Prog. Energy Combust. Sci. 30(6), 545–672 (2004)

4. Lu, F.K., Braun, E.M.: Rotating detonation wave propulsion: experimental challenges, model-

ing, and engine concepts. J. Propul. Power 30(5), 1125–1142 (2014)

5. Yao, M., Zheng, Z., Liu, H.: Progress and recent trends in homogeneous charge compression

ignition (HCCI) engines. Prog. Energy Combust. Sci. 35(5), 398–437 (2009)

6. Stanglmaier, R.H., Roberts, C.E.: Homogeneous charge compression ignition (HCCI): benefits,

compromises, and future engine applications. Technical report, SAE Technical Paper (1999)

7. Meyer, J.W., Cohen, L.M., Oppenheim, A.K.: Study of exothermic processes in shock ignited

gases by the use of laser shear interferometry. Combust. Sci. Technol. 8(4), 185–197 (1973)

8. Cai, L., Pitsch, H.: Tailoring fuels for a shockless explosion combustor. In: Active Flow and

Combustion Control 2014, pp. 299–315. Springer (2015)

5.1 Publication I: Effect of the Switching Times on the Operating Behaviour...

134 F. C. Yücel et al.

9. Reichel, T.G., Schäpel, J.-S., Bobusch, B.C., Klein, R., King, R., Paschereit, C.O.: Shock-

less explosion combustion: experimental investigation of a new approximate constant volume

combustion process. J. Eng. Gas Turbines Power 13(2), 021504 (2017)

10. Berndt, P.: On the use of the hll-scheme for the simulation of the multi-species euler equations.

In: Finite Volumes for Complex Applications VII-Elliptic, Parabolic and Hyperbolic Problems,

pp. 809–816. Springer (2014)

11. Bobusch, B.C.: Fluidic devices for realizing the shockless explosion combustion process (2015)

5 Publications

5.1 Publication I: Effect of the Switching Times on the Operating Behaviour...

Publication I: Summary and Contribution

The first publication included in this work experimentally investigates the controllability

of the flow motion inside the “SEC with bypass” as described in Ch. 2. Therefore, two

wastegates were attached at the combustor and bypass outlet, respectively, allowing

for switching the air flow path between both sections. The objective is to control

the position of the injected fuel–air package inside the combustor, and thus, enable

autoignition within the combustor section.

First, Laser Doppler Anemometry as described in Sec. 3.1 was applied to characterize

the wastegates in terms of valve opening and closing times. For this, measurements

without chemical reaction were conducted to measure the air flow velocity inside the

combustor during the switching process. The results revealed that there is a significant

delay, in the order of 200 ms, between the trigger signal and the actual wastegates

closing/opening event. A second delay was determined representing the time duration

for the air flow to accelerate/decelerate subsequent to the switching process.

Next, the determined parameters were applied for measurements with chemical reaction.

Here, five photomultiplier (PMTs) were used, distributed along the quartz combustor,

to detect the onset of autoignition. The SEC was operated at a frequency of 0.5 Hz.

At the beginning of each cycle, fuel was injected into the air flow. During operation

the wastegate timings were varied and the position of the autoignition was measured

using PMTs. The measured ignition timings were mapped onto the combustor position,

visualizing the position of autoignition with respect to the valve operation timings.

The results demonstrated that the flow motion and thus the position of the autoignition

inside the combustor is well controllable. However, this study also revealed that the SEC

process is highly sensitive towards various operating parameters. Also, the repeatability

suffered due to low frequency operation of 0.5 Hz, impeding steady-state operating

conditions. Thus, it became evident that for a repeatable and reliable SEC operation a

simplification of the operating cycle is essential. As a consequence, multiple steps were

taken in terms of test rig modification as described in Ch. 2 resulting in the “single

tube SEC” which has been subsequently investigated in the following publications

(publication II to IV).

5 Publications

5.2 Publication II: Investigation of the Fuel

Distribution in a Shockless Explosion Combustor

Contribution:

This work demonstrates the capability of the applied control strat-

egy of injecting a defined fuel profile into a continuous air flow.

Furthermore, the effect of boundary layers and turbulent fluctua-

tions on the preservation of the fuel concentration profile during

convection was evaluated.

Methods:

The fuel distribution inside the combustor is visualized applying

tunable diode laser absorption spectroscopy as time-resolved line-

of-sight measurement and acetone planar laser induced fluorescence

as a spatially resolved measurement technique.

Results:

The results reveal that the injection strategy is capable of injecting

a defined fuel profile within a given time span. The grade of

preservation was found to be mainly dependent on the spatial

width which is defined as the product of the air flow velocity

and the injection duration for a constant convection time. Two

dimensional measurements reveal a radial distortion of the fuel

profile due to boundary layer effects.

Reference:

Yücel, F. C., Habicht, F., Jaeschke, A., Lückoff, F., Oberlei-

thner, K., & Paschereit, C. O. (2021). Investigation of the

Fuel Distribution in a Shockless Explosion Combustor. Jour-

nal of Engineering for Gas Turbines and Power, 143(1). URL:

https://doi.org/10.1115/1.4049220

Fatma Cansu Y€

ucel

Chair of Fluid Dynamics,

Technische Universit€

at Berlin,

M€

uller-Breslau-Str. 8,

Berlin 10623, Germany

e-mail: [email protected]

Fabian Habicht

Chair of Fluid Dynamics,

Technische Universit€

at Berlin,

M€

uller-Breslau-Str. 8,

Berlin 10623, Germany

Alexander Jaeschke

Laboratory for Flow Instabilities and Dynamics,

Technische Universit€

at Berlin,

M€

uller-Breslau-Str. 8,

Berlin 10623, Germany

Finn L€

uckoff

Laboratory for Flow Instabilities and Dynamics,

Technische Universit€

at Berlin,

M€

uller-Breslau-Str. 8,

Berlin 10623, Germany

Kilian Oberleithner

Laboratory for Flow Instabilities and Dynamics,

Technische Universit€

at Berlin,

M€

uller-Breslau-Str. 8,

Berlin 10623, Germany

Christian Oliver Paschereit

Chair of Fluid Dynamics,

Technische Universit€

at Berlin,

M€

uller-Breslau-Str. 8,

Berlin 10623, Germany

Investigation of the Fuel

Distribution in a Shockless

Explosion Combustor

Shockless explosion combustor (SEC) is a promising concept for implementing pressure

gain combustion into a conventional gas turbine cycle. This concept aims for a quasi-

homogeneous auto-ignition that induces a moderate rise in pressure. Since the ignition is

not triggered by an external source but driven primarily by chemical kinetics, the homo-

geneity of the auto-ignition is very sensitive to local perturbations in equivalence ratio,

temperature, and pressure that produce undesired local premature ignition. Therefore,

the precise injection of a well-defined fuel profile into a convecting air flow is crucial to

ensure a quasi-homogeneous ignition of the entire mixture. The objective of this work is

to demonstrate that the injected fuel profile is preserved throughout the entire measure-

ment section. For this, two different control trajectories are investigated. Optical mea-

surement techniques are used to illustrate the effect of turbulent transport and dispersion

caused by boundary layer effects on the fuel concentration profile. Results from line-of-

sight measurements by tunable diode laser absorption spectroscopy indicate that the

transport of the fuel-air mixture is dominated by turbulent diffusion. However, compari-

sons to numerical calculations reveal the effect of dispersion toward the bounds of the

fuel concentration profile. The spatially resolved distributions of the fuel concentration

inside the combustor gained from acetone planar laser induced fluorescence (PLIF) rep-

licates a typical velocity distribution of turbulent pipe flow in radial direction visualizing

boundary layer effects. Comparing both methods provides deep insights into the transport

processes that have an impact on the operation of the SEC. [DOI: 10.1115/1.4049220]

1 Introduction

Improving the thermodynamic cycle of gas turbines is of major

interest for efficient power generation. Tremendous efforts in the

past decades, however, only led to minor improvements in cycle

efficiency. Implementing pressure gain combustion is a promising

approach to overcome this stagnation. For this, different

approaches are currently under investigation. The most popular

concepts are pulse detonation [1] and rotation detonation combus-

tion [2]. Both detonation-based concepts utilize a propagating det-

onation wave burning a flammable mixture quasi-instantaneously.

However, the sharp pressure rise occurring during the combustion

process goes along with considerable losses caused by high

entropy generation and might as well harm mechanical compo-

nents in the machine [3]. The shockless explosion combustion

(SEC) is a new approach that overcomes the aforementioned dis-

advantages of detonation-based combustors such as sharp pressure

peaks by aiming for homogeneous auto-ignition leading to a grad-

ual rise in pressure.

Zel’dovich [4] defined in his theory different modes of flame

propagation initiating from exothermic centers that can be classi-

fied by the spatial gradient in ignition delay time sai. This ignition

delay time is inversely proportional to the propagation velocity uai

of the auto-ignition front. Assuming a constant distribution in

temperature Tand pressure p, the ignition delay time can be

expressed as a function of the equivalence ratio uonly

uai ¼@sai

 1

¼@sai

 1

:(1)

The dimensionless parameter n, defined as

n¼a

uai

¼a@sai

@x(2)

with the speed of sound aand the propagation velocity of the

auto-ignition front uai, can be used to classify different flame

propagation modes as follows [4,5]:

n>1: subsonic auto-ignitive flame propagation or deflagration,

n1: coupling of a reaction front with a

pressure wave forming a detonation,

0<n<1: supersonic auto-ignitive flame propagation,

n¼0: thermal explosion (homogeneous auto-ignition).

If n>1 and uai exceeds the laminar flame speed, a subsonic

auto-ignition front can be observed. Otherwise a deflagration front

propagates through the unburned mixture. In case the propagation

velocity of the auto-ignition front is similar to the speed of sound

(n1), coupling of the heat release and the pressure rise may

lead to a developing detonation front. For n¼0, a homogeneous

Corresponding author.

Manuscript received August 31, 2020; final manuscript received September 21,

2020; published online December 24, 2020. Editor: Jerzy T. Sawicki.

Journal of Engineering for Gas Turbines and Power JANUARY 2021, Vol. 143 / 011008-1

C2021 by ASME

Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/143/1/011008/6610748/gtp_143_01_011008.pdf by Tu Berlin Universitaetsbibl.im Volkswagen-haus user on 30 January 2021

5.2 Publication II: Investigation of the Fuel Distribution in a Shockless Explosion Combustor

auto-ignition is observed which is referred as thermal explosion

by Zel’dovich. A thermal explosion is a theoretical construct

which is not feasible in experiments due to unavoidable perturba-

tions in temperature, pressure, or concentration, which cause fluc-

tuations in ignition delay time. However, in case n<1, a

supersonic auto-ignition front is induced that leads to auto-

ignition of each mixture particle before its state is affected by ear-

lier ignition of neighboring mixture particles. This phenomenon

can be described as quasi-homogeneous auto-ignition. The result-

ing gradual rise in pressure is similar to one that has been

observed in constant volume combustion [4]. Similar behavior has

been observed for multiple, spatially distributed ignition sources

igniting simultaneously and, thus, leading to numerous deflagra-

tion fronts propagating toward each other resulting in a confined

volume [6]. Therefore, the regime of 0 <n<1 is aimed for in

the periodic combustion process of the SEC that is sketched in

Fig. 1.

At the beginning of each cycle, the combustor is filled with an

axially stratified fuel-air mixture (Fig. 1(a)). After a certain delay

time, this mixture undergoes a homogeneous auto-ignition leading

to a gradual pressure rise. A pressure wave is initiated at the con-

tact surface of the burned gas and the previously injected air

buffer, which travels downstream the combustor (Fig. 1(b)). The

pressure wave is reflected at the acoustically open end of the com-

bustor and travels upstream as a suction wave (Fig. 1(c)). When

the suction wave reaches the tube inlet, the pressure inside the

combustor falls below the supply pressure, which supports the

refilling process. The initial inflow is used for purging and for sep-

arating the reactant mixture of the next cycle from the hot exhaust

gases by an air buffer. Subsequently, fuel is added to the air flow

(Fig. 1(d)) resulting in a stratified fuel-air mixture to restart the

cycle.

The fuel stratification is an essential part of the SEC process

which has to be precisely realized in order to compensate for the

spatial gradient in residence time to achieve homogeneous auto-

ignition. Since the ignition delay time is a function of pressure,

temperature, and equivalence ratio, s¼sðp;T;uÞ, the local igni-

tion delay time can be adjusted by the spatial variation of the fuel

distribution when assuming constant pressure and temperature.

The goal is to adjust the ignition delay time such that the sum of

ignition delay time and residence time result in a spatially con-

stant auto-ignition time sai and therefore, leading to a quasi-

homogeneous auto-ignition of the entire combustor volume.

In order to investigate this phenomenon, an atmospheric test rig

was designed by Bobusch et al. [3]. This test rig was used by

Reichel et al. [7] to increase the homogeneity of the observed

auto-ignition by applying closed-loop control. Y€

ucel et al. [8]

modified the setup in order to improve the fuel injection process.

These efforts led to a considerable improvement in terms of

homogeneity of the auto-ignition. However, remaining spatial

deviations in ignition time prevent reliable quasi-homogeneous

auto-ignition. Therefore, the repeatable and reliable injection of a

spatially stratified fuel-air mixture into a convecting air flow is a

major challenge in the evolution of the SEC.

In this paper, two example fuel profiles are injected into a con-

tinuous air flow, allowing for a spatial fuel stratification. Measure-

ments with tunable diode laser absorption spectroscopy (TDLAS),

which has been applied to the test rig before [7], are conducted to

examine the influence of the air flow velocity uair and the injection

duration sinj on the measured fuel stratification inside the combus-

tor. Subsequently, acetone planar laser induced fluorescence

(PLIF) is used to obtain deeper insights into the role of turbulent

diffusion and boundary layer effects causing radial dispersion.

2 Experimental Setup and Measurement Procedure

The SEC test rig used in this work consists of an injection sta-

tion, a convection tube with a length of 500 mm, a quartz combus-

tor with a length of 500 mm, and an exhaust tube with a length of

1200 mm, as sketched in Fig. 2. All tubes have an inner diameter

of 40 mm. The fuel is injected into a continuous air flow by ten

circumferentially distributed ports, each equipped with an individ-

ually controlled high-speed solenoid valve (Staiger VA 204-716)

with a maximum operating frequency of 250 Hz. By adjusting the

number of open valves during the injection period, the combustor

is filled with a stratified fuel-air mixture. Preliminary investiga-

tions showed no influence of different valve patterns on the

injected fuel profile which implies a sufficiently strong mixing in

radial direction during convection. The feeding line pressure is

controlled via a high-speed dome-loaded pressure regulator (Swa-

gelok RD6) in order to minimize the pressure drop during the fuel

injection. The operating pressure is set to pfuel ¼5 bar which

matches the feeding line pressure at reacting conditions. During

the injection, the fuel valves are choked to ensure a constant volu-

metric flow rate for reactive and non-reactive conditions.

For investigations with chemical reaction under atmospheric

pressure, dimethyl ether (DME) is applied as fuel due to its short

ignition delay time under atmospheric pressure and high tempera-

ture conditions. Thus, a preheater is applied upstream of the fuel

injection station to rise the air temperature up to 1023 K. By this,

the auto-ignition of the DME-air mixture occurs approximately

65–80 ms after the start of the fuel injection depending on the

local equivalence ratio. A convection tube is mounted between

injection station and combustor to allow for compensation of the

ignition delay time such that the ignition takes place inside the

combustor. However, for future applications, different fuels and

their composites are under investigation for application at ele-

vated pressure conditions [9].

The primary goal of this work is to demonstrate that a prede-

fined fuel profile injected into a continuous air flow stays largely

preserved throughout the measurement section. For this, experi-

ments are conducted under non-reacting conditions at ambient

pressure and temperature that give insights into transport proc-

esses that affect the shape of the fuel profile. All measurements

are conducted using methane as tracer fuel as the absorption char-

acteristics of methane coincide very well with the wavelength of

available continuous wave laser for TDLAS applications. Using

Fig. 1 Sketch of the SEC cycle

Fig. 2 Sketch of the SEC test rig with TDLAS setup. The mea-

surement position is xLaser 5550 mm downstream the injection

station.

011008-2 / Vol. 143, JANUARY 2021 Transactions of the ASME

Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/143/1/011008/6610748/gtp_143_01_011008.pdf by Tu Berlin Universitaetsbibl.im Volkswagen-haus user on 30 January 2021

5 Publications

methane as tracer fuel and subsequently draw conclusions to other

fuels (e.g., DME) is reasonable. This is motivated by the fact that

the Reynolds number for all examined configurations exceeds the

critical Reynolds number for turbulent pipe flows. Thus, the trans-

port of the injected mixture profile is expected to be mainly domi-

nated by the convection process and turbulent diffusion.

Therefore, the impact of molecular diffusion is expected to be

minor.

2.1 Tunable Diode Laser Absorption Spectroscopy. Tuna-

ble diode laser absorption spectroscopy, as described by Li et al.

[10] and Bluemner et al. [11], is applied to provide a time-

resolved line-of-sight measurement of the fuel concentration

inside the combustor. Methane is used as a tracer fuel with an

absorption maximum at a wavelength of 1.65 lm, which matches

the wavelength of the available laser. A fixed-wavelength absorp-

tion sensor is used in combination with a rapid tunable diode laser

as previously applied by Reichel et al. [7]. The laser is positioned

at xLaser ¼550 mm downstream of the injection station.

The measurements are conducted by applying wavelength mod-

ulation spectroscopy at a modulation frequency of 10 kHz and a

modulation amplitude of 0.03 V. The transmitted laser signal is

measured via a lock-in amplifier which determines the first and

second harmonic of the modulation frequency simultaneously.

The first harmonic is then used to normalize the second harmonic

in order to improve sensor performance and increase the signal-

to-noise ratio. The normalized output signal scales linearly with

the fuel concentration, which is calibrated by measuring the out-

put signal for different well-known steady-state fuel mass flow

rates.

2.2 Acetone-Planar Laser Induced Fluorescence. Acetone-

PLIF is a common and widely understood tracer-LIF technique

for concentration measurements in gaseous flows [12]. It was suc-

cessfully applied to investigate the spatial fuel distribution in vari-

ous technical applications such as swirl stabilized burners [13,14],

IC-engines [15], and shock tube facilities [16]. For this study, the

acetone-PLIF setup is integrated into the SEC test rig as shown in

Fig. 3to allow for spatially resolved fuel concentration measure-

ments inside the combustor.

Methane is led through a specifically designed seeder for ace-

tone saturation. Constant pressure and temperature conditions for

all measurements ensure equal acetone concentration in the

injected fuel mass flow. The amount of methane passing through

the acetone seeder has to be restricted to achieve acetone satura-

tion of the methane for a variety of fuel mass flow rates. Hence,

well separated intensity levels over the entire range of applied

equivalence ratios (0 u1:6) can be obtained. For this, a fuel

bypass is attached which allows the seeding concentration to be

tuned according to the required fuel-acetone ratio.

Excitation at 266 nm achieved by a frequency-quadrupled

pulsed high energy Nd:YAG laser (dual cavity, heated beta bar-

ium borate-crystal) yields a high absorption cross section with a

high signal-to-noise ratio at the fluorescence emission spectrum of

350–550 nm as proposed by Lozano et al. [17]. The remaining

part of 532 nm laser light in the frequency-quadrupled pulse is

separated by two 266/532 nm beamsplitters, transmitting the

266 nm share only. Both laser cavities are used for acetone excita-

tion with an average energy of 10 mJ per pulse. The laser pulses

are separated by 20 ns and grabbed as one frame since the cumula-

tive excitation time span is small compared to the residence time

of the fluid. The emitted fluorescence signal is imaged by a

1376 1024 pixels CCD-camera in combination with an intensi-

fier with S20 photocathode and separate shutter control. The cam-

era is equipped with a 50 mm f/1.4 lens and a notch filter to block

266 and 532 nm laser light. The laser light is transformed into a

thin light sheet of 250 mm width by means of a UV compatible

laser sheet optic. The measurement plane is excited by the laser

light sheet which is adjusted parallel to the flow on the center axis

of the combustor starting 500 mm downstream of the fuel injec-

tors. The imaged area is limited to 180 mm 28 mm and centered

inside the quartz combustor to avoid blurring image data and areas

of low laser intensity.

The valve control signal is used as a trigger for the synchro-

nizer, which provides the sampling frequency for the laser and the

camera. Convection delays are taken into account to match the

start of the recording to the fuel transport from the injection sta-

tion into the measurement plane. Due to the characteristic Gaus-

sian intensity distribution across the laser sheet, the resulting

spatial fuel profile is assembled from six image sections to capture

the entire fuel concentration inside the combustor at a similar

intensity level. For this, each image (representing a part of the

fuel profile) is measured with a trigger delay ranging from 80 to

130 ms with 10 ms increments. The difference between two con-

secutive trigger delays results in a window overlap of 50% of the

respective imaged fuel concentration profile, due to a convection

velocity of 9 m/s. Hence, all images are cropped in axial direction

by 25% on each side to provide a continuous fuel concentration

profile when merging the incrementally recorded images.

In order to account for spatial inhomogeneities in laser inten-

sity, steady-state calibration measurements are conducted with

continuous fuel injection for a number of 1–10 open valves to

allocate the detected intensity levels to well-known fuel concen-

trations for each spatial position of the imaged area. This allows

for a semiquantitative fuel concentration measurement without

metering the laser intensity during the measurement. Background

images are taken in order to account for the fluorescence of the

quartz tube. Prior to the postprocessing for cycle-averaged values,

a Gaussian filter with a two pixel standard deviation is applied to

each image. The application of a 10 10 kernel size median filter

turns out to provide good results for the calculated standard devia-

tions. Both filters are applied to reduce shot noise in the images.

Using the averaged calibration images, a correction curve in axial

direction is determined. This correction curve is then used to cor-

rect for the Gaussian distribution of the laser intensity.

2.3 Measurement Procedure. For measurements at reacting

conditions, the air mass flow rate is set to _mair ¼30 kg=h resulting

in an axial flow velocity of uair ¼18 m=s at an inlet temperature

of T¼1023 K. To ensure comparability to the TDLAS measure-

ments at non-reacting conditions, an air mass flow rate of 100 kg/

h is chosen to obtain a similar flow velocity. As for reacting con-

ditions, the injection duration is set to tinj ¼30 ms.

For gas turbine application of the SEC, the ignition delay time

is decreased due to an increase in the initial pressure [18,19].

Hence, the stratification in equivalence ratio results in a smaller

maximum variation of the ignition delay time which requires the

stratified fuel-air mixture to be injected within a smaller time

span. Therefore, TDLAS measurements are also conducted for

Fig. 3 Sketch of the SEC test rig with acetone-PLIF setup

Journal of Engineering for Gas Turbines and Power JANUARY 2021, Vol. 143 / 011008-3

Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/143/1/011008/6610748/gtp_143_01_011008.pdf by Tu Berlin Universitaetsbibl.im Volkswagen-haus user on 30 January 2021

5.2 Publication II: Investigation of the Fuel Distribution in a Shockless Explosion Combustor

tinj ¼15 ms. The spatial width of the injected fuel profile, which

is estimated by the parameter vas the product of air flow velocity

and injection period

v¼uair tinj (3)

is found to have a large influence on the measured concentration

distribution inside the combustor.

For acetone-PLIF measurements, the air mass flow rate is lim-

ited to 50 kg/h due to the available seeder capacity which provides

a maximum flow velocity of uair ¼9m=s. Additional TDLAS

measurements with this reduced air mass flow rate are conducted

in order to verify the ability of the fuel injection strategy to obtain

similar fuel stratification profiles for different air flow velocities.

In order to assure similar spatial widths of the injected fuel-air

mixture, the injection duration is increased to 30 and 60 ms in the

TDLAS measurements with uair ¼9 m/s. TDLAS measurements

reveal that a decrease in vleads to less steeper gradients of the

fuel concentration profile implying a reduced preservation of the

injected profile. Therefore, PLIF measurements are conducted at

tinj ¼30 ms to examine secondary, two-dimensional mixing and

diffusion effects that cannot be resolved by line-of-sight TDLAS

measurements. The conditions for reacting measurements as well

as all measurement settings applied in this work are summarized

in Table 1.

Two different fuel trajectories are investigated: (i) an ascending

ramp ( -profile), where the number of valves is successively

increased from 1 to 10 during the injection period and (ii) a

descending ramp ( -profile), where all ten valves are opened at

the start of the injection and closed one after the other. Every

trajectory is injected with a cycle frequency of 5 Hz. Thus, each

cycle has a total length of tcycle ¼200 ms, which can be separated

in two parts: the injection and the purging time with

tcycle ¼tinj þtpurge. A total number of 100 cycles is injected for

TDLAS measurements and 300 cycles for PLIF measurements,

respectively.

3 Results and Discussion

In this section, first, the data of time-resolved line-of-sight

measurements are examined. Subsequently, these measurements

are used to validate the observations made by acetone-PLIF.

Finally, the two-dimensional measurements are evaluated, and the

findings are combined to reveal improvements for advanced injec-

tion strategies.

3.1 One-Dimensional Line-of-Sight Measurements. The

control trajectories of two example fuel profiles (left) and the nor-

malized fuel concentration ^

caveraged over 100 cycles (right) are

shown in Fig. 4. The normalized fuel concentration is defined as

the actual fuel concentration divided by the fuel concentration

measured at steady-state conditions with all valves opened. The

graphs clearly show the capability of the injection strategy to

inject the predefined fuel profile within the desired time frame

which stays largely preserved throughout the measurement

section. However, turbulent mixing and dispersion as a result of

the radial velocity distribution caused by boundary layer effects

lead to a more gradual course of the injected profile in axial direc-

tion during the convection process. To further investigate the

impact of turbulent diffusion on the fuel stratification, the air flow

velocity and the injection period are varied.

The obtained normalized fuel concentrations are shown in

Fig. 5for two different air flow velocities and two different spatial

widths. The time window visualized in Fig. 5(left) is twice the

time window for Fig. 5(right). This allows for comparison of the

spatial distribution of the measured fuel concentration. Two con-

clusions can be drawn from the profiles shown in Fig. 5: (i) when

decreasing the spatial width vof the injected fuel profile, the max-

imum normalized fuel concentration decreases, and gradients are

less steep throughout the entire measured concentration profile

and (ii) increasing the air flow velocity does not effect the shape

of the measured profile when ensuring a constant v. Reasons for

these two phenomena will be discussed in the following.

Assuming gradients of the fuel concentration in radial direction

to be minor, the evolution of the injected profile can be described

by the one-dimensional diffusion equation

@t¼D@2c

@x2(4)

with the local concentration cand the diffusion coefficient D.

Thus, the steepness of the measured gradients in concentration is

mainly a function of three parameters: the second spatial deriva-

tive of the fuel concentration @2c=@x2, the diffusion coefficient D,

and the convection time Dt. Although convection is not consid-

ered in Eq. (4), the convection time Dt¼xLaser=uair defines the

time for which the injected profile is exposed to diffusion. As

mentioned above, the flow is assumed to be fully turbulent.

Table 1 Settings of air flow velocity and duration of the injec-

tion for reactive measurement conditions, TDLAS and acetone-

PLIF measurements

Applied method uair (m/s) tinj (ms) T(K) v(m)

Reactive measurements 18 30 1023 0.54

TDLAS 18 30 293 0.54

TDLAS 18 15 293 0.27

TDLAS 9 60 293 0.54

TDLAS 9 30 293 0.27

Acetone-PLIF 9 30 293 0.27

All measurements are conducted under atmospheric pressure conditions.

Fig. 4 Control sequence (left) and normalized fuel concentra-

tion averaged over 100 cycles (right) for two example curves:

ascending and descending ramp. The air flow velocity is

uair 518 m/s, and the fuel injection time is tinj 530 ms.

Fig. 5 Time-resolved normalized fuel concentration measure-

ments. The measured profiles are shown for different air flow

velocities and fuel injection times. The x-axis is scaled to obtain

a constant fuel profile width for a constant v.

011008-4 / Vol. 143, JANUARY 2021 Transactions of the ASME

Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/143/1/011008/6610748/gtp_143_01_011008.pdf by Tu Berlin Universitaetsbibl.im Volkswagen-haus user on 30 January 2021

5 Publications

Hence, the transport is dominated by turbulent diffusion:

DDturb. In the following, the parameters @2c=@x2;Dturb, and Dt

for each setting with respect to the reference case shown in Fig. 4

are determined for the -profile. The calculated ratios are

summarized in Table 2. Analogue examination of the -curve

produces equal results.

For both applied air mass flow rates, the measured fuel concen-

tration is more distinct for larger v. This results from the underly-

ing aspect that a decrease in the spatial width vleads to an

increase in @2c=@x2by the factor of 4. The turbulent diffusion

coefficient Dturb and the convection time Dtare assumed to be

constant for a given air flow velocity uair. A numerical approxima-

tion is conducted using the discretized form of Eq. (4) with

cnþ1¼cnþdc(5)

dc¼Dc

xx dt(6)

where c

is the second discrete spatial derivative and dtrepre-

sents the time-step. The evolution of the -profile throughout the

combustion tube is visualized in Fig. 6for three different time

instances. The fuel profile at the laser position is obtained after tex

when the maximum normalized concentration of the calculated

profile is equal to the measured value of ^

cmax (Fig. 6(c)). The ini-

tial concentration profiles are shown in Fig. 6(a)for t¼0. Figure

6(b)visualizes the obtained fuel profiles for t¼tex=4 which corre-

sponds to the time instance, when the maximum value of the pro-

file with lower spatial width matches ^

cmax of the profile obtained

after tex with higher spatial width. A smaller spatial width leads to

an increase in @2c=@x2by the factor of four, resulting in a four

times faster decay in ^

cmax.

Results show that the difference in the shape of the measured

profiles for a decrease in vcan be explained by the mentioned

increase in @2c=@x2.

Additionally, a decrease in air flow velocity results in an

increase in fuel concentration due to the constant fuel mass flow

rate and a reduced dilution by the air flow. Hence, decreasing the

air flow velocity from 18 to 9 m/s with a constant spatial width

v¼0:54 results in an increasing @2c=@x2by the factor of 2.

Considering the absolute fuel concentration for uair ¼9 m/s to

be twice as large as for uair ¼18 m/s requires an increase of dcby

the same ratio (Eq. (5)) in order to obtain the same normalized

concentration profile in equal number of time steps. The convec-

tion time, however, is doubled for the mentioned decrease in uair

which leads to an increase in dtby the factor of 2. Taking this into

account allows to assess the ratio of the diffusion coefficient.

From Eq. (6), it can easily be extracted that Dis decreased by the

factor of 1/2 when reducing the air flow velocity by half. This

hypothesis is validated by numerical simulations for both investi-

gated spatial widths. These findings agree well with the available

literature. According to Speziale [20], the turbulent diffusivity can

be linked to the turbulence eddy viscosity. Following the mixing

length model from Prandtl, this viscosity scales linearly with the

mean flow velocity, and thus, Dturb is proportional to uair.

All observed effects of the investigated variations in uair and v

on the measured time-resolved fuel concentration profiles shown

in Fig. 5and can thus be explained by deviations in the parameters

of the diffusion equation (Eq. (4)). Although the numerically

approximated concentration distribution of the final performed time-

step showed overall good agreement with the measured profile of ^

the results generally indicate steeper gradients of the normalized

fuel concentration toward the bounds of the profiles. This elevated

level of diffusion in these regions of the measured distributions is

likely caused by dispersion in radial direction due to boundary layer

effects. As the data from TDLAS measurements does not allow for

analyzing these effects in further detail, a two-dimensional visual-

ization of the spatially stratified fuel concentration by acetone-PLIF

is conducted and analyzed in Secs. 3.3 and 3.4.

In order to determine the impact of the valve response on the

fuel mass flow rate, the integrated area under each measured

profile is calculated for both fuel trajectories. The results reveal

that the total amount of fuel is constant for a constant injection

duration. Moreover, normalizing the calculated value by the injec-

tion duration tinj results in a constant injected amount of fuel per

time unit for all configurations with a maximum relative deviation

of 3% from the average. Hence, the normalized volumetric fuel

flow rate does not depend on the valve control trajectory or the

injection time. Therefore, it is valid to assume the impact of the

response lags of the fuel valves on the shape of the measured con-

centration profiles to be minor.

3.2 Comparison of Tunable Diode Laser Absorption Spec-

troscopy and Acetone-Planar Laser Induced Fluorescence.

This section links the acetone-PLIF measurements to those

obtained by TDLAS. Mean fuel concentrations and cycle-to-cycle

relative standard deviations are evaluated.

In order to compare the acetone-PLIF with the line-of-sight

measurements, the intensity is averaged over the tube diameter

resulting in a 1-D-array, representing the axial distribution of the

fuel concentration (as shown in Fig. 7). This methodology con-

denses the information that can be extracted from the image data.

As mentioned earlier, the fuel concentration inside the combustor

is recorded incrementally and subsequently merged. A line-wise

polynomial fit is chosen in order to overcome discontinuities of up

to 16% at the interfaces, which arise from differences in laser

intensity and uncertainties in the assumed flow velocity. The

resulting spatial distribution of each fuel profile is then transferred

into time-resolved profiles by taking the air flow velocity into

account.

The results shown in Fig. 7indicate that both measurement

methods capture similar fuel concentration profiles in the mea-

surement section. This is illustrated by the resembling shape and

position of the measured profiles for a given injection trajectory.

Both methods show a lower maximum fuel concentration for

the -profile. This is caused by an increased time window in

which the maximum gradient in concentration is exposed to the

turbulent flow compared to the -profile. As shown in Fig. 7, the

maximum fuel concentration for profiles from TDLAS measure-

ments is obtained at two different temporal positions t

and t

due

to a fixed position of the laser. Hence, the maximum fuel concen-

tration injected for the -profile during 27–30 ms is exposed to

the turbulent flow for 70 ms whereas the maximum fuel mass flow

in the -profile, which is injected during 0–3 ms, is exposed to

turbulence for 82 ms. Thus, the gradients in concentration are less

steep for the latter case. The ratio of these exposure times matches

well with the ratio of time steps in the conducted numerical

calculations.

The fuel concentration profiles obtained with acetone-PLIF

show a greater correspondence with the injection trajectories.

When applying a line-of-sight measurement technique such as

TDLAS, the radially averaged value is measured solely, causing a

superimposition of the dispersion effect in the obtained data. This

leads to a decreased steepness in the concentration gradient than

indicated by the 2-D-images of the acetone-PLIF measurements.

Beside the cycle-averaged fuel profiles shown above, the stand-

ard deviation of the measured fuel concentration reveals important

information on the accuracy and reliability of the injection

Table 2 Parameters influencing the shape of the injected fuel

profile for the descending ramp with respect to the reference

case (uair518 m/s, tinj530 ms)

uair (m/s) tinj (ms) v^

cmax

@2c

@x2



@2c

@x2



ref

Dref

Dtref

18 30 0.54 0.66 1 1 1

18 15 0.27 0.43 4 1 1

9 60 0.54 0.66 2 1/2 2

9 30 0.27 0.43 8 1/2 2

Journal of Engineering for Gas Turbines and Power JANUARY 2021, Vol. 143 / 011008-5

Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/143/1/011008/6610748/gtp_143_01_011008.pdf by Tu Berlin Universitaetsbibl.im Volkswagen-haus user on 30 January 2021

5.2 Publication II: Investigation of the Fuel Distribution in a Shockless Explosion Combustor

strategy. Examination of the probability density function of the

measured concentration profiles revealed that the cycle-to-cycle

variation is similar to a Gaussian distribution, which legitimates

the selection of the standard deviation as characteristic parameter.

The cycle-to-cycle relative standard deviation for both example

curves (left) and -profile (right) show an increase toward the

profile edges (Fig. 8). For the -profile, the calculated standard

deviation varies less then 10% in the center of the injection profile

and up to 18% toward the edges. For the -profile, there is a high

peak at the beginning of each injection up to 38%. This remark-

able difference in standard deviation for the bounds of each curve

is again related to the different exposure time windows of the

maximum gradient in concentration in the turbulent air flow.

Comparing the results obtained by acetone-PLIF and TDLAS,

there is a fairly good agreement. The standard deviation measured

by applying TDLAS is lower in the center of each profile, whereas

the values at the bound that is linked to the start of the injection

match almost perfectly.

3.3 Two-Dimensional Concentration Measurements. The

spatially resolved distribution of the mean fuel concentration and

cycle-to-cycle standard deviation shown in Figs. 9and 10 are

assembled of six individually recorded images and subsequently

merged. To avoid non-physical discontinuities at the interfaces

due to systematic measurement errors, a linewise polynomial fit is

applied.

Both images in Fig. 9show a characteristic turbulent velocity

profile which is characterized by a higher velocity in the center of

the pipe and decreases toward the wall. Comparing both injection

trajectories, the maximum fuel concentration for the -profile is

located further upstream than for the -profile, which is in good

agreement with the findings shown in Fig. 7. The pipe flow shape,

which is visible at all axial positions for both injection profiles,

leads to a nonuniform radial fuel distribution.

The spatial distribution of the standard deviation shown in

Fig. 10 is calculated similar to the mean concentration. The stand-

ard deviation of both ramps has a quite distinct appearance in

terms of spatial distribution. For the -profile, the spatial distribu-

tion is largely constant with a maximum located at the upstream

part of the tube.

The -profile shows an almost constant cycle-to-cycle standard

deviation along the centerline across the entire fuel-air mixture

with a very distinct maximum at x=D16 that corresponds to the

highest absolute value of fuel concentration gradient in axial

direction. Furthermore, the results reveal a strong nonuniform spa-

tial distribution over the tube diameter. The standard deviation

increases toward the wall in radial direction starting from the tube

center, which illustrates the influence of the radial dispersion on

the observed cycle-to-cycle deviations of the fuel concentration

profile.

The two-dimensional concentration measurements imply that

the axially stratified fuel profile is affected by the non-uniform

radial distribution of the mean flow velocity. The fuel profiles in

Fig. 11 represent the fuel concentration averaged over the tube

diameter (equal to Fig. 7) and the concentration along the center-

line of the 2-D-images shown in Fig. 9. The profiles indicate that

the maximum fuel concentration ^

cmax is higher for the centerline

when comparing to the radially averaged fuel profile. Addition-

ally, the absolute gradients in ^

care slightly lower for the averaged

Fig. 6 Results from numerical approximations of the diffusion equation for the descending ramp with an air flow

velocity of 18 m/s inside the combustor. (a) Initial concentration profiles at t50, (b) concentration profiles at t5tex/4,

(c) concentration profile at t5tex overlayed with measured concentration profiles. The spatial distribution is calculated

by accounting for the mean air flow velocity.

Fig. 7 Mean fuel concentration measured applying TDLAS and

acetone-PLIF for uair 59 m/s and v50:27 m

Fig. 8 Cycle-to-cycle relative standard deviation from TDLAS

and acetone-PLIF for uair 59 m/s and v50:27 m

011008-6 / Vol. 143, JANUARY 2021 Transactions of the ASME

Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/143/1/011008/6610748/gtp_143_01_011008.pdf by Tu Berlin Universitaetsbibl.im Volkswagen-haus user on 30 January 2021

5 Publications

profile, which is caused by the non-constant convection velocity

across the tube diameter. As mentioned above, line-of-sight mea-

surement techniques, such as TDLAS, measure a superimposition

of the dispersion effect which implies less steep gradients in con-

centration. This corresponds well with the findings of the previous

comparisons of TDLAS measurements and numerical calculations

which reveal that the measured gradients toward the profile

bounds are less steep compared to the ones extracted from calcu-

lations. In conclusion, the injected fuel profiles stay more pre-

served throughout the measurement section than expected from

the line-of-sight measurements.

4 Conclusion

Two different fuel concentration measurement techniques,

TDLAS and acetone-PLIF, were applied in order to gain insights

into transport processes of a predefined fuel profile that has been

injected into a continuous air flow. Results obtained by both meth-

ods reveal that the injected fuel profile stays largely preserved

throughout the measurement section. Time-resolved line-of-sight

measurements indicated that the shape of the measured fuel pro-

file is mainly dependent on the spatial dimension vof the fuel-air

mixture. Furthermore, it could be shown that a reduction of the

injection duration at a constant air flow velocity of 18 m/s from

tinj ¼30 ms (equal to conditions for reactive measurements) to

tinj ¼15 ms results in less steep gradients in fuel concentration

due to an increase in the second spatial derivative of the fuel con-

centration the factor of 4. Numerical approximations of the diffu-

sion equation showed that a reduction of uair by half results in a

decrease in the diffusion coefficient by the same ratio.

To obtain deeper insights into boundary layer effects that cause

radial dispersion acetone-PLIF measurements were conducted. It

was shown that line-of-sight measurements applying TDLAS and

data obtained from acetone-PLIF measurements are in good

agreement. Furthermore, acetone-PLIF provides access to investi-

gate two-dimensional features that cannot be extracted from

TDLAS measurements. A characteristic profile of a turbulent pipe

flow was identified. Additionally, the data revealed that the actual

axial gradients in concentration along the centerline of the

combustor are steeper and thus, more preserved than expected

from line-of-sight integrated measurements. However, large

cycle-to-cycle variations near the tube walls were observed.

Although, considerable errors are included in the quantitative

examination of the reconstructed fuel profiles, acetone-PLIF

allows for the observation of two-dimensional fuel concentration

distribution that were found to reveal important information on

the two primary processes that cause less steep gradients of the

fuel distribution: (i) radial distributed mean axial flow velocity

due to the pipe flow profile and (ii) turbulent mixing near the tube

walls degrading the fuel-air stratification. These observations

allow for future improvements of the injection strategy by con-

stantly injecting fluid through the tube wall in order to separate

the fuel-air mixture from boundary layers and to minimize the

effect of the radial gradient in axial flow velocity.

Fig. 9 Mean concentration distribution of ascending and descending ramp

Fig. 10 Absolute standard deviation of the concentration of ascending and descending ramp

Fig. 11 Mean fuel concentration determined by (i) the average

along the tube diameter and (ii) centerline of each snapshot

applying acetone-PLIF

Journal of Engineering for Gas Turbines and Power JANUARY 2021, Vol. 143 / 011008-7

Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/143/1/011008/6610748/gtp_143_01_011008.pdf by Tu Berlin Universitaetsbibl.im Volkswagen-haus user on 30 January 2021

5.2 Publication II: Investigation of the Fuel Distribution in a Shockless Explosion Combustor

Acknowledgment

The authors gratefully acknowledge support by the Deutsche

Forschungsgemeinschaft (DFG) as part of Collaborative Research

Center SFB 1029. “Substantial efficiency increase in gas turbines

through direct use of coupled unsteady combustion and flow

dynamics” on project A03, as well as within the DFG project

247226395 (OB 402/4-3). Special thanks go out to Andy G€

ohrs

and Thorsten Dessin for their technical support.

Funding Data

Deutsche Forschungsgemeinschaft (DFG) (SFB 1029) (Grant

No. 247226395; Funder ID: 10.13039/501100001659).

Nomenclature

a¼speed of sound

c¼fuel concentration

c¼normalized fuel concentration

tinj ¼injection time

uai ¼propagation velocity of autoignition

uair ¼air flow velocity

Dt¼convection time

n¼parameter for propagation of autoignition

sai ¼ignition delay time of autoignition

u¼equivalence ratio

v¼parameter for profile width

References

[1] Gray, J., Lemke, M., Reiss, J., Paschereit, C., Sesterhenn, J., and Moeck, J.,

2017, “A Compact Shock-Focusing Geometry for Detonation Initiation: Experi-

ments and Adjoint-Based Variational Data Assimilation,” Combust. Flame,

183, pp. 144–156.

[2] Bluemner, R., Bohon, M. D., Paschereit, C. O., and Gutmark, E. J., 2018,

“Single and Counter-Rotating Wave Modes in an RDC,” AIAA Paper No.

1608.

[3] Bobusch, B. C., Berndt, P., Paschereit, C. O., and Klein, R., 2014, “Shockless

Explosion Combustion: An Innovative Way of Efficient Constant Volume Com-

bustion in Gas Turbines,” Combust. Sci. Technol.,186(10–11), pp. 1680–1689.

[4] Zeldovich, Y. B., 1980, “Regime Classification of an Exothermic Reaction

With Nonuniform Initial Conditions,” Combust. Flame,39(2), pp. 211–214.

[5] Gu, X., Emerson, D., and Bradley, D., 2003, “Modes of Reaction Front Propa-

gation From Hot Spots,” Combust. Flame,133(1–2), pp. 63–74.

[6] Bartenev, A., and Gelfand, B., 2000, “Spontaneous Initiation of Detonations,”

Prog. Energ. Combust.,26(1), pp. 29–55.

[7] Reichel, T. G., Sch€

apel, J.-S., Bobusch, B. C., Klein, R., King, R., and Pascher-

eit, C. O., 2017, “Shockless Explosion Combustion: Experimental Investigation

of a New Approximate Constant Volume Combustion Process,” ASME J. Eng.

Gas Turb. Power,139(2), p. 021504.

[8] Y€

ucel, F. C., V€

olzke, F., and Paschereit, C. O., 2019, “Effect of the Switching

Times on the Operating Behavior of a Shockless Explosion Combustor,” Active

Flow and Combustion Control 2018, Springer, Berlin, Germany, Sept. 19–21,

2018, pp. 121–134.

[9] Vinkeloe, J., Zander, L., Szeponik, M., and Djordjevic, N., 2019, “Tailoring the

Temperature Sensitivity of Ignition Delay Times in Hot Spots Using Fuel

Blends of Dimethyl Ether, Methane and Hydrogen,” Energy Fuels,34(2), pp.

2256–2259.

[10] Li, H., Wehe, S. D., and McManus, K. R., 2011, “Real-Time Equivalence Ratio

Measurements in Gas Turbine Combustors With a Near-Infrared Diode Laser

Sensor,” P. Combust. Inst.,33(1), pp. 717–724.

[11] Bluemner, R., Paschereit, C. O., and Oberleithner, K., 2019, “Generation and

Transport of Equivalence Ratio Fluctuations in an Acoustically Forced Swirl

Burner,” Combust. Flame,209, pp. 99–116.

[12] Schulz, C., and Sick, V., 2005, “Tracer-Lif Diagnostics: Quantitative Measure-

ment of Fuel Concentration, Temperature and Fuel/Air Ratio in Practical Com-

bustion Systems,” Prog. Energy Combust. Sci.,31(1), pp. 75–121.

[13] St€

ohr, M., Arndt, C., and Meier, W., 2015, “Transient Effects of Fuel-Air Mix-

ing in a Partially-Premixed Turbulent Swirl Flame,” Proc. Combust. Inst.,

35(3), pp. 3327–3335.

[14] Galley, D., Ducruix, S., Lacas, F., and Veynante, D., 2011, “Mixing and Stabili-

zation Study of a Partially Premixed Swirling Flame Using Laser Induced Fluo-

rescence,” Combust. Flame,158(1), pp. 155–171.

[15] Trost, J., L€

offler, M., Zigan, L., and Leipertz, A., 2010, “Simultaneous Quanti-

tative Acetone-Plif Measurements for Determination of Temperature and Gas

Composition Fields in an IC-Engine,” Phys. Procedia,5(12), pp. 689–696.

[16] Yoo, J., Mitchell, D., Davidson, D. F., and Hanson, R. K., 2010, “Planar Laser-

Induced Fluorescence Imaging in Shock Tube Flows,” Exp. Fluids,49(4), pp.

751–759.

[17] Lozano, A., Yip, B., and Hanson, R. K., 1992, “Acetone: A Tracer for Concen-

tration Measurements in Gaseous Flows by Planar Laser-Induced Fluo-

rescence,” Exp. Fluids,13(6), pp. 369–376.

[18] Djordjevic, N., Rekus, M., Vinkeloe, J., and Zander, L., 2019, “Shock Tube and

Kinetic Study on the Effects of co2 on Dimethyl Ether Autoignition at High

Pressures,” Energy Fuels,33(10), pp. 10197–10208.

[19] Burke, U., Somers, K. P., O’Toole, P., Zinner, C. M., Marquet, N., Bourque,

G., Petersen, E. L., Metcalfe, W. K., Serinyel, Z., and Curran, H. J., 2015, “An

Ignition Delay and Kinetic Modeling Study of Methane, Dimethyl Ether, and

Their Mixtures at High Pressures,” Combust. Flame,162(2), pp. 315–330.

[20] Speziale, C. G., 1991, “Analytical Methods for the Development of Reynolds-

Stress Closures in Turbulence,” Annu. Rev. Fluid Mech.,23(1), pp. 107–157.

011008-8 / Vol. 143, JANUARY 2021 Transactions of the ASME

Downloaded from http://asmedigitalcollection.asme.org/gasturbinespower/article-pdf/143/1/011008/6610748/gtp_143_01_011008.pdf by Tu Berlin Universitaetsbibl.im Volkswagen-haus user on 30 January 2021

5 Publications

5.2 Publication II: Investigation of the Fuel Distribution in a Shockless Explosion Combustor

Publication II: Summary and Contribution

Publication II

investigates the capability of the applied injection strategy of injecting

a desired fuel profile into a convecting air flow and evaluates its preservation based on

experimental data. The main objective is to identify processes that dominate the flow,

and thus, affect the fuel distribution inside the combustor.

The measurements were conducted using the “single tube SEC” as described in Ch. 2.

Two model injection trajectories were injected into a continuous air flow at a frequency

of 5 Hz. The fuel distribution inside the combustor was measured using a time-resolved

line-of-sight measurement (TDLAS), as described in Sec. 3.1 in addition to a spatially-

resolved measurement technique (acetone PLIF), as described in Sec. 3.1. By applying

both methods, temporal and spatial variations in the fuel distribution were evaluated.

TDLAS measurements were conducted at different air mass flow rates and fuel injection

times for two model injection trajectories. The measured fuel concentration profiles were

subsequently compared to calculated profiles, determined by solving the one-dimensional

diffusion equation. The results show good agreement of the measured and calculated

concentration profiles. It was found that the steepness of gradients in concentration

are impacted by turbulent diffusion and boundary layer effects, causing a “blurring” in

the measured concentration. The spatial width of the injection profiles was found to

be a driving parameter for the preservation of the injection profiles. For a sufficiently

high spatial width the injection strategy is capable of injecting a defined fuel profile

which stays largely preserved throughout the measurement section during convection.

Numerical approximations also showed that by reducing the air mass flow by a certain

value, the diffusion coefficient is reduced by the same ratio.

Spatially-resolved measurements using acetone-PLIF revealed a characteristic turbulent

velocity profile inside the combustor. This velocity profile causes a distortion of the

concentration profile in radial direction which can not be extracted from the line-of-sight

measurements. When comparing the concentration profiles extracted from the centerline

of the spatially-resolved measurements with the concentration profiles obtained from the

line-of-sight measurements, it is evident that the actual axial gradients in concentration

along the centerline are steeper and thus more preserved than expected from line-of-sight

integrated measurements.

The ability to control the fuel distribution inside the combustor is vital for the imple-

mentation of the SEC. These investigations proof that the applied injection strategy

is capable of injecting a defined fuel profile which stays largely preserved throughout

the measurement section. Therefore, this work serves as a basis for measurements

5 Publications

with chemical reaction in which a stratified fuel–air mixture was injected into an

continuous air flow using the same injection strategy. The spatial width of the in-

jection profiles is considered according to the results in this study when conducting

measurements with chemical reaction. Further discussion of the results in context with

publication III and IV will be provided in Ch. 6.

5.3 Publication III: Autoignition in Stratified Mixtures for Pressure Gain Combustion

5.3 Publication III: Autoignition in Stratified Mixtures

for Pressure Gain Combustion

Contribution:

This work investigates the correlation between the fuel distribution

and autoignition in a convecting air flow.

Methods:

Pressure sensors were applied to measure the pressure rise induced

by the autoignition. Ionization probes were used to detect the

autoignition event at distinct axial positions which allows for the

evaluation of the homogeneity of the ignition front. Additionally,

flame propagation and process variability were studied using high-

speed images of OH* chemiluminescence.

Results:

The results reveal a close correlation between the maximum pressure

rise and the time delay between the detection of the pressure rise

and the detection of the ignition front. For a decreased time

delay an increase in maximum detected pressure was observed.

Furthermore, it is shown that this value can be very well controlled

by the injected fuel trajectory. Although there is a visible cycle-to-

cycle variation, a notable shift to different pressure amplitudes for

the applied injection trajectories can be observed.

Reference:

Yücel, F. C., Habicht, F., Bohon, M., & Paschereit, C. O. (2020).

Autoignition in stratified mixtures for pressure gain combustion.

Proceedings of the Combustion Institute. URL:

https://doi.org

/10.1016/j.proci.2020.07.108

Available online at www.sciencedirect.com

Proceedings of the Combustion Institute 38 (2021) 3815–3823

www.elsevier.com/locate/proci

Autoignition in stratified mixtures for pressure gain

combustion

Fatma Cansu Yücel

∗, Fabian Habicht , Myles D. Bohon ,

Christian Oliver Paschereit

Technical University of Berlin, Institute of Fluid Dynamics and Technical Acoustics, Müller-Breslau-Straße 8,

Berlin 10623, Germany

Received 7 November 2019; accepted 12 July 2020

Available online 2 October 2020

Abstract

The reliable generation of quasi-homogeneous autoignition inside a combustor fed by a continuous air flow

would represent a milestone in realizing pressure gain combustion in gas turbines. In this work, the ignition

distribution inside a stratified fuel–air mixture is analyzed. The ability of precise and reproducible injection

of a desired fuel profile inside a convecting air flow is verified by applying tunable diode laser absorption

spectroscopy in non-reacting measurements. High-speed, static pressure sensors and ionization probes allow

for simultaneous detection of the flame and pressure rise at several axial positions in reactive measurements

with dimethyl ether as fuel. A second, exchangeable combustion tube enables optical observation of OH

∗in-

tensity in combination with pressure measurements. Experiments with three arbitrary fuel profiles show a

set of ignition distributions that vary in shape, homogeneity, and the number of simultaneous autoignition

events. Although the measurements show notable variation, a significant and reproducible influence of the

fuel injection on the ignition distribution is observed. Results show that uniform autoignition leads to a cou-

pling of the reaction front with the pressure rise and, therefore, induces a greater aerodynamic constraint

than non-uniform ignition distributions, which are dominated by propagating deflagration fronts.

Keywords: Homogeneous autoignition; Fuel stratification; Pressure gain combustion; Shockless explosion combustor

1. Introduction

Replacing the conventional isobaric combus-

tion of the gas turbine cycle by pressure gain com-

bustion is a promising concept to achieve improve-

ments in cycle efficiency. Different approaches have

been investigated, such as pulse detonation com-

∗Corresponding author.

E-mail address: f[email protected] (F.C. Yücel).

bustion (PDC) [1] and rotation detonation com-

bustion (RDC) [2] . Both concepts utilize propagat-

ing detonation wave(s), the high temperatures and

pressures of which present significant engineering

challenges. An alternative approach, called shock-

less explosion combustion (SEC) [3] , overcomes

these challenges by using quasi-homogeneous au-

toignition to achieve approximately constant vol-

ume combustion without the presence of a detona-

tion wave or mechanical constraints.

https://doi.org/10.1016/j.proci.2020.07.108

5 Publications

3816 F.C. Yücel, F. Habicht, M.D. Bohon et al. / Proceedings of the Combustion Institute 38 (2021) 3815–3823

quasi-homogeneous

reﬁlling of the tube

burned gas

stratiﬁed fuel-air mixture

burned gas

air-buﬀer quasi-homogeneous autoignition

p>p

τ= const.

expansion wave

pressure wave

autoignition

Fig. 1. Sketch of the SEC cycle.

Homogeneous autoignition occurring in well

mixed combustible mixtures was originally referred

to as thermal explosion by Zel’dovich [4] as a con-

cept of spontaneous flames, leading to an increase

in pressure similar to constant volume combustion.

However, non-uniformities in the mixture, such as

variations in temperature, pressure or equivalence

ratio, can cause deviations in ignition delay time

and result in a propagating autoignition front in-

stead. The propagation velocity of this autoigni-

tion front u

ai is inversely proportional to the spa-

tial gradient of the ignition delay time τai

. Assum-

ing constant temperature T and pressure p across

the mixture, the ignition delay time τai

is a function

of the equivalence ratio ϕsuch that u

can be ex-

pressed as:

∂τai

∂x

−1

∂τai

∂ϕ

∂x

−1

. (1)

Comparing u

to the speed of sound a leads to the

dimensionless parameter ξ, where

ξ=

= a

∂τai

∂ϕ

∂x

. (2)

This allows for the description of different combus-

tion modes [4,5] . A subsonic propagation of the re-

action front occurs when ξ> 1. For ξ= 1 the prop-

agation velocity of the autoignition front is equal to

the speed of sound, allowing amplification by cou-

pling and enabling deflagration-to-detonation tran-

sition (DDT). The ideal process of thermal explo-

sion, which occurs for ξ= 0 , is not practical in ex-

periments due to unavoidable perturbations of the

initial conditions and mixture inhomogeneity, lead-

ing to gradients in reactivity. However, ξ< 1 results

in a quasi-homogeneous autoignition leading to

an approximate constant volume combustion while

avoiding DDT. Similar behavior in pressure rise can

be observed in the case of multiple separate ignition

sources that ignite quasi-simultaneously [6] . This

regime of quasi-homogeneous autoignition and/or

many distributed ignition points is the objective for

the implementation of SEC.

Along these principles, the realization of homo-

geneous charge compression ignition (HCCI) has

been investigated intensively in the past [7] . Pre-

venting engine knock is a major challenge in HCCI

and has been addressed by applying different meth-

ods, such as modification of oxidizer and fuel prop-

erties, equivalence ratio, exhaust gas recirculation,

or engine parameters [8] . However, the implemen-

tation of this concept in a gas turbine cycle for pres-

sure gain combustion is a new approach.

The SEC is based on a periodic combustion pro-

cess as sketched in Fig. 1 . The cycle begins with a

stratified, autoignitable fuel–air mixture through-

out the combustor ( Fig. 1 (a)). This stratification

has been tailored to compensate for the gradient in

residence time, resulting in a quasi-homogeneous

autoignition ( Fig. 1 (b)). The pressure rise during

combustion induces a pressure wave propagating

downstream which is reflected as an expansion

wave when reaching the acoustically open com-

bustor outlet ( Fig. 1 (c)). When this wave reaches

the combustor inlet, the refilling process begins

( Fig. 1 (d)) and the cycle restarts.

The objective of this work is to investigate the

ignition processes within a stratified fuel–air mix-

ture. First, the ability of injecting a defined mixture

profile in a convecting air flow is analyzed. Sub-

sequently, the homogeneity of measured autoigni-

tion times and pressure rise as a function of the

fuel stratification are examined. Finally, the pro-

cesses of autoignition homogeneity, flame propa-

gation, and process variability are studied in high-

speed images of OH

∗chemiluminescence.

2. Experimental setup and measurement procedure

A sketch of the test rig for the experimental in-

vestigation of autoignition of an axially stratified

fuel–air mixture in a convecting flow is shown in

Fig. 2 . The test rig is composed of several sections,

including reactant injection, convection (0.5 m),

combustor (0.5 m), and exhaust (1 m) section. All

5.3 Publication III: Autoignition in Stratified Mixtures for Pressure Gain Combustion

F.C. Yücel, F. Habicht, M.D. Bohon et al. / Proceedings of the Combustion Institute 38 (2021) 3815–3823 3817

Fig. 2. Sketch of the test rig. Sensors: low-speed, static pressure sensors ( F

, F

), thermocouples ( T

, T

), high-speed,

static pressure sensors ( P

–P

), ionization probes ( I

–I

). The inset subfigure (b) shows the exchangeable version of the

combustor tube with optical access.

sections have an inner diameter of 40 mm. The rig

was originally designed by Bobusch et al. [3,9] and

later used by Reichel et al. [10] . However, in these

works, reproducibility was limited and consistent

homogeneous autoignition was difficult to achieve.

The control of the injection process has since been

improved by Yücel et al. [11] .

In the reacting cases, a preheater is used to

raise the temperature of the constant air flow to

1023 K measured at T

. Downstream of the pre-

heater, the air flow is forced through a restriction

in order to prevent backflow of hot gases into the

preheater due to ignition. Fuel is injected via ten ra-

dial ports with 1 mm diameter each, which are in-

dividually controlled by high-speed solenoid valves

(Staiger VA 204-716). A dome-loaded pressure reg-

ulator (Swagelok RD6) is installed upstream of

the injection station to control the fuel supply

pressure. Two static pressure sensors F

A and F

(Festo SPTW) are installed to monitor the air and

fuel supply pressures. The fuel injection duration

is t

inj

= 50 ms , and is divided into ten time win-

dows, each with a length of 5 ms. The number of

open valves is individually set for each time window

defining the injected fuel profile.

The modular setup allows for exchanging

the stainless steel combustor for a quartz tube

( Fig. 2 (b)) in order to achieve optical access. This

configuration is used for fuel concentration mea-

surements and OH

∗chemiluminescence imaging of

the ignition distribution.

2.1. Fuel concentration measurements

Fuel concentration measurements using near-

infrared tunable diode laser absorption spec-

troscopy (TDLAS) are conducted as proposed by

Li et al. [12] . These measurements are used to

validate the control of the injection geometry to

achieve a desired mixture profile within a defined

time frame. This technique has been used previ-

ously for time-resolved fuel concentration measure-

ments in a similar configuration [10,13] . While the

combustion experiments in this work are conducted

with dimethyl ether (DME) as fuel, the concen-

tration measurements are done with methane to

match the absorption features around a wavelength

of 1654 nm utilizing the available laser. In the scope

of this work, we expect the variation in the in-

jected mixture fraction profile to be primarily con-

trolled by turbulent diffusion and mixing. Since tur-

bulent fluctuations scale with the Reynolds number,

it is considered as the dominant mixing parame-

ter rather than molecular diffusion (which is of the

same order for both fuels). However, when com-

paring the Reynolds numbers for the non-reacting

(approx. 48,000) to reacting (approx. 6000) cases, it

is expected that the non-reacting cases will exhibit

much greater turbulent diffusion and a blurring of

the mixture profile. Lastly, the residence time is kept

constant for all measurements by matching the flow

velocities, allowing an equal amount of time to

diffuse. Considering this, it is reasonable to con-

clude that for the reacting cases the resulting gra-

dients are steeper than TDLAS measurements re-

veal. While this prevents quantifying the exact lo-

cal equivalence ratio for the reacting DME cases, it

does allow for a qualitative measure of the repro-

ducibility and accuracy of the injection scheme.

2.2. Reactive measurements

For reactive measurements, the initial temper-

ature is monitored via two Type-K thermocou-

ples. Five water-cooled, high-speed pressure sen-

sors are installed in the combustor with a distance

of 100 mm to record the static pressure variation.

The flame is detected via 8 ionization probes that

are mounted in the combustor and the exhaust

tube. Figure 2 shows the naming convention for

each sensor.

The temperature at the injection station remains

constant during the measurements. At the begin-

ning of each measurement, a gradient in wall tem-

perature of about 50 K between sensors T

and T

is observed. Heating during the run increases T

by approximately 50 K. However, the measurement

data show no correlation between the ignition time

throughout the measurement region and the mea-

sured wall temperature. Therefore, it is reasonable

to assume the impact of the transient wall temper-

ature on the ignition process to be negligible.

5 Publications

3818 F.C. Yücel, F. Habicht, M.D. Bohon et al. / Proceedings of the Combustion Institute 38 (2021) 3815–3823

DME is used as fuel resulting in ignition times

in the range of 60 ms to 80 ms for the applied

conditions ( p = 1 atm, T = 1023 K and 1 ≤ϕ≤2).

This assures autoignition of the convecting mix-

ture inside the combustor. The fuel supply pressure

is F

= 5 . 7 bar and the equivalence ratio is con-

trollable from ϕ = 0 (all valves closed) to ϕ = 2

(ten valves open). The average fuel mass flow rate

was measured under steady state conditions using a

Coriolis mass flow meter. To assure a gaseous state,

the fuel is vaporized and guided through a heated

pipe (330 K) before injection.

The ignition behavior of DME is characterized

by a negative temperature coefficient (NTC) region,

which is studied in more detail by Burke et al. un-

der high pressure conditions [14] . However, calcu-

lating the relevant ignition delay times with Can-

tera [15] for a zero-dimensional constant volume re-

actor using the mechanism AramchoMech2.0 that

has been validated for DME-kinetics in previous

works, reveal that all tests were conducted outside

the NTC region of DME. Operating in the NTC

region of DME would require accounting for addi-

tional non-linear behavior of DME autoignition,

and is therefore avoided.

The optically accessible section is composed of

a series of four quartz tubes, each 120 mm long,

supported by stainless steel flanges fitted with one

pressure sensor each. An optical band-pass filter

( CWL = 310 nm , FWHM = 10 nm ), an intensifier

(Lambert Instruments HiCATT) and a high-speed

camera (Photron Fastcam SA-Z) are used to detect

the reaction zones by light emission of OH

∗inten-

sity. The recorded high-speed images allow for ob-

servation of the ignition distribution at 87500 fps

and a spatial resolution of 2.7 px/mm.

3. Results and discussion

The results will be broken into two sections.

First, the control of the fuel injection profile will

be investigated, and three example contours will

be discussed. The second section will then examine

the autoignition characteristics of these profiles, fo-

cusing on the pressure rise (as representing aerody-

namic confinement) and correlate the variation in

pressure rise with direct observations of the homo-

geneity of autoignition.

3.1. Fuel injection

The mass flow rates of fuel and air are set to

match the mixture bulk flow velocity for react-

ing experiments ( u

bulk

= 18 m / s ). Each control se-

quence is injected for 150 cycles with an operating

frequency of 5 Hz. Three different injection profiles

are investigated at ambient pressure and temper-

ature: (i) –curve, (ii) V–curve and (iii) ࣵ–curve.

The control sequences and the respective TDLAS

measured, cycle averaged fuel concentration for the

three trajectories are shown in Fig. 3 .

The averaged results clearly show the capabil-

ity of replicating a desired fuel profile within the

given time span t

inj

. The measured fuel profiles

of individual cycles show a standard deviation (std)

of less than 5 % throughout the injection. There is

a clear smoothing effect, especially in the regions

of high gradient. This is expected and can be at-

tributed to two effects: (i) turbulent diffusion and

(ii) shear layer effects. Diffusion will smooth the

sharp features of the injection profile (beginning

and end of injection period). The shear layer ef-

fect near the wall causes a variation in the velocity

profile through the tube and induces a spatial dis-

tortion of the injection profile. This phenomenon

can only be measured as an integrated value across

the tube through the line-of-sight measurement. It

is also important to mention that due to inertia

of the valves, there is a hysteresis to the valve re-

sponse when opening and closing, the effects of

which are difficult to account. Currently, there is

no way to avoid these effects, and must instead

be accounted for when interpreting the reacting

results.

For reacting tests, it is important to maintain

a constant cycle-averaged fuel flow rate, otherwise

differences in pressure rise might occur due to vari-

ations in total heat release. For this, the total valve-

open time is maintained constant through the injec-

tion period and averaged to eight open valves. This

corresponds to an average equivalence ratio of the

fuel–air mixture of ϕ = 1 . 6 . The variation of the

integrated area of the measured fuel concentrations

shown in Fig. 3 (b) is within the measurement uncer-

tainty. Therefore, the reactant injection profile can

be maintained and controlled with reasonable cer-

tainty.

3.2. Autoignition control

For the following investigations, the air mass

flow is held constant at 30 kg/h, resulting in a mean

mixture bulk velocity of u

bulk

= 18 m / s . The three

injection profiles shown in Fig. 3 (a) are applied

to the reactive measurements for 150 cycles with

a firing frequency of 5 Hz. Figure 4 shows sam-

ple pressure histories of the five distributed pres-

sure sensors for a single cycle. The ignition begins

approximately 72 ms after the start of the fuel in-

jection. At peak 1 in Fig. 4 the maximum pres-

sure rise is reached and then starts decreasing due

to expansion of the burned gas in upstream and

downstream directions. Pressure waves that travel

upstream are reflected at the acoustically closed in-

let and result in a sharp rise in pressure ( 2 ). Pressure

waves propagating downstream are reflected at the

tube outlet as an expansion wave ( 3 ) that travels up-

stream. This expansion wave can be used to support

the refilling of the combustor.

5.3 Publication III: Autoignition in Stratified Mixtures for Pressure Gain Combustion

F.C. Yücel, F. Habicht, M.D. Bohon et al. / Proceedings of the Combustion Institute 38 (2021) 3815–3823 3819

0204060

tin ms

number of open valves

a) valve commands

0 20406080

tin ms

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

XCH4

b) measured injection curves

V-curve

Λ-curve

-curve

Fig. 3. Injection curve commands for V-, - and ࣵ-curve (a) and measured methane concentration 50 mm downstream

of the combustor inlet averaged over 150 cycles (b).

Fig. 4. An example pressure history during ignition event for pressure sensors P

to P

resulting from the injection of the

-curve.

For each cycle, the maximum relative pressure

increase p

max is calculated as the difference be-

tween the mean maximum pressure in peak 1 and

the pressure before ignition. The expression ‘igni-

tion time’ is introduced and is calculated from the

pressure data τp

as the time delay between the start-

ing point of the injection until the first increase in

pressure exceeding a threshold of 0.03 bar. A sim-

ilar τi

is derived from the ionization probe data as

labeled in Fig. 4 . The difference between these two

ignition times is labeled τip

. ‘Ignition delay time’

is ambiguous due to the long injection period.

The amplitude of the pressure rise is plotted

against the coherency of the ignition ( τ ip

) in

Fig. 5 for the three injection profiles. There is quite

a lot of cycle-to-cycle variance for each specific fuel

profile. However, there is a notable shift towards

higher pressure amplitudes for lower τ ip

. Also,

the three profiles tend to cluster throughout this

graph, indicating that the –curve tends towards

smaller τ ip

while the V-curve is less homogeneous.

The ࣵ–curve falls generally in between. Low τip

indicates a more quasi-simultaneous detection of

ignition and pressure rise, whereas higher τip

im-

plies a later or less homogeneous ignition com-

pared. With decreasing τ ip

, the resulting pressure

amplitude is increasing significantly. According to

Oppenheim [16] , two characteristic modes of ig-

nition can be generally observed: (i) mild ignition

and (ii) strong ignition. The latter case can be de-

scribed as a reaction front with a detonation-like

structure characterized by a sharp increase in pres-

sure. A mild ignition appears due to multiple au-

toignition sources that propagate chaotically and

individually. All observed ignitions that are shown

in Fig. 5 can be categorized as mild ignitions dif-

fering in the number of autoignition sources. This

conclusion is based on the comparison of the grad-

5 Publications

3820 F.C. Yücel, F. Habicht, M.D. Bohon et al. / Proceedings of the Combustion Institute 38 (2021) 3815–3823

Fig. 5. Pressure amplitude over delay τip

between autoignition and pressure rise for each cycle.

Fig. 6. Dominant POD mode calculated from pressure data for regime 2 (a) and regime 1 (b).

ual pressure increase ( t ≈73 ms in Fig. 4 ) subse-

quent to the autoignition when comparing to the

steep increase of the reflected wave ( t ≈76 ms in

Fig. 4 ) as suggested by Bartenev and Gelfand [6] .

However, comparing regime 1 to regime 2 shows

that under certain conditions, multiple autoignition

sources can initiate a reaction front that is more

likely to couple with the pressure rise resulting in

a greater pressure.

To characterize the pressure response within

the combustor, proper orthogonal decomposition

(POD) was applied to the pressure data of 50 cycles

for both regions identified in Fig. 5 . The dominant

POD modes of the pressure histories of sensors P

and P

are shown in Fig. 6 .

Figure 6 (a) shows a much weaker pressure wave

traveling upstream the combustor with the speed

of sound detected by sensor P

5 and later by sen-

sor P

respectively. The propagation velocity of the

ignition front is slower such that limited coupling

between the pressure wave and heat release occurs.

Figure 6 (b) shows a nearly simultaneous detection

of the pressure increase by sensors P

1 and P

. A

gradual, non-sharp (compared to a shock wave)

rise in pressure is noticeable. The period of elevated

pressure was calculated to a mean value of 7.4 and

6.7 ms for V-curve and -curve respectively with a

standard deviation of this period of 1.1 ms for both

injection profiles. The prolonged duration of pres-

sure increase results from a wider spatial distribu-

tion of the ignition event for the V-curve whereas a

more homogeneous ignition for the -curve results

in a more distinct pressure rise. Due to the higher

pressure amplitude in regime 1 , the expansion wave

that arises from the reflection at the tube outlet is

more pronounced resulting in a lower pressure for

57 ms < t < 62 ms. The period of low pressure was

calculated to be 5.5 and 5.8 ms with a standard de-

viation of 0.3 ms. These calculations indicate that

there is little correlation between the duration of

5.3 Publication III: Autoignition in Stratified Mixtures for Pressure Gain Combustion

F.C. Yücel, F. Habicht, M.D. Bohon et al. / Proceedings of the Combustion Institute 38 (2021) 3815–3823 3821

0 100 200 300

xin mm

c) V-curve, regime 2

0 100 200 300

xin mm

-1

Δtin ms

a) Λ-curve, regime 1

0 100 200 300

xin mm

b) Λ-curve

0.0

0.5

1.0

OH∗intensity

-0.33

0.00

0.33

0.66

Δpin bar

Fig. 7. x - t -diagrams for OH

∗intensity and pressure histories for P

and P

for three example shots. The spatial range of

each figure covers the downstream part Section 1 and the Sections 2 and 3 of the optical accessible combustor ( Fig. 2 (b)).

The white line represents the spatial distribution of the ignition time τo

that is defined by the OH

∗intensity. The temporal

positions of the three figures are adjusted by t = t −τo , min

the high pressure and the time span of low pres-

sure. Hence, this low pressure region is caused by

reflection of pressure waves at the acoustically open

end of the combustor and is therefore a function of

acoustic time scales only, which do not change for

the conducted measurements.

Based on the observed variations in pressure

and ionization probe histories for the ignition

events as shown in Fig. 5 , the homogeneity in au-

toignition was found to be primarily responsible for

larger pressure rises. Because the specific fuel injec-

tion profiles used in this study are not optimized

to achieve consistent, homogeneous autoignition,

they therefore show significant variation in the like-

lihood of a uniform autoignition which results in

the broadened distributions shown above.

To further explore the ignition process, the setup

with an optical access as shown in Fig. 2 (b) is used

to measure OH

∗intensity and the pressure simulta-

neously. Figure 7 shows x –t –diagrams of the OH

∗

intensity for the downstream part of section 1 and

sections 2 and 3 of the optical setup for three sam-

ple shots. Every time step in the x –t –diagrams rep-

resents an average OH

∗intensity for all pixels of

a single image at the respective axial position. 649

snapshots are aligned for each figure. The vertical

shadows in the images correspond to the combus-

tor supports and the pressure traces ( P

2 and P

)

at these locations are overlaid. The white line rep-

resents the ignition time τo for each axial posi-

tion that is defined by the OH

∗intensity exceed-

ing a threshold of 0.079. The temporal position

is aligned to the earliest ignition point within the

frame. Figures 7 (a) and (b) shows two different cy-

cles for the –curve while the Fig. 7 (c) on the right

results from measurements with the V–curve. These

three cycles have been selected to be representa-

tive of the various visible ignition events observed

for many cycles. Lastly, it is important to recall,

that there is a constant bulk flow in the positive x -

direction before ignition.

A uniform, highly distributed ignition front

with multiple points of ignition is shown in

Fig. 7 (a). Compared with the other examples, the

region after ignition shows the highest OH

∗inten-

sity. Furthermore, the largest amplitude in the pres-

sure signals are observed. The majority of the high

intensity OH

∗occurs within a period of about t =

2 ms . There also appears to be a propagation of the

gas in the upstream direction for x < 200 mm and in

downstream direction for x > 200 mm respectively.

The distributed ignition results in an aerodynamic

confinement, which serves to increase the pressure

more than in the other examples. As the products

expand, the pressure falls and eventually reaches a

local minimum for t ≈3 ms. As previously men-

tioned, the pressure wave propagating upstream is

reflected at the tube inlet and propagates down-

stream which is visible as a second pressure peak

in Fig. 7 (a) at t ≈3.5 ms. At the acoustically open

tube outlet, this pressure wave is reflected as an ex-

pansion wave that causes a low pressure region for

t > 5 ms.

The ignition front in Fig. 7 (b) is less uniform,

and instead propagates primarily from a single igni-

tion point at x ≈100 mm. From this source, the ig-

nition front propagates in both axial directions. The

propagation velocity in the lab frame reference can

be estimated using the slopes of the fronts in the x –

t –diagram as u

a , US

≈−68 m / s and u

a,DS

≈112 m/s

for upstream (US) and downstream (DS) direction,

respectively. Accounting for the bulk flow velocity

of u

bulk

≈18 m/s, a nearly constant propagation ve-

locity of u

≈−86 and 94 m/s can be found, which

suggests that the propagation of the ignition front

5 Publications

3822 F.C. Yücel, F. Habicht, M.D. Bohon et al. / Proceedings of the Combustion Institute 38 (2021) 3815–3823

0 50 100 150 200 250 300 350

std(τo)

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Δpmax in bar

Λ-curve

V-curve

-curve

regime 1

regime 2

Fig. 8. Pressure amplitude over standard deviation of the ignition time std( τo

) plotted for V-curve, -curve and ࣵ-curve.

The highlighted markers represent the three shots shown in Fig. 7 .

in Fig. 7 (b) is a propagating turbulent flame, rather

than an autoignition front.

Figure 7 (c) shows the distribution of OH

∗inten-

sity for a measurement with the V–curve. The first

ignition point occurs close to or just outside of the

image frame. Compared with the first two exam-

ples, the autoignition occurs much further down-

stream in the combustor due to the richer mixture

at the beginning of the injection. Subsequently, a

deflagration propagates upstream in the combustor

at u

a , US

≈−70 m / s in the lab frame reference. Si-

multaneously, a weaker pressure wave can be seen

in the pressure histories travelling at the speed of

sound. At t = 4 ms , a second deflagration front

is propagating downstream in the combustor which

indicates a second autoignition event has occurred

further upstream of the combustor (presumably

also due to the second high equivalence ratio re-

gion at the end of the V–curve injection profile).

In this case, the reduced aerodynamic confinement

due to multiple, separated ignition points results in

the smallest increase in pressure of the three exam-

ples. The acceleration of the flow due to the up-

stream ignition event results in the two propagat-

ing fronts, and the curved trajectories in the fixed

laboratory frame.

From Fig. 7 (a)–(c), the inhomogeneity of the ig-

nition front τo increases, which can be expressed

by an increase of the standard deviation of the ig-

nition time std( τo

). This increase in std( τo

) corre-

sponds to a decrease in the maximum pressure am-

plitude p

max

. Simultaneously, a decrease in the

∗intensity and the amplitude of the subsequent

low pressure region p

min

is visible.

Figure 8 shows p

max as a function of std( τo

)

for all measured cycles with the three injection pro-

files. The three examples presented in Fig. 7 are

highlighted. Shots with high variance in the ig-

nition front (large std( τo

)) correspond to a lower

pressure rise than those with more uniform igni-

tion. The time delay between detecting a pressure

rise and a combustion event (from either an ion-

ization probe or OH

∗), as shown in Figs. 5 and 8 ,

shows a consistent trend. When the time delay be-

tween the pressure rise and the combustion event is

coupled (as in regime 1 ), the distributed heat release

results in a greater pressure rise due to the aerody-

namic confinement compared with the uncoupled

regime 2 . Achieving this type of autoignition event

consistently and routinely is the objective of utiliz-

ing the SEC for pressure gain combustion. Towards

this aim, on-going work is focused on tailoring the

specific injection profile to maximize this pressure

rise as well as the consistency in autoignition ho-

mogeneity.

4. Conclusion and outlook

This work presented a novel combustion rig

equipped with a controllable reactant injection

system designed to deliver a prescribed, stratified

charge of autoignitable mixture. The reproducibil-

ity of the system for three injection profiles was

demonstrated. Then, the correlation between pres-

sure rise and distribution of ignition events was

shown.

It was observed that a temporal decoupling be-

tween pressure rise and flame detection by ion

probe was associated with a lower overall pressure

rise. Observing the ignition process for different ig-

nition regimes, broadly classified as regimes 1 and

2 , supported the conclusion that a distributed, uni-

form autoignition event results in a greater overall

pressure rise. This work has shown that these phe-

nomena are repeatable and measurable, which is an

important requirement for the use of a shockless

explosion combustor as a pressure gain combustion

5.3 Publication III: Autoignition in Stratified Mixtures for Pressure Gain Combustion

F.C. Yücel, F. Habicht, M.D. Bohon et al. / Proceedings of the Combustion Institute 38 (2021) 3815–3823 3823

device, where the concept of aerodynamic confine-

ment during autoignition is comparable with con-

stant volume combustion.

Using three arbitrary injection profiles in this

work demonstrated a variety of ignition distri-

butions. This work therefore serves as a starting

point for the refinement of injection profiles in or-

der to maximize the pressure rise and minimize

the variability in ignition occurrence as well as to

later implement closed-loop control of the injection

scheme. Despite the variability in the profiles stud-

ied here, it is already seen that a significant portion

of ignition events follow the process in regime 1 and

that the injection profile is very controllable. Both

are important results for progressing the controlled

autoignition and studying these combustion events

in stratified mixtures.

Declaration of Competing Interest

The authors declare that they have no known

competing financial interests or personal relation-

ships that could have appeared to influence the

work reported in this paper.

Acknowledgments

The authors gratefully acknowledge the support

of the Deutsche Forschungsgemeinschaft (DFG)

as part of Collaborative Research Center CRC

1029 “Substantial efficiency increase in gas turbines

through direct use of coupled unsteady combus-

tion and flow dynamics”. The authors also wish to

thank Andy Göhrs and Thorsten Dessin for their

technical support.

References

[1] J. Gray , M. Lemke , J. Reiss , C. Paschereit , J. Sester-

henn , J. Moeck , Combust. Flame 183 (2017) 144–156 .

[2] R. Bluemner , M.D. Bohon , C.O. Paschereit ,

E.J. Gutmark , AIAA Paper 2018-1608 (2018) .

[3] B.C. Bobusch , P. Berndt , C.O. Paschereit , R. Klein ,

Combust. Sci. Technol. 186 (10–11) (2014)

1680–1689 .

[4] Y. B. Zeldovich , Combust. Flame 39 (2) (1980)

211–214 .

[5] X. Gu , D. Emerson , D. Bradley , Combust. Flame 133

(1–2) (2003) 63–74 .

[6] A. Bartenev , B. Gelfand , Prog. Energ. Combust. 26

(1) (2000) 29–55 .

[7] F. Zhao, T.N. Asmus, D.N. Assanis, J.E. Dec, J.A.

Eng, P. M . Najt, Technical Report, SAE Technical Pa-

per , 2003.

[8] J. Eng, Characterization of pressure waves in HCCI

combustion, Technical Report, SAE Technical Paper ,

2002.

[9] B.C. Bobusch , TU Berlin, 2015 Ph.D. thesis . Fluidic

devices for realizing the shockless explosion combus-

tion process

[10] T.G. Reichel , J.-S. Schäpel , B.C. Bobusch , R. Klein ,

R. King , C.O. Paschereit , J. Eng. Gas Turb. Power 139

(2) (2017) 021504 .

[11] F.C. Yücel , F. Völzke , C.O. Paschereit , in: Active

Flow and Combustion Control 2018 , Springer, 2019,

pp. 121–134 .

[12] H. Li , S.D. Wehe , K.R. McManus , Proc. Combust.

Inst. 33 (1) (2011) 717–724 .

[13] J.-S. Schäpel , R. King , F. Yücel , F. Völzke ,

C.O. Paschereit , R. Klein , Vol. 51128, Ameri-

can Society of Mechanical Engineering, 2018 .

[14] U. Burke , K.P. Somers , P. O’Toole , C.M. Zinner ,

N. Marquet , G. Bourque , E.L. Petersen , W. K . Met-

calfe , Z. Serinyel , H.J. Curran , Combust. Flame 162

(2) (2015) 315–330 .

[15] D.G. Goodwin, R.L. Speth, H.K. Moffat, B.W. We-

ber, ( https://www.cantera.org ), 2018, Version 2.4.0.

doi: 10.5281/zenodo.1174508

[16] A.K. Oppenheim , Philos. Trans. R. Soc. A 315 (1534)

(1985) 471–508 .

5 Publications

5.3 Publication III: Autoignition in Stratified Mixtures for Pressure Gain Combustion

Publication III: Summary and Contribution

Publication III

investigates the correlation between the fuel distribution inside the

combustor and the autoignition characteristics in terms of ignition homogeneity and

pressure rise. The “single tube SEC” as described in Ch. 2 was used to investigate three

fundamentally different injection trajectories.

First, concentration measurements applying TDLAS were conducted, demonstrating

the ability of injecting model fuel profiles into the convecting air flow, which stay

largely preserved. Subsequently, these fuel profiles were investigated under reacting

conditions. Pressure transducers and ionization probes were used to measure the

ignition homogeneity and pressure rise induced by the autoignition simultaneously.

It was shown that by injecting a defined fuel profile the ignition homogeneity can

be greatly impacted. Furthermore, the results reveal that an increase in ignition

homogeneity strongly correlates with the pressure rise inside the combustor. Thus,

a more uniform and distributed ignition induces a greater aerodynamic confinement

compared to a less homogeneous ignition. Two regimes were classified with regard to

ignition homogeneity and pressure rise representing different modes of autoignition:

homogeneous and non-homogeneous.

A second combustor with optical access was used for simultaneous measurement of

the OH* intensity in combination with pressure measurements. Single ignition cycles

were investigated in terms of shape, homogeneity and number of simultaneous ignitions.

The results confirm the previous observations using the steel combustor. It was found

that an increased number of simultaneous autoignitions is accompanied by a high

pressure amplitude compared to a deflagration, which is associated with a low rise in

pressure. Although a distinct shift in terms of pressure rise can be achieved by the

proper adjustment of the injection profile, a cycle-to-cycle variance remains which can

be attributed to turbulence effects dominated by stochasticity.

This work proves that autoignition, which is mainly driven by chemical kinetics and

highly sensitive to perturbations, remains well controllable on a cycle-averaged basis.

This is indicated by the clear correlation between ignition distribution and pressure

rise which can be exploited as a basis for control algorithms. Hence, as a next step, a

control approach was developed to allow for maximizing the pressure rise through fuel

stratification. Further discussion of the results in context with

publication II and IV

will be provided in Ch. 6.

5 Publications

5.4 Publication IV: Controlled Autoignition in

Stratified Mixtures

Contribution:

This work shows the capability of controlling an autoignition event

in terms of pressure rise and ignition homogeneity using an online

optimization approach. The controller performance was analyzed

based on two different cost functions. Furthermore, single ignition

cycles were analyzed to identify mechanisms causing cycle-to-cycle

variations.

Methods:

Pressure sensors and ionization probes were used to characterize

the autoignition. The data was further used as controller input.

Chemiluminescence measurements using photomultiplier and high-

speed imaging were applied to analyze single cycles.

Results:

The results reveal that the applied control algorithm is capable of

controlling autoignition characteristics by fuel stratification. This

optimization however is cycle-averaged and thus cycle-to-cycle

variations remain. Further investigations on single cycles revealed

a complex interaction between heat release and pressure waves

influenced by low and high temperature chemistry of the applied

fuel promoting different modes of autoignition.

Reference:

Yücel, F. C., Habicht, F., Arnold, F., King, R., Bohon, M. and

Paschereit, C.O. (2021). Controlled autoignition in stratified mix-

tures. In Combustion and Flame, 232, p.111533. (open access

publication) URL:

https://doi.org/10.1016/j.combustflame.2

021.111533

Combustion and Flame 232 (2021) 111533

Contents lists available at ScienceDirect

Combustion and Flame

journal homepage: www.elsevier.com/locate/combustflame

Controlled autoignition in stratiﬁed mixtures

Fatma Cansu Yücel

∗, Fabian Habicht

, Florian Arnold

, Rudibert King

, Myles Bohon

Christian Oliver Paschereit

Chair of Fluid Dynamics Technische Universität Berlin Straße des 17. Juni 135, Berlin 10623, Germany

Chair of Measurement and Control Technische Universität Berlin Straße des 17. Juni 135, Berlin 10623, Germany

Chair of Pressure Gain Combustion Technische Universität Berlin Straße des 17. Juni 135, Berlin 10623, Germany

a r t i c l e i n f o

Article history:

Received 16 December 2020

Revised 25 May 2021

Accepted 26 May 2021

Keywords:

Pressure gain combustion

Shockless explosion combustion

Fuel stratiﬁcation

Controlled autoignition

a b s t r a c t

The shockless explosion combustion (SEC) is a recently proposed concept aiming for pressure gain com-

bustion through an unsteady process of multiple, distributed autoignitions occurring simultaneously. For

this, a stratiﬁed fuel proﬁle of dimethyl ether is injected into a continuous air ﬂow. This proﬁle is tailored

such that the mixture residence time and ignition delay time are matched, allowing multiple ignition ker-

nels to initiate simultaneously leading to an aerodynamic conﬁnement during heat release. This work ﬁrst

presents an injection strategy for injecting a deﬁned mixture proﬁle into a convection air ﬂow to control

the local equivalence ratio throughout the combustor. Line-of-sight measurements are applied to visu-

alize the concentration proﬁle and subsequently used to develop a one-dimensional tool to predict the

local equivalence ratio before ignition. Next, an extremum seeking control algorithm is applied to an ex-

isting SEC test rig to control the cycle averaged formation of different autoignition modes by optimizing

the fuel supply. Pressure and ionization probe data indicate the successful initiation of speciﬁc modes of

ﬂame propagation by adjusting the fuel injection trajectory. The previously developed simulation tool is

applied to the injection trajectories optimized by the controller. Correlating the fuel concentration distri-

bution and the obtained autoignition modes reveal that the ignition process can be very well controlled

by the fuel injection trajectory. Lastly, single representative ignition cycles are further investigated by ap-

plying optical measurement techniques to obtain OH

∗and CH

∗chemiluminescence. The results reveal a

complex interaction between heat release and pressure waves inﬂuenced by temperature-dependent igni-

tion behavior of the applied fuel. As a conclusion, four different ﬂame propagation modes are identiﬁed,

namely: turbulent deﬂagration, subsonic autoignition, supersonic autoignition and aerodynamic conﬁne-

ment by multiple simultaneous autoignition fronts.

This is an open access article under the CC BY-NC-ND license

( http://creativecommons.org/licenses/by-nc-nd/4.0/ )

1. Introduction

Implementation of pressure gain combustion (PGC) into a con-

ventional gas turbine is a promising way to achieve an increase

in thermal eﬃciency. One concept among others for realizing the

PGC is through a shockless explosion combustion (SEC). The SEC is

based on a periodic combustion process which aims for a quasi-

homogeneous autoignition inside a combustor. Figure 1 illustrates

the different phases (a, b, c, d) of a single SEC cycle. A stratiﬁed

fuel proﬁle is injected into a convecting air ﬂow ( Fig. 1 a). This

fuel proﬁle is tailored to compensate for the variations in residence

time throughout the injection duration. Therefore, the equivalence

∗Corresponding author.

E-mail address: [email protected] (F.C. Yücel).

ratio ϕis axially stratiﬁed along the combustor length, resulting in

a gradient in ignition delay time τ. Hence, simultaneous autoigni-

tion at multiple ignition locations is achieved, leading to an aero-

dynamically conﬁned volume during heat release and therefore a

rise in pressure similar to constant volume combustion ( Fig. 1 b). A

pressure wave is then observed traveling downstream in the com-

bustor. At the open end of the combustor, the pressure wave is re-

ﬂected as an expansion wave traveling upstream ( Fig. 1 c). Once the

expansion wave reaches the combustor inlet, the pressure drops

below the supply pressure, supporting the reﬁlling process and the

restart of the cycle ( Fig. 1 d).

Although autoignition as the driving mechanism for combusting

fuel is a new approach for gas turbine application, it already has

been investigated in terms of internal combustion engines, such

as diesel or homogeneous charge compression ignition engines.

https://doi.org/10.1016/j.combustﬂame.2021.111533

( http://creativecommons.org/licenses/by-nc-nd/4.0/ )

5.4 Publication IV: Controlled Autoignition in Stratified Mixtures

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

Fig. 1. Sketch of the SEC cycle.

Table 1

Modes of autoignition.

ξ> 1 : subsonic autoignitive ﬂame propagation or deﬂagration,

ξ≈1 : coupling of a reaction front with a pressure wave forming

a detonation,

0 <

ξ< 1 : supersonic autoignitive ﬂame propagation,

ξ= 0 : thermal explosion (homogeneous autoignition).

Particularly with respect to low temperature combustion, the au-

toignition provides several advantages compared to the spark igni-

tion in terms of emissions [1] .

Since autoignition is primarily driven by chemical kinetics, it is

highly dependent on the initial state of the mixture such as pres-

sure, temperature, and fuel–air ratio making it susceptible to high

stochasticity in practical approaches. Gradients in the initial state

lead to deviations in mixture reactivity and thus, induce different

ﬂame propagation modes. The ability of controlling the occurrence

of these modes would be a leapfrogging step in combustion re-

search.

The characterization of autoignition phenomena originating

from gradients in reactivity has been intensively studied in the

past in terms of detonations and explosions. An overview on this

topic is provided by Bartenev et al. [2] . This phenomenon was orig-

inally investigated by Zel’dovich in his theoretical work deﬁning

different modes of ﬂame propagation that initiate from ignition

kernels [3] . Zel’dovich deﬁned different regimes based on the gra-

dient in ignition delay time τai

resulting in different autoignition

propagation velocities u

∂τai

∂x −1

∂τai

∂ϕ

∂x

−1

. (1)

The dimensionless parameter ξ, is introduced

ξ=

= a

∂τai

∂ϕ

∂x

(2)

with the speed of sound a and the equivalence ratio ϕ. This allows

for the description of four regime classiﬁcations [3,4] , summarized

in Table 1 :

For ξ> 1 a deﬂagration or an autoignition front propagating

with a velocity below the speed of sound is observed. Propaga-

tion velocities occurring close to the speed of sound ( ξ≈1 ) might

eventually lead to a coupling of the ﬂame front and the gener-

ated pressure wave resulting in the onset of a detonation. Perfectly

homogeneous ignition (thermal explosion) is achieved for homo-

geneous initial conditions in pressure, temperature, and mixture

composition leading to a constant ignition delay time throughout

the mixture. This autoignition mode, however, is a theoretical con-

sideration corresponding to ξ= 0 . In practice, inhomogeneities in

temperature or mixture ﬂuctuations caused by turbulence lead to

gradients in reactivity, and thus, deviations in ignition time. For

ξ< 1 a supersonic ﬂame is observed, burning the reactive mixture

quasi-homogeneously, which means each particle autoignites with-

out being affected by neighboring particles. A comparable ignition

behavior is observed in the case of multiple spatially distributed

kernels igniting quasi-simultaneously leading to a conﬁned vol-

ume [2] . Both mechanisms result in a gradual increase in pres-

sure similar to constant volume combustion rather than inducing

sharp pressure peaks associated with shock waves. With his the-

ory, Zel’dovich highlighted the importance of reactivity gradients in

mixtures. In most applications these gradients are caused by tur-

bulent ﬂuctuations. For this, underlying aspects of the impact of

turbulent ﬂuctuations, which are highly stochastic in their nature,

on the autoignition process have been studied in terms of temper-

ature ﬂuctuation and turbulence intensity [5,6] .

Besides the effect of turbulence, recent studies focus on the

interaction of low and high temperature carbon chemistry lead-

ing to different combustion modes. Especially for fuels with nega-

tive temperature coeﬃcient (NTC) characteristics, such as dimethyl

ether (DME), these ﬂames show differences in ignition behav-

ior at low and high temperature conditions, resulting in either

a two-stage or a single-stage autoignition [7] . Direct numerical

simulations were performed by Luong et al. [8,9] for fuels with

NTC behavior exposed to ﬂuctuations in temperature and con-

centration. Moreover, a tendency towards the formation of so-

called double ﬂames is observed. Such double ﬂames are char-

acterized by the appearance of hot and cool ﬂame segments

simultaneously [10–13] . A detailed analysis is provided by

Ju [14] in his recent studies. However, most studies are based on

numerical simulations and experimental data is scarcely available

due to the diﬃculty of measuring temperature, pressure and mix-

ture composition simultaneously with very high resolution in time

and space. Optical measurement techniques help to gain funda-

mental insights to understand the underlying effects by allowing

for measurement of electronically excited species during hydrocar-

bon combustion [15–18] .

This work experimentally investigates the applicability of a con-

trol scheme to control the stochastic nature of an autoignition

event. First, a fuel injection strategy is presented and its ability

to generate an axially stratiﬁed fuel–air mixture throughout com-

bustor is veriﬁed by concentration measurements without chemi-

cal reactions. Subsequently, a numerical tool is introduced, which

allows for approximating the fuel concentration distribution inside

the combustor for a deﬁned fuel injection trajectory with suﬃcient

accuracy. As a next step, an extremum-seeking controller is applied

which adjusts the fuel injection trajectory to optimize the cycle-

averaged operation. The objective is to increase the probability of

a targeted combustion mode by generating the prescribed gradi-

ent in mixture reactivity. However, a certain amount of stochas-

ticity remains, which is visible in the cycle-to-cycle variation. For

this, single cycles are analyzed using optical chemiluminescence

and pressure measurements in order to better understand the

causality for the observed variations in the chemical and physical

processes.

2. Experimental setup

In the scope of this work three different –but inherently linked

– measurements were conducted using an existing SEC test rig (see

Fig. 2 ). The conducted measurements are categorized as fuel strat-

iﬁcation, autoignition control, and single cycle analysis of the ig-

nition distribution. For each measurement the combustor section

of the test rig is slightly modiﬁed in order to allow for changes in

instrumentation and optical accessibility. In this section, the over-

5 Publications

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

Fig. 2. Sketch of the test rig. Sensors: low-speed, static pressure sensors (FA, FF), thermocouples (T1, T2), high-speed, static pressure sensors (P1–P5), ionization probes

(I1–I8).

all experimental setup and the subsequent modiﬁcations made for

each measurement will be described.

The SEC test rig consists of a preheater, an injection station, a

convection tube, and an exchangeable combustor section. The con-

vection tube and combustor have a length of 500mm, respectively.

The length of the exhaust tube is 1200mm. All tubes have an inner

diameter of d = 40 mm. The injection station is designed with ten

circumferentially distributed ports, each with a diameter of 1mm.

Each port is equipped with an individually controlled high-speed

solenoid valve (Staiger VA 204–716), which is designed for an oper-

ation frequency up to 250 Hz. A high-speed dome-loaded pressure

regulator is mounted into the fuel supply line to minimize pressure

ﬂuctuations during fuel injection. While the concentration mea-

surements are conducted at non-reacting conditions with methane,

the reacting experiments are conducted with DME as fuel due to

its low ignition delay times under atmospheric pressure conditions.

2.1. Fuel stratiﬁcation

The instrumentation for investigating the fuel stratiﬁcation in

the experimental apparatus is highlighted in blue in Fig. 2 . Tun-

able diode laser absorption spectroscopy (TDLAS), as described by

Li et al. [19] and Blümner et al. [20] , is applied to the combustor at

0.65 m downstream of the fuel injection plane allowing for a line-

of-sight concentration measurement of methane in air. Methane

is used as tracer fuel as its absorption feature at 1.6537 m coin-

cides well with the wavelength of the available laser. Due to the

high Reynolds number, molecular diffusion is expected to play a

minor role. Thus, it is reasonable to conclude that gradients in

fuel concentration obtained with methane are reasonably compa-

rable to reactive measurement using DME. Wavelength modulation

spectroscopy with a modulation frequency of 10kHz and an ampli-

tude of 0.04nm is applied, resulting in a larger signal-to-noise ra-

tio compared to direct absorption spectroscopy. The output signal,

which is calculated as the second harmonic of the modulation fre-

quency divided by the ﬁrst harmonic, scales linearly with the fuel

concentration. Calibration measurements were conducted with nu-

merous well-deﬁned steady state mass ﬂow rates of fuel and air.

Hence, the applied TDLAS is used as an accurate, quantitative, and

time-resolved measure for the fuel concentration at one distinct

axial position in the combustor. After injection, the fuel proﬁle is

exposed to turbulent mixing while convecting downstream from

the injection station until reaching the measurement position.

A previous study [21] demonstrated that the main effects on

the fuel concentration proﬁle can well be captured by TDLAS mea-

surements. Further, it was shown that the evolution of fuel con-

centration in the combustor is dominated by diffusion in the axial

direction, which was successfully replicated by numerical simula-

tions of the one-dimensional diffusion equation given by

∂c

∂t

= D

∂

∂x

. (3)

Here, tand x are the respective variables in time and space, cis the

local fuel concentration, and D is the diffusion coeﬃcient. Compar-

ing the numerical results to the measured fuel concentration pro-

ﬁles revealed that D linearly depends on the mean ﬂow velocity

air

. Moreover, the spatial width of the injected fuel–air package

was observed to largely affect the attenuation of gradients in the

fuel concentration distribution.

In the scope of this work, fuel concentration measurements

were conducted, at different air ﬂow temperatures starting from

293 K up to 800 K with 100 K increments. This allows for evalu-

ating the transferability of the mentioned ﬁndings from investiga-

tions at ambient temperature conditions to combustion measure-

ments, which are conducted at elevated temperatures. To ensure a

constant convection time for all measurements, the air mass ﬂow

rate ˙

air

is adjusted for each temperature setting to provide a con-

stant volumetric air ﬂow rate, and thus, a constant air ﬂow velocity

air

= 18 m/s. This velocity was chosen to match the reference case

with chemical reaction ( T = 1023 K, p = 1 atm and ˙

air

= 30 kg/h).

By this, the impact of the Reynolds number, and thus, turbulent

ﬂuctuations can be evaluated. An internal proportional-integral-

derivative controller allows a maximum ﬂuctuation in air mass

ﬂow rate of about 0.5kg/h. The air temperature is measured at

T1 by a coated thermocouple (type K), which extends into the air

ﬂow by approximately 10 mm. The fuel supply pressure is set to a

constant value of p

fuel

= 4 . 5 bar for all measurements ensuring a

constant injection velocity. For a given combustor diameter the re-

quired mass ﬂow rate ˙

air

can be calculated for each applied tem-

perature as

air

= ρair

(T )

air

πd

. (4)

With an increasing temperature, the density ρair

(T ) decreases

while the dynamic viscosity ηair

(T ) increases resulting in an over-

all reduced Reynolds number Re

air

ρair

(T )

air

ηair

(T )

. (5)

For each applied temperature, three different fuel injection tra-

jectories, shown in Fig. 3 , are injected within a constant injection

duration of t

inj

= 30 ms. Each trajectory is deﬁned by ten time win-

dows with an independently set number of open valves. A series

of 150 injection cycles is performed for every applied trajectory

with a frequency of 5 Hz. In order to examine the inﬂuence of the

air temperature on the evolution of the fuel concentration, the ob-

5.4 Publication IV: Controlled Autoignition in Stratified Mixtures

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

Fig. 3. Normalized fuel concentration at the injection station for an ideal injection and the applied model. The parameters of the valve response were optimized to achieve

optimum agreement of the model results and the measured proﬁles.

tained proﬁles for a given injection trajectory are analyzed for a

range of temperature conditions.

To allow for the assessment of the fuel distribution inside the

SEC for an arbitrary fuel injection trajectory, a tool for the numer-

ical calculation of the spatial distribution of the fuel concentration

was implemented based on the work published in [21] . The tool

aims for reproducing three mechanisms: valve response, diffusion

effects and convection time. The calculations are performed for the

normalized fuel concentration ˆ

c , which is deﬁned as the actual fuel

concentration cdivided by the concentration obtained when con-

tinuously operating a single valve.

The fuel injection in the experiments is affected by inertia in

the valve response. Thus, the fuel concentration at the injection

station was modeled by a polynomial Bézier curve allowing for a

gradual change in the injected fuel ﬂow rate. The initial opening

speed and the time until the valve is completely open are used as

parameters for the adjustment of the valve behavior. Additionally,

dead-times for opening and closing are applied. The parameters

were identiﬁed to replicate all measurement data by the simula-

tion with different fuel injection curves. The normalized fuel con-

centration ˆ

c at the injection station is shown in Fig. 3 for an ideal

(black) and the modeled (red) fuel injection, respectively. Three

different fuel injection trajectories are shown: top hat, local maxi-

mum and local minimum curve.

Equal parameter values for the polynomial Bézier curve and the

dead-times were applied for all injection trajectories. The shown

proﬁles indicate that the triggering of the valves results in a grad-

ual increase in fuel mass ﬂow after a certain dead time. The same

can be observed for the closing of the valves. For all trajectories,

this results in a certain skewness of the injected fuel proﬁle.

To model dispersion effects, the numerical simulation of the

one-dimensional diffusion equation ( Eq. (3) ) was implemented on

a grid with a spatial resolution of x = 10 mm. In a preliminary

study with varying spatial resolution of the simulation area, this

grid size was found to be suﬃcient to obtain accurate results while

minimizing the computational effort. The spatial derivative ∂ /∂ x

was implemented by a central difference scheme. The time steps of

d t = 0 . 05 ms are discretized by a 4th order Runge–Kutta scheme.

For a given time step t

, the fuel concentration at the injection sta-

tion ( x = 0 m) is increased by the appropriate value obtained from

the valve opening. Subsequently, the spatial distribution in fuel

concentration for the next time step t

i +1

= t

+ d tis calculated. For

this, ﬁrst, the simulation domain is shifted by d x = ¯

air

d taccount-

ing for convection due to the mean ﬂow velocity. Then, the fuel

concentration proﬁle of the next time step is determined by ap-

plying the discretized diffusion equation. The diffusion coeﬃcient

D was determined iteratively to maximize the congruence of the

fuel concentration proﬁles for simulations and experiments.

The temporal evolution of the fuel proﬁle was generated by

observing the fuel concentration at a ﬁxed spatial position ( x =

0 . 65 m) during the entire simulation time. This procedure is simi-

lar to the line-of-sight measurement using TDLAS and is therefore

expected to produce similar results for a given fuel proﬁle.

2.2. Autoignition control

An extremum seeking control algorithm is deﬁned to ﬁnd the

optimum fuel injection trajectory to achieve reliable operation.

Two different cost functions are used to formulate the target of the

controller based on cycle-averaged measurement data of pressure

or ionization probes.

In this section, the injection optimization algorithm is intro-

duced and subsequently applied to the SEC test rig in Section 3.2 .

We basically extended the idea of a cyclic extremum seeking con-

trol [22] to the integer-constraint case and applied the Broyden-

Fletcher-Goldfarb-Shanno (BFGS) scheme with gradient informa-

tion estimated from measurements for the underlying optimiza-

tion. As the integer-valued actuation is an uncommon restriction

for extremum seeking control, the applied approach is brieﬂy sum-

marized in the following.

The fuel injection curve of a batch k , deﬁned by n steps N

k,i

can be described by the vector

N

k, 1

, . . . , N

k,i

, . . . , N

k,n

T

. (6)

Accordingly, the pressure signals p

k,j

(t) of injection period k are

functions of the injection curve n

k,j

(t) = f

) ∀ j ∈ { 1 , 2 , 3 , 4 } (7)

The pressure rise can be quantiﬁed by a cost function

= J

p

k, 1

(t) , p

k, 2

(t) , p

k, 3

(t) , p

k, 4

(t)

= J

(

) (8)

1 Note that in this section we describe the approach only for pressure signals

for easier readability. Nevertheless, the approach can be extended to all parame-

ters that are inﬂuenced by the injection curve n

. An example is the ignition time

detected by the ionization probes as described below.

5 Publications

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

that maps the pressure signals to a scalar value J

. The online op-

timization aims to minimize the cost J

over multiple injection

batches.

As a suﬃcient model for the mapping of the fuel injection curve

to the pressure signals is not available, an approximation of the

cost function is necessary. For such an approximation we consider

a Taylor series up to the second order term

J(n

k

) ≈J(n

) + ∇J(n

)

n +

n

)n (9)

with the deviated fuel injection curve n

k

= n

+ n . In Eq. (9) ,

∇J(n

) denotes the gradient of the cost function at the fuel injec-

tion curve of batch k and H

) denotes the corresponding Hes-

sian matrix. Without any model information, an analytical calcula-

tion of the gradient and the Hessian matrix is impossible. Hence, a

measurement-based estimation is required. For that, the fuel injec-

tion curve from batch k is perturbed in the upcoming 2 ·n batches.

The fuel injection curves of these 2 ·n batches are deﬁned by

k +2 i −1

N

k, 1

, . . . , N

k,i

+ 1 , . . . , N

k,n

= n

k,i

(10)

k +2 i

N

k, 1

, . . . , N

k,i

−1 , . . . , N

k,n

= n

−

k,i

. (11)

Accordingly, the pressure signals

k,j,i

(t) = f

k,i

) ∀ j ∈ { 1 , 2 , 3 , 4 } (12)

−

k,j,i

(t) = f

−

k,i

) ∀ j ∈ { 1 , 2 , 3 , 4 } (13)

are obtained for the corresponding fuel injection curve deﬁned by

Eqs. (10) and (11) . The pressure measurements are subsequently

used to calculate the costs

k,i

= J

p

k, 1 ,i

(t) , p

−

k, 2 ,i

(t) , p

k, 3 ,i

(t) , p

k, 4 ,i

(t)

= J

n

k,i

(14)

−

k,i

= J

p

−

k, 1 ,i

(t) , p

−

k, 2 ,i

(t) , p

−

k, 3 ,i

(t) , p

−

k, 4 ,i

(t)

= J

n

−

k,i

. (15)

After the perturbation phase, before batch k + 11 the gradient at

the fuel injection curve from batch k can be calculated using cen-

tral difference scheme:

∇J (n

) ≈1

⎡

⎢

⎣

k, 1

−J

−

k, 1

k,i

−J

−

k,i

k,n

−J

−

k,n

⎤

⎥

⎦

. (16)

In what follows we will refer to those batches where a new gra-

dient information is available as control steps with the counter

p. Note that if a perturbation as deﬁned in Eqs. (10) and (11) is

not possible due to the restricted number of valves, both the per-

turbation and the ensuing calculation of the gradient, have to be

adapted appropriately.

To avoid unnecessary delay during the experiments, as well as

to ensure a quick response to disturbances in the operation condi-

tion, the second order derivative is estimated iteratively from the

gradient information without applying a central difference scheme.

For that, the common BFGS algorithm as described in [23] is ex-

ploited. The Hessian matrix in control step pcan be approximated

) ≈(I −μz (n

p−1

)

) ·H

p−1

) ·

(I −μ(n

p−1

) z

) + μzz

T (17)

with

n

p−1

= n

−n

p−1

z = ∇J(n

) −∇J(n

p−1

)

μ=

(n

p−1

)

This estimation is subsequently applied within the approximated

cost function from Eq. (9) .

Thus, the optimal adjustment of the fuel curve in control step

pis obtained by solving the integer-valued quadratic program:

n

= arg min

n n

) n + ∇J(n

)

n (18)

subject to n ∈ N

n (19)

A n ≤c

. (20)

Besides the restriction to an integer domain, Eq. (20) is used to

incorporated additional linear inequality constrains like the mini-

mum and maximum averaged number of opened valves in one in-

jection cycle as well as the restricted number of valves. The con-

straint for the average number of opened valves is translated to

linear constraints

(

)

p+1

≤n ·8 and −(

)

p+1

≤−n ·6 (21)

with respect to a fuel injection curve in the upcoming control step

p + 1 . In Eq. (21) , 1

n denotes a size n column vector of ones. Ac-

cordingly, the limited number of valves for the fuel injection can

be expressed by

p+1

≤1

·10 and −I

p+1

≤1

·0 (22)

with the identity matrix I

n of size n ×n . Eqs. (21) and (22) is

stacked together to obtain the linear constraint

⎡

⎢

⎣

−I

(

)

−(

)

⎤

⎥

⎦

  

p+1

≤⎡

⎢

⎣

·10

·0

n ·8

−n ·6

⎤

⎥

⎦

  

. (23)

As the quadratic program in Eq. (18) is formulated for the adjust-

ment of the fuel injection curve n between the control steps, the

constraint from Eq. (23) is transformed to a formulation compliant

to Eq. (20) :

A n

p+1

≤c (24)

A (n

+ n ) ≤c (25)

A n ≤c −A n

  

. (26)

The quadratic program in Eq. (18) can thus be applied to optimize

the fuel injection curve from one control step to the next.

The experimental setup as described earlier is modiﬁed to al-

low for reactive measurements under atmospheric pressure con-

ditions. For this, the optically accessible combustor is exchanged

for a stainless steel combustion chamber, as shown in Fig. 2 . DME

is used as the fuel due to its suﬃciently low ignition delay times

( 45 ms < τign

< 80 ms) under high temperature ( T

air

= 1023 K) and

atmospheric pressure conditions for the applied equivalence ratios

( ϕ ∈ [1 . 15 , 1 . 45] ). In order to assure gaseous injection, DME is led

through a vaporizer and guided through a heated fuel line at 330K.

The fuel supply pressure is set to p

fuel

= 4 . 5 bar which matches the

supply pressure for non-reactive measurements. By this, the volu-

metric ﬂow rate of the fuel is kept constant in both reacting and

5.4 Publication IV: Controlled Autoignition in Stratified Mixtures

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

Fig. 4. Optically accessible combustor and measurement setup consisting of a high-

speed camera equipped with an intensiﬁer and an image doubler, four PMTs, a

spectroscope, pressure sensors P1–P4 and ionization probes I1–I5.

non-reacting cases. The air mass ﬂow rate is set to a steady state

value of ˙

air

= 30kg/h resulting in a ﬂow velocity of 18m/s for a

temperature of 1023K.

Two static pressure sensors (Festo SPTW) are installed to mon-

itor the fuel (labeled FF ) and air (labeled FA ) supply pressure dur-

ing each measurement. Four high-speed, water-cooled, static pres-

sure sensors (Kulite EWCTV-312) are mounted inside the combus-

tor with 100 mm distance to measure the pressure rise subsequent

to the ignition event. Seven ionization probes are installed to lo-

cate the ignition front inside the combustor. The data is sampled

with an acquisition rate of 10 kHz.

The test rig is operated with a frequency of 5Hz. At the be-

ginning of each cycle a well-deﬁned fuel proﬁle is added to the

continuous air mass ﬂow. The total injection duration is 30 ms for

all measurements. This duration is subdivided into injection sec-

tions of either 3 or 6 ms, resulting in either 10 or 5 valve operation

steps, respectively (example given in Fig. 3 for 10 operating steps).

The average number of open valves during the injection period is

limited to a range between 6 and 8. By this, the global fuel mass

ﬂow rate during each cycle is restricted. For each fuel injection tra-

jectory applied by the controller, the test rig is operated for 40

cycles of which the last 30 are used to calculate a cycle-averaged

value for the given cost function. The applied 40 consecutive cycles

are referred to as batch. A constant fuel proﬁle with seven valves

open is chosen as an initial injection trajectory for the ﬁrst batch.

2.3. Single cycle analysis of distributed autoignition

To allow for a detailed analysis of single cycles, chemilumines-

cence measurements are conducted. For this, the setup is again

slightly modiﬁed by integrating an optically accessible combustor

assembled from four quartz tubes each with a length of 120 mm

and held together by ﬁve ﬂanges as shown in Fig. 4 . A highspeed

camera (Photron SA-Z) and intensiﬁer (Lambert Instruments) in

combination with an image doubler is used to record the two-

dimensional line-of-sight chemiluminescence distribution in the

measurement section around two center wavelengths (CWL) simul-

taneously. 307 nm was chosen to capture the light emission by

∗species and a 435 nm slightly off the main CH

∗peak emission

feature allows for the detection of CH

∗and formaldehyde simul-

taneously. Note that the two-dimensional measurements are over-

layed with broad band emission spectrum of CO

. 700 snapshots

are recorded with a sample rate of 50,0 0 0 Hz resulting in a total

recording time of 14 ms.

In addition to the two-dimensional measurements, time-

resolved ﬁber-coupled line-of-sight measurements were con-

Table 2

Emission spectra of excited species during the com-

bustion of dimethyl-ether in air [24,25] .

Emitting species Emission wavelength in nm

∗350–505

∗300–600

∗307

∗431

∗470, 516

ducted. For this, a multi-connector optical cable is positioned in

the center of the measurement section (see Fig. 4 ). This enables

chemiluminescence measurements for multiple wavelengths at one

distinct position. The multi-connector is connected to four photo-

multiplier (PMT) sensors and a spectroscope, respectively. The PMT

data is sampled at a sample rate of 10 kHz, while the spectroscope

operation frequency is limited to 5 Hz. The ﬁlter wavelengths be-

fore each PMT are 307 nm, 343 nm, 430 nm and 450 nm. The sig-

nals obtained at wavelengths 343 nm and 450 nm are used to cor-

rect the CH

∗and OH

∗emissions from the broadband CO

∗and CH

∗emission. Four highspeed pressure sensors as previously applied

are mounted into the ﬂanges to measure the pressure rise subse-

quent to the autoignition.

During the combustion of DME in air, several light emitting

species are excited, including CH

∗, CO

∗, OH

∗, CH

∗, and C

∗.

In the emission spectrum CH

∗and CO

∗are characterized by

broad-band emission feature, while OH

∗, CH

∗, and C

∗show dis-

tinct single or multiple peaks (Swan band) in the emission spectra.

The characteristic wavelengths are summarized in Table 2 .

The formation of formaldehyde has been observed during low

temperature combustion (LTC) [24] and is used as an indicator for

the onset of a ﬁrst-stage ignition [18,26] . Its emission spectrum

overlaps the emission features of other excited species such as CH

∗

and C

∗. For temperature ranges below 10 0 0 K emitted chemilu-

minescence around the wavelengths of CH

∗can be generally in-

terpreted as formaldehyde formation while CH

∗and OH

∗are typ-

ically observed during high temperature combustion (HTC) [18] .

Thus, the appearance of CH

∗in absence of OH

∗may reasonably be

interpreted as formaldehyde emission. The formation of C

∗pri-

marily occurs under fuel rich conditions. Additionally, according to

Becker et al. [27] , so-called hot and cold OH

∗are formed, differing

mainly in the chemical path by which they are formed. Hot OH

∗is

mainly formed by the reaction of CH with molecular oxygen. OH

∗

formed without the presence of CH molecules is therefore consid-

ered cold OH

∗.

3. Results

In this section ﬁrst a simulation tool is developed based on the

TDLAS measurements described in Section 2.1 for replicating the

fuel distribution inside the combustor for an arbitrary injection tra-

jectory. Next, the control approach introduced in Section 2.2 is ap-

plied to the SEC test rig and the controller performance is evalu-

ated based on the optimization of two different cost functions. Fi-

nally, a detailed analysis of single representative cycles is provided

based on chemiluminescence data as described in Section 2.3 .

3.1. Fuel stratiﬁcation

The measured fuel concentrations for three model injection tra-

jectories, introduced in Fig. 3 , were compared at different temper-

atures which revealed no signiﬁcant affect of the temperature on

the steepness of the gradients in concentration within the consid-

ered range. Hence, it is assumed that the diffusion processes which

5 Publications

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

Fig. 5. Temporal evolution of the measured and modeled normalized fuel concentration at the measurement position for TDLAS.

are predominating the ﬂow are mainly independent of the tem-

perature conditions. Therefore, when injecting a given trajectory in

reactive experiments, a similar distribution in fuel concentration

throughout the combustor as obtained by TDLAS measurements is

expected.

Further, the measured fuel concentration proﬁles are used to

validate the model parameters to predict the fuel concentration

distribution in the combustor for an arbitrary injection trajectory,

as discussed in Section 2.1 . A comparison of the measured fuel pro-

ﬁles with the modeled fuel concentration at the measurement po-

sition is shown in Fig. 5 .

Excellent agreement can be seen for the ﬁrst and second fuel

injection trajectories. Slightly greater deviations are observed for

the third curve due to increased curvature of the fuel proﬁles re-

sulting in promotion of inaccuracies. However, the main features

including the position and amplitude of the maximum fuel con-

centration are well reproduced by the simulations. The presented

results show, that the numerical model allows for the prediction

of the fuel concentration distribution inside the combustor. In the

following, this tool is used to estimate the fuel distribution inside

the SEC before ignition in measurements with chemical reaction.

In these cases, optical fuel concentration measurements cannot be

applied, which makes the reliable simulation of the injection pro-

cess a valuable tool for the interpretation of measurement data at

these conditions.

3.2. Autoignition control

The control algorithm formulated in Section 2.2 is applied to

the test rig aiming for a maximization of the pressure rise through

distributed autoignition. Earlier observations showed a distinct cor-

relation between the maximum pressure and the standard devi-

ation of the autoignition front [28] . A low variation in ignition

times across the combustor indicate a higher number of igni-

tion points that occur simultaneously and/or a fast propagating

ﬂame front(s). This distributed ignition was found to result in an

aerodynamic conﬁnement leading to an overall increase in pres-

sure amplitude. Based on these preliminary observations, a con-

trol approach is developed that optimizes the fuel injection as a

function of time-resolved output signals of ionization probes and

pressure sensors. The ﬁrst optimization goal is to maximize the

pressure rise subsequent to autoignition whereas the second opti-

mization goal is to maximize the homogeneity of the autoignition

front.

The controller performance is analyzed based on two different

cost functions J

1 and J

. The ﬁrst cost function is deﬁned by

(

)

= −1



m =11

sum

p

j, max

(

, m

)

(27)

with p

j, max

(

, m

) denoting the maximum pressure amplitude of

sensor jfor a given trajectory n

in cycle m . sum

m indicates the

sum of these amplitudes obtained by all sensors for cycle m . This

approach aims at maximizing the cycle-averaged peak pressure

amplitude generated by the autoignition events. The second ap-

proach aims at the simultaneous detection of the ignition front by

the installed ionization probes. The cost function J

is deﬁned by

(

)



m =11

std

τio ,j

(

, m

)

(28)

with τio ,j

(

, m

)

representing the time when the reaction front in

cycle m is detected at ionization probe j, and std

m the standard

deviation among these seven ionization probes. This approach aims

for maximizing the homogeneity of the ignition event.

Figure 6 illustrates the maximum pressure amplitude J

1 ob-

tained by each sensor for all applied control steps. The order of the

detection indicates whether a propagating pressure wave or simul-

taneous rise of the pressure throughout the combustor is achieved.

The presented pressure data is averaged over 30 cycles. These data

are complemented by the value of standard deviation of the igni-

tion time obtained by the ionization probes J

( n )

as a measure

of the autoignition homogeneity with respect to the control step

k . Comparing the graphs for the two cost functions it can be seen

that the choice of the optimization parameter has a signiﬁcant

impact on the controller performance. As intended, using J

results

in an overall increase in pressure amplitude throughout the entire

operation duration. Simultaneously, the adjustment of the injection

trajectory by the control algorithm leads to an increase in the ho-

mogeneity of the ignition event. From this, it is observed that large

pressure amplitudes tend to correlate with an increased ignition

homogeneity as indicated by the overall decrease in the variation

of the observed ignition times. The second approach, which focuses

on directly minimizing the standard deviation of the observed

ignition times, turned out to be less effective and routinely failed

to monotonically decrease the cost function. Here, a signiﬁcant

increase in the ignition homogeneity is only observed from control

step 3 to 4, and from 7 to 8, respectively. Simultaneously, this

decrease in J

2 is correlated to a clear increase in the pressure am-

plitude, which agrees well with the found correlation for measure-

5.4 Publication IV: Controlled Autoignition in Stratified Mixtures

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

Fig. 6. Maximum pressure measured by pressure sensors P1–P4 (black line) and standard deviation obtained by ionization probes I1–I7 (red line) for each control step

averaged over 30 cycles when minimizing cost function J

a) and J

b). (For interpretation of the references to color in this ﬁgure legend, the reader is referred to the web

version of this article.)

Fig. 7. Maximum pressure over standard deviation of the ignition timing for each cycle optimizing cost function J

. The chosen control steps 1, 4, 3 and 7 represent a broad

range of values of the cost functional. (For interpretation of the references to color in this ﬁgure legend, the reader is referred to the web version of this article.)

ment with J

. However, control steps 3 and 7 result in less homo-

geneous ignitions with a minor increase in pressure. Presumably,

the controller performance becomes more sensitive to stochasticity

of the system when applying cost function J

, which makes it in-

capable of converging to a solution. However, in what follows, the

results will help to shed more light upon the underlying processes.

To better understand the causality of the applied control tra-

jectory and the resulting ignition behavior, four example control

steps (1, 3, 4 and 7) from Fig. 6 b that distinctly differ in their ob-

tained pressure amplitudes are further examined. The correlation

between maximum pressure amplitude and standard deviation of

the ignition time is shown in Fig. 7 for each cycle when applying

the respective fuel injection trajectories. A small standard deviation

in τio ,j

indicates a more homogeneous ignition which is correlated

to a greater rise in pressure. Although there is a notable cycle-to-

cycle variation for a given injection trajectory, a distinct shift of

the data points by the application of different trajectories can be

observed. For example, the trajectory injected during control step 4

results in a more homogeneous ignition and higher pressures com-

pared to control step 3 or 7. Additionally, both injection trajectories

are scattering throughout the x -axis, which will be discussed later.

Control step 1 shows intermediate values for both, pressure ampli-

tude and ignition homogeneity. The cycle-to-cycle variation is as-

cribed to a stochastic process that cannot be predicted. Therefore,

the calculation of the cost function based on cycle-averaged values

is necessary.

To further investigate the results, the respective fuel injection

trajectories applied by the controller are correlated to the observed

ignition characteristics. For this, the estimated fuel concentration

distribution inside the combustor at the time of the ignition is

5 Publications

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

Fig. 8. Applied fuel injection trajectory (black), calculated fuel proﬁle using the simulation tool (shaded area) and gradient in concentration (red) for each control step. The

simulation time was set to match the respective ignition timings. The average number of open valves during the injection duration representing the global fuel mass ﬂow in

one cycle is indicated. (For interpretation of the references to color in this ﬁgure legend, the reader is referred to the web version of this article.)

determined by applying the previously introduced 1D simulation

tool. Figure 8 visualizes the injected trajectories for each applied

control step shown in Fig. 6 b. The fuel distribution is visualized by

the shaded area. The average number of open valves during the

injection period, represented by a horizontal dashed line, varies

between 6.2 and 8, which corresponds to the restriction of the

total number of open valves. The red line shows the gradient in

fuel concentration along the tube axis. A constant injection trajec-

tory with seven valves open (control step 1) is given as a starting

point for the controller, resulting in a broad region of gradually

increasing fuel concentration along the tube axis. As the ﬁrst op-

timization step (control step 2) the controller increases the total

fuel mass ﬂow rate without changing the shape of the injection

pattern. This leads to an increase in the maximum pressure, while

resulting in similar values for the standard deviation of the igni-

tion timing, as shown in Fig. 6 . The injection trajectory for con-

trol step 3 differs remarkably from the previously applied cases.

The resulting fuel concentration proﬁle contains two distinct max-

ima and a local minimum in the center of the proﬁle. As shown

in Fig. 6 , the measured data reveals a greatly reduced maximum

pressure. The application of the injection trajectory in control step

4 results in the highest measured increase in pressure while hav-

ing the lowest total fuel mass ﬂow rate together with control step

3. The ignition timings indicate a more homogeneous ignition. The

fuel distribution is characterized by a single distinct peak followed

by a region with a smooth decrease in fuel concentration. This in-

duces a gradual increase in fuel concentration along the tube axis

inside the combustor. Control step 5 induces a very large maxi-

mum value with rather steep gradients. This distribution results in

a signiﬁcant decrease in pressure compared to control step 4, al-

though a high value of averaged number of open valves leads to

large amount of injected fuel. Control step 6 has a ﬁrst ’bump’ in

fuel concentration followed by a dominant global maximum. Com-

pared to control step 5 the total fuel mass ﬂow rate is slightly

lower. Nevertheless, there is a notable increase in pressure with

a slight increase in homogeneity. The fuel concentration distribu-

tion for control step 7 shows similar features as the control step

3 although the respective injection trajectories differ signiﬁcantly.

Therefore, both cases result in comparable ignition behavior, as in-

dicated by the similar distributions of the respective data in Fig. 7 .

Most likely, the downstream peak in fuel concentration triggers

an early ignition leading to a pressure wave that does not couple

with the ﬂame front. The second peak ignites only after the pres-

sure wave initiated by the ﬁrst ﬂame front propagates through the

mixture. By this, no aerodynamic conﬁnement is achieved and the

mixture undergoes a deﬂagrative combustion. Moreover, the loca-

tion of the ignition fronts at more than one separate locations in

the tube result in the detection of either one ﬂame front or two

subsequently appearing ﬂame fronts, which explains the scattering

in ignition delay time as earlier observed in Fig. 7 for control steps

3 and 7. In contrast, the fuel proﬁle with one distinct maximum

followed by a region of gradually decreasing fuel concentration, as

achieved in control step 4, results in a more homogeneous ignition,

and thus, in a signiﬁcantly increased pressure amplitude. All these

observations reveal that by adjusting the fuel injection trajectory,

the gradient in fuel distribution inside the combustion tube can

be controlled such that multiple ignition kernels are initiated lead-

ing to a quasi-homogeneous autoignition. The total fuel mass ﬂow

rate and the maximum local fuel concentration are found to have

a rather minor impact on the pressure rise. Furthermore, the con-

trol algorithm is a good approach for cycle-averaged optimization.

However, in order to achieve a reliable controller performance, a

suitable cost function is vital.

3.3. Detailed analysis of distributed autoignition

In the previous section, it was observed that the cost func-

tion J

, by attempting to minimize the variation in ignition timing,

generated several fundamentally different ignition phenomena by

the applied injection trajectories. Although the ﬁnal goal of pres-

sure maximization was missed, the obtained results can be used

to study the underlying processes. In this part, a detailed exami-

nation of single ignition cycles with respect to the chemilumines-

cence, pressure, and ionization measurements is investigated. The

objective is to gain a better understanding of the ignition phenom-

ena and how they contribute to the grouping or scattering of the

individual events as discussed in Fig. 7 . For this, the previously ap-

plied test rig is modiﬁed to allow for the optical examination of

underlying aspects that trigger different modes of autoignition and

how those are distinguishable in terms of pressure and ignition

characteristics.

The ignition behavior is examined based on four different in-

jection trajectories as shown in Fig. 9 . These trajectories have been

chosen since the predicted fuel distributions in the combustor

5.4 Publication IV: Controlled Autoignition in Stratified Mixtures

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

Fig. 9. Applied fuel injection trajectory (black), calculated fuel proﬁle using the simulation tool (shaded area) and gradient in concentration (red) for four example injection

trajectories. (For interpretation of the references to color in this ﬁgure legend, the reader is referred to the web version of this article.)

Fig. 10. Maximum pressure rise over standard deviation of the ignition timing for curve green, blue and red. (For interpretation of the references to color in this ﬁgure

legend, the reader is referred to the web version of this article.)

show similar characteristics that were found in the operation with

the cost function J

. The air mass ﬂow rate of ˙

air

= 30 kg/h and

the temperature of T

air

= 1023 K are set to match the conditions

for measurements with the controller. Due to the low heat con-

duction property of the quartz tubes compared to the stainless

steel combustor, a slightly decreased temperature is observed in

the combustor. However, the goal is not to exactly replicate the

measurements conducted in the previous section but rather to in-

vestigate fuel injection trajectories that are similar to those ob-

served in the controller optimization.

The applied fuel injection trajectories resemble the fuel distri-

bution derived from control steps 1, 3, 4 and 7 that are discussed

in the previous section (see Fig. 8 highlighted in green, blue and

red). The corresponding injection curves are visualized in the same

color and referred to as green, blue and red curve in this section.

The gray curve will be discussed later. When comparing the igni-

tion behavior of the chosen curves to the respective control steps

(shown in Fig. 7 ) with regard to the cycle-to-cycle variation, a sim-

ilar correlation between the maximum pressure peak and the stan-

dard deviation of the ignition timing is observed (shown in Fig. 10 ).

An increase in maximum pressure goes along with an increase in

ignition homogeneity indicated by the decreasing standard devi-

ation in ignition time. Here, the green curve indicates a less ho-

mogeneous ignition and, thus a minor rise in pressure compara-

ble to control steps 3 and 7. The red curve ignites more homoge-

neously leading to notable increase in pressure amplitude which is

comparable to control step 4. The blue curve shows similar char-

acteristic as the control step 1 and can be assigned somewhere

in between. This is a clear indication that these speciﬁed trajecto-

ries are representative of the respective trajectories applied by the

controller.

In the following ﬁrst, single cycles obtained from three example

fuel injection trajectories (green, blue and red, as shown in Fig. 9 )

are compared. Later, three cycles obtained from injecting the gray

fuel trajectory will be compared allowing for the examination of

stochasticity.

For each curve one example cycle is shown in Fig. 11 . Two nor-

malized x –tdiagrams are shown representing the temporal evo-

lution of the CH

∗(ﬁrst column) and OH

∗(second column). For

this, each snapshot is averaged over the tube diameter resulting

in a one-dimensional array. The third and forth columns repre-

sent the time-resolved, CO

∗corrected chemiluminescence signals

at a wavelength of 307 and 430 nm obtained by the PMTs and

the pressure traces for each cycle. Column 5 shows the data mea-

sured by the spectroscope. Due to a limited sampling frequency

of the spectroscope, the data is accumulated over the entire ig-

nition event, and is not time-resolved. The obtained peaks are

therefore not quantitatively comparable to the data obtained from

PMTs and high-speed imaging. In order to compare the tempo-

ral evolution of the ignition obtained from two-dimensional data

to the one-dimensional measurements, ﬁve data points in the x –t

diagrams at the respective position of the optical sensor are av-

eraged resulting in a 1D array. The resulting time-resolved data

(white line) as well as the uncorrected PMT data (dotted line)

are overlayed in the x –tdiagrams. For convenience, the fuel in-

jection trajectory is visualized in the bottom left for each sample

shot.

When comparing the one-dimensional temporal evolution of

the chemiluminescence data obtained from the uncorrected PMT

5 Publications

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

Fig. 11. Three example cycles for injection curve green (top), blue (middle) and red (bottom). (For interpretation of the references to color in this ﬁgure legend, the reader

is referred to the web version of this article.)

data to the x –tdiagrams, it can be seen that both are in good

agreement. This is a good validation of the measurement method-

ology and reveals that both, PMT and high-speed imaging, result

in similar measurement signals. The two-dimensional snapshots

are impacted by CO

∗chemiluminescence in the background,

which has to be considered when interpreting. Comparing the

pressure traces of each cycle, it can be stated that the maximum

obtained pressure rise applying the green is the lowest and red

curve is the highest. The blue curve can be categorized in the

intermediate range. Similar behavior regarding the shape of the

fuel distribution and the resulting maximum pressure amplitude

was observed in the controller optimization for control steps 1, 3,

4 and 7, respectively.

Two propagating ﬂame fronts can be observed when ap-

plying the green curve (see Fig. 11 a). The shape of the fuel

concentration distribution is characterized by two spatially

separated maxima. These peaks ignite inside the combustor

at two different ignition times and locations initiating two

propagating ﬂame fronts. Both ﬂame fronts propagate through

the reactive mixture independently leading to a non-conﬁned

volume and therefore a low pressure amplitude. When com-

paring the temporal position of the ﬁrst pressure peak to the

temporal position of the ﬁrst peak in the PMT data, an increased

time delay can be observed. This agrees well with the previ-

ous observation shown in Fig. 10 in which an increased time

delay is associated with a less homogeneous autoigni-

tion/deﬂagration. The pressure traces reveal three independent

pressure rises corresponding to each individual ignition event

and a downstream propagating pressure wave reﬂected at the

combustor inlet. Both ﬂame fronts are characterized by the pres-

ence of CH

∗and OH

∗. In general, the simultaneous presence of

∗and OH

∗is an indicator of the reaction zone. When further

investigating the PMT data, it can be seen that the second ﬂame

front is characterized by a strong peak in the CH

∗data. This corre-

lates with the larger propagation velocity than the ﬁrst observed

ﬂame front, which is linked to an increased reaction rate. The

second increase in OH

∗is more gradual and remains for a longer

period of time compared to the emitted OH

∗by the primary ﬂame

front. OH

∗without the presence of CH

∗, as visible in the PMT and

x –tdata, is considered cold OH

∗and is characteristic for purged

exhaust.

The ignition observed for the blue curve (see Fig. 11 b) is charac-

terized by a fairly homogeneous ignition while showing a consider-

able spatial variation in CH

∗and OH

∗chemiluminescence along the

combustor axis indicated by the x –tdiagrams. The pressure traces

reveal a slight increase in the pressure prior to the main pressure

peak, which occurs before the detection of CH

∗and OH

∗chemilu-

minescence. Potentially, a ﬁrst stage ignition is taking place lead-

ing to slow rise in pressure. Terashima et al. [13] numerically in-

vestigated the role of low-temperature combustion in end-gas au-

toignition and pressure wave generation. They found out that low

temperature combustion in DME–air mixtures leads to a consid-

erable heat release rate which is correlated to the formation of

O and promotes pressure waves. If the heat release of the ﬁrst

stage ignition is large enough, a transition to HTC is likely to oc-

cur. This two-stage ignition behavior was further investigated by

Savard et al. [11] . Their results reveal that a reduced reactivity in

the products of the cool ﬂame can limit subsequent HTC. Thus, if

LTC is observed over an extended time frame without transition-

ing into HTC, the hot ﬂame speed is retarded due to the decrease

in reactivity of the cool ﬂame products. In the present cycle, this

decrease in reactivity prior to HTC, which is indicated by OH

∗and

∗chemiluminescence, leads to a limited pressure increase. How-

ever, this can not be deﬁnitively veriﬁed in the chemiluminescence

data due to weak chemiluminescence of formaldehyde which is an

indicator for the ﬁrst stage ignition. Future work is planned to fur-

ther investigate the role of formaldehyde in the observed ignition

phenomena.

5.4 Publication IV: Controlled Autoignition in Stratified Mixtures

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

The latter case (see Fig. 11 c) reveals similar features to the pre-

vious case in Fig. 11 b with the difference that no pre-pressure rise

is observed. Assumingly, there is a faster transition to the second

stage ignition, which is characterized by a more rapid heat release.

Zhang et al. [12] investigated the interaction and transition of cool

and hot DME ﬂame chemistry numerically. They conclude that a

double-ﬂame structure can evolve due to HTC that takes place in a

propagating cool ﬂame. Since hot ﬂames are characterized by much

higher ﬂame velocities, both ﬂames eventually merge resulting in

an increased ﬂame propagation speed compared to a single hot

ﬂame. In the presented data of Fig. 11 , coupling of the heat release

and the pressure rise promotes HTC and leads to an aerodynamic

conﬁnement of the volume. This eventually results in a higher rise

in pressure. In the spectroscope data, the species OH

∗, CH

∗and C

∗

appear in distinct peaks for all curves.

Even though different modes of ﬂame propagation can be pro-

moted by the fuel injection, a notable cycle-to-cycle variation re-

mains, as illustrated in Fig. 10 . The examples presented above do

not allow for quantitative comparisons of the PMT data due to dif-

ferent fuel injection trajectories leading to different fuel distribu-

tions inside the combustor before ignition. Thus, to further under-

stand the variation in the combustion phenomena for a given fuel

concentration proﬁle, as visualized in Fig. 7 , three sample shots are

compared for the same injection trajectory, the gray curve from

Fig. 9 . This curve is chosen because the induced fuel concentration

distribution in the combustor is similar to control step 4 for ap-

plying the cost function J

2 in Fig. 8 , which was posited to result

in more homogeneous autoignition. As in the previously presented

cases, Fig. 12 visualizes the measured data for three different cy-

cles when applying the gray curve as injection trajectory.

In cycle 1 ( Fig. 12 a), CH

∗and OH

∗formation occurs simulta-

neously. The maximum concentration in CH

∗is located at an ear-

lier temporal position compared to the maximum peak in OH

∗.

Additionally, the obtained CH

∗chemiluminescence in the x –tdi-

agram and the PMT data is less than that of cases 2 and 3. How-

ever, the pressure signals reveal a pre-pressure rise, which indi-

cates a ﬁrst stage ignition prior to main ignition. Similar to the cy-

cle presented in Fig. 11 b, this LTC reduces the reactivity of the mix-

ture, and thus, leads to a limited pressure rise in the second stage

ignition.

Cycle 2 represents an aerodynamic conﬁnement resulting from

two propagating autoignition fronts as indicated in Fig. 12 b. The

temporal evolution in each snapshot that is used for assembling

the x –tdiagram is attached in the appendix to illustrate the merg-

ing process of two propagating ﬂames ( Fig. A1 ). The initial ﬂame

propagating upstream provides a compression to the autoignition

that occurs simultaneously. Thus compared to the previous case,

the LTC takes place under increased pressure conditions leading

to a shift in the relative prominence of the peaks obtained in the

spectroscope data, which matches the observation in the PMT data

(high CH

∗and low OH

∗compared). Furthermore, the data reveals

prominent peaks at the characteristic wavelength of C

∗. These

peaks are much smaller in the other examples and thus, an in-

dication of different reaction paths that are promoted.

In the ﬁnal scenario, the observed ignition phenomena is sim-

ilar to the previously shown cycle in Fig. 11 c. The data reveals

a homogeneous ignition with a high concentration in OH

∗and

∗resulting in a high pressure rise. This case can be classi-

ﬁed as single-stage ignition in which high temperature chemistry

dominates resulting in a high heat release rate which leads to

an increase in pressure amplitude. Compared to the second case

( Fig. 12 b), here, the aerodynamic conﬁnement is achieved by a sin-

gle supersonic autoignition front rather multiple propagating ﬂame

fronts. When comparing the spectroscope data for all three cases,

they quantitatively match the data obtained from the PMTs for OH

∗

and CH

∗.

The available data leads to the conclusion that four different

ignition modes, namely deﬂagration, subsonic autoignition, super-

sonic autoignition, and aerodynamic conﬁnement by multiple si-

multaneous autoignition fronts, can be triggered by the fuel in-

jection trajectory. As previously introduced (see Table 1 ) accord-

ing to the theory of Zel’dovich [3] subsonic and supersonic ﬂame

fronts can be characterized by the dimensionless number ξ=

a/u

. Figure 13 sketches the identiﬁed mechanisms responsible for

each mode. A low rise in pressure along with a slow propagation,

and thus, a less homogeneous ignition can be labeled as a deﬂa-

gration or turbulent ﬂame propagation. Here the ﬂame is led by

diffusion rather than autoignitive processes. An example for this

case is shown in Fig. 11 a. The ﬂame propagation speed is lower

than the speed of sound.

The second mechanism responsible for different modes was

found to be the two-stage ignition, which is characterized by a

ﬁrst ignition dominated by LTC followed by the second stage igni-

tion which is promoted by HTC. Example cycles showing two-stage

ignition characteristics are presented in Figs. 11 b and 12 a. The de-

lay x between both ignition stages is a function of the tempera-

ture [13] . For very small values of x between the ﬁrst and sec-

ond stage ignition, a fast propagating ﬂame induced by the HTC

chemistry eventually merges with the slow ﬂame induced by LTC.

By this, a fast propagating ﬂame is generated leading to a sharp in-

crease in pressure. In case x is high, the hot ﬂame does not catch

up with the initial ﬂame leading to a low overall rise in pressure

amplitude.

In this case, the propagation velocity of the ﬂame front is lower

than the speed of sound.

In the third scenario, a single-stage ignition ( x = 0 ) occurs

which is characterized by the rapid heat release leading to a

sharp increase in pressure. The ignition event seems homogeneous

throughout the measurement frame leading to a very high propa-

gating speed of the ﬂame front (e.g. Figs. 11 c and 12 c).

The last scenario is a combined phenomenon that is assembled

from the prior cases. Here, a high increase in pressure is observed

due to an aerodynamic conﬁnement induced by multiple ﬂame

fronts occurring simultaneously. Pressure waves that originate from

these ignition kernels merge and thus, result in an overall greater

pressure rise (e.g Fig. 12 b).

4. Conclusion and outlook

An SEC test rig was used for the investigation of the control-

lability of an autoignition phenomenon by the adjustment of the

fuel injection trajectory at atmospheric pressure and elevated tem-

perature conditions using dimethyl ether as fuel. First, the ability

of injecting a deﬁned fuel proﬁle into a convecting air ﬂow was

demonstrated by applying a line-of-sight measurement technique.

Second, an online control approach was applied to the test rig in

order to maximize the measured pressure rise subsequent to au-

toignition by adjusting the fuel injection trajectory. The optimiza-

tion results revealed an increase in maximum pressure for an in-

creased homogeneity of the ﬂame front. As the next step, a one-

dimensional simulation tool was used to derive the actual fuel dis-

tribution inside the combustor before ignition for selected trajecto-

ries applied by the controller. Correlating the calculated fuel distri-

bution with the obtained autoignition homogeneity and pressure

amplitude reveal a signiﬁcant impact of the injection trajectories

on the autoignition characteristics. Although, the applied control

approach is a good tool for cycle-averaged optimization, indicated

by a notable shift towards higher pressures for certain fuel injec-

tion trajectories, a visible cycle-to-cycle variation remains.

Detailed single cycle analyses of three different injection tra-

jectories were performed to further analyze the scattering of the

observed autoignition by the means of optical measurements as

5 Publications

100

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

Fig. 12. Three example cycles for the gray injection curve. (For interpretation of the references to color in this ﬁgure legend, the reader is referred to the web version of this

article.)

Fig. 13. Sketch of the pressure rise and heat release rate for different modes of ﬂame propagation triggered by diffusion, two-stage ignition, single-stage ignition or aerody-

namic conﬁnement by multiple simultaneous autoignition fronts.

well as pressure and ionization probe data. Two-dimensional high-

speed recording of the CH

∗and OH

∗enabled the examination

of the temporal ﬂame evolution. PMT and spectroscope data re-

veal correlation of amplitudes in pressure and chemiluminescence

of OH

∗and CH

∗, respectively. It was found that the stochastic-

ity of the observed ignition can be attributed to fuel character-

istics dominated by multi-stage ignition behavior of the applied

fuel. Four main ignition phenomena were identiﬁed, namely: deﬂa-

gration, subsonic autoignition, supersonic autoignition and aerody-

namic conﬁnement by multiple simultaneous autoignition fronts.

A deﬂagration is dominated by diffusion processes rather autoigni-

tion. This ignition mode was found to be characterized by a moder-

ate rise in pressure along with low propagation velocities. A sub-

sonic autoignition was mainly observed for two-stage ignition in

which the delay between ﬁrst and second-stage ignition was large

enough to prevent a merging of hot and cold ﬂame. A supersonic

ﬂame is observed for single-stage ignition in which HTC dominates

and for low values of x . An aerodynamic conﬁnement was ob-

served for multiple ﬂames that occur simultaneously leading to a

conﬁned volume. It was found that besides the gradient in reactiv-

ity, the autoignition characteristics of the fuel/air mixture greatly

impact the mode of ﬂame propagation. The results are crucial for

further improvement of control approaches based on the different

mechanism identiﬁed. The affects of the observed stochasticity on

long-term cyclic operation is seen to be important and will be ad-

dressed in future work.

Declaration of Competing Interest

The authors declare that they have no known competing ﬁnan-

cial interests or personal relationships that could have appeared to

inﬂuence the work reported in this paper.

Acknowledgments

The authors gratefully acknowledge the support of the

Deutsche Forschungsgemeinschaft (DFG) as part of Collaborative

Research Center CRC 1029 “Substantial eﬃciency increase in gas

turbines through direct use of coupled unsteady combustion and

ﬂow dynamics” through sub-projects A03 and A05. The authors

also wish to thank Andy Göhrs and Thorsten Dessin for their tech-

nical support.

Appendix A. Aerodynamic conﬁnement

5.4 Publication IV: Controlled Autoignition in Stratified Mixtures

101

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

Fig. A1. Temporal evolution of an aerodynamic conﬁnement visualized by single snapshots of the CH

∗and OH

∗chemiluminescence for cycle 2 in Fig. 12 .

5 Publications

102

F.C. Yücel, F. Habicht, F. Arnold et al. Combustion and Flame 232 (2021) 111533

References

[1] T.J. Jacobs , D.N. Assanis , The attainment of premixed compression ignition

low-temperature combustion in a compression ignition direct injection engine,

Proc. Combust. Inst. 31 (2007) 2913–2920 .

[2] A. Bartenev , B. Gelfand , Spontaneous initiation of detonations, Prog. Energy

Combust. Sci. 26 (20 0 0) 29–55 .

[3] Y.B. Zeldovich , Regime classiﬁcation of an exothermic reaction with nonuni-

form initial conditions, Combust. Flame 39 (1980) 211–214 .

[4] X. Gu , D. Emerson , D. Bradley , Modes of reaction front propagation from hot

spots, Combust. Flame 133 (2003) 63–74 .

[5] H. Im , J. Chen , C.K. Law , Ignition of hydrogen-air mixing layer in turbulent

ﬂows, Symp. (Int.) Combust., 27, Elsevier (1998), pp. 1047–1056 .

[6] R. Sankaran , H.G. Im , E.R. Hawkes , J.H. Chen , The effects of non-uniform tem-

perature distribution on the ignition of a lean homogeneous hydrogen–air

mixture, Proc. Combust. Inst. 30 (2005) 875–882 .

[7] G. Mittal , M. Chaos , C.-J. Sung , F.L. Dryer , Dimethyl ether autoignition in a rapid

compression machine: experiments and chemical kinetic modeling, Fuel Pro-

cess. Technol. 89 (2008) 1244–1254 .

[8] M.B. Luong , F.E. Hernandez Perez , A. Sow , H.G. Im , Prediction of ignition

regimes in DME/air mixtures with temperature and concentration ﬂuctuations,

AIAA Scitech 2019 Forum (2019), p. 2241 .

[9] M.B. Luong , F.E.H. Pérez , H.G. Im , Prediction of ignition modes of NTC-fuel/air

mixtures with temperature and concentration ﬂuctuations, Combust. Flame

213 (2020) 382–393 .

[10] J. Pan , H. Wei , G. Shu , Z. Chen , P. Zhao , The role of low temperature chemistry

in combustion mode development under elevated pressures, Combust. Flame

174 (2016) 179–193 .

[11] B. Savard , A. Wehrfritz , K. Lam , Q. Margerte , L. Ferney

, S. Farjam , Decreased

mixture reactivity and hot ﬂame speed in the products of diffusion-af-

fected autoignitive cool ﬂames in the NTCregime, Combust. Flame 222 (2020)

434–445 .

[12] W. Zhang , M. Faqih , X. Gou , Z. Chen , Numerical study on the transient evolu-

tion of a premixed cool ﬂame, Combust. Flame 187 (2018) 129–136 .

[13] H. Terashima , H. Nakamura , A. Matsugi , M. Koshi , Role of low-temperature ox-

idation in non-uniform end-gas autoignition and strong pressure wave gener-

ation, Combust. Flame 223 (2021) 181–191 .

[14] Y. Ju , Understanding cool ﬂames and warm ﬂames, Proc. Combust. Inst. (2020) .

[15] T. Kathrotia , Reaction Kinetics Modeling of OH

∗, CH

∗, and C2

∗Chemilumines-

cence, Universität Heidelberg, 2011 Ph.D. thesis .

[16] N. Djordjevic , M. Rekus , J. Vinkeloe , L. Zander , Shock tube and kinetic study

on the effects of co2 on dimethyl ether autoignition at high pressures, Energy

Fuels 33 (2019) 10197–10208 .

[17] M. Oguma , S. Goto , H. Hatsuzawa , M. Konno , Z. Chen , T. Watanabe , Spectro-

scopic analysis of combustion ﬂame fueled with dimethyl ether (DME), SAE

Trans. (2003) 1037–1043 .

[18] M.P. Mayo , A.L. Boehman , Ignition behavior of biodiesel and diesel under re-

duced oxygen atmospheres, Energy Fuels 29 (2015)

6793–6803 .

[19] H. Li , S.D. Wehe , K.R. McManus , Real-time equivalence ratio measurements in

gas turbine combustors with a near-infrared diode laser sensor, Proc. Combust.

Inst. 33 (2011) 717–724 .

[20] R. Bluemner , C.O. Paschereit , K. Oberleithner , Generation and transport of

equivalence ratio ﬂuctuations in an acoustically forced swirl burner, Combust.

Flame 209 (2019) 99–116 .

[21] F.C. Yücel , F. Habicht , A. Jaeschke , F. Lückoff, K. Oberleithner , C.O. Paschereit ,

Investigation of the fuel distribution in a shockless explosion combustor, J. Eng.

Gas Turb. Power (2020) .

[22] S.Z. Khong ,

D. Neši

c , M. Krsti

c , Iterative learning control based on extremum

seeking, Automatica 66 (2016) 238–245 .

[23] Y.-H. Dai , Convergence properties of the bfgs algoritm, SIAM J. Optim. 13

(2002) 693–701 .

[24] R.S. Sheinson , F.W. Williams , Chemiluminescence spectra from cool and

blue ﬂames: electronically excited formaldehyde, Combust. Flame 21 (1973)

221–230 .

[25] V. Smith , J. Robinson , Ultraviolet emission and absorption spectra produced

by organic compounds in oxyhydrogen ﬂames1: emission spectra, Anal. Chim.

Acta 48 (1969) 391–403 .

[26] N. Djordjevic , N. Hanraths , J. Gray , P. Berndt

, J. Moeck , Numerical study on

the reduction of nox emissions from pulse detonation combustion, J. Eng. Gas

Turb. Power 140 (2018) 041504 .

[27] K. Becker , D. Dley , R. Norstrom , Oh

∗chemiluminescence in hydrocarbon atom

ﬂames, Symp. (Int.) Combust., 12, Elsevier (1969), pp. 405–413 .

[28] F.C. Yücel , F. Habicht , M. Bohon , C.O. Paschereit , Autoignition in stratiﬁed mix-

tures for pressure gain combustion, Proc. Combust. Inst. (2020) .

5.4 Publication IV: Controlled Autoignition in Stratified Mixtures

103

5 Publications

Publication IV: Summary and Contribution

Publication IV

serves as a continuation of the previously presented publications and

combines their outcomes. In this work a closed-loop control approach was applied to the

existing single tube SEC. Two different cost functions

and

were defined aiming for

maximizing the pressure rise through autoignition by optimizing the injection profile.

Pressure and ionization probes were used to measure the pressure rise and ignition

homogeneity respectively. Based on the measurement data, the controller calculates

the injection trajectory for the next cycle. The results reveal that the controller is

capable of maximizing the cycle-averaged pressure rise by optimizing

. However,

an alternating controller performance was observed when optimizing cost function

Correlating the pressure rise and ignition data show greater pressure amplitudes for

increased ignition homogeneity.

To further investigate the controller performance the previously introduced simulation

tool based on the one-dimensional diffusion equation (see

publication II

), was further

developed and applied to each control trajectory applied by the controller. Three

injection trajectories applied by the controller inducing different pressure levels when

undergoing autoignition were chosen for further investigation. The calculation of

the fuel distribution using the simulation tool revealed that each trajectory results

in a fundamentally different distribution of fuel concentration along the combustor.

An optically accessible combustor as previously introduced was used to enable a

simultaneous measurement of light emission by photomultiplier, a spectroscope and a

high-speed camera as well as the pressure rise by pressure transducers and the ignition

time by ionization probes. Three injection profiles, resulting in a similar fuel distribution

as the ones applied by the controller, were injected into the convecting air flow.

The data obtained reliably and repetitively shows four different distinct types of

ignition: deflagration, subsonic autoignition, supersonic autoignition and aerodynamic

confinement through multiple autoignition fronts. A deflagration is mainly dominated

by diffusion processes and is associated by a moderate rise in pressure along with

low propagation velocities. A subsonic autoignition was found to correlate strongly

with a two-stage ignition characteristic of the applied fuel, as described in Sec. 1.3.

A supersonic flame propagation along with an high increase in pressure was observed

for single-stage ignition characterized by a rapid heat release leading to a confined

volume. The latter case was found to result in similar pressure levels observed for an

aerodynamic confinement induced by multiple propagating flame fronts. However, the

probability of the occurrence of either one of these modes can be greatly impacted by

the fuel injection and the resulting gradients in concentration. These observation are

104

5.4 Publication IV: Controlled Autoignition in Stratified Mixtures

fundamental in nature and greatly contribute to the understanding of the process of

autoignition in the context of the SEC. Further discussion of the results in context with

publication II and III will be provided in Ch. 6.

105

6 Discussion

In this chapter, the collective findings presented in the

publications II-IV

will be

summarized and placed into an overall context as visualized in Fig. 6.1. In Sec. 6.1, the

ability of the injection strategy to control the fuel distribution inside the combustor is

discussed, which has been investigated within the scope of

publication II

. Furthermore,

the impact of boundary layers and turbulent diffusion on the gradients in concentration

will be examined.

Next, the correlation of the fuel distribution and autoignition will be described, which

has been investigated within the scope of

publication III

. Here, the fuel injection

strategy previously presented is used to assess three model injection profiles, which are

subsequently correlated with autoignition characteristics in terms of pressure rise and

ignition homogeneity. It is shown that by injecting a prescribed charge of ignitable

mixture, the homogeneity of the autoignition accompanied by an increased maximum

pressure amplitude can be greatly impacted.

In Sec. 6.4, a control algorithm for optimizing the fuel distribution inside the combustor,

developed within the scope of

publication IV

, will be described. Here,

publica-

tions II

and

III

are linked through

publication IV

, as visualized in Fig. 6.1, where

the correlation between autoignition and fuel distribution is exploited to optimize the

injection trajectory. Major findings of

publication I

contributed to

publications II-

with regard to test rig improvements which are thoroughly discussed in Ch. 2 and,

therefore, will not be included in the following discussion. All investigations within the

context of publications II-IV are conducted using the “single tube SEC” setup.

6.1 Fuel Stratification Approach

As discussed in Sec. 1.3, shockless explosion combustion is defined as the simultaneous

autoignition of a stratified fuel–air mixture, which has been precisely tailored to

compensate for the fuel injection duration by varying the local autoignition delay

time

τidt

. The autoignition delay time of a reactive mixture is a function of the local

107

6 Discussion

Fuel stratification Autoignition characteristics

Fuel concentration

Pressure rise

Ignition homogeneity

Fuel concentration TDLAS

Acetone PLIF

TDLAS

Pressure transducers

Ionization probes

High-speed camera

Closed-loop control

input

output

Parameters Diagnostics Parameters Diagnostics

Publication II Publication III

Publication IV

Fuel concentration

Pressure rise

Ignition homogeneity

CH* & OH* distribution

TDLAS

Pressure transducers

Ionization probes

High-speed camera

PMT

Spectroscope

Parameters Diagnostics

Publication I

injection strategy

Figure 6.1: Overview and context of the publications.

temperature, pressure and equivalence ratio as discussed in Sec. 1.3. In practice however,

an accurate control of the spatial distribution of either temperature or pressure is not

feasible. Therefore, the equivalence ratio appears to be a reasonable parameter for

controlling local autoignition characteristics. Thus, in a defined measurement volume,

a stratification in equivalence ratio is sufficient for an accurate control of the ignition

delay time distribution.

Publication II

investigates the capability of the applied injection scheme of controlling

the equivalence ratio distribution throughout the combustor length. Here, different fuel

profiles, defined by the number of simultaneously operated fuel valves, are injected into

a continuous air flow at high frequencies. The fuel distribution inside the combustor is

obtained by applying two measurement methods allowing for the spatial (acetone-PLIF)

and temporal (TDLAS) resolution of the convective transport. Both methods, however,

reveal that by adjusting the number of open valves, a well-defined fuel profile can

be injected into a convecting air flow that remains largely preserved. For this study,

methane is used as a tracer fuel, compared to the reacting cases, in which DME is

used as the fuel. While this prevents quantifying the exact local equivalence ratio for

the reacting DME cases, it does allow for a qualitative measure of the reproducibility

and accuracy of the injection scheme. In addition to the shape of the injected fuel

108

6.1 Fuel Stratification Approach

0.5

ĉ (-)

20 50 80 110 20 50 80 110 20 50 80 110

open fuel valves (-)

0 30 60 90 0 30 60 90 0 30 60 90



-curve



-curve V-curve

t (ms) t (ms) t (ms)

t (ms)t (ms)t (ms)



-curve



-curve V-curve

control trajectory modeled injection

measurement simulation

Figure 6.2:

Model injection trajectories with a) valve control trajectory and respec-

tive modeled injection curve b) measurement and simulation data. Adapted from

Publication IV and included for convenience.

trajectory, the air flow velocity

uair

and the fuel injection time

tinj

are varied for TDLAS

measurements. When increasing the air flow velocity, while maintaining a constant fuel

injection time, the convection duration, which the fuel profile exhibits when traveling

from the injection station to the measurement point, is decreased. Hence, there is less

time to diffuse, which ultimately leads to an improved preservation of the injected fuel

profile indicated by steeper gradients in the measured fuel concentration. The same

effect, however, was observed when increasing the fuel injection time while maintaining

a constant air flow velocity. An increased steepness of the gradients in the measured fuel

concentration was observed, indicating a higher grade of preservation of the injected

fuel profile. Therefore, the spatial width

uairtinj

was found to have a significant

impact on the fuel distribution inside the combustor. Results reveal that the impact of

the air flow velocity does not affect the shape of the measured fuel profile when ensuring

a constant χ.

To confirm the validity of these observations, a numerical calculation tool is developed

in cooperation with Fabian Habicht for solving the one-dimensional diffusion equation

∂c

∂t =D∂2c

∂x2,(6.1)

where

is the local fuel concentration and

the diffusion coefficient, which is equal to

109

6 Discussion

the turbulent diffusion coefficient

Dturb

when considering a fully turbulent flow. However,

when assuming the radial diffusion to be negligible, the obtained fuel concentration

is a function of the second spatial derivative of the fuel concentration

∂2c/∂x2

, the

diffusion coefficient

, and the convection time ∆

. Figure 6.2a shows three model

injection profiles and the respective modeled injection profiles for a

-, Λ- and V-curve.

The black line represents the actual valve control signal, whereas the red line is the

modeled injection which arises when matching the valve operation parameters such

that all measurement data is resembled with a maximum accuracy replicating the

fuel injection and flow conditions. When comparing the measured fuel profiles to the

simulation (see Fig. 6.2b) it can be seen that by solving the one-dimensional diffusion

equation, the measured fuel profiles can be accurately replicated. Further investigations

demonstrated that by decreasing the spatial width

, the second spatial derivative of

the fuel concentration

∂2c/∂x2

is increased by a factor of four, resulting in a four times

faster decay of the obtained gradients in concentration.

Moreover, when decreasing the air flow velocity, the convection time d

is increased by

the same ratio. Thus, from Eq. (6.1), it can be concluded that the turbulent diffusion

coefficient

Dturb

is decreased by the same factor when reducing the air flow velocity.

This observation agrees well with the findings of Speziale [

], who states that the

turbulent diffusivity can be linked to the turbulent eddy viscosity which scales linearly

with the mean flow velocity and, thus, leads to proportionality between

Dturb

and

uair

To further increase the accuracy of the simulation tool, additional TDLAS measurements

are conducted at a variety of preheat temperatures, within the scope of

publication IV

Here, the air mass flow rate is adjusted for each applied temperature to obtain a constant

volumetric flow rate and, thus, a variation in the resulting Reynolds number. The

results reveal that the diffusion coefficient is not affected by temperature variations.

Assumingly, all existing flow conditions are largely independent of the Reynolds number.

Taking this into account, it can be stated that the developed tool is capable of replicating

the fuel distribution inside the combustor at reacting conditions. In

publication IV

this tool is used to predict the fuel distribution inside the combustor before ignition for

each control trajectory, which will be discussed in Sec. 6.4.

Acetone-PLIF measurements were conducted within the scope of

publication II

further explore the fuel concentration distribution inside the combustor and extract

flow features, such as boundary layer effects and radial dispersion, which are masked

when applying a line-of-sight measurement technique such as TDLAS. The results of the

study greatly contributed to a better understanding of the flow inside the combustor. A

characteristic profile of a turbulent pipe flow was identified, causing a radial distortion

110

6.2 Correlation of Fuel Stratification and Autoignition

Δpmax (bar)

Δτip (ms)

0.2

0.4

0.6

0.8

1.0

0 2 4 6 8 10

Λ-curve

V-curve

Π-curve

Figure 6.3:

Maximum pressure amplitude over delay time ∆

τip

for three different injec-

tion trajectories. Adapted from publication III and included for convenience.

of the injected fuel profile. Generally, the results obtained from both, TDLAS and

acetone-PLIF, are in good agreement. However, the fuel concentration obtained along

the centerline of the acetone-PLIF measurements, compared to the integrated values

obtained when applying TDLAS, revealed that the actual gradients in fuel concentration

are steeper and, thus, more preserved.

The ability to accurately apply fuel stratification is a key challenge towards the imple-

mentation of the SEC. Therefore, the presented fuel concentration measurements lay

the foundation for understanding the correlation of autoignition and fuel concentration

distribution.

6.2 Correlation of Fuel Stratification and Autoignition

Based on the findings in Sec. 6.1,

publication III

investigates the correlation between

fuel injection and autoignition characteristics. Therefore, three model injection profiles

(as shown in Fig. 6.2) are injected into a preheated, continuous air flow at high frequencies

using a stainless steel combustor configuration (see Fig. 2.4a). DME is applied as fuel

for all measurements. Here, the injection strategy as previously introduced was applied.

Each fuel trajectory is injected 150 times at a frequency of 5 Hz. Ionization and

pressure probes are used to measure the pressure rise and ignition times throughout

the combustor section. The maximum pressure amplitude ∆

pmax

and the delay time

111

6 Discussion

∆

τip

between the onset of ignition and pressure rise are determined for each operating

cycle, as described in Sec. 3.2 and Sec. 3.2. Figure 6.3 visualizes the amplitude of the

pressure rise as a function of the coherency of the ignition ∆

τip

. The results show that

there is a visible link between ∆

τip

and the maximum pressure amplitude ∆

pmax

. Low

∆

τip

indicate a more simultaneous detection of the ignition and pressure rise resulting

in a higher maximum amplitude, whereas higher values of ∆

τip

are associated with less

homogeneous ignition. Moreover, Fig. 6.3 clarifies that the applied fuel profiles tend to

cluster throughout the graph while indicating a notable shift of the Λ-profile towards

higher pressure amplitudes, and of the V-profile towards less homogeneous ignition.

The

-profile generally tends to fall in between. Even though a significant correlation

between the injection curve and ignition can be found, a notable cycle-to-cycle variation

remains.

In order to further analyze the autoignition process, single cycles are investigated.

Optical measurements to measure the OH* intensity distribution in combination with

pressure measurements are conducted using the optically accessible combustor (see

Fig. 2.4b). Here, the ignition homogeneity was determined by calculating the standard

deviation of the position of the autoignition front obtained in the

–

diagrams, assem-

bled from averaged single snapshots, as described in Sec. 4.3. The results confirm the

previous findings obtained using the stainless steel combustor. A decrease in standard

deviation, meaning a more homogeneous autoignition, was associated with a higher

pressure amplitude, while a high standard deviation correlated with a low rise in pres-

sure. The Λ-curve was found to generally result in a more spatially distributed ignition

characterized by multiple simultaneous ignition sources, leading to an aerodynamic con-

finement which constrains the gas expansion subsequent to ignition. Thus, the pressure

rise through an evenly distributed ignition is greater compared to an autoignition that

originates from a single ignition point as observed for the V-curve.

The results reveal a clear correlation between the fuel injection and the homogeneity of

the autoignition. However, a cycle-to-cycle variation remains which will be discussed in

the following in the context of publication IV.

6.3 Single Cycle Analysis

The ability to precisely control the autoignition process on a cycle-to-cycle basis would

represent a leap-frog towards a more reliable and repeatable operation of the SEC. To

begin with, a more profound understanding of the autoignition, not only with regard to

the stratification, but also its interplay with fuel properties, is necessary. Therefore,

112

6.3 Single Cycle Analysis

sub-/supersonic autoignition supersonic autoignition multiple autoignition frontturbulent deflagration

LTCHTC HTC

pressure

heat release rate

x xx x

00 0 p0



a) b) c) d)

heat release rate

x x

multiple deflagrations single supersonic autoignition

e) f) multiple autoignitions



Figure 6.4:

Different modes of autoignition. Adapted from

publication IV

and

included for convenience.

Figure 6.5:

Probability of the occurrence of autoignition modes for different control

steps.

simultaneous optical measurements in combination with pressure and ionization probe

measurements are conducted. In addition to the measured OH* intensity in

publication

III

, here, the CH* intensity was measured simultaneously using a high-speed camera

equipped with an intensifier and an image doubler. Furthermore, four PMTs and

a spectroscope where used to qualitatively measure the species concentration at a

distinct spatial position in the combustor. The obtained data indicate the occurrence of

deflagration, two-stage, and single-stage ignition as a function of the applied injection

trajectory leading to different modes of autoignition.

Four different autoignition modes were identified, namely deflagration, subsonic autoigni-

tion, supersonic autoignition, and aerodynamic confinement by multiple autoignition

fronts. Figure 6.4a–d sketches the heat release rate and pressure distribution of each

113

6 Discussion

mode with respect to the axial position along the combustor. The illustrated modes

originate primarily from the utilized fuel DME, which exhibits different ignition char-

acteristics depending on the temperature, as previously introduced in Sec. 1.3. These

characteristics originate from different chemical paths namely, low, intermediate and

high temperature oxidation which promote either two- or single-stage ignition behavior,

as discussed in Sec. 1.3. A single-stage ignition is associated with a high heat release

rate and predominated by high temperature chemistry (HTC). A two-stage ignition

is characterized by low temperature chemistry (LTC) in the first-stage followed by a

second stage ignition which is promoted mainly by HTC. Thus, the ignition behavior

can promote i) either a supersonic flame, which is associated with a high heat release

leading to consecutive quasi-simultaneous autoignitions and, thus, tends to be more

homogeneous in nature, or ii) a subsonic flame which is associated with a lower heat

release rate compared, leading to a less homogeneous autoignition. A very low rise in

pressure however is primarily associated with flames that are dominated by diffusion

rather than autoignition (e.g., turbulent flames). A forth mode, which induces pres-

sure amplitudes similar to a single-stage ignition, was found to result from multiple

simultaneously initiating autoignition kernels, causing an aerodynamic confinement

comparable to a supersonic propagating autoignition front, as illustrated in Fig. 6.4d–g.

The available data lead to the conclusion that the observed modes can be triggered by

the injected fuel trajectory. Therefore, four different injection profiles are analyzed set

via a closed-loop controller (which will be introduced in Sec. 6.4) and, thus, referred

to as control steps 1, 3, 4 and 7. It is important to note that control steps 3 and 7

show features in fuel concentration similar to the previously introduced V-curve and

are associated with a low rise in pressure. Control step 1 is characterized by a constant

fuel distribution comparable to the

-curve and a mid-range pressure rise, while control

step 4 exhibits features similar to the Λ-curve along with a high pressure amplitude.

Figure 6.5 shows the percentage-wise distribution of the different autoignition modes

evaluated based on the pressure history (as shown in Fig. 6.4) of each cycle for the

respective control steps. The results confirm the observations previously discussed

within the context of

publication III

. It is evident that a low pressure (control steps

3 and 7/V-curve) is solely dominated by turbulent deflagration. The pressure profiles

do not indicate the presence of a single- or two-step autoignition. An intermediate rise

in pressure (control step 1/

-curve) was found to result from two autoignition modes,

namely turbulent deflagration and two-stage ignition, where both modes occur in equal

amounts. The highest rise in pressure, obtained for control step 4/Λ-curve, is mainly

dominated by two-stage and partially by single-stage ignition behavior. The forth

autoignition mode (shown in Fig. 6.4d), which is defined as aerodynamic confinement

114

6.4 Autoignition Control

τ (ms)

P1 P2 P3 P4 std(I1-I7)

pmax (bar)

1347

control step

25 6 8

0.7

0.6

0.1

0.2

0.3

0.4

0.5

Figure 6.6:

Controller performance for optimizing cost function

. Adapted from

publication IV and included for convenience.

by multiple autoignition fronts, is not distinguishable solely from the pressure data and,

thus, generally falls into the single-stage ignition mode.

The results disclose underlying effects that are responsible for cycle-to-cycle variations in

conjunction with the applied fuel. Furthermore, it is shown that by properly adjusting

the injection trajectory, the ignition properties of the applied fuel can be used to trigger

different ignition modes.

6.4 Autoignition Control

Within the scope of

publication III

, a clear correlation between the injected fuel

profiles and the resulting autoignition homogeneity was observed (see Fig. 6.3). Based

on these observations in

publication IV

, an automated extremum seeking control

algorithm was developed and applied in cooperation with Florian Arnold from the

Institute of Measurement and Control to control the cycle-averaged formation of different

autoignition modes by optimizing the fuel supply. Therefore, two cost functions are

formulated, both aiming to find the optimum fuel injection trajectory for efficient

operation based on the measured ignition timings and maximum pressure amplitudes.

Hence, cycle-averaged data based on a batch of 20 cycles is considered. The gradient

information is determined by perturbing the initial trajectory incrementally. This

approach is an extended version of a cyclic extremum seeking control [

] to the

integer-constraint case.

The controller performance is examined based on two cost functions

and

, of which

results in an overall increase in pressure amplitude throughout the entire operation

duration. However,

turned out to be more sensitive toward the stochasticity of the

system and is, thus, chosen for further investigations. Figure 6.6 shows the controller

115

6 Discussion

performance for

. The red line is the average standard deviation of the ignition

times along the combustor calculated from ionization probe data. The black lines

represent the average maximum pressure amplitude measured by each sensor for the

respective operating cycle. It is evident that an increased maximum pressure amplitude

is accompanied by an increase in autoignition homogeneity, indicated by a low standard

deviation of the obtained ignition times. This confirms the earlier observations within

the context of

publication III

. The four control steps 1, 3, 4 and 7, highlighted in

Fig. 6.6, correspond to the injection curves discussed in the previous Sec. 6.3. The

resulting local fuel distribution throughout the combustor when applying the respective

control steps is calculated using the previously developed simulation tool (

publication

). Thus, a correlation of the fuel distribution and the obtained ignition characteristics

can be evaluated. The calculated fuel profiles match the previously applied fuel profiles

(Λ-,

- and V-curve) in

publication III

. Along with the observed pressure amplitude

and ignition homogeneity, as shown in Fig. 6.6, the results are in good agreement.

Furthermore, the cycle-averaged results allow for a more robust evaluation of the

ignition characteristics, compared to the single cycle analysis, which revealed different

autoignition modes occurring dependent on the fuel characteristics. These results

provide a solid basis for further improvement of control approaches based on the

different mechanism identified in context of the applied fuel and its characteristics.

116

7 Conclusion

This work investigates the controllability of autoignition characteristics through the

injection of a prescribed mixture profile experimentally, enabling shockless explosion

combustion (SEC).

First, a combustor was designed allowing for repeatable and reliable initiation of

autoignition at elevated temperature and atmospheric pressure conditions for a range

of equivalence ratios using dimethyl ether as fuel. An injection station was designed,

equipped with ten highspeed solenoid valves, allowing for the injection of a stratified

fuel trajectory at high frequencies. The combustor length and ignition delay time of the

applied fuel were matched to allow for autoignition inside the measurement section. For

a bulk velocity of

uair

= 18 m/s, ignition delay times around 39 to 67 ms are required to

enable ignition within the desired combustor section.

Next, the capability of the applied injection strategy of injecting a defined fuel profile

into a convecting air flow was investigated. Furthermore, the impact of turbulent

diffusion and boundary layer effects on the preservation of the injected fuel profile

during convection were evaluated. For this, optical measurements were conducted using

one-dimensional line-of-sight and two-dimensional measurements to obtain the fuel

concentration distribution inside the combustor for different injection trajectories. It

was demonstrated that the applied injection strategy is capable of injecting a defined fuel

profile which stays largely preserved until autoignition occurs. Moreover, it was found

that turbulent diffusion and boundary layer effects affect the gradients in concentration.

However, for a constant spatial width, which was identified as a driving parameter,

the injection profiles are sufficiently preserved enabling a precise control of the fuel

distribution inside the combustor.

Ignition experiments were conducted applying the injection strategy to inject three

model fuel trajectories and evaluate the resulting autoignition characteristics in terms

of pressure rise and ignition homogeneity. The test rig was, therefore, equipped with

a set of pressure transducers and ionization probes. It was shown that the injected

fuel trajectory strongly correlates with the observed autoignition characteristics. The

117

7 Conclusion

gradients in concentration have a significant impact on the number of simultaneously

initiating autoignition fronts as a measure of the autoignition homogeneity. Also,

the results clearly show that an increase in autoignition homogeneity leads to an

increase in the obtained pressure amplitude due to a higher aerodynamic confinement.

Because the SEC operation is impacted by stochasticity, next, optical chemiluminescence

measurements were conducted allowing for the evaluation of single cycles. The results

confirm the earlier observation that an increased ignition homogeneity is linked to a

greater rise in pressure.

These observations were subsequently used to develop a control algorithm to control the

autoignition characteristics by adjusting the fuel injection profile. It was successfully

demonstrated that different modes of autoignition can be initiated, on a cycle-averaged

basis, by the proper adjustment of the injection trajectory. Optical chemiluminescence

measurements of different species revealed multi-stage ignition behavior of the applied

fuel to be responsible for the initiation of different autoignition modes, which can be

triggered by the applied fuel profiles. These are important findings which provide a new

basis for improved control approaches enabling a reliable and repeatable SEC operation

with minimum cycle-to-cycle variation.

The presented results prove the experimental feasibility of the SEC. In this work,

underlying mechanism were successfully analyzed and explained. For the implementation

into a gas turbine application, higher operating frequencies are required. The SEC

offers an alternative to other pressure gain combustion approaches and is a promising

device for combustion applications.

118

8 Outlook

The obtained results are very promising and pave the way towards new fundamental

questions such as: does a supersonic propagating autoignition result in a similar pressure

rise compared to multiple ignition fronts that occur quasi-simultaneously (as illustrated

in Fig. 6.4e–g)? In case of multiple quasi-simultaneous ignition fronts how does the

chemistry promote the ignition process? How does the pressure rise scale with the

number of quasi-simultaneous occurring ignition fronts? Assuming the pressure rise to

be mainly a function of the heat release rate, is there an equation which relates the

number of simultaneous deflagrations to the number of fast autoignitions?

Alongside these more fundamental research questions, practical challenges remain that

need to be tackled in order to enable a reliable and sustainable operation of the SEC

test rig. In the following these challenges will be discussed and further improvements

will be proposed.

8.1 Operation under Elevated Pressure Conditions

preheater restriction

injection

combustor exhaust tube

plenum

FUS FDS

I1I2I3I4I5I6I7I8I9I10 I11 I12 I13 I14 I15

P1P2P3P4P5P6

m1m1,2

sensor/outlet

ports

bypass

Figure 8.1:

Sketch of the high pressure SEC test rig equipped with ionization probes

–I

5), high-speed pressure transducer (P

–P

) and low-speed pressure transducers

(FUS and FDS)

As discussed in Sec. 2, one major drawback of the current test rig is the limitation in

attainable temperature and pressure, which result in elongated ignition delay times of

the applied fuel. This ultimately leads to an increased duration in which the injected

fuel profile is exposed to turbulent diffusion and stochasticity of the flow, causing

119

8 Outlook

cycle-to-cycle variations. Therefore, reducing the ignition delay time would lead to a

great improvement of the SEC operation by an improved controllability of the fuel

distribution inside the combustor. Generally, a reduction in ignition delay time for

a given fuel–air mixture can be achieved by the application of elevated temperature

or pressure conditions. However, since the temperature limit is greatly exhausted,

increasing the initial pressure would be a next step towards lower ignition delay times.

Figure 8.1 shows a test rig which is designed to allow for an operation under elevated

pressure conditions. This test rig is derived from the “single tube SEC” and is composed

of a pressure-resistant preheater (15 bar) which can rise the air temperature up to

1023 K. The air flow is forced through a restriction which prevents backflow of hot gases

subsequent to ignition. An injection station, suitable for gaseous and liquid injection

of fuel is designed, followed by a stainless steel pressure-resistant combustor equipped

with a series of pressure sensor and ionization probes. Downstream of the combustor

an exhaust tube is attached followed by a spherical plenum [83].

The plenum is equipped with multiple ports which serve the purpose of adjusting the

pressure drop inside the combustor. By this, the pressure inside the combustor can be

controlled either by the cross section area of the plenum outlets or by adjusting the

air flow rate, respectively. The idea of a spherical plenum is to generate an increase in

cross section such that the pressure wave subsequent to ignition is expanded generating

an expansion wave traveling upstream. This expansion wave is exploited during the

refilling of the combustor preventing efficiency losses during fuel injection. At the same

time the boundary condition of an acoustically open end stays assured.

However, two obstacles remain that will be targeted in the following discussion. First,

when starting up the test rig, the pressure inside the combustor needs to match

autoignition conditions, such that a initial flame is initiated. This initial flame however,

will ultimately lead to an increase in pressure inside the combustor. Depending on the

firing frequency the pressure thus will gradually increase until an average steady-state

condition is reached. During this process, the pressure inside the combustor needs to

remain below the combustor design pressure. For this reason, a bypass flow is connected

to the plenum equipped with an proportional valve allowing for the precise adjustment

of the pressure inside the combustor. Before the first ignition, the plenum is fed with

an increased air flow rate. Once the ignition process starts, the proportional valve is

gradually closed decreasing the air supply through the bypass, and thus, adjusting the

pressure with regard to the pressure rise due to autoignition.

A second obstacle is the injection strategy at high frequency operation. In order to

take advantage of the suction wave, the fuel injection process needs to be realized

120

8.2 Further Improvements of the SEC Operation

within a few milliseconds. Thus, a gaseous injection, as applied at atmospheric pressure

conditions, is not feasible due to the increased number of valves which are required

in order to injected the required amount of fuel within the given time time. One

suggested solution is to switch to a liquid injection strategy, which allows for the

injection of an increased mass flow rate within short injection times. Possibly, existing

injection strategies which can be found in applications for internal combustion engines

e.g. common-rail technology, which are optimized to deliver a prescribed charge of fuel,

can be utilized. Most likely, piezo-injection valves operating at high inlet pressures, are

suitable for the SEC approach. This strategy however, needs to account for additional

time scales which are typical for spray flame e.g. atomization, evaporation and mixing.

8.2 Further Improvements of the SEC Operation

Up to now the SEC process has been experimentally investigated using dimethyl ether.

However, several investigations reveal that under elevated pressure conditions different

fuels and their composites can be used for an efficient SEC operation. For instance, it

was found that the SEC performance can be improved by tailoring the temperature

sensitivity of ignition delay times using blends of multiple fuels [

]. Another promising

concept is to increase the excitation time, meaning the heat release occurs over a

prolonged time duration. This can be achieved by dilution of the fuel, thus, enabling a

sufficient control in terms of hot spot reactivity. This approach has been intensively

studied in with regard to SEC application by Djordjevic et al. [

] and Zander et al. [

Furthermore, the efficient tailoring of different fuels (liquid or gaseous) in order to affect

the ignition delay time, temperature dependence and excitation time has been studied

in more detail by Cai et al. [13].

Furthermore, an improved performance can be achieved by reducing the pressure drop

at the combustor inlet. By this, the overall efficiency can be further increased. The

concept of a fluidic diode is a promising approach, despite its compromising effect of

creating a certain heat loss in the air path. The concept is based on creating a minimal

pressure loss in downstream direction while simultaneously achieving a high pressure

loss in upstream direction. Thus, a valveless operation is enabled while the backflow of

exhaust gases into the air supply is prevented simultaneously.

The ideas presented in this chapter, provide a basis for further improvements which are

required to implement the SEC into an existing gas turbine cycle, as well as, obtain an

increase in the cycle performance of the SEC operation. Nevertheless, these ideas still

need to be experimentally validated. The SEC performance under elevated pressure

121

8 Outlook

conditions in combination with the liquid fuel injection seems promising but also comes

along with challenges that need to be tackled. Specifically in terms of test rig ramp up,

as well as, liquid injection, which requires the consideration of additional chemical and

physical time scales compared to gaseous injection. Moreover, the impact of different

fuels and their composites must be experimentally analyzed to understand the impact

of the different fuel characteristics on the SEC operation. The same applies to the

fluidic diode, which needs to be characterized in terms of pressure loss and backflow of

hot gases during operation.

122

Bibliography

[1]

Tasneem Abbasi and SA Abbasi. Biomass energy and the environmental impacts

associated with its production and utilization. Renewable and sustainable energy

reviews, 14(3):919–937, 2010.

[2]

JB Abbiss, TW Chubb, and ER Pike. Laser doppler anemometry. Optics & Laser

Technology, 6(6):249–261, 1974.

[3]

Zoha Azizi, Mohsen Rezaeimanesh, Tahere Tohidian, and Mohammad Reza Rahim-

pour. Dimethyl ether: A review of technologies and production challenges. Chemical

Engineering and Processing: Process Intensification, 82:150–172, 2014.

[4]

Gal Berkooz, Philip Holmes, and John L Lumley. The proper orthogonal de-

composition in the analysis of turbulent flows. Annual review of fluid mechanics,

25(1):539–575, 1993.

[5]

Richard Bluemner, Myles D Bohon, Christian O Paschereit, and Ephraim J

Gutmark. Single and counter-rotating wave modes in an RDC. AIAA Paper

2018-1608, January 2018.

[6]

Richard Bluemner, Christian O Paschereit, and Kilian Oberleithner. Generation

and transport of equivalence ratio fluctuations in an acoustically forced swirl burner.

Combustion and Flame, 209:99–116, 2019.

[7]

Richard Blümner. Measurement of equivalence ratio fluctuations in the mixing

section of a swirl-stabilized burner using wavelength modulation spectroscopy.

Master’s thesis, TU Berlin, 2016.

[8]

Bernhard C Bobusch, Phillip Berndt, Christian O Paschereit, and Rupert Klein.

Shockless explosion combustion: An innovative way of efficient constant volume

combustion in gas turbines. Combust. Sci. Technol., 186(10-11):1680–1689, 2014.

[9]

Bernhard C Bobusch, Phillip Berndt, Christian O Paschereit, and Rupert Klein.

Investigation of fluidic devices for mixing enhancement for the shockless explosion

123

Bibliography

combustion process. In Active Flow and Combustion Control 2014, pages 281–297.

Springer, 2015.

[10]

Bernhard C Bobusch, René Woszidlo, JM Bergada, C Navid Nayeri, and Chris-

tian Oliver Paschereit. Experimental study of the internal flow structures inside a

fluidic oscillator. Experiments in fluids, 54(6):1559, 2013.

[11]

Bernhard Christian Bobusch. Fluidic devices for realizing the shockless explosion

combustion process. PhD thesis, Technische Universitaet Berlin (Germany), 2015.

[12]

D Bradley and Gautam T Kalghatgi. Influence of autoignition delay time char-

acteristics of different fuels on pressure waves and knock in reciprocating engines.

Combustion and flame, 156(12):2307–2318, 2009.

[13]

Liming Cai and Heinz Pitsch. Tailoring fuels for a shockless explosion combustor.

In Active Flow and Combustion Control 2014, pages 299–315. Springer, 2015.

[14]

Anindya Chatterjee. An introduction to the proper orthogonal decomposition.

Current science, pages 808–817, 2000.

[15]

Hubert Frank Coward and George William Jones. Limits of flammability of gases

and vapors, volume 503. US Government Printing Office, 1952.

[16]

HJ Curran, SL Fischer, and FL Dryer. The reaction kinetics of dimethyl ether. ii:

Low-temperature oxidation in flow reactors. International Journal of Chemical

Kinetics, 32(12):741–759, 2000.

[17]

HJ Curran, WJ Pitz, CK Westbrook, P Dagaut, J-C Boettner, and M Cathonnet.

A wide range modeling study of dimethyl ether oxidation. International Journal

of Chemical Kinetics, 30(3):229–241, 1998.

[18]

Peng Dai, Chengken Qi, and Zheng Chen. Effects of initial temperature on

autoignition and detonation development in dimethyl ether/air mixtures with

temperature gradient. Proceedings of the Combustion Institute, 36(3):3643–3650,

2017.

[19]

Dantec. Laser doppler anemometry. https://www.dantecdynamics.com/solutions-

applications/solutions/fluid-mechanics/laser-doppler-anemometry-

lda/measurement-principles-of-lda/. (accessed: 04.01.2021).

[20]

Neda Djordjevic, Mike Rekus, Johann Vinkeloe, and Lisa Zander. Shock tube and

kinetic study on the effects of co2 on dimethyl ether autoignition at high pressures.

Energy & Fuels, 33(10):10197–10208, 2019.

124

Bibliography

[21]

Lars Eriksson and Lars Nielsen. Ionization current interpretation for ignition control

in internal combustion engines. Control Engineering Practice, 5(8):1107–1113, 1997.

[22]

SL Fischer, FL Dryer, and HJ Curran. The reaction kinetics of dimethyl ether. i:

High-temperature pyrolysis and oxidation in flow reactors. International Journal

of Chemical Kinetics, 32(12):713–740, 2000.

[23]

David G. Goodwin, Raymond L. Speth, Harry K. Moffat, and Bryan W. Weber.

Cantera: An object-oriented software toolkit for chemical kinetics, thermodynamics,

and transport processes. https://www.cantera.org, 2018. Version 2.4.0.

[24]

JAT Gray, M Lemke, J Reiss, CO Paschereit, J Sesterhenn, and JP Moeck.

A compact shock-focusing geometry for detonation initiation: experiments and

adjoint-based variational data assimilation. Combust. Flame, 183:144–156, 2017.

[25]

JAT Gray, J Vinkeloe, J Moeck, CO Paschereit, P Stathopoulos, P Berndt, and

R Klein. Thermodynamic evaluation of pulse detonation combustion for gas turbine

power cycles. In Turbo Expo: Power for Land, Sea, and Air, volume 49767, page

V04BT04A044. American Society of Mechanical Engineers, 2016.

[26]

JF Griffiths and SK Scott. Thermokinetic interactions: Fundamentals of spon-

taneous ignition and cool flames. Progress in energy and combustion science,

13(3):161–197, 1987.

[27]

XJ Gu, DR Emerson, and D Bradley. Modes of reaction front propagation from

hot spots. Combust. Flame, 133(1-2):63–74, 2003.

[28]

GG Guilbault. Principles of luminescence spectroscopy. luminescent determination

of clinically and agriculturally important samples. Pure and Applied Chemistry,

57(3):495–514, 1985.

[29]

Krishna Kumar Gupta, Ameenur Rehman, and RM Sarviya. Bio-fuels for the gas

turbine: A review. Renewable and Sustainable Energy Reviews, 14(9):2946–2955,

2010.

[30]

Fabian E Habicht, Fatma C Yücel, Joshua AT Gray, and Christian O Paschereit.

Detonation initiation by shock focusing at elevated pressure conditions in a pulse

detonation combustor. International Journal of Spray and Combustion Dynamics,

12:1756827720921718, 2020.

[31]

George E Halkos and Eleni-Christina Gkampoura. Reviewing usage, potentials,

and limitations of renewable energy sources. Energies, 13(11):2906, 2020.

[32]

Arthur M Halpern and William R Ware. Excited singlet state radiative and

125

Bibliography

nonradiative transition probabilities for acetone, acetone-d 6, and hexafluoroacetone

in the gas phase, in solution, and in the neat liquid. The Journal of Chemical

Physics, 54(3):1271–1276, 1971.

[33]

Ronald K Hanson and David F Davidson. Recent advances in laser absorption and

shock tube methods for studies of combustion chemistry. Progress in Energy and

Combustion Science, 44:103–114, 2014.

[34]

Julian Heicklen. The fluorescence and phosphorescence of biacetyl vapor and

acetone vapor1. Journal of the American Chemical Society, 81(15):3863–3866,

1959.

[35]

Olivier Herbinet and Guillaume Dayma. Jet-stirred reactors. In Cleaner combustion,

pages 183–210. Springer, 2013.

[36]

IEA. Iea - international energy agency. https://www.iea.org/. (accessed:

11.01.2021).

[37]

Alexander Jaeschke. Master thesis development of an acetone-plif system for fuel

concentration measurements in a shockless explosion combustor. Master’s thesis,

TU Berlin, 2020.

[38]

K Kailasanath. Recent developments in the research on pulse detonation engines.

AIAA journal, 41(2):145–159, 2003.

[39]

Trupti Kathrotia. Reaction Kinetics Modeling of OH*, CH*, and C2* Chemilu-

minescence. PhD thesis, Fakultät für Chemie und Geowissenschaften, Universität

Heidelberg, 2011.

[40]

Trupti Kathrotia, Uwe Riedel, Angela Seipel, Kai Moshammer, and Andreas

Brockhinke. Experimental and numerical study of chemiluminescent species in

low-pressure flames. Applied Physics B, 107(3):571–584, 2012.

[41]

Sei Zhen Khong, Dragan Nešić, and Miroslav Krstić. Iterative learning control

based on extremum seeking. Automatica, 66:238–245, 2016.

[42]

H Li, Aamir Farooq, JB Jeffries, and RK Hanson. Near-infrared diode laser

absorption sensor for rapid measurements of temperature and water vapor in a

shock tube. Applied Physics B, 89(2):407–416, 2007.

[43]

Hejie Li, Gregory B Rieker, Xiang Liu, Jay B Jeffries, and Ronald K Hanson. Ex-

tension of wavelength-modulation spectroscopy to large modulation depth for diode

laser absorption measurements in high-pressure gases. Applied optics, 45(5):1052–

1061, 2006.

126

Bibliography

[44]

Hejie Li, Shawn D Wehe, and Keith R McManus. Real-time equivalence ratio

measurements in gas turbine combustors with a near-infrared diode laser sensor.

P. Combust. Inst., 33(1):717–724, 2011.

[45]

Yang Li, Chong-Wen Zhou, Kieran P Somers, Kuiwen Zhang, and Henry J Cur-

ran. The oxidation of 2-butene: A high pressure ignition delay, kinetic modeling

study and reactivity comparison with isobutene and 1-butene. Proceedings of the

Combustion Institute, 36(1):403–411, 2017.

[46]

JTC Liu, JB Jeffries, and RK Hanson. Wavelength modulation absorption spec-

troscopy with 2 f detection using multiplexed diode lasers for rapid temperature

measurements in gaseous flows. Applied Physics B, 78(3-4):503–511, 2004.

[47]

A Lozano, B Yip, and RK Hanson. Acetone: a tracer for concentration measure-

ments in gaseous flows by planar laser-induced fluorescence. Experiments in fluids,

13(6):369–376, 1992.

[48]

Frank K Lu and Eric M Braun. Rotating detonation wave propulsion: experimental

challenges, modeling, and engine concepts. Journal of Propulsion and Power,

30(5):1125–1142, 2014.

[49]

Mohammad S Masnadi, John R Grace, Xiaotao T Bi, C Jim Lim, and Naoko

Ellis. From fossil fuels towards renewables: Inhibitory and catalytic effects on

carbon thermochemical conversion during co-gasification of biomass with fossil

fuels. Applied Energy, 140:196–209, 2015.

[50]

Matlab. fitgeotrans. https://de.mathworks.com/help/images/ref/fitgeotrans.html.

(accessed: 05.01.2021).

[51]

Matlab. Matlab tutorial. http://kiwi.bridgeport.edu/cpeg540/Matlab%20Tutorials.pdf.

(accessed: 05.01.2021).

[52]

Kok Leong Ng, D Chadwick, and BA Toseland. Kinetics and modelling of dimethyl

ether synthesis from synthesis gas. Chemical Engineering Science, 54(15-16):3587–

3592, 1999.

[53]

Takashi Ogawa, Norio Inoue, Tutomu Shikada, and Yotaro Ohno. Direct dimethyl

ether synthesis. Journal of natural gas chemistry, 12(4):219–227, 2003.

[54]

Jiaying Pan, Haiqiao Wei, Gequn Shu, Zheng Chen, and Peng Zhao. The role

of low temperature chemistry in combustion mode development under elevated

pressures. Combustion and Flame, 174:179–193, 2016.

[55]

PH Paul, MP Lee, and RK Hanson. Molecular velocity imaging of supersonic flows

127

Bibliography

using pulsed planar laser-induced fluorescence of no. Optics Letters, 14(9):417–419,

1989.

[56]

Phillip H Paul and Habib N Najm. Planar laser-induced fluorescence imaging of

flame heat release rate. In Symposium (International) on Combustion, volume 27,

pages 43–50. Elsevier, 1998.

[57]

Sabrina Polierer, David Guse, Stefan Wild, Karla Herrera Delgado, Thomas N Otto,

Thomas A Zevaco, Matthias Kind, Jörg Sauer, Felix Studt, and Stephan Pitter.

Enhanced direct dimethyl ether synthesis from co2-rich syngas with cu/zno/zro2

catalysts prepared by continuous co-precipitation. Catalysts, 10(8):816, 2020.

[58]

TS Rähse, CO Paschereit, P Stathopoulos, P Berndt, and R Klein. Gas dynamic

simulation of shockless explosion combustion for gas turbine power cycles. In Turbo

Expo: Power for Land, Sea, and Air, volume 50831, page V003T06A005. American

Society of Mechanical Engineers, 2017.

[59]

Gregory B Rieker, Jay B Jeffries, and Ronald K Hanson. Calibration-free

wavelength-modulation spectroscopy for measurements of gas temperature and

concentration in harsh environments. Applied optics, 48(29):5546–5560, 2009.

[60]

Gregory Brian Rieker. Wavelength-modulation spectroscopy for measurements of

gas temperature and concentration in harsh environments. Stanford University

Stanford, 2009.

[61]

Laurence S Rothman, Iouli E Gordon, Yury Babikov, Alain Barbe, D Chris Benner,

Peter F Bernath, Manfred Birk, Luca Bizzocchi, Vincent Boudon, Linda R Brown,

et al. The hitran2012 molecular spectroscopic database. Journal of Quantitative

Spectroscopy and Radiative Transfer, 130:4–50, 2013.

[62]

J Rudloff, J-M Zaccardi, S Richard, and JM Anderlohr. Analysis of pre-ignition in

highly charged si engines: Emphasis on the auto-ignition mode. Proceedings of the

Combustion Institute, 34(2):2959–2967, 2013.

[63]

Nicolae Scarlat and Jean-François Dallemand. Recent developments of bio-

fuels/bioenergy sustainability certification: A global overview. Energy policy,

39(3):1630–1646, 2011.

[64]

Stephane Schilt, Luc Thevenaz, and Philippe Robert. Wavelength modulation

spectroscopy: combined frequency and intensity laser modulation. Applied optics,

42(33):6728–6738, 2003.

[65]

Christof Schulz and Volker Sick. Tracer-lif diagnostics: quantitative measurement of

128

Bibliography

fuel concentration, temperature and fuel/air ratio in practical combustion systems.

Progress in energy and combustion science, 31(1):75–121, 2005.

[66]

Troy A Semelsberger, Rodney L Borup, and Howard L Greene. Dimethyl ether

(dme) as an alternative fuel. Journal of power sources, 156(2):497–511, 2006.

[67]

NN Semenov. Thermal theory of combustion and explosion. 3; theory of normal

flame propagation. Progress of Physical Science, 24, 1942.

[68]

N.N. Semenov. Some Problems of Chemical Kinetics and Reactivity: Volume 1,

volume 1. Elsevier, 2013.

[69]

Ronald S Sheinson and Frederick W Williams. Chemiluminescence spectra from

cool and blue flames: Electronically excited formaldehyde. Combustion and Flame,

21(2):221–230, 1973.

[70]

John M Simmie. Detailed chemical kinetic models for the combustion of hydrocar-

bon fuels. Progress in energy and combustion science, 29(6):599–634, 2003.

[71]

György Sitkei. Kraftstoffaufbereitung und Verbrennung bei Dieselmotoren. Springer-

Verlag, 2013.

[72]

VJ Smith and JW Robinson. Ultraviolet emission and absorption spectra produced

by organic compounds in oxyhydrogen flames1: Emission spectra. Analytica

Chimica Acta, 48(2):391–403, 1969.

[73]

Charles G Speziale. Analytical methods for the development of reynolds stress

closures in turbulence. Final Report Institute for Computer Applications in Science

and Engineering, 1990.

[74]

Panagiotis Stathopoulos, Johann Vinkeloe, and Christian Oliver Paschereit. Ther-

modynamic evaluation of constant volume combustion for gas turbine power cycles.

In Proceedings of the 11th International Gas Turbine Congress, Tokyo, Japan,

pages 15–20, 2015.

[75]

Richard Stone. Introduction. In Introduction to Internal Combustion Engines.

Springer, 1992.

[76]

Jian Sun, Guohui Yang, Yoshiharu Yoneyama, and Noritatsu Tsubaki. Catalysis

chemistry of dimethyl ether synthesis. ACS Catalysis, 4(10):3346–3356, 2014.

[77]

Kunpeng Sun, Weiwei Lu, Fengyan Qiu, Shuwen Liu, and Xianlun Xu. Direct

synthesis of dme over bifunctional catalyst: surface properties and catalytic perfor-

mance. Applied Catalysis A: General, 252(2):243–249, 2003.

129

Bibliography

[78]

Chih-Jen Sung and Henry J Curran. Using rapid compression machines for chemical

kinetics studies. Progress in Energy and Combustion Science, 44:1–18, 2014.

[79]

James M Supplee, Edward A Whittaker, and Wilfried Lenth. Theoretical descrip-

tion of frequency modulation and wavelength modulation spectroscopy. Applied

Optics, 33(27):6294–6302, 1994.

[80]

Hiroshi Terashima, Hisashi Nakamura, Akira Matsugi, and Mitsuo Koshi. Role of

low-temperature oxidation in non-uniform end-gas autoignition and strong pressure

wave generation. Combustion and Flame, 223:181–191, 2021.

[81]

William Thomson. The bakerian lecture.â

€

”on the electro-dynamic qualities of

metals. Philosophical Transactions of the Royal Society of London, 146(3):649–751,

1856.

[82]

MC Thurber and RK Hanson. Simultaneous imaging of temperature and mole

fraction using acetone planar laser-induced fluorescence. Experiments in fluids,

30(1):93–101, 2001.

[83]

Giordana Tornow and Rupert Klein. A 1d multi-tube code for the shockless

explosion combustion. In Active Flow and Combustion Control 2018, pages 321–

335. Springer, 2019.

[84]

Alberto Varone and Michele Ferrari. Power to liquid and power to gas: An

option for the german energiewende. Renewable and Sustainable Energy Reviews,

45:207–218, 2015.

[85]

Johann Vinkeloe, Lisa Zander, Martin Szeponik, and Neda Djordjevic. Tailoring

the temperature sensitivity of ignition delay times in hot spots using fuel blends of

dimethyl ether, methane, and hydrogen. Energy & Fuels, 34(2):2246–2259, 2019.

[86]

Jurgen Warnatz, Ulrich Maas, Robert W Dibble, and J Warnatz. Combustion.

Springer, 2006.

[87]

Julien Weiss. A tutorial on the proper orthogonal decomposition. In AIAA Aviation

2019 Forum, page 3333, 2019.

[88]

Fatma C Yücel, Fabian Völzke, and Christian O Paschereit. Effect of the switching

times on the operating behavior of a shockless explosion combustor. In Active

Flow and Combustion Control 2018, pages 121–134. Springer, 2019.

[89]

Fatma Cansu Yücel, Fabian Habicht, Myles Bohon, and Christian Oliver Paschereit.

Autoignition in stratified mixtures for pressure gain combustion. Proceedings of

the Combustion Institute, 2020.

130

Bibliography

[90]

Fatma Cansu Yuecel, Fabian Enrico Habicht, and Christian Oliver Paschereit.

Investigation of different techniques for autoignition detection. In ICPCD 2020,

2020.

[91]

Lisa Zander, Giordana Tornow, Rupert Klein, and Neda Djordjevic. Knock control

in shockless explosion combustion by extension of excitation time. In Active Flow

and Combustion Control 2018, pages 151–166. Springer, 2019.

[92]

Lisa Zander, Johann Vinkeloe, and Neda Djordjevic. Ignition delay and chemical–

kinetic modeling of undiluted mixtures in a high-pressure shock tube: Nonideal

effects and comparative uncertainty analysis. International Journal of Chemical

Kinetics, 2021.

[93]

Ya B Zeldovich. Regime classification of an exothermic reaction with nonuniform

initial conditions. Combust. Flame, 39(2):211–214, 1980.

[94]

Ya B Zeldovich. To the question of energy use of detonation combustion. Journal

of propulsion and power, 22(3):588–592, 2006.

[95]

Weikuo Zhang, Mahdi Faqih, Xiaolong Gou, and Zheng Chen. Numerical study

on the transient evolution of a premixed cool flame. Combustion and Flame,

187:129–136, 2018.

[96]

Chong-Wen Zhou, Yang Li, Ultan Burke, Colin Banyon, Kieran P Somers, Shuiting

Ding, Saadat Khan, Joshua W Hargis, Travis Sikes, Olivier Mathieu, et al. An

experimental and chemical kinetic modeling study of 1, 3-butadiene combustion:

Ignition delay time and laminar flame speed measurements. Combustion and Flame,

197:423–438, 2018.

131

A Publications Associated with this

Thesis

During my time as a researcher at the TU Berlin I have been given the opportunity of

working on two projects. Besides my main project, the shockless explosion combustion,

I have been working on the project of pulse detonation combustion. Both projects aim

for a pressure gain combustion and thus contributed to this thesis.

Published peer-reviewed papers:

•

Schäpel, J. S., King, R.,

Yücel, F.

, Völzke, F., Paschereit, C. O., & Klein, R.

(2018, June). Fuel injection control for a valve array in a shockless explosion

combustor. In Turbo Expo: Power for Land, Sea, and Air (Vol. 51128, p.

V006T05A007). American Society of Mechanical Engineers.

URL: https://doi.org/10.1115/GT2018-75295

•Yücel, F. C.

, Völzke, F., & Paschereit, C. O. (2019). Effect of the switching

times on the operating behavior of a shockless explosion combustor. In Active

Flow and Combustion Control 2018 (pp. 121-134). Springer, Cham.

URL: https://doi.org/10.1007/978-3-319-98177-2_8

•

Völzke, F. E.,

Yücel, F. C.

, Gray, J. A., Hanraths, N., Paschereit, C. O., & Moeck,

J. P. (2019). The Influence of the Initial Temperature on DDT Characteristics in

a Valveless PDC. In Active Flow and Combustion Control 2018 (pp. 185-196).

Springer, Cham.

URL: https://doi.org/10.1007/978-3-319-98177-2_12

•Yücel, F. C.

, Habicht, F., Bohon, M., & Paschereit, C. O. (2020). Autoignition

in stratified mixtures for pressure gain combustion. Proceedings of the Combustion

Institute.

URL: https://doi.org/10.1016/j.proci.2020.07.108

•

Habicht, F. E.,

Yücel, F. C.

, Gray, J. A., & Paschereit, C. O. (2020). Det-

onation initiation by shock focusing at elevated pressure conditions in a pulse

133

A Publications Associated with this Thesis

detonation combustor. International Journal of Spray and Combustion Dynamics,

12, 1756827720921718. (open access publication)

URL: https://doi.org/10.1177/1756827720921718

•Yücel, F. C.

, Habicht, F., Jaeschke, A., Lückoff, F., Oberleithner, K., &

Paschereit, C. O. (2021). Investigation of the Fuel Distribution in a Shock-

less Explosion Combustor. Journal of Engineering for Gas Turbines and Power,

143(1).

URL: https://doi.org/10.1115/1.4049220

•

Habicht, F.,

Yücel, F. C.

, Hanraths, N., Djordjevic, N., & Paschereit, C. O.

(2020). Lean Operation of a Pulse Detonation Combustor by Fuel Stratification.

Journal of Engineering for Gas Turbines and Power.

URL: https://doi.org/10.1115/1.4048775

•Yücel, F. C.

, Habicht, F., Arnold, F., King, R., Bohon, M. and Paschereit, C.O.

(2021). Controlled autoignition in stratified mixtures. In Combustion and Flame,

232, p.111533. (open access publication)

URL: https://doi.org/10.1016/j.combustflame.2021.111533

•

Habicht, F.,

Yücel, F. C.

, Haghdoost, M., Oberleithner, K. & Paschereit, C.

O. (2020). Acoustic Modes in a Plenum Downstream of a Multi-tube Pulse

Detonation Combustor. AIAA Journal. (submitted 2020)

Published non-peer-reviewed papers:

•

Arnold, F., King, R.,

Yücel, F.

, Völzke, F., & Paschereit, C. O. (2019). Modelling

of Fuel Transport for a Shockless Explosion Combustion Process. In AIAA Scitech

2019 Forum (p. 1254).

URL: https://doi.org/10.2514/6.2019-1254

•Yücel, F. C.

, Habicht, F. E., & Paschereit, C. O. (2020, February). Investigation

of Different Techniques for Autoignition Detection. In ICPCD 2020.

•

Habicht, F. E.,

Yücel, F. C.

, Haghdoost, M. R., Oberleithner, K., & Paschereit,

C. O. (2020, February). Pressure fluctuations in air supply of a valveless PDC-

multitube. In ICPCD 2020.

134