The European Commission is planning to limit emissions under real driving conditions up to high

engine loads. RDE (real driving emissions) legislation demands the complete conversion of exhaust
gases in the catalytic converter which can only be achieved for spark-ignition engines at lambda=1.
High exhaust gas temperatures resulting from late centers of heat release caused by knock can
then no longer be limited by mixture enrichment. In addition, higher mean effective pressures are
needed to improve the efficiency of SI engines. A strong tendency to knock during stoichiometric
combustion in conjunction with high mean effective pressure places exacting demands on the SI
engine combustion process.

The focus of engine development consequently remains on reducing knock and on avoiding
irregular combustion events. In particular, phenomena such as pre-ignition, which is typically
observed in downsizing concepts, or extreme knock of the type frequently occurring in high-
Knocking in
compression lean-burn concepts, are immense challenges to developers.
Gasoline Engines

Kratzsch , Günther: Knocking in Gasoline Engines

Contents:
Potentials and limits of downsizing | Mega-knock in super-charged gasoline engines interpreted as a
localized developing detonation | A contribution to better understanding the pre-ignition phenomenon
in highly charged internal combustion engines with direct fuel injection | Minimising autoignition for
optimum efficiency in high specific output spark-ignited engines | Reduction in knocking intensity of Matthias Kratzsch, Michael Günther
an SI engine by in-cylinder temperature stratification | New approach to the determination of knock
onset | Cylinder pressure-based knock detection – challenges in cylinder pressure indication and (ed.)
application in a new engine-based fuel test method | Irregular combustion: development and
calibration of highly boosted SI engines | Optically diagnosing combustion anomalies as part of
designing the combustion process | Using surface thermocouples and light conductor
measurements to examine the thermal load on a gasoline engine’s components during knocking

fuel diesel-methane engines | LEC-GPN – a new Index for assessing the knock behavior of gaseous
engine operation | Comparative analysis of low-speed pre-ignition phenomena in SI gasoline and dual

fuels for large engines | A statistical modeling approach with detailed chemical kinetics for use in 3D-
CFD engine knock predictions | Investigation on knocking combustion with reaction kinetics for a
turbocharged SIDI engine | Knocking simulation at Mercedes-Benz – application in series production
development | Multizone simulation of gasoline engine combustion anomaly based on reaction
kinetics | The DELTA knocking control – the necessary paradigm shift for engines with high power
density | Artificial Intelligence for knock detection | Knock detection strategies based on engine
acoustic emission analysis | Continental’s pre-ignition and glow ignition function – detection and
avoidance of irregular combustions | Pre-ignition analysis on a turbocharged gasoline engine with
direct injection | Knock and irregular combustion – challenges for the new turbocharged, high-
performance four-cylinder AMG engine | Simulations and experimental investigations of intermittent
pre-ignition series in a turbocharged DISI engine

Target group:
This book addresses engine developers working for car manufacturers and suppliers. With regard
to knocking combustion in spark-ignition engines – irregular combustion – it provides an overview
of thermodynamic principals, approaches to measurement and computation together with current
trends for mass-production development.

Editors:
Dipl.-Ing. Matthias Kratzsch, Executive Vice President Development Powertrain, IAV GmbH
Dipl.-Ing. Michael Günther, Department Manager / Combustion/Thermodynamics SI Engines, IAV GmbH

ISBN 978-3-9449-7604-4

www.druckcenter.de
9 783944 976044
Matthias Kratzsch, Michael Günther

Knocking in Gasoline Engines

Knocking in
Gasoline Engines
Dipl.-Ing. Matthias Kratzsch
Dipl.-Ing. Michael Günther (ed.)

and 82 Co-Authors

With 286 Figures and 20 Tables

Bibliografische Information Der Deutschen Bibliothek
Die Deutsche Bibliothek verzeichnet diese Publikation
in der Deutschen Nationalbibliografie;
detallierte bibliografische Daten sind im Internet über
http://www.dnb.de abrufbar.

Bibliographic Information published by Die Deutschen Bibliothek

Die Deutsche Bibliothek lists this publication
in the Deutsche Nationalbibliografie;
detailed bibliographic data are available on the internet at
http://www.dnb.de

Bei der Erstellung des Buches wurde mit großer Sorgfalt vergegangen; troztdem lassen sich
Fehler nie vollständig ausschließen. Verlag und Autoren können für fehlerhafte Angaben und
deren Folgen weder eine juristische Verantwortung noch irgendeine Haftung übernehmen.
Für Verbesserungsvorschläge und Hinweise auf Fehler sind Verlag und Autoren dankbar.

© 2013 by DCM Druck Center Meckenheim GmbH

Werner-von-Siemens-Straße 13
53340 Meckenheim
Tel. : 0 22 25 88 93-550 / Fax: 88 93-558
Mail: dcm@druckcenter.de
www.druckcenter.de
Alle Rechte vorbehalten
Printed in Germany

Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung
außerhalb der Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlags unzu-
lässig und strafbar. Dies gitl insbesondere für Vervielfältigungen, Übersetzungen, Mikrover-
filmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen.
Preface
The European Commission is planning to limit emissions under real dri-
ving conditions up to high engine loads. RDE (real driving emissions) le-
gislation demands the complete conversion of exhaust gases in the ca-
talytic converter which can only be achieved for spark-ignition engines at
lambda=1. High exhaust gas temperatures resulting from late centers of
heat release caused by knock can then no longer be limited by mixture
enrichment. In addition, higher mean effective pressures are needed to
improve the efficiency of SI engines. A strong tendency to knock during
stoichiometric combustion in conjunction with high mean effective pressu-
re places exacting demands on the SI engine combustion process.

For this reason, the focus in engine development remains on reducing

knock and on avoiding irregular combustion events. In particular, phen-
omena, such as pre-ignition, which is typically observed in downsizing
concepts, or extreme knock of the type frequently occurring in high-com-
pression lean-burn concepts, are immense challenges to developers.

With the 3rd Conference on ‘SI Engine Knock – Irregular Combustion” ha-
ving placed a particular focus on pre-ignition events, this year’s conferen-
ce provides broad coverage of research results, approaches to simulation,
measurement technology and control algorithms.

Participants from various fields of engine development at manufacturers,

components suppliers and universities can expect to receive a comprehen-
sive picture of the latest research and development results and have the
opportunity to engage in intensive dialog on the subject matter covered.

Acknowledgements:

Our sincere thanks go to the members of the program committee for their
active involvement in preparing this event and for playing a key part in
making it a success.

We also wish to express our particular gratitude to the authors and co-au-
thors who have made this book possible with their truly outstanding con-
tributions.

Berlin, December 2013 Matthias Kratzsch

Michael Günther
Contents
1 Potentials and limits of downsizing 9
Hermann Rottengruber, Eike Christian Todsen

2 Mega-knock in super-charged gasoline­engines

interpreted as a localized ­developing detonation 23
Norbert Peters, Bruno Kerschgens, Bernhard Jochim,
Günter Paczko

3 A contribution to better understanding the pre-ignition

phenomenon in highly charged internal combustion
engines with direct fuel injection 41
Arndt Döhler, Stefan Pritze

4 Minimising autoignition for optimum e ­ fficiency

in high specific output spark-ignited engines 62
Paul Freeland, Marco Warth, Bernd Mahr

5 Reduction in knocking intensity of an SI engine

by in-cylinder temperature stratification 88
Yasuo Moriyoshi, Tatsuya Kuboyama, Taku Kondo,
Hiroshi Sono

6 New approach to the determination of knock onset 101

Sebastian Ohler

7 Cylinder pressure-based knock detection –

challenges in cylinder pressure indication and
application in a new engine-based fuel test method 119
Karl Huber, Johann Hauber

8 Irregular combustion: development and calibration

of highly boosted SI engines 142
Thomas Dobes, Alois Hirsch, Paul Kapus, Mario Ninaus,
Martin Ogris, Harald Philipp, Ernst Winklhofer

9 Optically diagnosing combustion anomalies as

part of designing the combustion process 159
Philipp Adomeit, André Brunn, Carsten Dieterich,
Marco Günther, Silja Klier
10 Using surface thermocouples and light conductor
measurements to examine the thermal load on a gasoline
engine’s components during knocking engine operation 177
Karsten Michels, Christopher Gessenhardt, Jörg Theobald

11 Comparative analysis of low-speed pre-ignition

phenomena in SI gasoline and dual fuel diesel-methane
engines 200
Jean-Marc Zaccardi, David Serrano

12 LEC-GPN – a new Index for assessing the knock

behavior of gaseous fuels for large engines 239
Andreas Wimmer, Franz Chmela, Martin Kirsten,
Gerhard Pirker, Peter Christiner, Christian Trapp,
Herbert Schaumberger

13 A statistical modeling approach with detailed chemical

kinetics for use in 3D-CFD engine knock predictions 255
Dirk Linse, Andreas Kleemann, Christian Hasse

14 Investigation on knocking combustion with reaction

kinetics for a turbocharged SIDI engine 273
Werner Holly, Michael Heiss, Nikola Bobicic, Thomas Lauer,
Stefan Pritze

15 Knocking simulation at Mercedes-Benz –

application in series production d
­ evelopment 293
Arnold Kaden, Alfred Frommelt, Helmut Gildein,
Eberhard Kraus, Uwe Schaupp

16 The DELTA knocking control – the necessary

paradigm shift for engines with high power density 308
Alexander Stahr, Philipp Langfritz, Michael Guenther,
Matthias Kratzsch

17 Artificial Intelligence for knock detection 331

Matthias Biehl, Elvira Perless, Robert Sloboda
18 Knock detection strategies based on engine
acoustic emission analysis 347
Filippo Cavanna, Nicola Garagnani, Riccardo Lanzoni,
Stefano Sgatti, Massimo Zanotti

19 Continental’s pre-ignition and glow i­gnition function –

detection and a
­ voidance of irregular combustions 360
Markus Kieberger, Dominik Biehl

20 Pre-ignition analysis on a turbocharged gasoline

engine with direct injection 380
Erik Schünemann, Andreas Witt, Markus Selder,
Christian Schwarz

21 Knock and irregular combustion – c ­ hallenges for the

new turbocharged, highperformance four-cylinder
AMG ­engine 394
Benjamin Kraus, Sebastian Wolf, Jürgen Fischer

22 Simulations and experimental investigations of

intermittent pre-ignition series in a turbocharged
DISI engine 414
Stefan Palaveev, Max Magar, Christian Disch, Robert Schießl,
Heiko Kubach, Ulrich Spicher, Ulrich Maas, Thomas Koch 414

The Authors 443

1 Potentials and limits of downsizing
Hermann Rottengruber, Eike Christian Todsen

Abstract
By introducing binding CO2-emission standards for light duty vehicles during
the last decade, a certain pressure was built upon SI-engine developers wor-
ldwide. One opportunity to reduce engine fuel consumption of engines, espe-
cially in the statutory low load testing cycles like the NEDC, was Downsizing.

Downsizing in this context means a reduction in displacement and mostly

also cylinder number, without a deduction in engine peak power and per-
formance.

Turbo charging in combination with gasoline direct injection were the key
technologies for enabling this optimization in fuel consumption for the sta-
tutory driving cycles.

Nowadays that numerous downsized powertrain concepts have reached

the market, a gap can be seen between fuel consumption in real-life ve-
hicle operation and the certified fuel consumption measures, in e.g. the
NEDC. Also new regulations for measuring fuel consumption and pollutant
emissions raise the question if downsizing as singular measure for rea-
ching CO2-emission standards is coming to its boundaries.

Kurzfassung
Mit der weltweiten Einführung von verbindlichen CO2-Emissionsgren-
zwerten für den PKW-Verkehr wurde im vergangenen Jahrzehnt der Druck
auf die Entwickler von Pkw-Ottomotoren erhöht. Eine wirksame Möglich-
keit den Verbrauch in den niedriglastigen gesetzlichen Zertifizierungszyk-
len, wie dem NEFZ, abzusenken, ist besonders beim Ottomotor das sog.
Downsizing. Damit ist die Reduktion von Hubraum und der Zylinderzahl
ohne Abstriche bei der Motorleistung zu machen, zu verstehen.

Die Abgasturboaufladung in Kombination mit Direkteinspritzung ist die

Schlüsseltechnologie, welche die Verbrauchsoptimierung in den Zertifi-
zierungszyklen möglich gemacht hat.

9
Da es nun sehr viele Downsizing-Antriebskonzepte auf dem Markt gibt,
öffnet sich die Schere zwischen realem Verbrauch beim Kunden und den
Zulassungswerten sehr stark. Mit Einführung neuer Regelwerke zur Be-
stimmung von Verbrauch und Schadstoffemissionen scheint das Downsi-
zing als einzige Maßnahme zur Erfüllung der Verbrauchsgrenzwerte an
Grenzen zu stoßen.

1. Introduction
Engine development has always been influenced by reduction of ener-
gy consumption with regard to limited resources. It is approx. 15 years
ago that CO2 was identified as important factor for the often cited clima-
te change. This fact has caused a growing debate in social and politi-
cal life about energy consumption and particularly with regard to perso-
nal an commercial traffic. Spark-ignited (SI) engines dominate the world
market for passenger cars, light-duty commercial vehicles and powered
two-wheeler. Therefore, it results high development needs for SI engines.
It means high effort for the engine and car manufacturers to meet statutory
regulations concerning carbon dioxide (see fig. 1).

Figure 1: CO2-emission limit values for new registered passenger cars

worldwide until 2020 - 2025 [2]

10
It is possible to improve efficiency by enhancing the vehicle design and by
electrifying the powertrain. But it is also essential to implement effective
measures at the internal combustion engine (ICE).

Downsizing is one example of such an improvement; it is particularly ef-

fective, due to the comparatively poor efficiency of SI engines in part load
range.

Regarding fuel-consumption smaller, lighter, turbocharged SI engines with

reduced cylinder number have a high advantage in test cycles with a high
amount of part load, e.g. NEDC, in comparison to ICE with equal power
but larger displacement.

The most important effect for this enhanced fuel economy behavior is not
the reduction of engine weight and friction, by reducing the cylinder num-
ber, although these effects are not unessential. The crucial point is that the
required load points for the test cycle are shifted upwards to higher load
points in the engine operating map of a downsized engine. This results in
more efficient operating conditions, i.e. less fuel consumption (see fig. 2).

The disadvantage of these measures is that these downsized, turboch-

arged engines operating even at relative moderate driving conditions, in
load ranges, which are equal in respect of mean effective pressure to
those of naturally aspirated engines under full load.

11
Figure 2: Upward shift of engine load and specific power by downsizing [3]

In these use-cases engine components must be protected from thermal

overload and irregular combustion effects. Especially engine knock has
to be controlled, due to massively increased displacement-specific torque
and power.

2. Fuel Economy Potentials of Downsizing

The specific load of a downsized engine will increase in a specified test
cycle if the engine load is constant and the displacement is reduced.

Figure 3 shows the reduced efficiency losses by raising the load points.
The engines feature direct-injection with homogenous and stoichiometric
air-fuel-mixture (λ=1) distribution.

The mean effective pressure increases from 2.0 bar up to 3.3 bar at the
1.2 l engine. This applies for both engines to achieve the required torque
of 32 Nm.

The increased efficiency, which was mentioned in the introduction, can be

partly explained by the thermodynamic loss analysis according to Witt [3]
as the thermodynamic losses are reduced.

12
Figure 3: Analysis of the fuel consumption potential by downsizing by
means of the thermodynamic loss division

The small-displacement engine has still the advantage of a higher effec-

tive efficiency ηe. That applies under the boundary conditions that the
losses of the imperfect combustion (HC and CO emissions) are constant
and the efficiency of the constant volume cycle decreases due to the dimi-
nished compression ratio ε, which is caused by turbocharging.

However, an improved efficiency is achieved by:

• decreased gas exchange losses thanks to enhanced dethrottling,

• diminished wall heat losses due to a compact combustion chamber de-

sign,

• decreased mechanical losses due to less masses in motion and less

friction of the crankshaft drive.

The more the real engine cycle differs from the constant volume cycle, the
higher are efficiency losses. This is influenced by the selection of the ideal
ignition point and the forming of the heat release. Both are aspects, which
have only small influence on azt part load; but they get relevant for high
load operation and under real world driving conditions.

13
A consequently downsized car reaches a fuel consumption advantage of
approx. 25 % in the NEDC. This applies for cars from the C-class accor-
ding to ECE [4], i.e. a passenger car from the middle-class, e.g. Audi A4,
BMW 3series, Mercedes C-class etc.; all cars in which a naturally aspira-
ted six-cylinder SI engine (λ = 1) is substituted by a highly turbocharged
3-cylinder SI engine in several steps (see fig. 4).

Figure 4: Fuel consumption potentials in the NEDC by downsizing under

constant engine output as boundary condition

Practical advantages and potentials of fuel efficiency are realistically esti-

mated by real and projected values and realistic estimations for the inner
and the mechanically efficiency.

These results show that a declined displacement and less cylinders are
advantageous thanks to less friction pairings. An inferior influence has the
diminishing of thermic masses because this is only relevant during the
engine warming-up.

Equal engine performance can be achieved by implementing up-to-da-

te turbocharging technologies only. The spontaneous response behavior,
which is typically for a naturally aspirated ICE, could not be realised by
a simple turbocharger for a long time. Consequently, concepts featuring
more than one turbocharger and combination of turbo- and supercharger
were discussed. A few of those concepts were provided to the market [5];
however, they could not remain.

14
Nowadays, powertrains, which feature turbocharged ICE with reduced
displacements, use usually single-turbocharger systems. These concepts
have been widely established due to its advantages in packaging, costs
and complexity. In addition, they have convincing driving performance and
can compete against naturally aspirated ICE thanks to high torque. The-
se downsized ICE‘s compensate the missing spontaneity during dynamic
load changes by the indeed delayed but sharper rising of torque. This is
also rewarded by customers, as can bee seen with a look at current sales
figures. Furthermore, the fuel efficiency in the NEDC can be enhanced by
lower engine speed (so-called down-speeding), which is achieved by an
prolonged gear- and axle-ratio.

3. Limits and Future Challenges

The above mentioned potentials have a powerful leverage factor with re-
gard to the fuel consumption in the current statutory driving cycles for the
car registration. The driving performance could be enhanced on a broad
front at the same time by introducing turbocharged SI engines for all pas-
senger cars on the market.

The improved driving performance is usually delivered outside of the test

cycle load points. The integration of those load points is the aim for fu-
ture test cycles doubtlessly and the derived official fuel consumption labe-
ling. This has the consequence that the advantages in the “off-cycle” fuel
consumption may differ considerably.

Figure 5 shoes the area of conflict, which results from the specific rise of
engine output for the downsizing concept.

Figure 5: Area of conflict for downsizing concepts

15
3.1. Irregular Combustion and Robustness
The risk of irregular combustion rises with the increase of the specific
engine output, which is a matter of principle of the downsizing concept.

The temperature in the combustion chamber rises partly drastically due to

the higher specific energy conversion and the hotter fresh mixture owing
to turbocharging. Only charge air cooling and another injection strategy
may help to reduce this problem.

The whole scope of irregular combustions (see fig. 6) is the result, as

for instance knocking and sudden occurring pre-ignitions (so-called super
knocking). The reasons for these phenomena are currently investigated in
several research projects [6].

Figure 6: usual combustion process in contrast to knocking and

pre-ignition

Those combustion cycles, which feature the above pictured pre-ignitions,

appear over the entire engine speed range under full load and may cause
severe engine damage. Many of these abnormal combustion phenomena
are very difficult to predict. If the resulting pressure gradients are high
enough and the thermal boundary conditions are adverse, the combustion
process results in a self-energizing effect to an irreversible damage of

16
the engine. Unlike the ordinary knock control, a continuously pre-ignition
cannot be stopped by shifting the ignition angle. It requires a complete and
foremost fast deactivation of fuel supply. This interruption is noticed during
driving unpleasantly, and can also be a mayor safety issue e. g. during
accelerations at overtaking, but it is necessarily needed.

The ordinary knocking of SI engines, which was already described by

Harry Ricardo during the 1920s as self-ignition of unburned natural gas
[7], increasingly occurs at highly turbocharged SI engines.

A satisfying control is possible with knocking control systems and a skilled

combustion chamber design combined with optimized heat dissipation out
of the combustion chamber. However, the above mentioned pre-ignitions
and engine knocking represents restrictions for realizable compression
ratios [11]. A too high chosen compression ratio is not feasible from the
point of view of operation reliability.

The thermal reliability with regard to an increased specific engine output

can be ensured by modifications of the basic design, e.g. crank drive,
cylinder head and the so-called hot-end, i.e. exhaust-gas line from the
cylinder heat outlet via the turbocharger to the close to the engine fitted
catalyst outlet.

The aerodynamic design of the coolant ducts, a water-cooled exhaust-gas

line up to the turbocharger and an appropriate choice of material enable a
high reliability to a large extend under real driving conditions. Unfortuna-
tely, the presented modifications are not cost-neutral, i.e. the manufactu-
ring costs are too high.

Additionally, the ignition system of the turbocharged SI engine is strained

more than of the naturally aspirated SI engine owing to the higher gas
density. It is evident that this means higher thermal stress, which arises
out of the larger convertible energy mass. Design modifications offer a
solution for this issue; a larger ignition plug is appropriate. The ignition
voltage considerably increases due to the increased counter pressure in
the combustion chamber, according to the general results published by
Paschen [8]. This requires more powerful ignition plugs, diminishes the
endurance of the ignition electrode and can lead to irregular flashovers
especially at smaller ignition plugs, which are common for engine con-
cepts with central ignition plug.

17
3.2. Drivability and Fuel Consumption under Real
­Driving Conditions
A high engine output requires a large mass flow, which has to be carried
into the combustion chamber and out.

This entails charging systems that can deliver these high mass flows. It
is common to use a single turbocharger system. The dimensioning un-
der stationary condition, which is the operation mode of impounded
turbo-charging, are usually possible without any major difficulty. However,
this charging principle an gas exchange layout cannot satisfy the torque
demands in every load point under real world driving conditions. Dynamic
effects have to be exploited, which have to be induced by every single
cylinder in the firing order. In addition, a disadvantageous interaction bet-
ween the single cylinders has to be prevented. The so-called TwinScroll
turbocharger concept with flow baffle until the turbine offers a suitable
option especially for the most common four and six-cylinder engines. The
dynamic response is additionally and in particular determined by mass
inertia θATL and therewith by the turbocharger size (see fig. 7).

Figure 7: TwinScroll turbocharger and running shaft with mass inertias [9]

18
Another issue concerns the building up of torque at low engine speed. A
turbocharger dimensioning has to deliver sufficient charge-air pressure for
mean effective pressures of 25 bar and more with low exhaust mass flow.
A turbocharger with a small turbine, which operates efficient already at
low engine speed, is eligible from this point of view. However, this dimen-
sioning is for high specific engine output contra-productive due to several
reasons and therefore it leads to a trade-off:

The reduction of the turbine diameter improves the characteristics at low-

end torque and the dynamic response, but it reduces the maximal realiz-
able mass flows at the same time. In addition, the exhaust back pressure
rises, which can cause a massive increase of exhaust residual share. This
is often the reason for the irregular combustion phenomena, which are
described above.

As a countermeasure it is possible to operate with a rich, over-stoichiome-

tric mixture. The air-fuel mixture is cooled by the enthalpy of vaporization
of the liquid gasoline in the combustion chamber.. The described cooling
effect is also used to diminish the thermal overstress of the exhaust valves
and the turbocharger during high load.

Although this is effective, but absolutely not fuel efficient.

This so called enrichment opens the gap between stated and real fuel
consumption during acceleration and operation modes, which requires a
high engine output. This mismatch is difficult to communicate to the custo-
mers and the legislation authorities.

3.3. New Legally Stipulated Test Cycles and Real

­Driving Emissions (RDE)
The fuel consumption of cars, which are approved in the European Union,
is measured in accordance with the New European Driving Cycle (NEDC).
SI engines that have implemented the downsizing concept can reach
exceptional low values in this cycle.

But at the moment legislature is modifying the test landscape cycles signi-
ficantly in Europe and in broad parts of the world.

The regulatory authorities in the respective countries will introduce test

cycles, which feature much more dynamic conditions and aggressive ac-

19
celeration parts, e.g. WLTC (see fig. 8). As a consequence, the load spec-
trum in the engine operating map shifts to a more efficient area but also
into new sectors not yet covered for example by the NEDC. There it is
partly needed to deviate from the optimal process, owing to the knocking
risk. These actions include enrichment or a variation of ignition angle.

Figure 8: Legally stipulated test cycles and load profiles in the engine
operating map

The focus of the regulatory authorities in Europe and USA has been pla-
ced onto the emissions at full load and during highly dynamic operation
modes. Car manufacturers in both markets have to prove that they do not
exceed the limited enrichment to protect engine components, for inner
cooling effects and to avoid pre-ignition. The emissions resulting at real
driving conditions of the customer have to be included into the European
approving process by 2017.

Especial issues arise out of the above mentioned changes, which signifi-
cantly constrict a further maximization of the specific engine output.

It is evident that it is getting more difficult to meet the law, performance

and reliability at the same time. It will not be possible to deviate from the
stoichiometric mixture, which is required for the optimal exhaust after-tre-
atment any longer. The opposite was done for enhanced engine dynamic

20
and component protection. For the discussed cases it will be irrelevant,
if the certification process incorporates portable emission measurement
systems (PEMS) or randomly chosen, synthetically generated load cycles.

4. Conclusion
Downsizing – or rather better said the optimization of the specific power of
ICEs for the customer-relevant operating mode – will remain an important
tool to minimize the fuel consumption.

Two current issues lead to lower growth rates of specific power of the SI
engine unlike usual growth rates over the last five years: On the one hand
it is required to control those operating points, which were not relevant for
the emissions so far. On the other hand the customer’s driving characte-
ristics shall accurately be represented in the fuel consumption labeling.

A specific power of 100 kW/l is currently reached and shows probably the
maximum for the mass market. An effective and efficient control of the
combustion process is required for these engines especially at full load
and dynamic acceleration. This is still a major challenge but practical so-
lutions are still needed urgently.

As it looks now, moderate downsizing concepts gain increasingly ac-

ceptance – what might be called “rightsizing”. However, there remain a
vast number of unsolved and ambitious issues concerning knocking and
pre-ignition.

These issues arise out of the necessary optimization of engine control fun-
ctions and to enhance their reliability. A further increase of the compressi-
on ratio or a variable compression ratio construction, which is discussed
since decades, for an SI engine needs extensive development.

The development of the combustion process and of the engine control

functions as well as the application has to ensure the reliability, the com-
ponent protection against thermal and mechanical overstress and the ma-
ximal efficiency of the ICE.

21
References
[1] The International Council On Clean Transportation: Reducing CO2
and fuel consumption from new cars: Assessing the near-term tech-
nology potential in the EU, www.theicct.org, January 2013
[2] van Basshuysen, R.: Ottomotoren mit Direkteinspritzung, Vieweg &
Teubner Verlag, 2008
[3] Witt A.: Analyse der thermodynamischen Verluste eines Ottomotors
unter den Randbedingungen variabler Steuerzeiten, Dissertation, TU
Graz 1999
[4] European Commission: REGULATION (EEC) No 4064/89 Merger
Procedure, Article 6(1 b), Date: 17.03.1999, http://ec.europa.eu/
competition/mergers/cases/decisions/m1406_en.pdf abgerufen am
14.10.2013
[5] Krebs R., et. al.: Neuer Ottomotor mit Direkteinspritzung und Dop-
pelaufladung von Volkswagen Teil 1 & Teil 2, MTZ 11 & 12 /2005,
Jahrgang 66
[6] Forschungsvereinigung für Verbrennungskraftmaschinen (FVV), Bis
zu 30 Prozent mehr Effizienz: Motorenforscher enträtseln Vorentflam-
mung, Pressemitteliung, Frankfurt am Main 8. April 2013, http://www.
fvv-net.de/cms/upload/aktuelles/FVV_Pressemitteilung_Vorentflam-
mung_2013-04-08.pdf abgerufen am 14.10.2013
[7] Ricardo H. R.: The Internal-Combustion Engine, Vol II, Blackie and
Son Limited, Glasgow and Bombay, 1923
[8] F. Paschen: Über die zum Funkenübergang in Luft, Wasserstoff und
Kohlensäure bei verschiedenen Drucken erforderliche Potentialdiffe-
renz, Annalen der Physik, vol. 273, no. 5, pp. 69 – 96, 1889.
[9] Vogt M.: Aufladesysteme für Ottomotoren im Vergleich, Dissertation,
TU Berlin 2010
[10] Kessel J.-A.: Modellbildung von Abgasturboladern mit variabler Tur-
binengeometrie an schnelllaufenden Dieselmotoren, Dissertation, TU
Darmstadt 2003
[11] Spicher U.: Research and Development Trends of Spark Ignition
Engines –Past, Present, Future, SIA Conference, Strasbourg, Nov.
/ Dec. 2011

22
2 Mega-knock in super-charged gasoline­
engines interpreted as a localized
­developing detonation
Norbert Peters, Bruno Kerschgens, Bernhard Jochim,
Günter Paczko

Abstract
Modern highly downsized spark ignition engines may produce a very high
IMEP and thereby reduce the fuel consumption considerably. A drawback
of this technology is the occurrence of violent mega-knock events which
can damage the moving parts of the engine. This abnormal combustion
phenomenon occurs randomly and is all the more dangerous as it cannot
be avoided by simply retarding the spark timing or reducing the fuel-to-air
ratio as it is generally done to avoid the classical knock phenomenon.

In this paper we ascribe the random occurrence of mega-knock to a se-

quence consisting of primary hot spots, flame propagation (which increa-
ses the pressure and thereby the temperature), and secondary hot spots.
The primary hot spots are known as ‚pre-ignition’ events which randomly
occur at the end of the compression cycle, while the secondary hot spots
initiate a localized detonation. The detonation develops when the reaction
front initiated by the secondary hot spot couples with the sound waves
which it emits. Whether this acoustic coupling occurs depends on

1. the initial temperature gradient in the vicinity of the hot spot and
2. the length over which this temperature gradient extends.

The joint probability distribution of these two parameters is determined by

using a refined theory of small scale turbulence which was developed by
the group of the first author. This theory is based on the new concept of
dissipation elements and uses results from Direct Numerical Simulations.
The joint probability distribution becomes universal by normalizing it with
traditional turbulence quantities such as k and as they are used in CFD
simulations based on the k- model.

By combining the joint distribution with the detonation diagram proposed

by Kalghatgi and Bradley one can determine the detonation probability,

23
i.e. its probability to occur in a large number of engine cycles. Numerical
CFD k- -type simulations of the flow and temperature field in a super-char-
ged engine were performed to predict the location where the detonation
is most likely to develop. To account for the increase of temperature after
pre-ignition and flame propagation the temperature is uniformly increased
by 30 K.

The simulations predict a different dependence of the detonation proba-

bility on intake temperature for different PRF fuels. This is in line with the
different dependence of the pre-ignition frequency on the coolant tem-
perature observed experimentally. The effect seems to appear for a low
engine speed with no residual gas content only.

1. Introduction
Dahnz and Spicher [1] have discussed the irregular combustion phen-
omena occurring in super-charged spark ignition engines, namely clas-
sical knock, extreme knock, surface ignition and pre-ignition, and have
evaluated them in terms of their severity. While knock sensors employed
in modern engines can reduce the occurrence of classical knock by retar-
ding the spark timing, extreme knock occurring at high engine speeds and
exhibiting significantly higher amplitudes, can not be avoided by standard
knock control. Surface ignition is the result of repeated knock events which
heat up the engine walls and consequently the combustion chamber. In
principle it can be avoided by appropriate cooling of the walls and the right
choice and position of the spark plug. The last phenomenon, pre-ignition,
which takes place prematurely before the spark and occurs particularly in
highly supercharged downsized engines, is described in [1] as relatively
new and the least understood.

In a comprehensive experimental and theoretical approach Zahdeh et al.

[2] have identified several parameters which may have a direct effect on
the onset of pre-ignition including liner wetting, injection targeting, mixture
stratification, mixture motion and oil formulation. Among these, injection
targeting is found to be most crucial. They suggest that oil mixing with the
fuel would lower the self-ignition temperature and would hence cause pre-
mature combustion. A 3-D model of piston ring dynamics was developed
to understand the stochastic nature of oil intrusion into the combustion
chamber. Nevertheless the authors note that the mechanism of oil and
fuel accumulation to produce self-ignition is not fully understood. They

24
also note that a sufficiently high end of compression temperature has to
be met for pre-ignition to take place.

Zaccardi et al. [3] investigate the effects of in-cylinder charge motions

on pre-ignition probabilities for direct injection (DI) and port fuel injection
(PFI) and for medium and high tumble cases and a mix of swirl and tum-
ble. Pre-ignition probabilities are found to decrease for increasing engine
speeds between 1000 and 2000 rpm for both injection modes and both
tumble configurations. For the high tumble case the DI mode reduces the
ignition probability for the lower IMEP cases. The probabilities were further
reduced to less than 1% when high swirl is added to the tumble (the swirl
level is 3.7 and the tumble level is 1.5).

Kalghatgi and Bradley [4] give a historical account of the damaging ef-
fects of pre-ignition and ‘super-knock’ (which we call mega-knock here).
We argue that pre-ignition is a necessary but not sufficient condition for
mega-knock. Bradley et al. [5] and Gu et al. [6] had developed their the-
ory of wave amplification and Bradley and Kalghatgi [7] had considered
the effect of different fuels on engine knock. The analysis is based on a
prediction by Zeldovich [8] who distinguishes between several modes of
propagation of reaction waves in mixtures with temperature gradients. A
developing detonation is postulated to occur as an acoustic resonance
phenomenon between the reaction wave velocity and the sound speed a.
The analysis requires that the temperature derivative of the ignition delay
time ∂ /∂T couples locally with a negative spatial temperature gradient
∂T/∂x in direction of the front propagation in such a way that a large (posi-
tive) front velocity u results

u = 1 / (d /d T * ∂T/∂ x).
(1)

If this velocity is of the order of magnitude of the sound speed a, there is

the possibility that a detonation wave develops.

The novel aspect of the theory developed in [5]-[7] is the consideration

of the original size of the ignition kernel as a measure for the region over
which the temperature gradient must extend. This size appears in the con-
text of a model of dynamic effects of auto-ignition kernels by Meyer and
Oppenheim [9]. The model had been tested in numerical calculations by
Lutz et al. [10] for a homogeneous mixture and shows how the gas origi-
nally confined in a radius r0 expands as the temperature increases due

25
to homogeneous auto-ignition. Results of a numerical simulation using an
elementary mechanism for iso-octane at 700 K initial temperature and 20
bar initial pressure p0 based on the system of equations given in [10] are
shown in figure 1. It is seen that after a relatively long induction time of
7.98 ms there occurs a first stage ignition with a small temperature and
pressure increase and a small increase of the normalized radius r/r0. The-
reafter, a second stage ignition occurs at 8.35 ms where the temperature
jumps to nearly 3000 K and the normalized radius increases to more than
4 times its initial value.

Figure 1. Auto-ignition of an ignition kernel of initial radius r0 showing its

in-crease due to temperature increase. The pressure increase at second
stage auto-ignition and the heat release rate are also shown.

The normalized pressure p/p0 also shows a jump to 1.7 times its original
value. This initiates sound waves which then may couple with the propa-
gating reaction wave along a negative temperature gradient as predicted
in [8]. Lutz et al. [10] have de-ined a so-called excitation time which
is the time from the instant of 5% heat release rate to the maximum heat
release rate. The normalized heat release Q/Qmax rate is shown in figure
1 on a logarithmic scale. At first stage ignition it is 1 % of the maximum
value, then drops again to cross the threshold value of 5 % only at the time
when both the pressure and the temperature rise very steeply. The time

26
from there to the maximum heat release is extremely short and therefore
cannot be shown in figure 1. Calculations of for an iso-octane-air mix-
ture plotted in figure 2 show that it is of the order of microseconds.

Figure 2. Excitation times e for a stoichiometric mixture of iso-octane

and air as a function of temperature and pressure.

With the local temperature gradient ∂T/∂x assumed to be given, the reac-
tion wave velocity from eq. (1) can be normalized by the sound speed a to
form a non-dimensional parameter

= a/u = a * ∂T/∂x * d /dT (2)

In addition, the length of the region over which the temperature gradient is
l
constant must be considered. If this length is equal or larger than the di-
stance over which the sound wave emitted from the ignition kernel travels
during the excitation time, a resonance can be established. The product
of the excitation time and the sound speed defines a length a* e which
corresponds to the distance the sound wave travels during the excitation
l
time. The ratio of and a* e defines another non-dimensional parameter
as

= l / (a * e) (3)

27
 Figure 3. Pressure profiles from a one-dimensional simulation as in [6]
for an n-heptane-air mixture. The ratio of the sound speed a to the front
propagation velocity u is = 3. For a non-dimensional length = 2 a de-
tonation develops while for the shorter length = 1 it dies out.

The two non-dimensional parameters have led the authors of [5]- [7] to
define a regime diagram for developing detonations shown in figure 3, in
which the regime of a resonant developing detonation is located within li-
l
miting values of , namely and u, which are a function of the value of .

To validate the detonation diagram given in [6] which had been developed
using a H2-CO – air mixture, several different propagation scenarios were
analyzed here for n-heptane-air as well as for iso-octane air mixtures. Fi-
gure 3 shows two typical results for = 3 and two different values of = 2
and = 1. For = 2 a detonation develops while for = 1 it dies out after
2 mm. This is in agreement with the transition boundary from a developing
detonation to a non-developing detonation at = 1.6 as specified in [5].
The transition boundaries for other values of were also confirmed.

The validated detonation diagram is shown in figure 4.

28
Figure 4: Detonation diagram of [6] with data points from simulations with
n-heptane added. The yellow points show cases where no detonation
develops while the red points show cases with a developing detonation.

The reason for this apparent fuel independence is due to the fact that only
high temperature chemistry which is essentially the same for all fuels,
determines the dynamics of a developing detonation.

DETONATION PROBABILTY
Based on the framework given by Kalghatgi and Bradley [4] we have
developed a prediction method to determine the detonation probabilty
in super-charged gasoline engines based on CFD k- -type simulations
[11]. The method uses results from 3D-DNS simulations of homogeneous
l
isotropic turbulence to obtain the joint pdf P( ,g) statistics of the two para-
meters which are needed in the detonation diagram of [4], namely the tem-
l
perature gradient g=∂T/∂x and the length over which it extends. Further
details on the methodology are given in [11]. With the length definded by

l= *a* e (4)

l
one can define a minimum temperature gradient g and a maximum gra-
dient gu

gu = u /(a* d /dT) (5)

g l = l /(a* d /dT) (6)

29
such that the detonation probability can be calculated by integrating over
l
the joint pdf P( ,g) within these limits

(7)

ENGINE SIMULATIONS
In order to test the predictive capability of the model with respect to pre-ig-
nition leading to ‘mega-knock’ we have performed motored engine simu-
lations in a super-charged gasoline engine.

Figure 5. Sketch of computational grid for intake simulation, including in-

take port and valves. Side walls are removed for illustrational purposes.
This mesh comprises ~ 97 000 cells.

The set-up of the engine was provided by the R&D department of Gene-
ral Motors Company. The specifications of the investigated engine are
summarized in Table 1. The topology of the engine [12] as well as the
computational grid are shown in figure 5. The engine is equipped with a
Siamese port, a four-valve pentroof head, and a contoured piston with val-
ve pockets. Only cycles with port fuel injection and a stoichiometric fuel/

30
air ratio are considered. Three engine operation cases shown in Table 2
are examined. All the cases are calculated with both, the PRF85 and the
PRF100 surrogate. Only part of the calculations will be shown below.

Table 1. Engine Specifications

Engine type Single-cylinder research engine
Stroke 94.6 mm
Bore 86 mm
Connecting Rod 152.2 mm
Compression ratio 12 : 1
Number of Valves 4
Inlet Valve Open 425° ATDC
Inlet Valve Close 564° ATDC

Table 2. Engine operating reference conditions

Intake pressure [bar] 2
Intake temperature [K] 300, 340, 380
Residual gas content [%] 0, 10, 20
Engine speeds [rpm] 1000, 2000, 3000
Valve lift [mm] 2, 3, 4
Piston temperature [K] 434
Cylinder bore temperature [K] 426
Cylinder head temperature [K] 426
Spark plug temperature [K] 500

The cases shown in table 2 model a super-charged case with the inlet
pressure of p0 = 2.0 bar and variations of the intake temperature, residual
gas content, engine speed and valve lift. In order to account for the increa-
se of temperature by pre-ignition and flame propagation the temperature
in the cylinder is uniformly increased by 30 K. To specify the time of igniti-
on tign of the secondary hot spots we have used, as suggested in [4], the
Livengood-Wu integral as

31
 (8)

Here (T(t), p(t)) is the time evolving ignition delay time calculated during
compression and expansion as the local temperature and the pressure
are changing. By setting the integral to unity one obtains an order of ma-
gnitude estimate for the time of ignition. The cases shown in table 2 are
calculated with two surrogates, pure iso-octane (PRF100) and an 85%
iso-octane, 15% n-heptane mixture (PRF85) using detailled kinetic me-
chanisms from the literature [13].

Figure 6 shows two plots of the mean temperature over crank angle for
the three different intake temperatures and the two surrogates, PRF100
and PRF85, respectively. The time of ignition is indicated by the vertical
line for each of the cases, its height denotes the maximum detonation pro-
bability. It is seen that in both cases the time of ignition is the earliest for
the 380 K case and the latest for the 300 K case, as expected. For PRF85
the Livengood-Wu integral becomes unity before TDC for the 380 K case
(which would indicate the likelihood of severe mega-knock [14]), while for
PRF100 this occurs just after TDC.

Figure 6. Temperature over degree Crank Angle for PRF100 (left) and
PRF85 (right). The time of ignition is indicated by the vertical line for
each of the cases, its height denotes the maximum detonation probabili-
ty.

However, the calculated maximum detonation probability has the opposite

tendency in both cases: For PRF100 it decreases as the intake tempera-

32
ture increases, while for PRF85 it increases with intake temperature. Plots
of the detonation probability for the two end cases T = 300 K and T = 380
K are shown as horizontal slices in the combustion chamber in figure 7
and in figure 8. As one moves from left to right in figure 7 one observes
a decrease of the detonation probability for PRF100, while there is an
increase for PRF85 in figure 8.

Figure 7. Plots of the detonation probability for the cases T = 300 K (left)
and T = 380 K (right) for PRF100.

Figure 8. Plots of the detonation probability for the case T = 300 K (left)
and the case T = 380 K (right) for PRF85.

The maximum detonation probability is shown in the upper row of figure

9 for the three cases and PRF100 and PRF85, respectively. Here the ho-
rizontal axis has been converted into °C to facilitate the comparison with
measurements by Zaccardi [15] which are shown in the lower row. Zac-

33
cardi had measured the pre-ignition frequency as a function of the cooling
water temperature. He found that in the range from 80°C to 100 °C for
higher grade fuels (E5 and RON91) the pre-ignition frequency decreases
while for lower grade fuels (RON87) it increase. For higher grade fuels
a similar effect of the cooling temperature on the pre-ignition rate was
reported in [15]. These observations are in line with the dependence on
intake temperature found above. However, while in [15] and [16] fuel/oil
interactions or fuel droplets are thought to be responsible for the tendency
our prediction method, in using motored engine data for port fuel injection,
relies entirely on the Zeldovich theory of acoustic resonance leading to a
localized developing detonation.

Figure 9. Dependence of the maximum detonation probability for the two

surrogates on intake temperature (upper row ) compared with the ex-
perimentally found dependence of the pre-ignition frequency on cooling
water temperature (lower row), Zaccardi [15].

34
There remains the question about the origin of this opposite tendency
predicted for the two surrogates. It was shown in [11] that the two domi-
nating parameters in the calculation of the detonation probability are the
temperature and the ratio k/ ob-tained from the turbulence model. Figure
10 shows for p = 50 bar phase diagrams of the detonation probability as
a function of these two parameters for both surrogates. The respective
values for the two end cases T = 300 K and T = 380 are also shown in
these diagrams. The surprising result is that in the upper plot in figure 10
for PRF100 the detonation probability increases dramatically at a cham-
ber temperature of T = 800 K where the 300 K intake temperature case is
located. The 380 K case falls into a blue region in this diagram where the
detonation probability is lower. This explains the tendency seen in figure
7. In the lower plot in figure 10 for PRF85 there is a monotonous increase
of the detonation probability with increasing temperature in the range of
chamber temperatures below 1100 K. This corresponds to the thermal
effect seen in figure 8.

Figure 10. Phase diagrams showing the detonation probability as a fun-

ction of the chamber temperature and k/ for both surrogates at p = 50
bar.

The occurrence of a region with a dramatically increased detonation pro-

bability in the upper diagram in figure 10 needs a further explanation. This
may be found in the scaling used in eq. (2) which says that for a fixed
range of - values in figure 4 there is a larger range of ∂T/∂x values in the
integral of eq. (7) when the derivative ∂ /∂T is small. Then the probability
computed from the integral becomes larger than in regions where ∂ /∂T
is larger. Therefore we have investigated whether the derivative ∂ /∂T
takes different values for the two surrogates in the temperature region
around 800 K. In figure 12 the ignition delay time (lower plots ) and its

35
derivative ( upper plots ) are shown for the two surrogates. It is seen that
the derivative ∂ /∂T which is always negative has a minimum at T = 800
K for PRF100 while no such minimum appears for PRF85. Small values of
the derivative ∂ /∂T in eq. (1) can therefore cause unexpected effects in a
prediction method based on the Zeldovich theory of acoustic resonance.

Figure 12. The ignition delay time (lower plots ) and its derivative (upper
plots) for the two surrogates at p = 50 bar.

The tendency observed above appears to be a particular phenomenon in

a small range of parameters. In our calculations it also occurs when the
valve lift is reduced from 4 mm to 1 mm and 2 mm as shown in the res-
ponse surfaces in figure 13.

Figure 13. Response surfaces showing the detonation probability at the

three intake temperatures for different valve lifts at 1000 rpm and 0 %
residual gas.

36
When 10 % residual gas remains in the cylinder, however, the effect lea-
ding to a smaller detonation probability with decreasing temperature for
PRF100 disappears, while the thermal tendency becomes dominant. This
is shown for the previous variations of valve lifts in figure 14.

Figure 14. Response surfaces showing the detonation probability at the

three intake temperatures for different valve lifts at 1000 rpm and 10 %
residual gas.

When the engine speed is increased to 2000 rpm, the predicted detona-
tion probability becomes zero for PRF100 in the entire range of parame-
ters, while for PRF85 it slightly increases with the intake temperature. This
is shown in figure 15.

Figure 15. Response surfaces showing the detonation probability at the

three intake temperatures for different valve lifts at 2000 rpm and 0 %
residual gas.

37
DISCUSSION AND SUMMARY
In this paper we have explored the potential of a methodology based on
the theory of acoustic resonance to predict mega-knock in super-charged
gasoline engines. The predictions reproduce for 1000 rpm the tendency of
decreasing mega-knock probabilities with increasing intake temperatures
in the range from 300 k to 380 K for higher grade gasoline fuels which
have been observed for cooling water temperatures experimentally. While
we do not claim to provide the only valid explanation for this effect we
believe that it occurs in a limited parameter range. For lower grade fuels
or higher combustion chamber temperatures due to residual gases the
thermal effect predominates. Mega-knock is not predicted at the higher
engine speed of 2000 rpm with the higher grade fuel for the engine geo-
metry investigated.

Acknowledgement: This work was founded by the Honda Initiation Grant

Europe 2013. The engine simulations were performed by Hüseyin Armut-
cu during his master thesis at the Institute for Combustion Technology of
the RWTH Aachen University.

38
LITERATUR
[1] Dahnz, C., Spicher, U., “Irregular combustion in supercharged spark
ignition engines-pre-ignition and other phenomena” ,Int .J .Engine
Res. 11:485-498,2010.
[2] Zadeh, A., Rothenberger, P., Nguyen, W., Anbarasu,,M.,
Schmuck-Soldan, S., Schäfer, J., Goebel, Th.,“Fundamental appro-
ach to investigate pre-ignition in boosted SI engines”, SAE paper
32011-01-0340, 2011.
[3] Zaccardi, J.-M., Laget,O., Pagot,A., Mokhtari,S., Leone, T., “Investi-
gations on the effects of incylinder charge motion and injection mode
on pre-ignition in highly boosted spark ignition engines”. 19. Aache-
ner Kolloquium Fahrzeug- und Motorentechnik ,1323-1349, 2010.
[4] Kalghatgi, G.T.,Bradley,D., “Pre-ignition and ‘Super-knock’ in
turbo-charged spark-ignition engines”, Int. J. Engine Res. 13: 399-
414, 2012.
[5] Bradley, D., Morley, C.,Gu, X.J., Emerson,D.R., ”Amplified Pres-
sure Waves During Autoignition: Relevance to CAI Engines” SA
Technical Paper 2002-01-2868, 2002.
[6] Gu X. J., Emerson D. R. and Bradley D. “Modes of reaction front
propagation from hot spots.” Combust. Flame 133: 63–74 (2003).
[7] Bradley, D., Kalghatgi, G.T., “Influence of autoignitiondelay time
characteristics of different fuels on pressure waves and knock in
reciprocating engines,” Combust. Flame 156: 2307–2318 (2009).
[8] Zeldovich,Ya.B. “Regime classification of an exothermic reacti-
on with nonuniform initial conditions“. Comb. Flame 39: 211-214
(1980).
[9] Lutz, A.E., Kee, R.J., Miller, J.A., Dwyer, H.A., Oppenheim,A.K., “Dy-
namic Effects of Autoignition Centers for Hydrogen and C1,2-Hy-
drocarbon Fuels,” Proc. Combust. Inst. 22: 1683-1693 (1988).
[10] Meyer,J.W., Oppenheim, A.K., “Dynamic Response of a Plane-Sym-
metrical Exothermic Reaction Center,“ AIAA J. 10 : 1509-1513(1972).
[11] Peters, N., Kerschgens, B., Paczko, G. ‘Super-knock predicti-
on using a refined theory of turbulence’ SAE paper 2013-01-1109
(2013).
[12] Dahms, R., Fansler, T.D., Drake, M.C., Kuo, T.-W., Lippert, A.M., Pe-
ters, N. “Modelling ignition phenomena in spray-guided spark-igni-
ted engines,” Proc. Combust. Inst. 32: 2743-2750 (2009).

39
[13] The PRF mechanism is available on https://www-pls.llnl.gov/?ur-
l=science and_technology-chemistry-combustion-prf an iso-octane
mechanism on https://www- pls.llnl.gov/?url=science and_technolo-
gy-chemistry-combustion-isooctane version 3
[14] Haenel,P., Seyfried, Ph., Kleeberg, H.,Tomazic,D., ‘Sastematic
approach to analyze and characterize pre-ignition events in
turbo-charged direct-injected gasoline engines’ SAE paper 2011-01-
0343 (2011).
[15] Zaccardi, J.-M. private communication (2013).
[16] Dahnz, C., Magar, M., Spicher, U., Maas, U., Kubach, H., Schießl, R.
‘Vorentflammung bei Ottomotoren’ 6th FVV Work Committee Mee-
ting, May 28 2009, Karlsruhe (2009).

40
3 A contribution to better understanding
the pre-ignition phenomenon in highly
charged internal combustion engines
with direct fuel injection
Arndt Döhler, Stefan Pritze

Abstract
Downsizing as the development path chosen in recent years allows for a
significant efficiency improvement of current gasoline engine concepts.
But with power densities beyond 20 bar BMEP the likelihood of combus-
tion irregularities is increasing. Especially stochastic pre-ignitions (SPI)
may occur during engine operation at full load and low speeds. They re-
present a real challenge because of the high damage potential of extraor-
dinary high in-cylinder pressure resulting from combustion onset prior to
calibrated spark timing.

At General Motors basic research projects focused on SPI occurrence

have been conducted for several years. With experience in the use of
optical measurement techniques grown over years we have chosen two
synchronized video endoscopy systems for the work presented in this pa-
per. This set-up enables an analysis of the combustion process with large
scale and high temporal resolution.

Numerous studies have already dealt with the possible root causes of SPI
occurrence. A widely accepted hypothesis traces the onset of pre-igniti-
on events back to fuel wall-wetting effects. The accompanying oil dilution
encourages the release of oil/fuel droplets from the top land area of the
piston. Due to highly inflammable components these droplets can trigger
pre-ignition under thermodynamic critical conditions.

In this study evidence is provided that an oil/fuel droplet is not causing

pre-ignition onset due to auto-ignition. Instead, the observations indica-
te that those droplets are entrained by the regular flame front and start
diffusive burning. Due to lack of oxygen a complete combustion of these
droplets is suppressed. As a consequence glowing particles remain in the
combustion chamber that can cause a pre-ignition in the subsequent com-
bustion cycle.

41
Based on these findings the well accepted hypothesis on SPI resulting
from droplet ignition could be refined. It now allows to give a reasonable
explanation of the not yet understood intermittent sequence of pre-igni-
tions events.

Kurzfassung
Downsizing ist die in den letzten Jahren verfolgte Entwicklungsrichtung
zur signifikanten Effizienzsteigerung aktueller Ottomotorenkonzepte. Bei
den damit verbundenen hohen Leistungsdichten mit Mitteldrücken über
20 bar steigt zunehmend die Wahrscheinlichkeit für das Auftreten von irre-
gulären Verbrennungszyklen. Besonders stochastische Vorentflammun-
gen (SPI –Stochastic Pre-Ignition) können bei einem Motorbetrieb unter
hoher Last bei niedrigen Drehzahlen auftreten. Sie stellen eine große He-
rausforderung aufgrund des enormen Schädigungspotenzials dar. Dies
ergibt sich aus den extrem hohen Zylinderdrücken, die bei einem Verbren-
nungsbeginn vor dem kalibrierten Zündzeitpunkt entstehen.

Bei General Motors sind in den letzten Jahren grundlegende Forschungs-

projekte mit Fokus auf die Entstehung von Vorentflammungen durchge-
führt worden. Aus der über die Jahre gewachsenen Erfahrung im Bereich
optischer Messtechniken wurden für die hier vorgestellten Untersuchun-
gen zwei synchronisierte Videoendoskopie-Systeme ausgewählt. Mit die-
sem Aufbau wird eine Analyse des Verbrennungsprozesses mit hoher ört-
licher und zeitlicher Auflösung möglich.

Zahlreiche Studien haben sich bereits mit den möglichen Ursachen für
das Auftreten von SPI beschäftigt. Eine allgemein akzeptierte Hypothese
führt die Entstehung von Vorentflammungsereignissen auf Mechanismen
im Zusammenhang mit der Wandanlagerung von Kraftstoff zurück. Die
damit verbundene Ölverdünnung befördert die Freisetzung von Öl/Kraft-
stoff-Tropfen aus dem Feuersteg des Kolbens. Durch die darin enthalte-
nen leicht entzündlichen Komponenten können unter thermodynamisch
kritischen Randbedingungen Vorentflammungen ausgelöst werden.

In der vorliegenden Studie werden Ergebnisse vorgestellt, die darauf hin-

deuten, dass ein Öl/Kraftstoff-Tropfen nicht aufgrund seiner Selbstzün-
dung eine Vorentflammung auslöst. Vielmehr weisen die vorgestellten
Beobachtungen darauf hin, dass ein solcher Tropfen von der regulären
Flammenfront entzündet wird und in der Folge beginnt, mit diffusiver

42
Flamme zu brennen. Aufgrund des Mangels an Sauerstoff wird die voll-
ständige Verbrennung dieser Tropfen abgebrochen. In der Konsequenz
verbleiben glühende Partikel in Brennraum, die eine Vorentflammung im
nachfolgenden Verbrennungszyklus auslösen können.

Mit diesen Ergebnissen wird es möglich, die allgemein akzeptierte SPI-Hy-

pothese zu verfeinern. Sie erlaubt nun auch eine einsichtige Erklärung für
das noch nicht verstandene Auftreten von intermittierenden Vorentflam-
mungs-Serien zu liefern.

1. Preface
Downsizing represents a technology to significantly increase efficiency
of SI engines. However, under low speed / high load conditions stocha-
stic occurrence of pre-ignitions was observed by development engineers.
With the thermodynamic conditions close to firing TDC being far below the
limits for auto-ignition there is no obvious explanation for this phenome-
non available.
Pre-ignition can lead to extremely high in-cylinder pressures causing se-
vere engine failures limiting the development of highly boosted engines.
Therefore, in the recent years a lot of investigations have been carried out
to understand the nature of this combustion phenomenon and to develop
appropriate mitigation measures.

2. State of Knowledge
There is a large variety of possible mechanisms which are comprehen-
sively summarized in [1]. In this work a hypothesis was developed which
considers the pre-ignition initiated by oil/fuel droplets being ejected from
the piston ring crevice into the combustion chamber and evaporating on
their trajectory. Since they contain highly ignitable components originating
from lubricant oil, the vapor will be compression ignited when in-cylinder
conditions reach critical levels of pressure and temperature. The released
energy is high enough to ignite the bulk air/fuel mixture.

43
 Figure 1: Release of Oil Dropletsfrom Top Land [1]

It is also possible that particles originating from combustion deposits can

induce ignition. That might be especially likely if the particles are soaked
with oil and fuel due to spray impingement. Under those conditions they
will behave similar to droplets with highly inflammable hydrocarbon com-
ponents being evaporated and finally compression ignited.
Furthermore it may be possible that particles big enough are heated up
over consecutive cycles to temperatures sufficient to induce pre-ignition
or that deposits may partially loose contact with the surface leading to a
local hot spot condition [2].

Hot spots in the combustion chamber resulting from hot surfaces like
spark plug or exhaust valves or due to trapped residuals may not be the
root cause for pre-ignition but strongly promote their occurrence by signifi-
cantly reducing ignition delay times. On the other hand increasing ignition
delay times e.g. by charge dilution with cooled external EGR can effecti-
vely suppress pre-ignition [3].

The stochastic, i.e. unpredictable nature of pre-ignition is an interaction of

possible ignition sources with locally changing conditions inside the com-
bustion chamber. The severity of the pre-ignition is determined by how far
before regular ignition timing a critical interaction will occur.

It is a characteristic feature of pre-ignitions that they occur not only as

single events but also as series of several pre-ignitions. In most of the ca-
ses this happens in an intermittent sequence with the pre-ignitions being
separated by at least one regular combustion event. However, consecu-
tive pre-ignitions have been observed as well. During the presented in-

44
vestigations series of up to 11 pre-ignitions partly interrupted by normal
combustion events could be recorded. The mechanism behind is still one
of the open questions of pre-ignition occurrence [1, 4].

At General Motors investigations on the pre-ignition phenomenon have

been started early and were comprehensively presented in [5]. Based on
experimental investigations including optical measurements as well as
CFD analysis and other simulation techniques the hypothesis on oil/fuel
droplets could be refined and basically confirmed.
Oil/fuel interaction at the liner was found to be the dominating driver for
pre-ignition. Spray targeting, volatility of fuel and additives as well as wall
temperatures were identified as the main drivers for the generation of oil/
fuel droplets in the piston top land crevice.
Beside the fuel impacts oil intrusion and formulation is also of importan-
ce since the highly inflammable components prone to compression igni-
tion originate from the oil. In [6] the significant effect of oil formulation on
pre-ignition is reported. Regarding oil intrusion oil control ring tension as
well as piston ring dynamics have to be considered.
Fuel impingement on the cylinder liner in combination with low volatility
fuels and low wall temperatures builds up high amounts of liquid hydrocar-
bons in the piston top land crevice. Droplets can be expelled from there
since viscosity compared to oil is reduced by fuel dilution. This mechanism
is confirmed by the fact that reducing the accumulation of hydrocarbons
by modifying piston crown geometry and allowing the flame to promote
into the crevices reduces pre-ignition frequency [7].
Factors affecting the thermodynamic state of charge were classified as
indirect drivers for pre-ignition occurrence. These include compression ra-
tio, boost pressure, air/fuel ratio and combustion timing. The presence of
hot surfaces like the spark plug ground electrode has an effect by locally
raising the thermodynamic state. Monitoring the thermodynamic state du-
ring engine operation can be an effective measure to prevent pre-ignition
occurrence [4].

The well accepted hypothesis on droplet release from piston top land cre-
vice considers the creation of the ignition source and the related pre-ig-
nition occurring in the same cycle. However, based on the video imaging
presented in this paper it can be concluded that this assumption has to
be revised. A refined hypothesis will be discussed which comprehends
the known dependencies and provides a reasonable explanation for the
intermittent nature of pre-ignitions.

45
3. Test Carrier
The main features of the 1.6 liter SIDI (spark-ignited direct injection) engi-
ne used for the investigations are shown in figure 2.
The engine applies a homogeneous combustion process and exhaust gas
turbocharging. The cylinder block is made of cast iron and for the cylinder
head AlSi7Mg is used. Variable cam timing on inlet and exhaust cams-
hafts with an actuator authority of 60° crank angle is realized. The cams-
hafts are driven by a toothed chain.

Figure 2: 1.6l SIDI, Technical Data

The combustion system consists of a centrally mounted 6-hole solenoid

injector, intake ports generating a high tumble charge motion, a piston with
a small bowl for improved cold start performance and a M12 spark plug.
The realized compression ratio is 10.5:1. The orientation of injector and
spark plug is in-line with the crankshaft axis.
Emphasis was put on a combustion process with robust performance
against irregularities. Therefore, the use of high-speed cameras has be-
come a standard within the development process.

4. Measurement Setup
General Motors has now gone for several years through basic research
on stochastic combustion phenomena both in Germany and the USA. Th-
erefore, the International Technical Development Center in Rüsselsheim,
Germany was able to build on years of experience for the present work.
Continuous investigations on pre-ignition occurrence have led to the con-
clusion that a single high speed camera is not sufficient for a profound

46
analysis of stochastic combustion irregularities. Therefore, the use of two
synchronized endoscopy systems was a logical step towards the design
of an in-house camera system. Two synchronized endoscopy systems
allow a large view into the combustion chamber. In particular, the analysis
of pre-ignition phenomena profits from this approach enabling a highly
time-resolved analysis of the combustion process.
Since the optical path consisting of boroscope, light intensifier and ca-
mera has a length of approximately one meter a rigid support of these
components is crucial. Based on the system described in [5] a third optical
access was installed on the intake side of the combustion chamber. The
camera support went through a thorough analysis to ensure stiffness and
lowest deformation, see figure 3. The support has been designed to reali-
ze a first order natural frequency of less than 60 Hz since the target speed
for the investigations was around 2000 rpm.

Figure 3: Camera Support, Stress Analysis

Table 1 describes the main components of the optical system. Based on

the experience within the preceding projects mentioned in [5] the design
of the combustion chamber window was improved. The window is now
capable to withstand multiple pre-ignition events with combustion peak
pressures above 300 bar.

47
 Combustion cham- Material Sapphire
ber window Shape Planar

Outer Diameter 7 mm
Observation Cone (nominal /
Boroscope 67° / ≈58°
real)
(Storz) Viewing Direction Straight

Cooling Air Cooled

Outer Diameter of Light Dis-
7 mm
persing Stick
Lighting – Xenon Light Field Cone 60°
Type
(Storz) Lighting Direction Straight

Cold light Source 300 W

< 200 –800
Spectral Response
nm
Image Intensifier 25 mm / 18
Output Diameter
(Lambert) mm
Resolution on Output (with 1:1
25 l/mm
Relay Lens)
Resolution 640 x 640

Color None
High Speed Camera Acquisition Rate 12000 fps
(Vision Research) 1 °CA @
Crank Angle Resolution
2000 rpm
80µs @
Exposure Time
2000rpm
Table 1: Optical System, Technical Data

Furthermore, it was possible to shield the light dispersing stick against

combustion gases to avoid glow ignitions or pre-ignitions triggered by
overheating. All of the optical components are still interchangeable, i.e.
every combustion chamber access can be used either for lighting or re-
cording.

48
Due to design limitations cylinder 4 was chosen for installation of the opti-
cal accesses. Besides two accesses from the flywheel side of the engine
a third access was realized on the intake side of the combustion chamber
(see figure 4, left). To minimize the area out of sight one camera has taken
pictures from the left exhaust valve to the right intake valve. The second
camera access has been located between the intake valves to enable
a direct view to the exhaust side of the combustion chamber. Figure 4
(center and right) shows the realized field of view through the boroscopes
inside combustion chamber of cylinder 4.

Each camera system contains a ring buffer which has been permanently
updated with images. Thus combustion events prior and after a pre-igni-
tion could be recorded. The pressure threshold triggering the recording is
calculated by a simple formula adding five times the standard deviation of
the cylinder pressure to its mean value.
Additionally each sequence of images was aligned with the indication data
during post-processing. The location of 2% MFB (mass fraction burned)
was compared against the average value and used as a secondary check
on start of combustion prior to the calibrated ignition point.

Figure 4: Cylinder Head, Viewing Directions on Cylinder No. 4

(Left: Position of Optical Accesses – L=Light, 1=View 1, 2=View 2;
Center: View 1 from Left Exhaust Valve to Intake Valves; Right: View 2
from Intake to Exhaust Side)

5. Test Program
Testing was conducted at full load conditions with 280 Nm respectively 22
bar BMEP (brake mean effective pressure) at engine speeds from 1650
rpm up to 2200 rpm. At 2000 rpm chosen parameter settings were global
lambda at 1.02, knock limited spark advance (KLSA) at 1.5°CA aTDC,
start of injection at 270°CA bTDC and fuel pressure at 200 bar. The coo-

49
lant was preconditioned at operating temperature prior to test start. Thus
it was possible to minimize fouling of the optical windows due to low soot
emissions at engine start. The test procedure for pre-ignition checks ac-
cording to GM standards was applied until the optical window became too
dirty after about 20 to 30 minutes, see figure 5.

Figure 5: Test Procedure for PI Check, Time Course

Since oil/fuel interaction at the liner was found to be the dominating driver
for pre-ignition [5] a specific test fuel with a reduced volatility was used
for the majority of tests to promote pre-ignition occurrence. Compared
to a typical market fuel approximately 20% less fuel is evaporated in the
temperature range of 130 to 150°C characteristic for the upper part of the
cylinder liner, figure 6.
Of course, comparison tests with representative RON95 (E5) fuel accor-
ding to EN228 have been carried out to make sure that the observed
mechanism is the same for both fuels.

Figure 6: Distillation Curve of Tested Fuels

50
6. Measurement Results and Discussion

6.1. Advantage of Two Optical Accesses

The two optical accesses have proven their value right at the beginning of
the investigations. Typically, an optical access at the flywheel side of the
engine offers a viewing direction from the left exhaust valve to right inta-
ke valve respectively from the left intake valve to the right exhaust valve
since at least one access will be used for lighting. In either case a huge
section of the combustion chamber will be left out of the visible area, see
also figure 4 (center).
Figure 7 shows the pressure trace of a typical pre-ignition event. It is cle-
arly visible that the combustion starts about 6°CA before the calibrated
point of ignition.

Figure 7: Typical Pre-Ignition Event, Pressure Trace

A sequence of pictures taken from the pre-ignition shown in figure 7 th-

rough both optical accesses is displayed in figure 8. Through the optical
access on the flywheel side (upper row) the pre-ignition can be seen for
the first time at 8°CA aKLSA. In comparison the optical access on the inta-
ke side (bottom row) allows to detect the same pre-ignition at 4°CA bKLSA
which is 12°CA earlier. Even more important it is possible to narrow down
the origin of the pre-ignition to a distinct location in the combustion cham-
ber due to the maximized field of view.

51
 Figure 8: Advantage of 2 Optical Access Drillings for Pre-Ignition
Detection
Upper Row: View 1 from Left Exhaust Valve to Intake Valves,
Bottom Row: View 2 from Intake to Exhaust Side

6.2. Formation of Pre-Ignition Sources

A homogeneous mixture guarantees a stable inflammation, robust com-
bustion and low particulate numbers. Figure 9 shows such a combustion
cycle at full load.
However, even under those conditions the video imaging recorded dif-
fusively burning lumps travelling through the combustion chamber and
assumingly originating from areas close to the cylinder liner. This supports
the hypothesis that oil/fuel droplets are expelled from the piston top land
crevice. However, those droplets are obviously not compression ignited
during the compression stroke but start diffusive burning when reached by
the regular flame front close to end of combustion. Due to a lack of oxygen
under stoichiometric or even enriched conditions they can not burn com-
pletely and glowing particles will remain.

52
 Figure 9: Typical Combustion Cycle at Full Load

Figure 10 shows a diffusively burning lump creating glowing particles. The-

se particles can be observed in the shown sequence around 60°CA after
KLSA which is well after end of combustion (90% MFB) at 38°CA aTDC.
They keep glowing during expansion and exhaust stroke and instead of
being expelled to the exhaust can remain in the combustion chamber.

Figure 10: Generation of Glowing Particles

Strong evidence has been found that those particles can be the source of
a pre-ignition in the succeeding cycle. Depending on the size of the par-
ticle sufficient temperature can be sustained during the induction stroke
with fresh air and fuel entering the combustion chamber. In this phase
lubricant oil formulation may become important since this will decisively
determine the properties of the particle in terms of size and composition.

53
For example it may be an indication that low ash oil reduces pre-ignition
frequency [5]. Other dependencies found regarding oil formulation may
deliver further insight when being evaluated in this focus.
During compression stroke the particle is heated up again and with oxy-
gen being now sufficiently available frozen chemical reactions can cont-
inue. Depending on the trajectory of the particle when approaching TDC
it can ignite the charge prior to the regular spark. It is obvious from video
imaging that hotter areas in the vicinity of exhaust valves or the spark plug
are preferred locations of particle induced ignition.

Figure 11 shows such a pre-ignition triggered by glowing particles. First

inflammation starts approximately 5°CA before the calibrated point of ig-
nition. Shortly thereafter a second particle ignites the charge at a different
location but still prior to sparking.

Figure 11: Glowing Particles as Multiple Pre-Ignition Source

It is known that already knocking can clean the combustion chamber by

removing hydrocarbons accumulated in crevices as well as burning and
dislodging deposits from the surfaces due to high pressure oscillations
and increased temperatures.
With severe pre-ignition this cleaning effect becomes even stronger. Se-
vere pre-ignition events as shown in figure 11 start well before the cali-
brated point of ignition. Since heat release rates during pre-ignition can
be extremely fast (less than half of regular combustion duration) there is
no oxygen left to burn anything of the deposits dislodged from the surfa-
ces and hydrocarbons released from crevices. Liquid hydrocarbons, e.g.
originating from piston top land crevice may be partly soaked by depo-

54
sits/particles close to rims. The removal of hydrocarbons correlates with
the observation that after a pre-ignition event the oxygen sensor in the
exhaust system detects a rich dip in the lambda trace, figure 12.

Figure 12: Intermittent Pre-Ignition Events, Lambda Trace (Run Time

Corrected)

After such an event the combustion chamber contains at exhaust valve

opening a lot of debris being dislodged from the surfaces. A considerab-
le amount remains in the combustion chamber during the following gas
exchange. After regular spark ignition these particles are entrained by the
flame front and start glowing due to the heat of combustion as shown in
figure 13.

Figure 13: Subsequent Combustion Cycle w/o Pre-Ignition, Directly

Following a Cycle w/ Pre-Ignition

And again many of these particles can be trapped in the combustion

chamber at the subsequent intake valve closing. A particle large enough
to sufficiently sustain the temperature and possibly further heated up by

55
exothermal reactions of still contained unburned hydrocarbons in the en-
vironment of fresh oxygen can then again induce a pre-ignition.
Breaking up the creation of the ignition source and the related pre-ignition
event in consecutive cycles provides an explanation for its intermittent
nature.
Figure 14 illustrates such a series of combustion cycles. Starting with the
spark of a regular cycle glowing particles are created from diffusively bur-
ning lumps (1st row). They keep glowing during the following induction
stroke and fuel injection (2nd row). A first pre-ignition is induced shortly
before the calibrated sparking (3rd row, 1st picture, arrow at left side).
Debris but no glowing particles are left over from that 1st pre-ignition (3rd
row, 2nd picture) creating in the subsequent regular combustion a huge
amount of glowing particles (3rd row, 4th picture). Part of these particles
is trapped in the combustion chamber and act as an ignition source in the
following cycle as shown in the 4th row of the picture sequence.

Figure 14: Intermittent Pre-Ignition Series

56
This mechanism of a pre-ignition series can continue over several cycles
as long as particles with critical size and temperature are remaining in the
combustion chamber. A statistical evaluation in [8] comes to the result that
the frequency of pre-ignition series is monotonically decreasing with its
length, i.e. the number of included pre-ignitions and in most cases only
single pre-ignition events occur. That indicates the diminishing likelihood
of particles with critical size remaining in the combustion chamber.
Normally in such intermittent series the pre-ignition events are separated
by at least one regular combustion cycle However, consecutive pre-igni-
tion events could be recorded as well. In these cases the video imaging
identified glowing particles being a trigger for the subsequent pre-igni-
tions, too. Part of these consecutive pre-ignition events are characterized
by an onset close to the calibrated spark advance resulting in relatively
low peak pressures and flame propagation similar to regular combustion.
Under those conditions particles might be exposed for a longer time to
high temperatures and therefore can achieve sufficient temperature to act
in the following cycle as an ignition source.

It was possible to estimate the size of particles which are acting as a

pre-ignition trigger due to a favorable coincidence shown in figure 15. A
glowing particle came to a rest for a few nanoseconds at the spark plug
electrode shortly before it induced a pre-ignition. Thus it became possib-
le to compare the size of the particle against the known diameter of the
center electrode tip. The diameter of this particle can be assessed with
about 0.3 to 0.4 mm. Being well aware of the limited accuracy this result
indicates that the critical size of particles lies in the range of tenths of a
millimeter.

Figure 15: Glowing Particle at Spark Plug Gap, Estimation of Particle

Size

57
Therefore, also considering the results presented in [5] soot particles with
several magnitudes smaller size can be excluded being a potential source
of pre-ignition.

6.3. Refined Hypothesis of Pre-Ignition Onset

Based on the video imaging conducted and the experimental investiga-
tions performed at General Motors and by other developers we propo-
se the following refined hypothesis for the occurrence of pre-ignitions in-
duced by oil/fuel droplets:

1. Initial pre-ignitions are induced by oil/fuel droplets being stochasti-

cally expelled from the piston top land crevice. These particles are
ignited by the flame front of a regular combustion and burn with a dif-
fusive flame. A hot particle remains when oxygen is used up. In case
the particle is not exhausted from the combustion chamber the heat
of this particle is sustained during the induction stroke due to its size
exceeding a critical threshold and limited convective heat transfer to
the fresh charge. With oxygen becoming available during intake stro-
ke frozen exothermal reactions will continue and during compression
the particle is heated up further.

2. In case a large enough particle acquires on its trajectory through the

combustion chamber sufficient heat and/or reaches areas with critical
thermodynamic conditions it will ignite the surrounding air/fuel mix-
ture. Depending on the energy of the particle as well as on the locally
changing thermodynamic state of the charge this can happen signifi-
cantly prior to regular spark timing, close to it or even later. Therefo-
re, the severity of the pre-ignition event will decrease from extremely
high cylinder pressures down to a point where it can be covered by
the regular combustion process.

3. After such an initial pre-ignition a series of pre-ignitions is driven by

particles dislodged from the combustion chamber surfaces and pos-
sibly attached by hydrocarbons being expelled from crevices. The
series may take as long as particles of sufficient size and temperature
will be trapped in the combustion chamber.

58
After one pre-ignition or a series of pre-ignition events it will take time to
accumulate a sufficient amount of hydrocarbons in the piston top land
crevice before an initial pre-ignition will occur again.

One could imagine a severe knocking event being able to initiate a pre-ig-
nition by either expelling and igniting droplets from the crevice or dislod-
ging particles from combustion chamber surfaces. However, during the
conducted measurements that has not been observed.

4. Summary and Outlook

Investigations on pre-ignition occurrence were conducted at a 1.6l SIDI
turbocharged engine with optical access for two high-speed cameras and
an additional light source.
The over the past years well accepted hypothesis on oil/fuel droplets re-
leased from piston top land crevice being a dominating driver for pre-igni-
tion occurrence could be confirmed.
However, deviating from known publications evidence is provided that the
source of a pre-ignition is not the compression ignition of highly inflam-
mable hydrocarbons evaporating from an oil/fuel droplet on its trajectory
through the combustion chamber. Instead, it could be observed that such
an oil/fuel droplet is ignited by the propagating flame of that cycle and
starts burning with diffusive flame. Under the given boundaries this can
create a glowing particle capable to induce an initial pre-ignition in the
subsequent cycle.
It was possible to estimate the critical particle size being in the range of
tenths of a millimeter.

The proposed refined hypothesis provides a reasonable explanation for

the intermittent nature of pre-ignition series. After an initial pre-ignition in-
duced by a glowing particle resulting from the diffusive burn of an oil/fuel
droplet subsequent pre-ignitions are driven by glowing particles resulting
from deposits being dislodged from combustion chamber surfaces during
the initial pre-ignition. The intermittent occurrence of pre-ignitions in a se-
ries results from the need to heat up these particles sufficiently in a regular
combustion cycle to become an ignition source for a next pre-ignition. Ho-
wever, heating up the particles can also occur in combustion cycles with
moderate pre-ignitions and burn rates close to regular cycles.

59
The presented results raise interesting questions requiring further inves-
tigations. So it is yet not fully understood what conditions are necessary
that a particle can sufficiently sustain temperature during gas exchange
and fuel injection.
Beside the size of the particle it can be suspected that oil formulation
and fuel composition are important drivers. Especially oil formulation may
determine the formation of particles in terms of size, heat capacity and
convective heat transfer.
In this regard it looks necessary to analyze the composition of the oil/fuel
droplets with their tendency to form particles when burning with diffusive
flame.

With the release of oil/fuel droplets being considered the prerequisite for
an initial pre-ignition, further investigations are necessary to develop hard-
ware measures avoiding accumulation of oil and fuel in the piston top land
crevice as well as to understand the mechanisms leading to a droplet
release from the crevice. This includes the geometry of the piston top land
crevice, piston ring dynamics and the flow conditions close to the crevice
resulting from global charge motion or locally from squish areas.

Beside further studies on the formation and release of droplets it is still

of high importance to control the state of the charge close to firing TDC
in terms of mixture preparation and homogeneity. The existence of hot
spots in the vicinity of critical components like spark plug or exhaust val-
ves locally facilitates the onset of pre-ignitions by decreased ignition delay
times and therefore has to be avoided.

5. Acknowledgements
The authors would like to thank the peers at Adam Opel AG and General
Motors Powertrain supporting the conducted investigations with design
work, execution of testing as well as valuable discussions: Thomas Gö-
bel, Wolfgang Vogel, Jörg Schäfer, Dirk Wieczorek, Richard S. Davis and
Achim Königstein.

60
References
[1] Dahnz, C.; Han, K.-M.; Spicher, U.; Magar, M.; Schießl, R.; Maas,
U.: Investigations on Pre-Ignition in Highly Supercharged SI Engines,
SAE Technical Paper 2010-01-0355, 2010
[2] Willand, J.; Daniel, M.; Montefrancesco, E.; Geringer, B.; Hof-
mann, P.; Kieberger, M.: Grenzen des Downsizing bei Ottomotoren
durch Vorentflammung, MTZ – Motortechnische Zeitschrift, vol. 70
(05|2009):422-428, 2009.
[3] Amann, M.; Alger, T.; and Mehta, D.: The Effect of EGR on Low-Speed
Pre-Ignition in Boosted SI Engines, SAE Int. J. Engines 4(1):235-245,
2011, doi: 10.4271/2011-01-0339.
[4] Günther, M. et al.: Enthalpy-based Approach to Quantifying and Pre-
venting Pre-Ignition, MTZ 04/2011 Vol.72, 2011
[5] Zahdeh, A. et al.: Fundamental Approach to Investigate Pre-Ignition
in Boosted SI Engines, SAE Technical Paper 2011-01-0340, 2011
[6] Takeuchi, K.; Fujimoto, K.; Hirano, S.; Yamashita, M.: Investigation
of Engine Oil Effect on Abnormal Combustion in Turbocharged Direct
Injection – Spark Ignition Engines, SAE Technical Paper 2012-01-
1615, 2012
[7] Amann, M.; Alger, T.; Westmoreland, B.; Rothmaier, A.: The Effects of
Piston Crevices and Injection Strategy on Low-Speed Pre-Ignition in
Boosted SI Engines, SAE Technical Paper 2012-01-1148, 2012
[8] Dahnz, C. et. al.: Vorentflammung bei Ottomotoren (Abschlussbe-
richt), FVV-Heft 907 – 2010

61
4 Minimising autoignition for optimum
­efficiency in high specific output spark-
ignited engines
Paul Freeland, Marco Warth, Bernd Mahr

Abstract
From the very outset of the development of spark-ignited IC engines, the
need to avoid autoignition has limited both the geometric compression
ratio and the boost pressure that can be employed, thereby effectively
restricting both engine efficiency and specific performance levels that can
be achieved.

The current trends for downsizing in the meantime require ever increasing
specific power and torque output levels on the one hand, and lead to in-
creased engine residency time at higher load on the other. These factors
mean that a greater proportion of real-world driving time is spent within re-
gions of the operating map where combustion phasing is limited by knock.
Compounding this, the combination of high boost pressures and retarded
combustion phasing have also significantly increased the propensity for
pre-ignition at low engine speeds.

Due to these effects, the fuel efficiency of vehicles is becoming increa-

singly sensitive to the knock limit, and reducing the tendency for fuel/air
mixtures to auto-ignite is key to delivering improvements in real-world fuel
consumption. To this end, a systematic and holistic approach to minimi-
sing knock is required and a detailed understanding of the interaction bet-
ween engine systems in this respect is vital.

This paper focuses on selected examples of MAHLE’s approach, taken

from development work conducted with its own agressively downsized 1.2
litre three-cylinder engine, to illustrate the effectiveness of the integrated
technologies in providing optimum resilience to auto-ignition. It concludes
with an outlook on future powertrain developments that will be required to
ensure that engine operating efficiency can be maintained with increased
specific power outputs.

62
Summary
The occurrence of autoignition and the subsequent phenomenon (and de-
structive potential) of knocking has limited both the compression ratio, and
also the combustion phasing that can be employed within spark-ignition
engines. Both of these factors reduce the efficiency of the combustion and
expansion process, and limit the potential fuel consumption / CO2 emissi-
on levels that can be achieved.

Continued developments over the past century, in both engine and fuels
technology have led to a steady increase in the compression ratio that
can be used in production engines. However achievable values are still
below the empirical optimum of around 16:1 Indeed, while current trends
towards downsizing in engines have shown tangible benefits at part-load
in terms of operating point shifting and de-throttling, the need to boost the
engine to maintain the required power output has pushed further the res-
trictions on compression ratios and combustion phasing that can be used.

Thus, now as much as ever, the need to develop ways to minimize or even
eliminate autoignition is critical to achieving reduced fuel consumption and
CO2 emissions for the next generation of engines.

In this respect, it has been shown that careful attention to detail at the
design stage can give excellent cylinder head cooling potential, which has
demonstrated that increased compression ratios can be used even at very
high power output levels of up to 100 kW/l or more.

Furthermore, when combined with good scavenging, advanced charge

air cooling and cooled EGR open the potential to increase compression
ratios yet further, and it is envisaged that compression ratios for gasoline
engines will continue to rise, even with highly boosted engines.

List of Abbreviations:
A/C Air Conditioning
ATDC After Top Dead Centre
BMEP Brake Mean Effective Pressure
BMTS Bosch-MAHLE Turbo Systems
BSFC Brake Specific Fuel Consumption
CAS Charge Air Subcooler

63
CFD Computational Fluid Dynamics
CO2 Carbon Dioxide
CR Compression Ratio
EGR Exhaust Gas Recirculation
EGRC EGR Cooler
GDI Gasoline Direct Injection
HT High Temperature
I2CAC Integrated Indirect Charge Air Cooler
IMEP Indicated Mean Effective Pressure
ISAC Indicated Specific Air Consumption
IVC Inlet Valve Closing point
KI Knock Index
LT Low Temperature
MFB Mass Fraction Burned
MON Motor Octane Number
OH Hydroxide Radical
RON Research Octane number
TDC Top Dead Centre

Introduction
From the very earliest developments of the internal combustion engine,
autoignition of part of the charge during the combustion process has limi-
ted the compression ratio and power output potential of engines. Indeed
the phenomenon of autoignition of end gas regions was known even in hot
tube / hot bulb engines before “positive ignition” systems became com-
monplace, and even in those days methods had to be developed to retard
the ignition point of the main charge to give acceptable combustion with a
variety of fuels available [1].

Over time, developments in combustion technology and engine cooling,

together with developments in fuel technology have led to a steady increa-
se in the compression ratios that are capable of being employed. Howe-
ver, in practice production engine compression ratios still lie significantly
below the empirical optimum value of around 16:1 to 17:1 [2]

64
 Figure 1: Trends in compression ratio for gasoline engines over the last
century

Indeed, with recent trends towards downsizing, the need to boost engines
to achieve the required power has necessitated reductions in the maxi-
mum compression ratio that can be used, resulting in a corresponding
degradation in potential BSFC.

These downsizing trends, whilst giving tangible de-throttling benefits un-

der light-load operation, also result in a greater proportion of real world
driving occurring at higher engine loads, and thereby a greater portion of
the operating time occurs in areas where the knock limit has a significant
impact on operating efficiency.

A recent study conducted by MAHLE [3] based on a large sample of dri-

ving records illustrates the engine residency trends of modern downsized
engines very well, for different driving styles over different types of road.
This data is shown graphically in figure 2 below, and highlights the fact
that, even under “gentle” driving, downsized engines operate within the
regions of the speed-load range where engine efficiency is reduced due
to the knock limit.

65
 Figure 2: Operating residency points for a 1600kg ‘C’ Class Downsized
Vehicle for different driving styles

These two factors have meant that the sensitivity of fuel consumption to
the knock limit during real world driving has become more significant, and
minimizing an engine’s propensity to knock is more important than ever in
achieving low CO2 vehicles.

1. Measures to mitigate autoignition

The occurrence of autoignition of the end gases within IC engines is a
complex process involving interactions between aspects of both engine
hardware and the characteristics of the fuel being used. However, funda-
mentally it can be simplified to be a function of three main factors:

66
 Figure 3: The three main practical factors influencing autoignition

Any effort to reduce the tendency for autoignition within an engine has to
address one or more of these factors.

1.1. Fuel Composition:

Simplistically, the chemical composition, and thereby the anti-knock or oc-
tane rating of a fuel is generally considered to be the responsibility of the
oil refinery / fuel blenders. Even though in most of the world, the fuel com-
position is subject to legislated standardization norms (such as EN228 for
gasoline fuels in Europe) in practice such standards are relatively broad
and still allow for wide ranges of blends to be sold. However, whilst there
is scope to increase the octane rating of fuels within the wider markets,
typically the chemical compounds available that give an increase in knock
resistance (such as aromatic compounds, ethers and alcohols) tend to
be relatively expensive to include, and potentially harmful. Therefore it is
generally recognized that there is a cost penalty to increasing the octane
rating of gasoline.

For these reasons, the octane rating of fuel has remained fairly stable
over the past forty years, and engine manufacturers currently focus their
efforts on making engines perform to the required level with fuels that are
available in each of the markets in which they are to be sold.

67
 Figure 4: Available Octane rating in the UK over the last century

However, with increasing demand for engines to operate at increased

boost levels without compromising compression ratios and part-load ef-
ficiency, there is definitely an appetite for higher octane rated fuels within
the market. There is also a strong political desire among most of the wor-
ld’s leading nations to increase the renewable energy component within
gasoline fuels for reasons of increased self-sufficiency regarding fuel sup-
ply, and in order to limit reliance on fossil fuels as much as possible. For
these reasons, it may well be that there is the motivation on both parts to
see an increase in the octane rating of standard fuels in the future, brought
about by an increased use of alcohol blends.

In this respect, one interesting fact to emerge over recent years is that
when alcohol blends are used as octane enhancing components in gaso-
line, the charge cooling effect brought about by the high latent heat of
evaporation gives additional anti-knock characteristics, particularly under
RON test conditions (which are closer to those experienced by highly
boosted engines [4]). This effect becomes even more pronounced when
these fuels are used in GDI engines, where the cooling effect of fuel
vaporisation is applied directly within the engine cylinders [5]. This means
that the effective resistance to knock that these fuels give when used in
direct-injection engines is increased disproportionally beyond the nominal
RON / MON rating.

68
1.2. Charge temperature:
In general, the rate of chemical reactions increase with temperature, and
this certainly true for the chain branching reactions that form the interme-
diate species that lead to autoignition [6]. An increase in the charge tem-
perature prior to combustion reduces the time required for these reactions
to complete, and makes it more likely that required conditions for autoig-
nition can exist in the unburned mixture before the normal flame front can
progress to engulf it.

In this respect, any measure that reduces the temperature of the remai-
ning unburned mixture will reduce the tendency to knock. Such measures
may be applied to the incoming charge prior to the start of combustion,
such as minimizing the heating of the charge air from the intake duct surfa-
ces, or through active charge cooling. However, there are also measures
that can be applied to reduce the heating of the charge within the cylinder
itself, such as better cooling of the metal surfaces, reducing the amount of
hot residual gases left over from the previous cycle, introduction of cooled
residual gasses (by virtue of the higher specific heat capacity, and the re-
duction in flame temperatures during combustion) and as engine speeds
increase, by a reduction in the time that the gas spends in contact with the
hot metal surfaces prior to the passage of the combustion flame.

1.3. Time available:

As implied above, because the chemical reactions that occur when a fuel-
air mixture is compressed and heated require finite time to take place, the
faster the combustion process is, the less time there is for the required
pre-reactions to complete, and the less likely autoignition is to occur.

This means that any effects that speed up combustion and reduce the
time it takes for the flame to traverse the combustion chamber will also
reduce the tendency for autoignition.

Measures that are available to assist in this respect include increasing the
charge motion at the time of ignition, reducing the bore diameter, creating
a more compact combustion chamber shape, and even using faster-bur-
ning mixtures (such as effected by introducing hydrogen into the charge).

In this respect, high-speed engines are generally more tolerant to kno-

cking than those designed to give peak power at lower speeds, for al-

69
though high speed engines are still required to operate at lower engine
speeds, the requirement for longer cam durations for high-speed volu-
metric efficiency means that low-speed volumetric efficiency is reduced.
Hence at low speeds, trapped mass is less (and therefore charge pres-
sure and temperature as the cylinder volume reduces) and in the speed
region where knock has time to occur, higher compression ratios can still
be tolerated.

2. Practical Experience

2.1. Base Engine Design:

In 2006, work was started to design a bespoke 1.2 litre 3 cylinder down-
size engine, in order to investigate the requirements for extreme downsi-
zing, and to trial new components and systems [7].

Since then, this engine has been used to trial and demonstrate new tech-
nologies that were considered advantageous to improving the efficiency
of future engines.

This exercise has also provided the chance to share openly some of the
research and development work conducted, free of the constraints of
commercial confidentiality.

A summary of the engine specifications is given in figure 5 below:

Figure 5: MAHLE Downsize Engine: basic details

70
From the start, all the factors mentioned in section 2 were taken into con-
sideration when designing this engine. Initially, to give the most compact
combustion chamber and shortest flame path, an “oversquare” bore to
stroke ratio was considered. However, after reviewing the required valve
sizing, the packaging of a central injector and a spark-plug in their opti-
mum locations, and the desire to give the cylinder head the best cooling
capability, these requirements led to the need for a greater bore size, and
the basic cylinder dimensions were relaxed accordingly.

Notwithstanding this, a compact four-valve combustion chamber layout

was achieved that would enable a range of compression ratios to be ac-
commodated without giving too narrow a flame path. A sectional view of
this design is shown in figure 6:

Figure 6: Combustion Chamber cross-section

Two types of inlet port were designed – one to give tumble rates that
matched the highest seen at that time within the available database (the
“tumble port”) and an alternative that gave “moderate” levels of tumble in
comparison, but with increased flow rates (the “flow port”). The flow and
tumble characteristics of these ports are shown in figure 7:

71
 Figure 7: Inlet port flow and Tumble Characteristics

In practice, little difference was noticed regarding peak performance and

efficiency between these two designs, and the higher tumble port was re-
tained for further use for development testing and vehicle demonstration.

2.2. Metal Cooling:

With regard to cooling the metal temperatures, a split system was desi-
gned that fed two separate engine cooling circuits – one for the cylinder
block and one for the head. The circuit for the head was arranged in a
“cross flow” pattern, with the coolant entering a gallery on the left-hand
side of the block and feeding three passages that, in turn, fed into the
cylinder head below the exhaust ports. Particular attention was given to
the flow around the exhaust valve seats and guides, and the combustion
chamber skull to ensuring good flow velocities in these areas. The cylin-
der head coolant passages are shown in figure 8:

72
 Figure 8: Cylinder Head coolant passages

The result of this attention to detail to the cylinder head cooling has been
excellent cooling of the hottest parts of the combustion chamber, the
exhaust valves and the spark-plug. This has enabled peak metal tempe-
ratures to be controlled to remarkably low values, despite the very high
specific power output of the engine. The distribution of metal temperatures
measured in the cylinder head is illustrated in figure 9:

73
 Figure 9: Cylinder Head metal temperatures

The overall benefits of these design measures have been a very good
resistance to knock in the base engine design, and an engine concept that
delivers excellent full-load fuel efficiency, achieving BSFC values as low
as 265 g/kWh at 30 bar BMEP, and 285 g/kWh at 100 kW/litre.

2.3. Compression Ratio Potential

Testing to explore the compression ratio limit for this engine has demons-
trated that the 120kW peak power target can still be achieved with com-
pression ratios as high as 12.2:1, even when using 95RON pump grade
fuel.

Figure 10 shows results obtained for three different compression ratios

with this engine. In the left-hand column is the crank angle at which 50%
of the charge mass is burned (50% MFB). Generally, engine efficiency can
be regarded as starting to be compromised once the 50% MFB point has
to be retarded to more than 12° ATDC (this value shown as the thick blue
dashed contour).

At 10.2 CR, it can be seen that combustion phasing became compromi-

sed at engine loads greater than 10bar BMEP at 1000rev/min, and 15bar
BMEP above 2500rev/min. The most retarded combustion phasing oc-
curred between 1500rev/min and 2000rev/min, where the 50% MFB point
was just over 30° ATDC. At the compression ratio was increased, so the

74
propensity to knock required combustion phasing to be retarded further,
such that at 12.2 CR the 50% MFB point was 34° ATDC.

Nonetheless, these values are not unusual when considering boosted en-
gines, and are remarkable when the actual BMEP levels and compression
ratios involved are taken into account.

In the right hand column, the Indicated Specific Air consumption (ISAC)
achieved at 10.2 CR, and the delta from this baseline value for the other
two compression ratios is shown. These values are calculated from the
gross IMEP (ie the area within the “high-pressure” loop only) and therefore
give a direct measure of the efficiency of the heat addition and expansion
processes, without influences from friction or pumping work. It can be cle-
arly seen that where combustion phasing is more retarded, ISAC values
became greater (indicating worse efficiency).

As compression ratio was increased from 10.2 to 12.2, up to 5% benefits in

ISAC were seen in the part-load region as cycle efficiency was improved.
However, at full load where combustion had to be further retarded due to
knocking, ISAC values became worse, by between 2 to 8% at 11.2 CR,
and between 7 and 15% at 12.2CR. While it is clear that the higher com-
pression ratios require combustion phasing to be retarded further from the
optimum due to the knock limit, the degree to which this was required, and
the effect of this on engine efficiency was found to be surprisingly small
in this case, and the fact that the original peak power target could still be
achieved at the higher compression ratio without unacceptable combusti-
on was considered remarkable.

75
 Figure 10: The effect of compression ratio on knock limit and operating
efficiency

2.4. Charge Air Temperature Management:

Reducing the temperature of the air before it enters the cylinder has two
distinct advantages in reducing the tendency of an engine to knock.

• Firstly, the direct effect of cooler charge temperatures is to lower the

bulk gas temperature as it comes into the cylinders, and therefore the
temperature of the remaining unburned mixture towards the end of
the combustion process.

76
• Secondly, the cooler charge is more dense, and therefore the boost
pressure required for a given charge mass is reduced.

When these effects are combined, the benefit to thermal efficiency of the
engine means that the boost requirement is reduced yet further. This re-
sults in less turbine work and a corresponding reduction in pre-turbine
exhaust pressure, thereby also reducing the mass of residual gas left in
the cylinder, which gives yet further benefits in reducing the tendency to
knock.

Practical methods of reducing charge air temperatures as much as pos-

sible were therefore sought. Two potential systems were designed and
steady-state dynamometer tests were conducted to evaluate these.

The first system, called the cascaded charge air cooler, incorporates two
coolers arranged in series in the inlet air flow – a conventional heat-exch-
anger using engine coolant to give an initial reduction in charge tempe-
rature, and a subsequent heat exchanger with its own dedicated coolant
loop that operates at a lower temperature, to further reduce the charge
temperature. This layout is shown in figure 11:

Figure 11: Cascaded Charge Air Cooling Layout

77
A production feasible manifold assembly was designed as shown in figure 12:

Figure 12: Cascade Charge Air Cooling Design

With this system, it was possible to achieve charge air temperatures of

around 30° C (with ambient test-cell air at 25° C) ie, very close to those
obtained in naturally aspirated engines.

To extend the cooling capacity further, a charge air sub-cooler module

was designed, which incorporated a refrigerant evaporator. This system
is shown in figure 13:

Figure 13: Charge Air Sub-Cooling Layout

78
This system was initially designed to make use of spare capacity from
the vehicle air conditioning circuit. However, a dedicated refrigerant circuit
could be used if the application warranted the expense and complexity.
The module designed is shown in figure 14:

Figure 14: Charge Air Subcooling Module Design

The results from these tests are summarized in figure 15 below, showing
engine speed plotted on the X-axis, and full-load charge temperature on
the Y-axis.

79
Figure 15: Effect of Charge Air Temperature on knock limit & engine effi-
ciency

The original integrated charge air cooler was capable of maintaining 40°C
temperatures in normal operation, and this level is marked with the red
dashed line.

The upper diagram shows the 50% MFB angle, whilst the middle diagram
shows the corresponding BSFC values. The lowest diagram shows the
percentage difference in BSFC from that obtained at 40° C, at different
engine speeds.

80
It is interesting to note that at low engine speeds (below about 2000 rev/
min) the effect of charge temperature on the knock limit and BSFC is
almost negligible. It is presumed that this is a result of two time-related
effects.

• Firstly at the lower speeds there is (in relative terms) significantly

more time for the incoming gas to pick up heat from the inlet ports,
valves and cylinder walls, and radiated heat from the initial stages of
combustion. Therefore the resulting end gas temperatures are not
reduced directly in proportion to the reduction in charge temperature
in the plenum. However, it would be expected that there would still be
some relationship between plenum temperature and end gas tempe-
rature under these circumstances.

• Therefore secondly, it is presumed that at these speeds the time re-

quired for the intermediate chain-branching reactions to complete is
sufficient to give rise to knock, somewhat irrespective of the actual
end gas temperatures.

As engine speed increased, the gradient of the combustion phasing con-

tours became more horizontal, and the expected sensitivity of the knock
limit to charge temperature became established. In this region, a 10°C
reduction in charge temperature allowed the combustion phasing to be
advanced by approximately 2° crank angle, resulting in an improvement
of approximately 2% BSFC.

At higher engine speeds the sensitivity of combustion phasing to charge

air temperature became even greater, presumably as a result of the re-
duced residency time spent by the charge mixture in the cylinder prior to
combustion, and to the reduction in boost pressure required to give target
torque. Here, an even greater reduction in BSFC was achievable, as the
more advanced combustion phasing reduced exhaust gas temperatures,
and less overfuelling was required to maintain safe turbine operation.

2.5. Cooled EGR:

Whilst hot residual gasses are distinctly detrimental to the knock limit,
if exhaust gases are cooled prior to being re-introduced into the engine
cylinder then the increased specific heat capacity of the exhaust gas frac-

81
tion has a measurably beneficial effect by reducing combustion tempera-
tures, and hence the propensity to knock.

In order to control and cool the Exhaust gas, specific EGR coolers and
potentially, cooled EGR valves are required. For this purpose a bespoke
integrated EGR cooler and valve assembly was designed as shown in
figure 16:

Figure 16: Integrated EGR Cooler & Valve Assembly

Two cooled EGR circuits were investigated: a low-pressure circuit whe-

reby the exhaust gases are taken from downstream of the turbine (either
pre- or post-catalyst) and fed into the inlet stream upstream of the com-
pressor; and a high pressure circuit whereby the exhaust gases are taken
from before the turbine, and introduced into the inlet manifold. These cir-
cuits are shown in figure 17:

Figure 17: High-Pressure (left) and Low-Pressure (right) EGR Circuits

82
The advantages of the low-pressure circuit are that the exhaust gas is
well mixed with the intake air, and the products of combustion are more
completely reacted and thereby more inert, however the disadvantages
are that there is a lower pressure differential available to drive the exhaust
gas into the inlet stream and hence less EGR can be introduced this way.
Also, the exhaust gas flow has to be accommodated by the compressor,
and therefore extra demand is placed on the charging system. Typically
with such layouts, the run-up line (where engine torque is limited by the
boosting system capacity) is worse, as is transient response.

The high-pressure circuit, in comparison offers a greater potential pres-

sure difference, especially at high speeds, and significantly greater quan-
tities of exhaust gas can be introduced. Also, the compressor capacity is
not compromised by the need to handle the inert exhaust gas. However,
extra attention is required to ensure the exhaust gas is well distributed
between the cylinders, and there is the potential that some of the products
are not fully reacted, and may in themselves play a part in causing knock,
or pre-ignition. Also, the pressure differential at lower engine speeds is
often negative (as intake boost pressure exceeds the pre-turbine exhaust
pressure). Therefore EGR is prevented unless an EGR pump is used.

The results from these investigations are summarised in figure 18, which
shows the amount of EGR introduced with each system, the difference in
the 50%MFB angle given from the case with no EGR, and the correspon-
ding reduction in BSFC:

83
 Figure 18: Effect of LP and HP EGR on the knock limit and BSFC

The difference in EGR rates available with the two systems is evident,
especially at higher engine speeds. The low pressure curcuit was able to
introduce up to 20% externl EGR, and gave the best flow rates at relatively
low engine speeds. The high pressure circuit was able to introduce up
to 35% external EGR at higher speeds, though more noticeably with the
high-pressure circuit, the capacity to flow EGR reduced at load increased
due to the increase in intake plenum pressure opposing the flow.

For this engine, 20% to 30% EGR reduced the propensity to knock such
that combustion phasing could be advanced by approximately 5° to 6°
crank. In turn, this led to an improvement in high-load BSFC of appro-
ximately 4% to 8%. At the higher engine speeds, the increased specific
heat capacity of the cooled exhaust gas, together with the advanced com-
bustion phasing had an additional direct effect on reducing exhaust gas

84
temperatures, and thereby reducing the amount of overfuelling required to
keep gas temperatures below the turbine limit. Thereby further reductions
in fuel consumption were realised.

3. Conclusions and Future Outlook

The design and development of spark-ignition engines regarding cooling
and combustion processes over the past century has led to a steady in-
crease in the compression ratio that can be employed, yet values achie-
vable are still below the theoretical and practical optimum for best thermal
efficiency. To improve this yet further requires constant work at reducing
in-cylinder charge temperatures, increasing flame velocities at low engine
speeds, and attention to detail in designing cylinder head cooling arrange-
ments and combustion chamber layouts.

These approaches in themselves are not new, but advances in design

tools, materials and production techniques are enabling the knock limit to
be constantly reduced. In this respect, it has been shown that:

• A base engine design can be created that enables compression ratios

of more than 12:1 to be adopted, whilst still achieving 30bar BMEP
and 100kW/litre

• Increased cooling of the charge air gives the biggest benefits at hig-
her engine speeds where the charge has less time to pick up radiated
and conducted heat from the engine cylinder and flame propogation
process

• Cooled EGR can reduce the tendency to knock by approximately 1°

per 4% to 5% EGR – again, more prevalent at higher engine speeds

• For lower engine speeds, it would appear that the dominant factor is
time that the unburned mixture sits within the cylinder before the fla-
me reaches it. In this respect, any measure that improves the speed
of combustion (eg increased charge motion) and reduces the flame
path (eg, reduced bore size) will have the greatest effect.

Continued developments within the automotive industry, together with

continued political pressure to increase the percentage of renewable or-
ganic components used in gasoline blends means that the emergence of

85
fuels with increased octane rating for use in highly boosted engines would
seem sensible.

Therefore, it is envisaged that the trend for ever increasing compression

ratios will continue – at least to the point where the practical optimum of
around 16:1 to 17:1 is reached, allowing the best thermal efficiency to be
achieved whilst specific power outputs are pushed yet further.

Looking to the future, it is apparent that the trend in downsizing will conti-
nue into the next generation of engines at least. MAHLE have demonstra-
ted that 50% downsizing is quite a feasible concept, and with increasing
electrification, options to improve the dynamic response of heavily down-
sized engines such as e-boosting and mild „torque-assist“ hybridisation
mean that downsizing can be taken a stage further.

In order to accommodate even higher boosting, it is expected that the

useage of cooled EGR and increased charge air cooling capacity will play
an important role. In this respect, detailed thermal management is already
becoming an area of increasing importance for improving the thermal effi-
ciency of gasoline engines.

Another approach to increasing engine efficiency – the use of the „Miller

Cycle“ by employing either early or late IVC allows the effective compres-
sion ratio of an engine to be kept at a level where knock can be controlled,
whilst the expansion ratio can be increased to extract more useful work.
This philosophy has demonstrated significant benefits with regard to full-
load efficiency, although it places yet further demands on the boosting
system as the trapping efficiency of the engine is reduced, and a practical
limit still applies regarding combustion chamber shape and the flow losses
caused by the exhaust turbine etc.

It is also possible that with increased reliance on highly boosted engines,

there will be sufficient demand for a „premium“ fuel grade of more than
100 RON, for this grade to be made more widely available again. The
mounting political and economical pressure to increase the renewable
energy content of future fuels means that increased alcohol blends are a
convenient enabler in this respect.

86
Literature
[1] Crawford, MacKeand; Sparks and Flames, Ignition in Engines An His-
torical Approach; Tyndar Press 1997, pp126-128
[2] Taylor C. F; The Internal Combustion Engine in Theory and Practice;
The M.I.T Press 1992, pp443-445
[3] Taylor J, et al; Analysis of the Actual Operation with Substantially
Downsizing Engine; MTZ 0512013 vol 74
[4] Andrae et al.; Combustion and Flame 155 2008, p696
[5] Oudenijeweme D et al; Downsizing and Biofuels: Synergies for Sig-
nificant CO2 Reductions; Aachen Kolloquium Automobile and Engine
Technology 2011
[6] Griffiths J F; Flame and Combustion; Chapman & Hall, 1995
[7] Korte V et al; Das MAHLE-Bosch Downsizing Demonstrator Fahr-
zeug; Aachen Kolloquium Fahrzeug- und Motorentechnick 2010

87
5 Reduction in knocking intensity of an
SI engine by in-cylinder temperature
stratification
Yasuo Moriyoshi, Tatsuya Kuboyama,
Taku Kondo, Hiroshi Sono

Abstract
Many works have been made on knocking phenomenon in spark igni-
tion engines. The purpose of the works focuses on how to control the
autoignition following knocking. However, the important problem due to
knocking is serious damages of engine components such as piston and
valves by heavy pressure oscillation. Thereby, the aim of this study is to
suppress the strong pressure oscillation, even if autoignition occurs. The
effect of spatial temperature distribution of a mixture inside the cylinder on
knocking intensity was examined using a rapid compression and expan-
sion machine (RCEM). Results of experiments show that controlling the
knocking intensity can be realized by controlling the spatial temperature of
unburned mixture at the onset timing of autoignition.)

1. Introduction
Knocking is a resonance phenomenon as the pressure wave caused by
the local autoignition with locally high pressure of unburned end-gas com-
pressed by the burned gas with flame propagation comes back and forth
inside the cylinder. When the pressure wave propagates in sound speed,
the amplitude of wave becomes double at the wall as the phase of wave
turns over at the wall end. As a result, heat transfer to the wall or the oppo-
site-side wall will be enhanced, leading to a damage of the engine by this
abnormal combustion[1-4]. Since knocking is the most serious problem
to increase thermal efficiency of a spark ignition engine, many kinds of
studies have been made. However, most of them are engaged in how to
avoid autoignition. The real issue is the damage of engine by the strong
pressure oscillation not by the autoignition itself. For example, HCCI en-
gines follow autoignition that occurs simultaneously in the whole space of
cylinder, but not cause knocking. (Of course, the steep pressurerise brings
about noise and vibration issues.)

88
The objective of this study is to propose a combustion control method to
prevent a severe knocking even causing autoignition. The idea is to make
a spatial distribution of temperature using a RCEM (Rapid Combustion
and Expansion Machine). The effect of temperature distribution was ob-
served by using a high speed video camera.

2. Experimental Apparatus and Conditions

As a real engine is not suitable for basic study, the RCEM was employed
in this study to simulate a single combustion in safe and with easy data
acquirement. Figure 1 shows the schematics and specifications.

Fig. 1 Schematics and specifications of RCEM

The operation starts first by releasing compressed air into the driving unit
and the driving piston connected to the cam quickly moves. The horizontal
movement of cam makes vertical motion of piston connected to the rod
contacting to the cam. The cylinder volume is 480 cc and piston speed
corresponds to 600 rpm.

The experimental apparatus is shown in Fig. 2. Around the cylinder liner, 8

water paths run vertically as indicated in Fig. 2. Each path is independent,
so different water temperature can be set. Relatively cool water and hot
water are prepared using two tanks. A fuel-air mixture prepared in a tank
will be supplied instantaneously to the bottom of cylinder by electro-mag-

89
netic valve. The displacement of piston is measured using a laser sensor.
In the cylinder head, two pressure sensors (Kistler 6052C) are installed
near the spark plug and its opposite side. Pressure data sampling was
made in 1 μs while other data were sampled in 10 μs.

Experimental conditions are indicated in Table 1. Compression ratio is 9.1

and nbutane was used as fuel that has similar octane number to gasoline
and its boiling point is suitable for experimental setup. The equivalence
ratio was set to stoichio-metric and intake pressure was set to 90 kPa, a
little bit lower than atmospheric pressure. The wall temperatures were set
as indicated in Table 2. Each temperature was measured at 2 mm left in-
side from the cylinder wall using themo-couple. The average temperature
was set to 65 deg.C in each condition. Four conditions, A to D, are tested.

Fig. 2 Experimental apparatus

90
 Table1 Experimental conditions

Bore x stroke Ф 80.4×95 mm

Compression ratio 9.1
Fuel (RON, Boiling point) n-butane (94, -0.5C)
Initial intake pressure (at BDC) 90 kPa
Spark timing 20, 30 deg. BTDC
Temperature of mixture 303 K
Fuel/air equivalence ratio 1.0

Figure 3 shows predicted in-cylinder gas temperature distributions on the

middle plane at 20 and 30 degs. BTDC using FLUENT 3D code. The star
indicates the position of spark plug. In experiments, the cylinder head was
replaced to a quartz window to visualize the phenomenon.

Table2 Experimental settings of wall temperature on cylinder

91
 (a) 20deg.BTDC

(b) 30deg.BTDC

Fig.3 Temperature distribution inside cylinder by numerical calculation

3. Analytical Method
The pressure data obtained by the sensor located in the end gas regi-
on was used for combustion analysis. Knocking Intensity (KI) was deter-
mined as the band-passed component between 3 and 15 kHz indicated in
Fig. 4. When the intensity exceeds 0.05 MPa, the timing was determined
as the onset of autoignition.

92
 Fig.4 Definition of knocking intensity KI

The mass rate of unburned fuel at autoignition onset timing was determined by
the following equation, where the maximum accumulated heat release is ,
and the instantaneous accumulated heat release at autoignition onset is
.

The first wave in knocking is an instantaneous pressure-rise and an im-

portant factor to see the scale of autoignition. In this study, a mean pres-
sure-rise rate during a certain period indicated in Fig. 5 is a factor for exa-
mination. In Fig. 5, the pressure data is low-pass filtered at 5 kHz and the
pressure difference between 80% of Pmax and 20% of Pmax is devided
by the time period.

93
 Fig.5 Definitions of dp and dt

4. Results and Discussion

Figure 6 shows the mass fraction of unburned mixture at autoignition
onset. The fraction at the ignition timing of 30 deg. BTDC is very small in
case D. Generally, when the fraction is small, knocking intensity is small.
Thus, consideration was made using the data with ignition timing at 20
deg. BTDC that show mostly similar unburned fraction. Typical instanta-
neous histories of pressure and heat release are shown in Fig. 7 while
independent KI data and its averaged value are plotted in Fig. 8. The
averaged data show that case C has the largest KI followed by cases B,
D and A.

94
Fig.6 Mass fraction of unburned mixture at the time of autoignition

Fig.7 Histories of pressure and heat release for cases A~D

95
 Fig. 8 Individual runs and the average knocking intensity

Let us consider the result. In case B, mixture with relatively high tempera-
ture is consumed by the flame propagation and mixture with relatively low
temperature remains in the end-gas zone. Oppositely, in case C, mixture
with relatively low temperature is consumed by the flame propagation and
mixture with relatively high temperature remains in the end-gas zone. Me-
anwhile, case A shows a gentle temperature gradient with slightly higher
temperature inside the cylinder and lower temperature near the wall. In
case D, mixture with relatively low and high temperatures burn simulta-
neously, burn simultaneously, mixture with relatively low and high tempe-
ratures may exist in either end-gas region.

Figure 9(a) indicates calculated distribution of Livengood-Wu integral in

mass fraction at each onset timing of autoignition while (b) does its cumu-
lative mass fraction. Figure (a) shows that higher mass fraction is found
large in case C at high Liven-good-Wu integral due to relatively homoge-
neous temperature distribution, while the mass fraction is flat and wide
in cases A and D due to relatively large temperature distribution. These
results can lead an assumption that KI is high when temperature in the
end-gas zone is nearly constant while KI is low when temperature gradient
in the end-gas zone exits.

96
 Fig. 9 (a: left) Histograms of mixture mass fraction vs Livengood-Wu
integral
(b: right) Cumulative of mixture fraction vs. Livengood-Wu integral

Next, in order to visualize the in-cylinder phenomenon, direct photography

was adopted as shown in Fig. 10. Figure 11 are temporal variations of the
pictures in each case.

Fig.10 Setup for in-cylinder visualization

97
 Fig.11 Visualized images of autoignition

High speed photography was conducted with 30000 fps and the autoig-
nition timing was determined when the first radiation of light was found.
In cases B and C, autoignition occurs at a point around the center of the
endgas region, propagating to the whole space of end-gas region in 0.2
ms after the first autoignition. While in cases A and D, autoignition occurs
at multiple positions, especially in case D, time difference of 0.1 ms was
found between the first autoignition and the second. Moreover, in cases
A and D, the combustion speed after autoignition is slower than in B or C,
remaining unburned gas region after 0.2 ms. This visualization clarified
that when autoignition occurs locally at multiple positions and in different
timings, KI can be restrained by the thermal stratification in lateral direc-
tion perpendicular to flame propagation direction. Figure 12(a) indicates
the pressurerise rate, dp/dt between 20 and 80 % to the maximum pres-
sure. Figure 12(b) shows the autoignition timing. In both figures, each run
and the average are plotted with diamond and circle, respectively.

98
Fig.12(a) Pressure-rise rate Fig.12(b) Autoignition onset timing

Figure 12(a) indicates that dp/dt is remarkably decreased in case D while

the autoignition timing is almost the same among the cases. This means
that not only the maximum knocking amplitude but also the pressurerise
rate can be controlled and that the restrained knocking intensity in case D
is not due to the retard of autoignition.

5. Conclusions
(1) Some modifications were made to the RCEM (Rapid Compression and
Expansion Machine) to enable the thermal stratification inside the cylinder
by using eight lateral cooling water paths inside the cylinder and also to
charge a mixture quickly at BDC position.

(2) Combustion tests with the thermal stratifications were conducted using
n-butane as a fuel. As a result, making the thermal stratification in late-
ral direction perpendicular to flame propagation direction can drastically
decrease the knocking intensity.

(3) To examine the above result, visualization was conducted. Usually,

autoignition occurs in the end-gas region simultaneously where compres-
sed unburned mixture with high pressure and temperature, and leading to
a strong pressure oscillation. However, autoignition occurs locally and at
different timings, leading to weak pressure-rise rate.

99
References
[1] G. Konig et al.: Role of Exothermic Centers on Knock Initiation and
Knock Damage, SAE paper No. 902136
[2] J.B. Heywood: Internal Combustion Engine Fundamentals, McGra-
wHill (1988) 457.
[3] C.S.Draper: Pressure Waves Accompanying Detonation in the Inter-
nal Combustion Engine, J. Aeronautical Sci. 5-6 (1938)
[4] J. Pan, M. Berzines: End Gas Inhomogeneity, Autoignition and Knock:
SAE Paper, 982616 (1998)
[5] S.Wong, T.Kuboyama, Y.Moriyoshi: Proceedings of JSAE (in Japane-
se) No.20125048 (2012)

100
6 New approach to the determination of
knock onset
Sebastian Ohler

Abstract
In the development of new gasoline engines, detailed knowledge of
misfiring is crucial to attempts to reduce CO2 emissions and thus fuel
consumption. In addition to the knock intensity and knock location, the
start of knock is an important characteristic in the study of knocking com-
bustion. With conventional determination of the knock onset by analysing
the cylinder pressure signal the results are subject to inaccuracies due to
the measurement and analysis methods used.

In this paper a new computational approach is presented with which the

knock onset can be determined based on the cylinder pressure signal
with a greater degree of accuracy. This approach is based on the mathe-
matical analysis of the net heat release rate of a low-pass filtered cylinder
pressure signal, sampled at 0.1° of the crank angle. The results of the new
method are evaluated with accompanying experiments on a four-cycle
single-cylinder research engine using a fibre optic measurement spark
plug. In this case, the knock onset is determined by the light intensity cha-
racteristics of the fibre optic cable connected to the test engine.

By varying the compression ratio, the ignition point and the use of fuels
with different octane numbers, there is a linear correlation at different
knock intensities between the knock onset determined by the presented
net heat release rate knock criterion and by the optical measurement
technology.

Kurzfassung
Bei der Entwicklung neuer Otto-Motoren ist hinsichtlich der Reduzie-
rung des CO2-Ausstoßes und damit des Kraftstoffverbrauchs die genaue
Kenntnis der irregulären Verbrennungsvorgänge von entscheidender
Bedeutung. Neben Klopfintensität und Klopfort gilt der Klopfbeginn als
wichtiges Charakteristikum zur Untersuchung der klopfenden Verbren-
nung. Bei der bisher üblichen Bestimmung des Klopfbeginns durch die

101
Auswertung des Zylinderdruckverlaufs treten aufgrund der verwendeten
Mess- und Analysemethoden Ungenauigkeiten im Ergebnis auf.

Im Beitrag wird ein neuer rechnerischer Ansatz vorgestellt, mit welchem

der Klopfbeginn anhand des Zylinderdruckverlaufs mit einem geringeren
Fehlereintrag bestimmt werden kann. Dieser Ansatz beruht im Wesent-
lichen auf der mathematischen Untersuchung des Heizverlaufs eines
tiefpassgefilterten, bei einer Abtastung von 0,1°KW aufgenommenen Zy-
linderdruckverlaufs. Die Ergebnisse der neuen Methode werden durch
begleitende Versuche an einem Viertakt-Einzylinderforschungsmotor mit
Hilfe einer Lichtwellenleiter-Messzündkerze überprüft. Hierbei wird der
Klopfbeginn anhand der Lichtintensitätsverläufe der am Versuchsmotor
applizierten Lichtwellenleiter bestimmt.

Durch die Variation des Verdichtungsverhältnisses, des Zündzeitpunkts

und die Verwendung von Kraftstoffen unterschiedlicher Oktanzahl zeigt
sich bei unterschiedlichen Klopfintensitäten eine lineare Korrelation zwi-
schen dem nach dem vorgestellten Heizverlauf-Klopfkriterium und dem
anhand der optischen Messtechnik bestimmten Klopfbeginn.

1. Introduction
In the development of new gasoline engine generations, misfiring, in parti-
cular due to a shift in the operating point to higher loads for turbocharged
engine concepts, represents a major hurdle to achieving higher engine
efficiency. Types of misfiring that limit efficiency include knocking combus-
tion among others. The pressure and shock waves caused by knocking
combustion in the combustion chamber, which emanate from compres-
sion-ignition centres, can be detected by different detection methods. In
addition to the structure-borne noise analysis, knocking combustion can
be detected using the cylinder pressure signal. The pressure waves are
superimposed onto the normal cylinder pressure signal in the form of
high-frequency oscillations (see Figure 1).

102
Figure 1: Cylinder pressure signal of a heavily knocking combustion cyc-
le [1]

These high-frequency oscillations are evaluated in greater detail in order

to study knocking combustion. Using high-pass or band-pass filtering, the
high-frequency signal component of knocking combustion is separated
and further analysed (see Figure 1). Based on this signal component, the
knock intensity and strength of the knocking combustion is characterised
and evaluated with respect to its engine-damaging effect. In addition,
using different analysis methods, the start of the knocking combustion
is determined with the filtered signal. With multiple pressure measuring
points, triangulation based on the filtered signal components is used to
approximate the location of knock in the combustion chamber [8; 6].

For analysis of the filtered signal components, the properties of the signal
filter and the location of the pressure sensor must be considered careful-
ly for the determination of the onset of knocking. The different transient
response of the types of filters leads to inaccuracies in the increase of
the filtered cylinder pressure signal. There are also differences in the ma-
ximum values due to the use of different types of filters [5]. The signal of
the cylinder pressure sensor describes the pressure ratios at the sensor

103
location in the combustion chamber and does not provide information for
the entire combustion chamber [6].

In addition to the analysis of the cylinder pressure signal, the location of

knock and onset of knocking can be determined using fibre optic measu-
rement equipment. The measured light intensity characteristics contain
both the primary flame radiation as well as secondary signal components.
The secondary flame radiation is superimposed with high-frequency light
intensity fluctuations onto the signal of the primary flame propagation, in a
similar manner as with the cylinder pressure signal (see Figure 2, left) [6].
These high-frequency signal components are induced by pressure waves
in the combustion chamber [3] and are thus comparable with the results
of the cylinder pressure indication. The primary flame radiation, however,
can be used for elapsed time analyses which means that the time and
location of the occurrence of the compression ignition centres can be ac-
curately approximated. The evaluation of the primary flame radiation (see
Figure 2, right) thus represents a criterion of the knocking combustion that
is independent of the pressure waves that occur.

Figure 2: Light intensity characteristics of the fibre optics

104
The fast calculation of the net heat release rate commonly used in com-
bustion process development can be used as an additional method of
analysis for knocking combustion. Here, the unfiltered cylinder pressure
signal is used to calculate the net heat release rate. This net heat release
rate can then be filtered and analysed [8; 2].

In this paper a method for determining the knock onset is presented for
the first time that analyses the net heat release rate of a low-pass filtered
cylinder pressure signal. The results of this procedure are compared with
the results of the analysis of light intensity characteristics of a fibre optic
measurement spark plug.

2. Determination of the filtered cylinder pressure

signal
The presented analytical method for determining the knock onset was de-
veloped using a four-stroke single-cylinder gasoline engine with variable
compression ratio.

Cylinder pressure indication is performed with a cylinder pressure sen-

sor that is positioned flush with the combustion chamber boundary in the
cylinder head. In addition, the corresponding pressure signals were recor-
ded as a function of the crank-angle using an intake and exhaust pressure
sensor. The rotation angle incremental encoder on the test unit had a step
size of 0.1° of the crank angle.

Filtering of the cylinder pressure was performed with a 10th order Butter-
worth filter. For low-pass filtering, a cut-off frequency of 1.9 kHz was cho-
sen (see Figure 1). This cut-off frequency represents an optimum across
the entire engine speed band.

3. Fast net heat transfer calculation

To depict the progression of the combustion and generate combusti-
on-specific characteristics from the cylinder pressure signal, the net heat
release rate can be calculated. While the gross heat release rate dQB
represents the total heat released by the combustion, the net heat release
rate dQH represents only the component that is effectively transferred to
the gas. This component, which is transferred to the gas, differs by the

105
component of the wall heat transfer dQWÜ [4] and the assumption of cons-
tant thermophysical properties (see equation (1)):

(1)

With

= gross heat release rate

= net heat release rate
= wall heat transfer

As the calculation of the wall heat transfer requires extensive boundary

information which can only be calculated exactly with a great deal of effort,
the cylinder pressure can be evaluated with sufficient accuracy and low
computing time, even on the test stand, by applying the fast heat transfer
calculation according to Hohenberg.

Figure 3: Thermodynamic state change for the heat transfer calculation

according to Hohenberg 1

This fast heat transfer calculation is based on the fact that each state ch-
ange in the cylinder is divided into two defined state changes for each step
of the crank angle. There is adiabatic compression or expansion from 1 to
2´ and isochoric heat supply from the state change from 2´ to 2 (see Fi-
1 Own representation based on Hohenberg, G.: Der Verbrennungsablauf - ein Weg zur
Beurteilung des motorischen Prozesses. In: Hans Peter Lenz (publisher): 4. Wiener Mo-
toren-Symposium. Düsseldorf: VDI-Verl (Technischen Universität Wien), p. 71–88, 1982,
[3], p.82

106
gure 3) [4]. The calculation is performed within a defined interval during a
combustion cycle, which covers the time period after closing of the intake
valve up to opening of the exhaust valve.

For the adiabatic state change from 1 to 2´ the following applies, where
the polytropic expansion coefficient n is assumed to be constant during
the entire calculation interval:

(2)

With

p = pressure in the cylinder

v = volume of the combustion chamber
n = polytropic expansion coefficient, const.
= crank angle

For the isochoric heat transfer from 2´ to 2 the following applies:

(3)

With

(4)

and AOtto = 0.1 = konst. 2

Q = heat
T = temperature in the cylinder
cv = specific heat capcity of the process gas
mcyl = mass of the cylinder load

2 According to Hohenberg, G.: Der Verbrennungsablauf - ein Weg zur Beurteilung des
motorischen Prozesses. In: Hans Peter Lenz (publisher): 4. Wiener Motoren-Symposi-
um. Düsseldorf: VDI-Verl (Technischen Universität Wien), p. 71–88, 1982, [3], p.78.

107
With the assumption of an ideal gas (mcyl, R = const.), the following relati-
onships apply for the state variables:

(5)

with

R = specific gas constant.

Thus using equations (2) to (5) the net heat release rate can be described
by:

(6)

with (7)

The real change of state deviates from the described two-stage state ch-
ange by the surface 1-2´-2-1 (see Figure 3). With a small calculation step
and thus high angle resolution of the incremental encoder (≤ 1° of the
crank angle), this deviation is reduced greatly, allowing it to be neglected
[4].

If the integral net heat release rate or net heat release is put into relation to
its maximum, the phases of the combustion can be determined accurately.
In this process, the 5% mass fraction burned can be used to determine
the start of combustion (Qh5%) the 50% mass fraction burned can be used
to determine the combustion centre (Qh50%) and the 90% mass fraction
burned can be used to determine the end of combustion (Qh90%) for each
combustion cycle.

In the context of this paper, the heat transfer calculations were performed
with the following parameters for the studied test engine:

108
Test engine: n = 1.32
Start = 300 °CA
End = 500 °CA

4. Determination of knock onset

The determination of knock onset is based on an analysis of the net heat
release rate dQh of the low-pass filtered cylinder pressure signal. The
cylinder pressure signal was recorded with a resolution of 0.1° CA and
then filtered with a low-pass filter (see Chapter 2). Similar to a non-kno-
cking cylinder pressure signal, which provides a good approximation for
the pressure ratios throughout the combustion chamber even when unfil-
tered, a low-pass filtered cylinder pressure signal of a knocking operating
point can provide information about the pressure conditions in the entire
combustion chamber. The signal components relevant for engine damage
are not included here (see Figure 1).

For the calculation of the fast net heat release rate with a step size of 0.1°
CA large oscillation amplitudes show up in the result due to the high-fre-
quency oscillations of the unfiltered cylinder pressure signal (see Figure 4,
left). The analysis of the net heat release of an unfiltered cylinder pressure
signal with respect to the end of combustion is thus very difficult. In contrast,
determination of the end of combustion can be easily performed using the
net heat release of a low-pass filtered pressure signal (see Figure 4, right).

Figure 4: Net heat release rate and net heat release of the low-filtered
and unfiltered pressure signal

109
The determination of knock onset is based on an analysis of the slope of
the net heat release rate curve at the start of combustion. As seen in Fi-
gure 5, the net heat release rates of a knocking (black) and non-knocking
combustion cycle (grey) differ in their slope in the signal rise. The diffe-
rence in the progression of a heavily knocking combustion cycle (black,
dashed) and non-knocking combustion cycle (grey) is clearly visible in
Figure 5. On the other hand, the differences between the combustion cy-
cles at the knock limit can only be discerned in the details (see Figure 5,
black or grey).

Figure 5: Net heat release rate of knocking and non-knocking combusti-

on cycles

As a criterion, the gradient of the net heat release rate was used (see Fi-
gure 6). The effective evaluation algorithm does not compare the current
value of the net heat release rate with the next value, but instead compa-
res it to the value after that. Once the signal value of the second interval
value after the current value is recognised as more than five percent hig-
her than the current value, the corresponding crank angle of the current
value is defined as the start of the knocking. At the same time, the distin-
ction is made between a knocking and a non-knocking combustion cycle.
The analysis of the net heat release rate is limited to an interval of the net

110
heat release rate which covers the crank-angle range from the 2% mass
fraction burned of the net heat release to the maximum of the net heat
release rate (see Figure 6).

Figure 6: Knock onset detection by evaluating the net heat release rate

5. Evaluation of the light intensity signals

Validation of the results of the evaluation of the net heat release rate was
performed using a fibre optic measuring spark plug in the test engine. This
spark plug is fitted with twelve integrated optical probes and could be used
in the engine without further modifications.

The optical probes can be divided into two groups. Four optical probes are
positioned vertically downard onto the piston. The eight additional optical
probes are positioned at a 60° angle to the spark plug axis (see Figure
7). The recorded light intensity decreases with the arched opening cone,
resulting in a usable, arched angle of the opening cone of approxima-
tely 6° for the four 0°-optical probes and approximately 10° for the eight
60°-probes.

111
The four vertical optical probes of the measuring spark plug can be used
to determine the flame front velocity, making it possible to extrapolate the
possible further flame propagation [9].

For determination of the knock location eight optical probes are used,
positioned at an angle of 60° to the axis of the spark plug. By using a
temporal comparison of the light intensity signals, a spatial distribution of
the knock location could be calculated. To do so, the optical probes were
combined into groups as follows. These groups differ in their spatial ar-
rangement, with one 0°-optical probe and two adjacent 60°-optical probes
evaluated together. If one of the 60°-optical probes detects a light signal
rise before it is detected by the associated 0°-optical probe, this may be
due to a misfire.

Figure 7: Opening cones of fibre optic spark plug in the combustion

chamber [1]

Detection of the light signal was performed by analysing the primary sig-
nal of the light intensity. Above a threshold value of 3 % of the maximum
light intensity of the individual light sensor, an event was detected. At the
same time, a check was performed as to whether the light intensity signal
was detected before ignition. To do so, instead of using the ignition spark
timing set in the electronic control unit, the exact ignition point was deter-
mined using an optical ignition current sensor on the spark plug. When it
was detected before the measured ignition point, there was a glow igni-
tion or pre-ignition. When the light intensity signal was detected after the
ignition point and before the arrival of detection of a light signal rise of the
0°-optical probe, this was registered as a knock event. The sensor that
detected the knock event indicated with its observation space the area
in the combustion chamber in which the knock event occurred. After 100
combustion cycles were analysed, the spatial knock distribution could be

112
determined in the form of a frequency distribution by evaluating each com-
bustion cycle for the investigated operating point.

Determination of the knock location was expanded to include the knock

onset. When a knock event was successfully detected, the associated
crank-angle value was recorded at which the threshold of the signal rise
was exceeded. A comparison of the results of the sensor groups made it
possible to identify the earliest start of knocking, which was then regarded
as the knock onset for the entire combustion chamber of a combustion
cycle. The statistical analysis of the thus determined timing of the knock
onset was used for the presentation of the results.

6. Results
With the help of optical measurement equipment, the results of the pre-
sented method were evaluated. For the measurements, 100 combustion
cycles were recorded simultaneously per measurement using the cylin-
der pressure indication and the optical measurement equipment. The
knock onset determined by the analysis of the net heat release rate of
the knocking combustion cycles was averaged over all knocking combus-
tion cycles. For comparison with the results of the optical measurement
equipment, the mean knocking onset KOmean,HR and the standard deviati-
on KOstd,HR of the net heat release rate criterion of knock were specified.

The evaluation of the optical measurement technology is presented in

the result using the spatial distribution of the knock location for the 100
combustion cycles recorded per measurement. This provides a percen-
tage distribution of the detected knock locations with respect to the re-
corded number of combustion cycles per measurement. The knock onset
determined while the knock location was detected was averaged over all
knocking combustion cycles. The mean knock onset KOmean,opt and the
standard deviation KOstd,opt of the optical evaluation are presented for
comparison (see Figure 8 to Figure 11).

For a weakly knocking operating point at the knock limit, the results of the
optical measurements indicate a good distribution of the knock locations
in the area of the intake valve. The thus determined knock onset shows
good agreement with the analysis of the net heat release rate. The de-
viation of the differently identified events of the knock onset is within the
standard deviation (see Figure 8).

113
Figure 8: Mean knock onset KOmean and spatial distribution of the knock
location based on 100 combustion cycles at the knock limit

For a strongly knocking operating point, there is a more even distribution

of the knock locations over the circumference of the combustion chamber,
although the area around the intake valve is denser. At a higher knock
intensity, the number of knocking combustion cycles per measurement
rises. For this reason, the comparison of the different evaluation methods
at a higher knock intensity demonstrates a lower deviation of knock onset
than at the knock limit (see Figure 9).

Figure 9: Mean knock onset KOmean and spatial distribution of the

knock location based on 100 combustion cycles of a weakly knocking
operating point

The measurement shown in Figure 9 was recorded at an operating point

with an ignition point of 17° CA before TDC, while the measurement in
Figure 8 had an ignition point of 20° CA before TDC. The shift of the igni-

114
tion point by 3° CA can also be seen in the results of knock onset analysis. The
timing of the knock onset also shifts along with the ignition point for both analy-
sis methods by approximately 3° CA (see Figure 8 and Figure 9).

Figure 10: Mean knock onset KOmean and spatial distribution of the knock
location based on 100 combustion cycles at an operating point with medium
knock intensity

For an operating point of average knock intensity, the results are similar to those
for low knock intensity. The formation of the knocking combustion is also con-
centrated around the area of the intake valve. A comparison of methods used to
determine the knock onset also demonstrates good agreement (see Figure 10).

Figure 11: Mean knock onset KOmean and spatial distribution of the knock
location based on 100 combustion cycles of a heavily knocking operating point

115
The agreement of the knock onset can also be observed for heavily kno-
cking operation (see Figure 11). For a larger number of knocking combus-
tion cycles, which could be analysed for strongly knocking operation, there
is similarly good agreement for the timing of the knock onset. The shift
of the knock onset for different ignition points for heavy knocking agrees
with the previously presented results of the measurements for low knock
intensity (see Figure 8 to Figure 11).

Figure 12: Linear correlation between the results of optical measurement

equipment and knock onset of the analysis of net heat release rate

The results of all analyses are shown in Figure 12. For the analysis of over
120 measurements, for different compression ratios, ignition points and
with fuels with different octane numbers and thus at different knock inten-
sities, there is a linear correlation between the knock onset determined
using the analysis of the net heat release rate and the knock onset deter-
mined using the optical measurement equipment. A linear regression was
calculated with a coefficient of determination R² of 0.8. The lower range
up to a crank angle of 354° CA demonstrates particularly good correlation.
In this area, due to the higher knock intensity, a larger number of knocking
combustion cycles could be evaluated per measurement. In the upper
range there is variation, as this area is associated with a weakly knocking
operating point. Even with low-octane fuel, it was not possible to create a
heavily knocking operating point in this area.

116
7. Summary
In this paper, a new approach is presented for the determination of the
knock onset. This method is based on the analysis of the net heat release
rate of a low-pass filtered cylinder pressure signal, which was recorded
with a resolution of 0.1° CA. For identification of knocking combustion and
knock onset, the gradient of the net heat release rate is evaluated. Using
experimental testing and a fibre optic measurement spark plug, it was
possible to determine both the knock location and knock onset using a
second independent method. The comparison of the results led to a linear
dependence of the results of optical measurements to the knock onset of
the analysis of the net heat release rate.

117
Literature
[1] Ohler, S.: Numerical and Experimental Determination of Combustion
Chamber Resonances. In: Sens, Marc, Baar, Roland (publisher): En-
gine processes. Renningen: expert-Verl. (Automotive engineering),
p. 260–272, 2013.
[2] Ando, H.; Takemura, J.: A Knock Anticipating Strategy Basing on the
Real-Time Combur. In: SAE Technical Paper Series (890882), 1989.
[3] Dahnz, C.; Palaveev, S. et al.: Faseroptische Untersuchungsmetho-
den zur Analyse von Verbrennungsanomalien in Ottomotoren. In: M.
Kratzsch und M. Günther (publisher): Ottomotorisches Klopfen - Ir-
reguläre Verbrennung. Renningen: Expert-Verl (Reihe Technik), p.
134–153, 2011.
[4] Hohenberg, G.: Der Verbrennungsablauf - ein Weg zur Beurteilung
des motorischen Prozesses. In: Hans Peter Lenz (publisher): 4. Wie-
ner Motoren-Symposium. Düsseldorf: VDI-Verl (Technischen Univer-
sität Wien), p. 71–88, 1982.
[5] Lindemann, M.; Funk, T.: Einfluss verschiedener Filterverfahren auf
die Generierung druckbasierter Klopfmerkmale. In: M. Fischer (pub-
lisher): Klopfregelung für Ottomotoren. Trends für Serienentwickler.
Renningen: expert-Verl., p. 139–147, 2004.
[6] Michels, K.; Hoffeyer, H. et al.: Analyse klopfender Verbrennungen
anhand von Druckverläufe. In: M. Kratzsch und M. Günther (publis-
her): Ottomotorisches Klopfen - Irreguläre Verbrennung. Renningen:
Expert-Verl (Reihe Technik), p. 28–41, 2011.
[7] Spicher, U.; Kneifel, A. et al.: Methoden zur Erkennung und Analyse
klopfender Verbrennung. In: M. Fischer (publisher): Klopfregelung für
Ottomotoren. Trends für Serienentwickler. Renningen: expert-Verl.,
p. 22–42, 2004.
[8] Worret, R.: Zylinderdruckbasierte Detektion und Simulation der Klopf-
grenze mit einem verbesserten thermodynamischen Ansatz. Univ.,
Karlsruhe, 2002.
[9] Wytrykus, F.; Geiser, F.: Combustion Control with Optical Fibre Fit-
ted Produktion Spark Plug. In: SAE Technical Paper Series (980139),
1998.

118
7  ylinder pressure-based knock detection
C
– challenges in cylinder pressure indicati-
on and application in a new engine-based
fuel test method / Zylinderdruckbasierte
Klopferkennung – Herausforderungen bei
der Indizierung und Anwendung in einem
neuen Kraftstoffprüfverfahren
Karl Huber, Johann Hauber

Abstract
Knocking represents an insurmountable barrier for the combustion in SI
engines, which limits the power density and efficiency of petrol engines.
Due to the transgression of the ignition limits applicable for the respective
mixture composition, the knocking combustion leads to a spontaneous
self-inflammation of the yet unburned residual air-fuel mixture, combined
with instant energy conversion, which is expressed in form of a respective
high pressure gradient and may cause major engine damages. Based on
the location of the knocking, the combustion shock wave subsequently
spreads throughout the combustion chamber with the velocity of sound
and is reflected on the walls of the combustion chamber. These shock wa-
ves, modulated to the medium combustion chamber pressure, particularly
appear in the rim areas of the cylinder in form of amplitude peaks, which
can be measured by way of combustion chamber pressure indication.

Typically, a signal separation of these knock shock waves can be achie-

ved with a pressure pattern analysis in form of high or band-pass filtering,
whereby the filter limits have to be synchronized with the respective com-
bustion chamber, depending on its dimensions. Consequently, this provi-
des the option to statically assess the knocking appearance during statio-
nary operation, whereby the determination of the cumulative frequency of
the knock wave amplitude has proved to be productive.

In the context of a research project over several years, the described

knock detection in a current serial engine as well as in single cylinder
aggregates used for fuel testing were applied according to the standards

119
ASTM D2699 & D2700, in order to develop a new test procedure to deter-
mine knock resistance.

Based on extensive knock investigations with various fuels it is shown why

a new fuel test procedure is sensible and necessary and how this should
be designed and implemented. It is furthermore shown that shock waves
typical for the knocking in a current serial motor can also occur in case
of verifiably knock-free engine operation, which significantly complicates
the detection and assessment of the knocking. By way of parallel shock
measurement via an additionally introduced measuring spark plug, the
assumption that the observed phenomenon was a oscillation excitation of
the volume in the quench zone between piston crown rim and the roof of
the combustion chamber as a result of the high burning rate in case of this
engine was confirmed.

Kurzfassung
Das Klopfen stellt für die ottomotorische Verbrennung eine nicht zu über-
windende Grenze dar, welche Leistungsdichte und Wirkungsgrade von
fremdgezündeten Motoren limitiert. Bei der klopfenden Verbrennung
kommt es aufgrund einer Überschreitung der für die jeweilige Gemisch-
zusammensetzung gültigen Zündgrenzen zu einer spontanen Selbstent-
flammung des noch unverbrannten Gemischrestes, verbunden mit schlag-
artigem Energieumsatz, welcher sich in Form eines entsprechend hohen
Druckgradienten äußert und kapitale Motorschäden verursachen kann.
Ausgehend vom Klopfort breitet sich sodann die Verbrennungsdruckwelle
mit Schallgeschwindigkeit im Brennraum aus und wird an den Brennraum-
wänden reflektiert. Besonders in den Randbereichen des Zylinders treten
diese auf den mittleren Brennraumdruck aufmodulierten Druckschwingun-
gen in Form einer Amplitudenüberhöhung in Erscheinung, welche durch
Brennraumdruckindizierung gemessen werden können.
Typischerweise kann durch Signalanalyse der gemessenen Druckver-
läufe in Form von Hoch- oder Bandpassfilterung eine Signalabtrennung
dieser Klopfdruckschwingungen erreicht werden, wobei die Filtergrenzen
auf den jeweiligen Brennraum in Abhängigkeit von dessen Abmessungen
abzustimmen sind. In Folge besteht die Möglichkeit bei stationärem Be-
trieb die Klopferscheinungen statistisch auszuwerten, wobei sich die Be-
stimmung der Summenhäufigkeit der Klopfdruckamplitude als zielführend
herausgestellt hat.

120
Im Rahmen eines mehrjährigen Forschungsprojektes wurde die beschrie-
bene Klopferkennung sowohl an einem aktuellen Serienmotor, als auch
an den für die Kraftstoffprüfung nach den Normen ASTM D2699 & D2700
verwendeten Einzylinderaggregaten eingesetzt, um ein neues Prüfverfah-
ren zur Ermittlung der Klopffestigkeit zu entwickeln.
Auf der Basis umfangreicher Klopfuntersuchungen mit unterschiedlichen
Kraftstoffen wird gezeigt, warum ein neues Kraftstoffprüfverfahren sinnvoll
und notwendig ist und wie dieses aussehen und umgesetzt werden kann.
Darüber hinaus wird gezeigt, dass an einem aktuellen Serienmotor für das
Klopfen typische Druckschwingungen auch bei nachweislich klopffreiem
Motorbetrieb auftreten können, was die Klopferkennung und Bewertung
maßgeblich erschwert. Durch parallele Druckmessung über eine zusätz-
lich eingebrachte Messzündkerze bestätigte sich die Annahme, dass es
sich bei dem beobachteten Phänomen um eine Schwingungsanregung
des Volumens im Quench-Bereich zwischen Kolbenkronenrand und dem
Brennraumdach infolge der bei diesem Motor vorhandenen hohen Brenn-
geschwindigkeit handelt.

1. Introduction
In view of finiteness of fossil energy sources keywords like economize,
substitute and replace became more and more important. The limited
supply of fossil fuels leads to increasing diversification of drive types and
energy sources in the area of motor vehicles. In particular for liquid hy-
drocarbons, it must be assumed that the offer of different fuels with bioge-
nic admixtures already implemented will continue to increase worldwide.
The importance of fuel analytics is growing at the same ratio as fuel quality
is a decisive factor for an efficient and low-emission combustion.
The knock resistance of the fuel is a highly important quality feature in
gasoline engines as it directly influences the achievable efficiency via
compression ratio as early as in constructional design and also by the
ignition angle application of the knock control during engine operation.
Engine knock in petrol engines is understood as an abnormal combustion
phenomenon with occurring pressure oscillations, which are likely to cau-
se thermal and mechanical engine damage.
Gasoline´s knock resistance has been measured and described for more
than 80 years by octane numbers. They are determined in a so-called Co-
operative Fuel Research engine according to the standards ASTM D2699
(RON) and D2700 (MON) by bracketing with standardized reference fuels.

121
However, what significance can be reached with these fuel quality num-
bers for today´s engine technology, especially since the test engine´s mix-
ture formation is realized by a carburettor while engine speed is compa-
ratively low and last but not least, using a completely analogous signal
processing?

2. Single Cylinder Measurement

2.1. State of the Art

The octane number is determined at a CFR engine developed by Wau-
kesha. It is a single-cylinder test engine with two-valve technology and
lateral spark plug position. The engine is mainly characterised by the ad-
justable compression ratio. In this engine, the mixture is formed externally
by a carburettor. It also includes air preheating and a mixture heating for
the MON procedure. In contrast to MON-condition, RON is measured at
lower engine speed of 600 min-1 only using intake air preheating (see also
table 1 for further details).

122
 Research Octane Motor Octane
Number Number
RON MON
CFR F-1-Method CFR F-2-Method
Standard ASTM D 2699 ASTM D 2700

Engine speed [min-1] 600 900

[°CA b.
Ignition angle 13 f(CR)
TDCF]
Bore [inch] 3.250 3.250

Stroke [inch] 4.500 4.500

Mixture formation Carburettor Carburettor

not explicitely not explicitely
defined defined
Air/fuel ratio
„max. knock inten- „max. knock inten-
sity“ sity“
Intake air tempe-
[°F] f(pamb) 100
rature
Mixture tempera-
[°F] not specified 300
ture
Cooling type and boiling cooling, boiling cooling,
coolant tempera- [°F]
ture 212 212

Table 1: Test conditions for RON and MON determination according to

[1, 2]

The octane number of a fuel is standard-compliantly determined by bra-

cketing with reference fuels. Proper function of the test engine must be
inspected with standardized fuels with, among others, the intake air tem-
perature to be set. To determine the octane number of an unknown fuel,
the user must set the air ratio like that the knock meter reaches the ma-
ximum. Two fuel mixtures with known octane numbers are needed for
bracketing in. For octane numbers below 100, two-component mixtures of
the more knock-resistant eponymous isooctane (2,2,4-trimethylpentane)
and the less knock-resistant n-heptane are used. One of these mixtures

123
must show a higher, the other a lower deflection of the knock meter, to
linearly interpolate the octane number of the fuel sample between them.
However, one problem is determination of RON-values above 100, since
this requires reference fuels with lead addition (isooctane with additions of
tetraethyl lead TEL) that pose a considerable health risk for the operating
staff and therefore are only available with limitations in Europe. In the
most laboratories toluene base fuels (TSF, Toluene Standard Fuels) are
used for bracketing in fuels above 100 RON. These fuels should be used
for calibration purpose only.

2.2. Functionality of the Standardized Test Method

The knock meter reading is the most important value for determination of
the RON and MON. Therefore the following deals with signal recognition
with the pickup sensor, signal analysis in the detonation meter and kno-
cking strength at the knock meter in detail.

Figure 1: Transmission behavior of the detonation meter

The original versions of the test engines constitute a measuring chain im-
plemented with analogue technology only. The “XCP panels” marketed by
Waukesha since 2010 to replace the detonation and knock meter merely
map the behaviour described in the standard by digital signal recording,
data processing and control according to [1, 2, 3]. Therefore the panel
could be introduced without essential standard change.

124
The pickup which is connected with the combustion chamber consists
mainly of a magnetostrictrive rod within a coil. Deformation of the rod due
to the combustion chamber pressure induces a voltage in the coil. This is
the only measuring value that is included in the detection logic of the deto-
nation meter. The detonation meter is an analogue circuit that has the task
of processing the pickup signal and generating a voltage, which is a mea-
surement value for knocking strength and displayed by the knock meter.
To determine the transmission behaviour of the detonation meter experi-
mentally, synthetic pickup replacement signals of a defined form, frequen-
cy and amplitude were generated by a signal generator and submitted to
the detonation meter as a signal input. Then the knock meter value was
read. Like figure 1 shows, the pass band of the detonation meter ends at
2.85 kHz and thus below the frequency range of 3.5 to 15 kHz, in which
knocking pressure oscillations are usually found.
The complete circuit chart can be used to determine the transmission be-
haviour of the detonation meter with a cut-off frequency (amplitude dam-
pening by 3 dB) of approx. 2.2 kHz by a simulation calculation with ng-
spice. It corresponds with the measurements in figure 1 well.
For further analysis, 120 working cycles are evaluated for two different
fuels and the respective maximum of the pickup signal is compared to the
voltage drop at the relevant capacitator.

Figure 2: Correlation between pickup signal height, relevant capacitator

voltage and knocking pressure amplitudes

125
Figure 2, left, shows the result confirming the expected linear correlation
between the pickup signal height and relevant capacitator voltage.
Simultaneously to the pickup sensor, the pressure curves were also re-
corded with a piezoelectric pressure sensor that was installed close to the
pickup. To avoid pressure oscillation due to the sensor position, the setup
is adapted to be flush with the combustion chamber [4].
The pressure curves recorded with this sensor are used to determine
the maximum knocking pressure amplitudes by band pass filtering (3.5 -
15 kHz). They are compared to the voltage at the relevant capacitator for
these 120 working cycles respectively in the right-hand half of figure 2.
It can be seen easily, that the correlation between relevant capacitator
voltage and maximum knocking pressure amplitude for both fuels is com-
paratively bad. It also shows that the premium fuel (MON 85.3) has lower
maximum knocking pressure amplitude in spite of the lower MON.

The fact that the knocking pressure amplitude is not assessed by the deto-
nation meter is not surprising in the light of the low pass filter with a cut-off
frequency of approx. 2.2 kHz that was founded out by the circuit analysis
with ngspice.
These insights lead to the conclusion that the octane number is not a relia-
ble measure for the knock-resistance of fuels, in particular since knocking
intensity is not assessed based on knocking pressure oscillations, as it is
usual today.

2.3. Variation of Air/Fuel Ratio at RON-comparable con-

dition
As described in chapter 2.2, in order to evaluate the occuring knock in-
tensity in the test engine, the cylinder pressure was measured for 300
working cycles. The high frequency knock pressure oscilliations could be
extracted from the recorded cylinder pressure data by band- or high-pass
filtering subsequently the maximum knock amplitude per working cycle
being statistically classified [6].

126
 Figure 3: Relative cumulative frequency of knock pressure amplitudes
with 5 % and 15 % residual quantiles

An exemplary cumulative frequency distribution of the knock pressure

amplitudes is shown in figure 3. The statistically characterization of knock
intensity by cumulative frequency focusing on the part of the graph with 5
to 15 % residual quantile has been carried out to be an excellent measure
for knock strength and was chosen as the relevant criterion for the follo-
wing investigations described.

According to the valid standards, the combustion air/fuel ratio must be set
for each fuel, as noted initially, so that maximum knocking occurs.
For this, the operator must adjust the carburettor (fuel level). This requires
lots of experience and greatly influences the determined octane number,
as the charts in figure 4 show for different fuels.
The measurements were performed under RON-like conditions while the
combustion air/fuel ratio was varied. The knock meter reading and cu-
mulative frequency (95 %) of the maximum knocking pressure amplitude
were applied across the air/fuel ratio. For comparison, the fuels were set
for similar knocking intensity at a stoichiometric air/fuel ratio by changing
the compression ratio. Spread and meter reading at the detonation meter
were kept constant for all fuels, so that the knock meter readings as well
as the knocking pressure amplitudes are comparable to each other.

127
 Figure 4: Connection between knocking intensity and air/fuel ratio

The maximum positions of the curves mainly deviate from each other in
case of the paraffinic reference fuel isooctane, while the other fuels show
good correspondence.
The relation of the knock amplitudes and the knock meter reading is par-
ticularly important. For the primary reference fuel isooctane, comparably
large knocking pressure amplitudes cause low knock meter deflections,
while all other fuels show the opposite. One reason for this is in particu-
lar the lower burning rate of the reference fuels [7] that leads to a lower
evaluation by the detonation meter (see also chapter 2.2).
Regarding the usage in the full engine the comparison of isooctane accor-
ding to definition with RON 100 and premium plus at RON 99.5 is suitable.
Based on the stoichiometric air/fuel ratio, a richer premium plus fuel leads
to a reduction of the knocking pressure amplitudes (desired behaviour at
full load enrichment), while a richer mixture of isooctane leads to much
stronger knocking.

3. Examinations at the Full Engine

3.1. Description of Measurement Program

The knock-resistance investigations according to the RON- and MON-pro-
cedures at the CFR-test engine were accompanied by measurements at

128
a 2.0 l 4-cylinder serial engine. The engine is characterised by a state of
the art combustion procedure with direct injection and turbo charging. It
reaches a maximum mean effective pressure of 22.2 bar at full load.

This engine was operated at pre-defined characteristic points of the engi-

ne performance graph with a focus on high and full load operating points,
as engine knock can be expected there. Additionally, an engine operation
point at 900  min1 and 10  bar mean effective pressure was chosen for
good comparability with low engine speeds at the CFR engine at MON
respectively RON conditions (see figure 5).

These measuring program was enforced with a selection of 8  fuels (for

details: table 2) of different compositions.

Figure 5: Engine map with analyzed operating points

The range of fuels used comprises commercial premium E5 and premium

plus as well as E85 and two ethanol splash blends. One methanol and
isobutanol fuel each with approx. 15 % vol. of alternative alcohols and a
two-component mixture of 95  %  vol. isooctane and 5  %  vol. n-heptane
complete the range.

129
 isooc-
tane / pre- pre-
n-hep- mium mium E20 M15 BuOH15 E40 E85
tane plus
E5
95 / 5

RON [-] 94.9 96.4 99.5 100.6 101.3 101.9 104.7 106.9

MON [-] 94.9 85.4 88.6 87.6 88.5 88.9 89.6 91.1

Sensitivity [-] 0 11.0 10.9 13.0 12.8 13.0 15.1 15.8

Vapor pres-
[kPa] 14.1 56.4 50.5 52.1 71.4 52.8 52.5 51.7
sure
Density (at
[kg/m3] 695.5 749.9 755.9 760.5 762.8 761.7 767.2 775.7
15 °C)

Olefins [% vol.] 0 10.8 5.6 9.0 7.1 2.0 6.2 3.6

Aro-
[% vol.] 0 35 36 30.6 33.6 27.3 19.6 4.9
matics
composition

Methanol [% vol.] 0 0 0 0 15.1 0 0 0

Ethanol [% vol.] 0 4.9 0.5 19.5 0 0 44.3 76.9

i-Butanol [% vol.] 0 0 0 0 0 16.9 0 0

ETBE [% vol.] 0 2.8 12.9 2.3 0 0 1.7 0.8

Table 2: Overview of the fuels used

3.2. Results of Cylinder Pressure based Knock Detection

The ignition angle, at which the engine knocking of a defined intensity
appears, was consulted as measure for the fuel quality in the full engine.
For this purpose, the ignition angle was varied at the different operation
points for each fuel. A desired advancing adjustment of the ignition angle
is limited by the knock threshold to achieve the best possible efficiency.
In order to evaluate the knock limit, the maximum knock pressure amplitu-
de per working cycle determined by cylinder pressure measurement was
consulted and classified. In addition, this result was compared with the
synchronously recorded value of the knock control depth, an ECU-internal
value for occurring knock intensity.

130
3.2.1. Pressure Oscillations at non-knocking Engine
Operation
The knock limit was determined in the engine operation points described
under 3.1 by way of gradual early adjustment of the ignition angle at cons-
tant engine output. The respective analysis of the indexed measurement
data by way of band and/or high-pass filter with subsequent threshold
value observation showed that this criterion led to the triggering of the
knock detection also in case of a verifiably knock-free operation at engine
speeds above 2000 min1.
Due to their amplitude-faithfulness throughout the entire frequency range,
filters of the type Butterworth were used for the separation of the knock
pressure oscillations from the pressure pattern.
Graphic 6 shows typical individual work cycles at an operating point of
6000 min1 and full load with E40 fuel.
It shows the pressure pattern of cylinder 2 (indicated twice) and 3. Pres-
sure oscillations, which are not detectable on the spark plug sensor, are
clearly modulated at the measurement channels positioned at the cylinder
rim.

Figure 6: Pressure pattern in cylinder 2 and 3: Recorded with a cooled

M8, an uncooled M5 and a spark plug sensor; fuel: E40, engine speed:
6000 min1

131
The question arises whether these pressure oscillations, which are loca-
ted in the knock-relevant frequency range, are in fact based on knocking
combustion.
Such pressure oscillation can also be caused due to the sensor´s ins-
tallation position as a result of acoustic oscillation [4], if the sensor has
been installed recessed or via a shot channel. The sensor position in this
engine was flush with the combustion chamber in the case of the default
M8-sensor; in comparison, the M5-Sensor was installed front-sealed abo-
ve an adapter. The resulting shot channel was 3 mm long and also had a
diameter of 3 mm.
The position of the cylinder pressure sensor was at the outer rim of the
combustion chamber in the sector between intake and exhaust valve. The
spark plug of the test engine is located centrally in the four-valve combus-
tion chamber.
From own experience and [4] it is known, that the central measuring point
position for the detection of slight knocking is less suitable, which is why
an actually knocking work cycle cannot be excluded in this case. In cont-
rast, the two sensors, located radial in the edge of the bore, indicate pres-
sure oscillations, which could quite possibly be based on knocking. This
necessitated further investigations to obtain a conclusive assessment.
In image group 7, high-pass filtered cylinder pressure pattern of respecti-
vely 240 working cycles are displayed above each other. The additionally
shown vertical lines reflect the positions for the maximum pressure gra-
dient, the centre of combustion as well as the peak pressure.
The following findings can be derived from a comparison of these illust-
rations:
A significant increase of the cylinder pressure oscillations in the sector of
the centre of combustion is detectable at both operating points with the
displayed fuels.
If, in fact, the self-ignition of the remaining unburned mixture were to occur
at this moment, the maximum knock pressure amplitude to be measured
would be many times higher.

132
Figure 7: High pass filtered cylinder pressure pattern at two different en-
gine speeds and fuels (240 working cycles, filter border: 8 kHz) [12]

The illustration of the filtered pressure pattern of one exemplary cylinder

at 6000 min1 with E85 (fig. 7, right lower corner) illustrates strong pres-
sure oscillations of approx. up to 2 bar amplitude in the relevant knocking
frequency range, which are most clearly recognizable in the sector of the
peak pressure.
The characteristic of the filtered cylinder pressure pattern (the supposed
“knock signal”) with its commencement of the amplitude peaks arising
around the position of the maximum pressure increase allows the con-
clusion that the pressure oscillations in the combustion chamber at the
position of the sensor bore are stimulated by the pressure increase due to
the fast combustion process occurring in the series engine used.
If one compares the two illustrations at 1750 min-1, it is clearly evident that
ignition is possible quite earlier in the case of the more knock-resistant
E85 fuel.
Upon further observation of the full load operation point in case of
1750  min1 and premium fuel, some individual, approx. double-in-size
amplitudes occur after the peak pressure position rather than immediately
prior, which is due to knocking only in this instance. A comparison with the

133
recorded knock control depth from the engine control unit confirms this
result.
The following figure exemplary illustrates a frequency analysis of the cylin-
der pressure at 6000 min1 full load with E85 fuel. Even though the engine
was operated knock-free in this instance, amplitude peaks are evident in
the relevant frequency range at approx. 7 kHz, which confirms the descri-
bed state of facts.

Figure 8: Fast fourier transformation of cylinder pressure pattern [12]

The fact that the observed pressure oscillations cannot be separated from
real knock pressure amplitudes on a measurement and physical level,
may lead to errors in the detection and quantitative assessment of kno-
cking combustion.
In order to circumvent this problem, operating points at low revolutions
were predominately consulted in the context of this analysis, where this
effect did not appear quite as significantly and is positioned below the
threshold for knock detection.

3.2.2. Classification of Fuel´s Knock Resistance by Full

Engine
To determine the knocking limit the ignition angle was varied actively at
selected full- and high-load points. Cylinder pressure was measured at all
cylinders with subsequent knocking intensity. The knocking intensity [5]

134
and maximum knocking pressure amplitude per working cycle and asso-
ciated classification [6] were evaluated.
Figure 9 shows the fuels according to knock-resistance (along x-axis) re-
sulting from the earliest possible crank angle for the centre of combustion
as it could be found in full engine.

Entering the MON- and RON-values of the test engine above the centre
of combustion, the evaluation of fuel quality through the full engine (cen-
tre of combustion) corresponds with increasing RON-values, while the
MON-values do not show any clear trend.

Note that the two fuels E40 and E85 can be operated nearly or entirely
knock-free to the point of best performance so that knock-resistance
evaluation was not possible for them (see x-axis label).

Figure 9: Relation between the reachable centres of combustion that can

be found
in the full engine and the respective RON and MON values of the fuels;
n = 1750 min-1, BMEP = 20 bar, charge air temp. = 30 °C

135
4. Proposal of a new Test Procedure
For the results described in the following, ignition angle, intake air tempe-
rature and type of mixture composition were changed gradually to exami-
ne the influences on the assessment of knock resistance separately.

Figure 10: Determination of a fuel´s maximum compression ratio

The mixture heating was not considered, because the MON-values sho-
wed a very bad correlation to the full engine measurements. Therefore,
the test conditions of the CFR-F1-method as shown in table 1 were used
as a starting basis and the following modifications were performed:

• Stoichiometric air-fuel mixture.

• Constant intake air temperature of 50 °C.

• Ignition angle adjustment at centre of combustion with best efficiency

[8].

136
 Figure 11: Modified CFR test engine [11]

The compression ratio was finally increased up to consistent knock

strength (e.g. 1.2 bar), as shown in figure 10. Knock strength is defined
by the 95 %-cumulative frequency value of the maximum knocking pres-
sure amplitude [6, 10, 11]. The compression ratio was calculated via on-
line measurement of the relative movement of cylinder head and crank
case by magnetostrictive position sensor, which is illustrated in figure 11
in mounted position.

The respective values of the compression ratio are compared to the

RON-values of the fuels in figure 12 (as dark squares) [11].
To compensate for barometric pressure changes due to meteorological
causes or the height above sea level of the setup site, the intake manifold
pressure was furthermore throttled moderately to a consistent 930 mbar.
The load point lowering led to an increase of compression by about 0.4 to
0.6 units (light grey squares).

Finally, the carburettor (including venturi nozzle) was replaced by a port

fuel injection. This made it possible to increase precision of the mixture
formation per working cycle. Due to the more efficient utilisation of the

137
enthalpy of evaporation by intake-stroke-synchronous injection, another
increase of compression between approx. 0.3 and 0.9 units was possible,
which essentially contributes to the knock resistance of a fuel in the full
engine as well (light grey triangles) [9].
The examined fuels comprise the fuels table 2 and were supplemented by
additional fuel qualities in a wide RON range. Mixtures of toluene, isoocta-
ne and n-heptane, as well as E100, were mainly used.

Figure 12: Comparison of RON and possible compression ratio at the

knock limit in different operating points, according to [11]

The enthalpy of evaporation is decisive for cylinder inside charge cooling

effects and the reason for the clear difference between the measurements
with carburettor and PFI-operation (throttled), showing a comparatively
high enthalpy of evaporation with E85 and a much lower one with isooc-
tane.
The dark grey solid line marks the compression that, according to [1],
leads to the standard knock strength in the test engine at a specific RON
at an absolute pressure of 970 mbar (corresponds to the standard pressu-
re at the measuring location).
When observing this nominal compression line, we notice that there is
no linear correlation between the octane number and compression. The

138
curve is degressive in the range up to approx. RON 102 and slightly pro-
gressive above it.
A comparison to our own measurements (dark grey squares) now shows
that the knock-resistance of the fuels above the curve, such as isooctane
and 95-octane, are evaluated as too good according to the RON method.
If the points are below, such as E85, the knock-resistance of the fuels is
higher than the nominal RON value.
The light grey triangles show that the new test procedure has a very wide
measuring range that far exceeds the range of isooctane without requiring
associated reference fuels (light grey background, separated by the das-
hed line). It covers the available liquid fuels.

5. Summary and Outlook

During the analysis of the octane number determined at the CFR engine
as the parameter for the knock resistance of a fuel, some deficits of the
standard test procedures (RON and MON) came to light.
The assessment of engine knocking is not based on knock-pressure
amplitudes, as is common in today’s engine technology. The air/fuel ratio
to be adjusted is different depending on the fuel; thus, the transferability
of the results to engines with a lambda1 concept is only limitedly possible.
The calibration of the correct air/fuel ratio furthermore depends on the
abilities of the operator and directly impacts the test result.
The standardized determination of octane numbers above 100 requires
leaded fuels, which harbor a significant health risk for the operator.
Fuels with high vapor pressure result in high cycle to cycle variations due
to vapor bubble formation; this also occurs in case of fuels with a high
enthalpy of evaporation due to wall film formation.
Concerning the mixture heating at MON determination bad correlation
with the results of the full engine evaluations were found thus MON is no
longer applicable for current engines with PFI or DI.

The framework conditions for a new test procedure presented above can
be defined with the developed knowledge. It was particularly important to
maintain the proven CFR-basic engine, which is robust and available in
the fuel test labs around the world. The compression adjustment option in
operation and the benefits of simple and highly precise engine cooling by
thermo siphon effect could be used. Use of direct injection was delibera-
tely dispensed with to avoid concurrent problems with mixture formation
and combustion of many different fuels.

139
The use of reference fuels, especially purely paraffinic ones, for bracke-
ting should be dispensed with completely. Retrofitting of present systems
to the new procedure was considered.
The new test procedure proposed is characterised by the low RON-com-
parable engine speed of 600 min-1 for high resolution in knock intensity
as well as a contemporary electronic engine management. The use of an
ECU in combination with online- pressure pattern analysis enables fea-
tures like stoichiometrically controlled air/fuel ratio realized by port fuel
injection, ignition angle adjustment to point of best performance and knock
intensity regulation by setting the targeted compression ratio as a measu-
re for fuel´s knock resistance.
To ensure comparable engine operation at different test laboratories in a
wide range of ambient pressure and temperatures without impact on the
result the test engine is equipped with an intake air preheating system
(50 °C) slightly throttled to 930 mbar intake manifold pressure.
To verify perfect engine condition and ensure its proper functioning, selec-
ted reference fuels containing aromatics besides paraffins will be used.

The new test procedure is to be submitted as a suggestion to the stan-

dards committee for liquid fuels. In addition, the required round robin tests
will be performed to secure the results and prove it´s repeatability and
comparability.

140
Reference
[1] ASTM D 2699-12: Standard Test Method for Research Octane Num-
ber of Spark-Ignition Engine Fuel. 2012
[2] ASTM D 2700-12: Standard Test Method for Motor Octane Number of
Spark-Ignition Engine Fuel. 2012
[3] Waukesha gas engines: CFR F1/F2 XCP Digital Octane Panel, and
Octane Analyzer operation, maintenance parts catalog. Form 947-2,
2nd edition, 2011
[4] Bertola, A.; Spicher, U.; et al.: Druckindizierung bei klopfender Ver-
brennung. Kistler special print 920-349d-11.06, 2006
[5] Spicher, U.; Kollmeier, H.-P.: Detection of Flame Propagation During
Knocking Combustion by Fiber Diagnostics. SAE 861532, 1986
[6] Fischer, M; et al.: Klopferkennung im Ottomotor – Neue Tools und
Methoden in der Serienentwicklung. In: MTZ 3-2003, p. 186 et seq.
[7] Nehse, M.: Automatische Erstellung von detaillierten Reaktionsme-
chanismen zur Modellierung der Selbstzündung und laminarer Vor-
mischflammen von gasförmigen Kohlenwasserstoff-Mischungen.
Dissertation, Universität Heidelberg, 2001
[8] Bargende, M.: Schwerpunkt-Kriterium und automatische Klingeler-
kennung – Bausteine zur automatischen Kennfeldoptimierung bei
Ottomotoren. In: MTZ 10-1995, p. 632 et seq.
[9] Stein, R. A.; et al.: Effect of Heat Vaporization, Chemical Octane, and
Sensitivity on Knock Limit for Ethanol-Gasoline Blends. SAE 2012-
01-1277, 2012
[10] Huber, K.; Hauber, J.; Raba, A.: Entwicklungsschritte zu einem neuen
Prüfverfahren zur Bestimmung der Klopffestigkeit von Ottokraftstof-
fen. Ludwigsburg, 2013, published in:
[11] Leipertz, A.: Engine Combustion Process – Current Problems and
Modern Techniques, XI. Congress, ISBN 978-3-931901-87-5
[12] Huber, K; Hauber, J.: New Test Procedure to determine Fuel´s Knck
Resistance. In MTZ 7/8-2013, p. 62 et seq.
[13] Hauber, J; Huber, K.: Final report of the research project regarding
engine-based fuel numbers; non-published report, Ingolstadt, 2012

141
8 Vorentflammung: Maßnahmen in der
Entwicklung und Kalibrierung hoch
­aufgeladener Motoren
Irregular combustion: development and
calibration of highly boosted SI engines
Thomas Dobes, Alois Hirsch, Paul Kapus, Mario Nin-
aus, Martin Ogris, Harald Philipp, Ernst Winklhofer

Abstract
Boosted SI engines have most attractive torque characteristics, the risk of
pre-ignition, however, with related irregular combustion events imposes
strict limitations to engine operation and vehicle driveability. Root causes
are found in a combination of components’ and media temperatures, che-
mical reactivity of in-cylinder charge and residuals and time available to
establish self-sustained chemical reactions.

This imposes twofold tasks for engine and vehicle development:

• In combustion system development: to find measures to reduce and

limit the risk of pre-ignition
• In vehicle application: to identify pre-ignition events and activate cont-
rols in order to inhibit escalation of riskful irregular combustion events.
Such controls however, must not interfere with vehicle driveability.
We use examples derived from development of boosted SI engines to di-
scuss the systematic analysis of PI combustion events. Specific procedu-
res and methods are applied to the identification of root causes governing
irregular ignition and combustion in normal high load engine testing. This
results in advice to focus development efforts on components’ cooling,
gas exchange and mixture formation or on measures affecting lube oil
and lubrication.

In vehicles, functionalities of engine diagnostics are used to identify with

utmost reliability the appearance of PI events. Such identification is fol-
lowed by adapting fuel injection parameters to suppress potential “run-
away” PI combustion modes.

142
Kurzfassung
Aufgeladene Motoren erzielen hervorragende Drehmomentverläufe, un-
terliegen aber dem Risiko von Vorentflammungen und damit verbundenen
irregulären Verbrennungszuständen. Deren Ursachen liegen in der Kom-
bination von Bauteil- und Medientemperaturen, der Reaktivität der im
Brennraum vorhandenen Medien und der verfügbaren Reaktionszeiten.

Für die Motor- und Fahrzeugentwicklung ergeben sich daraus zweierlei

Aufgaben:

• durch geeignete Maßnahmen in der Brennverfahrensentwicklung da-

für zu sorgen, das Risiko für Vorentflammung in akzeptablen Grenzen
zu halten, und
• in der Fahrzeugapplikation Algorithmen zu verwenden, die den Motor
bei Auftreten von Vorentflammung in den regulären Betrieb zurückfüh-
ren ohne daß der Fahrzeuglenker durch die Motorreaktion irritiert wird.
Der Beitrag zeigt anhand von Beispielen aus Entwicklungsaufgaben an
aufgeladenen Motoren die Systematik der Versuchsführung und die Me-
thoden der Verbrennungsanalyse, aus denen die jeweils vorliegenden Ur-
sachen für irreguläre Zünd- und Verbrennungsvorgänge erkannt werden.
Maßnahmen zur Abminderung der Risiken setzen an diese Ursachenana-
lyse an. Sie betreffen je nach Analyseergebnis Eingriffe in die Bauteil-
kühlung, Ladungswechsel und Gemischbildung, und den Ölkreislauf des
Motors.

Für den Motorschutz im Serienfahrzeug werden Funktionen der Motor-

diagnose dazu verwendet, das Auftreten von Vorentflammung mit hoher
Sicherheit zu erkennen. Als Reaktion darauf wird über die Motorsteue-
rung durch geeignete Einspritzparameter in den nachfolgenden Zyklen
eine Eskalation des irregulären Motorbetriebs unterdrückt.

1. Regular and Irregular Combustion

In regular combustion mode, ignition of in-cylinder charge is initiated by
the spark discharge plasma, its consumption by turbulent propagation of
the flame front. Heat transfer from combustion chamber surfaces together
with compression heating, may, however, initiate uncontrolled self-igni-
tion. This appears subsequent to regular ignition in areas of unburned
endgas (the normal “spark knock” events, see the schematic of Fig. 1). As

143
endgas ignition is under influence of combustion pressure heating, retar-
ded ignition is most effective to reduce such spark knock.

Under extreme thermal conditions, self ignition may occur as a result of

thermochemical reactions prior to the well timed spark ignition event.
Such pre-ignition can result in very high cylinder pressures and result in
mechanical damage of combustion chamber components

There is usually a combination of mechanisms leading up to pre-ignition:

• In highly boosted engines, two or even three times as much charge

is burned as compared to a naturally aspirated engine. Combustion
chamber surfaces, consequently, are under influence of significantly
higher heat flux.
• More fuel injected can result in more deposits formed out of fuel, lube
oil or combustion residuals.
• Especially at low engine speed, sufficient time might be available to
initiate thermochemical reactions prior to spark ignition.
• Fluctuation of residual gas can contribute to pre-ignition
• Free moving combustion residuals might not pass out through the
exhaust valves, but stay in-cylinder and ignite fresh charge
Such root causes result in typical classes of irregular combustion, see Fig.
1, with cylinder pressure traces defining pre-ignition and “mega knock” as
opposed to the usual spark knock. A simple Arrhenius and time influence
relationship is given in Fig. 2 to highlight the functional interdependence of
chemical species properties, temperature and time scales.

Figure 1: Schematic of regular and irregular combustion. „Megaknock“ is

a consequence of uncontrollable pre-ignition

144
Figure 2: Self-ignition is under influence of reactants and of engine ope-
ration. Ignition delay (tau) according to an Arrhenius relation and ignition
activation periods as per Livengood and Wu (see [1]).

2. Development Tasks
Reducing the risk of pre-ignition, first of all, requires efforts in the phase
of basic combustion system development. Relevant root causes must be
identified in order to guide improvement in heat transfer, gas exchange or
fuel injection and lube oil flux. Examples for such root cause analysis and
components development are discussed in chapter 3 of this paper.

In the vehicle, however, the engine controller must be capable of selecting

operation parameters which avoid or at least reduce the occurrence of
critical combustion events to acceptable levels. Sensors available in the
engine are used to identify irregular combustion in order to enable control
algorithms to respond to fuel injection and load actuators. These topics
are addressed in chapter 4.

3. Engine Development

3.1. Diagnostics and Analysis of Irregular Combustion

In an engine speed – load diagram, irregular combustion are found in
typical event classes. A summary from numerous projects’ classifications
is given in Fig. 3. Relevant mechanisms and root causes have been pub-
lished in detail in past years [e. g. 2,3,4]. The task for combustion diagno-
stics is the application of measurement techniques to normal engines or
vehicles under high load operating conditions and recording data to en-

145
able identification of pre-ignitions’ root causes. This is followed by results
analysis to identify most suitable measures for engine improvement. In
engine calibration, improvements are confined to selection of actuator pa-
rameters in engine development the degrees of freedom may furthermore
include selection or adaptation of hardware components.

There are 3 typical questions to start such combustion analysis:

1. At which in-cylinder position does pre-ignition start?

2. Can we operate the engine to initiate pre-ignition events?

3. What is the in-cylinder status of cycles leading up to and following any

irregular combustion event?

Sensors and measurement techniques to address above questions must

be suitable for application in normal engine operation or even in a vehicle.
This is accomplished with fiber optic spark plug sensors to record flame
radiation together with cylinder pressure sensors for thermodynamic com-
bustion analysis.

Signal recording requires continuous measurement of combustion cham-

ber data. As soon as real time pressure analysis identifies irregular com-
bustion, a series of combustion cycles before and after the event is stored
in a data memory. By means of continuous data recording in a ring memo-
ry configuration and real time cylinder pressure evaluation, measurement
procedures can be maintained over long periods for automatic recording
of irregular combustion cycles. This is especially beneficial for measure-
ment of sporadic combustion events, which may as well be captured as
cycles initiated by specific engine operation.

Details of measurement systems capable of such tasks together with ty-

pical examples have been frequently described [2,3]. With access to data
on location of pre-ignition, their distribution statistics and relation to mode
of engine operation, understanding of root causes and consequent impro-
vements is achieved.

146
Figure 3: Classes of irregular combustion events and their appearance in
a speed – load map. A: LSPI – low speed pre-ignition, B: thermal pre-ig-
ntiion, C: sporadic pre-ignition, D: ignition by flame residuals at engine
start

3.2. Irregular Ignition Classification

The speed – load map in Fig. 3 identifies groups A to D with characteristic
features derived from mechanisms giving rise to pre-ignition:

A: LSPI – low speed pre-ignition: Ignition results from combination of high

temperature (surfaces, deposits, gas) and long time period at low engine
speed available to initiate thermochemical reactions.

B: Thermal pre-ignition: Heat flux results in high temperature of in-cylin-

der components which act as hot spot ignition source.

C: Sporadic pre-ignition: free moving deposits act as ignitor for subse-

quent cycle

D: Irregular ignition in cold engine: residual fuel continues to burn into

next cycle and ignites fresh charge.

147
In groups A and B, combustion chamber temperature and, consequently,
engine load, is one dominant factor, in C and D, it is glowing residuals or
residual burning fuel igniting the next cycle’s charge. Each one of these
groups needs engine improvements related to the specific cause of ignition.

3.3. LSPI – Pre-Ignition at Low Speed and High Load

Main parameters are temperatures of combustion chamber surfaces, of
gas, liquids and deposits and the time available to establish thermochemi-
cal reactions of these media. Reducing engine speed at high load can thus
be used to drive the engine into such LSPI combustion mode. Improvement
is found with reduction of temperature, and with avoiding formation of de-
posits from fuel or lube oil. As reactivity of the charge is furthermore under
the influence of residual gas, gas exchange parameters are to be observed.

In an engine analysis, questions arise as to

1. Fuel injection influence:

a. Is pre-ignition related to fuel injection parameters?
b. Is there a correlation to soot emissions?
2. Thermal ignition on surfaces:
a. Is pre-ignition located in specific combustion chamber areas?
b. Is it under influence of cooling?
3. Residual gas influence
a. Valve timing?
b. Turbo charger speed variations?
4. Lube oil influence
a. Cylinder liner,
b. Functionality of crank case ventilation, blowby, valves

Fuel injection in direct injection engines takes influence on charge cooling,

formation of deposits and, at very late injection, on charge stratification.
Numerous tests in GDI engines with either central or side mounted injectors
show PI trends as given in the schematics of Fig. 4: retarded injection (most
often a secondary injection) reduces PI risks by means of cooling effects.
Late injection, however, also gives rise to enhanced spray wall contact and
formation of deposits with the growing risk of deposits induced ignition.

148
 Figure 4: LSPI trends under influence of fuel injection. A: spray cooling
effect, B: deposit formation from spray – wall contact with subsequent
pre-ignition, C: injector influence

Injection timing effects on PI events is first of all tested in stationary ope-

ration with injection timing sweeps. Significant correlation is found bet-
ween injection timing and PI risk with IMEP improvements of a few bar at
borderline PI conditions, provided that late injection is effectively used for
local cooling. The trend example of Fig. 5 shows a typical BMEP response
to such injection variation.

Figure 5: LSPI step 1: fuel injection tuning to exploit spray cooling with
retarded injection timing.

Exploitation of such local cooling effects, however, needs precise selecti-

on of injector and fuel injection parameters to counteract the risk of depo-

149
sit formation and deposits induced thermochemical ignition. Diagnostics
of spray – wall interaction is accomplished with flame analysis techniques.
Even at very high load, wall film formation gives rise to local diffusion
flames. Such diffusion combustion is identified by means of fiber optic
spark plug sensors. A high resolution multichannel sensor configuration
is shown in Fig. 6 together with typical flame signal patterns comparing
diffusion flames with a perfect premixed flame example.

Figure 6: LSPI step 2: Fuel injection is tuned via flame analysis to reduce
risk of deposit related pre-ignition. Flame diagnostics is done with mul-
tichannel fiber optic spark plug sensors.

3.4. Thermal Pre-Ignition

The very same mechanisms giving rise to Low Speed Pre Ignition (LSPI)
are as well active at medium or high engine speed as are spray-wall inter-
action and deposit formation. Here, of course, temperature becomes the
ever more predominant factor in the basic kinetic relationships of Fig. 2
and requires specific attention to reduce the risks of hot spot ignition.

Temperature borderline behavior is tested with engine speed, load and

ignition parameter variations. Pre-ignition is initiated with load sweeps or
with aggressive knocking cycles [3]. Identification of hot spot ignition loca-
tions is accomplished with spark plug sensors already shown in Fig. 6.
Analysis of hot spot ignition statistics then gives guidance for heat transfer
improvements. Testing and diagnostic methods have been published in
[3,4]

150
3.5. Sporadic Pre-Ignition
Engines with deposits buildup on either combustion chamber or intake
port surfaces are at risk of ignition via such deposits. As long as deposits
are in close contact to a surface, risk of ignition is more related to the
features of high load engine operation as described in section 3.4. There
are, however, drive situations giving rise to deposits’ breakaway. They
are heated in the combustion stroke and – if not passing out through the
exhaust valves - may act as hot ignition source for the subsequent cycle.

As deposits formation and their breakaway is hardly controllable by engine

operation, diagnostics of such sporadic events requires specific measure-
ment procedures. They include continuous data recording, real time pres-
sure analysis and automated data storage for long hours of engine tests.
Test and analysis techniques have been described in reference [3,4].

3.6. Irregular Combustion at Engine Start

In cold engines, especially at engine start, the combination of excess fuel,
retarded ignition and low combustion speed can result in ongoing com-
bustion throughout the exhaust and intake stroke. This ignites the fresh
charge of the subsequent cycle and results in excessive emissions peaks
and near misfire cycles.

4. In Vehicle Challenges and Solutions

Whenever pre-ignition occurs in vehicle operation, engine calibration en-
gineers can access actuator parameters to either block riskful operation
regimes, or to respond to such events by instantaneous control of suita-
ble actuators. Calibration engineers’ tasks include detection of pre-igni-
tion events and definition of fast response actuator parameters to inhibit
escalation of irregular combustion cycles. In a vehicle environment, such
identification and response must exploit sensors and actuators which are
accessible to the ECU.

We use examples collected in the calibration of boosted engines to de-

scribe systematic testing and methods available for event detection in ve-
hicles. Improvement of pre-ignition behavior is demonstrated in vehicle
tests.

151
4.1. Task Analysis
A common method in responding to pre-ignition is to cut fuel injections
in order to protect the engine from “run-away” combustion cycles. Such
fuel cut response, however, affects engine torque and is noticeable when
driving the vehicle.

In any vehicle calibration project, questions arise relating to:

• correct detection,
• wrong detection and
• missed detection
of pre-ignition events. These detection and identification issues are ad-
dressed with mobile on-board combustion measurement techniques in
parallel to the standard ECU-based methodology. This enables vehicle
testing in normal drive situations.

Data analysis in a particular fleet test project has shown that vehicles were
subject to pre-ignition, ECU based identification was erroneous and vehic-
le response to pre-ignition events was highly irritating. Measures had to be
found to significantly improve such vehicle performance.

4.1.1. Mobile On-Bord Combustion Measurement

The necessity for mobile on-board combustion measurement arises from
vehicle driving situations which are specific to local markets. An examp-
le is given in Fig. 7 with obvious influence of fuel quality on combustion
chamber and catalyst degradation.

Figure 7: Fuel influence, left side China, right side European fuel.

Addressing such market specific vehicle and engine behavior is accom-

plished with on-board combustion diagnostics equipment. These “stand
alone” diagnostic modules enable long time data acquisition in normal
drive situations. Evaluation of recorded data and identification of specific
phenomena is achieved with automatic data reduction procedures.

152
Synchronization of ECU and combustion measurement signals ensures
correct comparison of data sequences.

4.1.2. Procedures on Test Bed and in Vehicle

The boundary conditions:

A normal engine was instrumented with spark plug pressure sensors to

provide combustion reference data. Lube oil was changed for each test
cycle. The ECU application system and combustion measurement system
were connected via an OHI - “open hardware interface”

This system configuration provides synchronized data recording in both

ECU and combustion measurement system. The ECU application system
is the master unit to generate trigger conditions for data acquisition. The
instrumentation setup as given in Fig. 8 is used to compare ECU preigni-
tion diagnostics (yes/no) with combustion pressure data as recorded with
the combustion measurement system.

Figure 8: Configuration of measurement equipment

153
Test bed evaluation procedure:

Testing was performed in stationary as well as in transient operation at

normal engine temperature (90 °C) and under forced cooling condition
(30°C) with a variety of fuel qualities. Pre-ignition was either detected
by means of relevant ECU variables and by cylinder pressure traces as
recorded and evaluated in the combustion measurement system. Each
event of “pre-ignition detected” was followed by a trigger to start data ac-
quisition including 10 seconds before and 20 seconds after the event.
Multiple events within each recording window added to the post event
record streams.

In vehicle test procedure:

Vehicles have been operated in two specific testing modes

Mode 1: 60% idle, 30% inner urban, 10% extra urban

Mode 2: 70% highway, 30% mountain roads

Fuel influence was evaluated from daily fuel tank refills, with fuel sample
analyses. One vehicle war furthermore tested with a fuel additive. Tests
were completed with oil analysis and engine inspections using an endos-
cope.

4.2. Data Evaluation Method

The LSPI and thermal pre-ignition events as described in section 3.3 and
3.4 most often arise with strong knock amplitudes, however still moderate
pressure peaks. Subsequent cycles can drift into more critical pressure
regimes with the risk of engine damage related to these pressure regimes.
It is useful to classify such cycles with respect to pressure levels at TDC.
Fig. 9 identifies 3 classes with high – medium – and moderate PI pressure
regimes

154
 Figure 9: Classification of 3 categories of pre-ignition cycles.

Combination of ECU and combustion measurement data provides stati-

stics on

• number of pre-igntion cycles

• Pre-igntion class
• Correct detection
• Wrong detection
• Missed detection

4.3. Vehicle Calibration Improvement

Improved calibration should

1. identify pre-ignition cycles

2. react in real time to prohibit further escalation of pre-ignition
3. find reaction measures to avoid torque fluctuations as is the case
with otherwise successful fuel cut strategies

Task 3 is accomplished with cylinder selective response to fuel injection:

Identification of a PI event is followed with fuel enrichment in subsequent
cycles. The mechanism is evident in the fuel injection and cylinder pressu-
re traces of Fig. 10. It is obvious that this measure is best suited to avoid
torque fluctuations as would occur with a fuel cut strategy.

155
Response parameters include degree of enrichment and number of rich
cycles. Evaluation of best suited injection parameters is supported with
flame analysis techniques given in Fig. 6.

Figure 10: Reaction to pre-ignition: cylinder specific fuel enrichment re-

duces charge temperature and prevents subsequent pre-ignition cycles.

Prerequisite to the widespread application of this fuel enrichment method

is the correct and reliable identification of pre-ignition cycles. This is achie-
ved with a refinement of ECU functionalities for identification of pressure
related combustion events.

4.4. Vehicle Test Results

Adaptation of ECU algorithms to evaluation of PI-related combustion
pressure events has significantly enhanced identification rates of irregu-
lar combustion cycles. Fig. 11 gives statistics examples with data sets
evaluated with “old” and “new” parameter settings. A typical test in Fig.
11(A) shows that 30% of PI events are undetected with “old” parameter

156
settings. “New” parameter settings applied to a control test, avoid such
identification failures.

As soon as a PI event is identified, the ECU responds with fuel enrichment

of the relevant cylinder. The test case given in Fig. 11 (B) shows a better
than 10-fold reduction of sequential pre-ignition events. Primary pre-igni-
tion, of course, is unaffected by this measure and can only be handled in
upfront hardware development.

Fig. 11: Pre-ignition test statistics

A: Detection quality: “new” method shows 100% detection quality. Refe-

rence is taken from pressure traces.

B: Detection and Response quality: “new” method yields significant reduc-

tion of sequential pre-ignition cycles.

5. Summary
Cylinder selective fuel enrichment has been shown to be a successful
method in responding to pre-ignition events via ECU functionalities. It pro-
vides significant reduction of secondary pre-ignition cycles, whilst main-
taining required engine torque for good driveability.

Prerequisite to this fuel injection response is the reliable detection and

identification of pre-ignition cycles. Enhanced functionality of ECU identifi-
cation algorithms has been shown to accomplish this task.

Engine hardware development for PI resistant engines is supported with

analysis of flame features by means of fiber optic spark plug sensors.

157
Literature
[1] Heywood J. B.: Combustion engine fundamentals, Mc Graw Hill,
p450 ff, 1988
[2] Zahdeh A. et al. “Fundamental Approach to Investigate Pre-Ignition in
Boosted SI Engines”, SAE 2011-01-0340
[3] Hirsch, A., Kapus, P., Philipp, H., Winklhofer, E.: Irregular ignition
events in TC GDI engines: phenomenology, analysis and engine de-
velopment, PTNSS Poland, 2010.
[4] Winklhofer E., Kapus P., Knorz C., Moik J.: „Ottomotoren im Hoch-
lasttest“, 10. Tagung „DER ARBEITSPROZESS DES VERBREN-
NUNGSMOTORS“ 22./23. September 2005.

158
9 Optische Diagnostik von Verbrennungs-
anomalien als Teil der Brennverfah-
rensauslegung / Optically diagnosing
combustion anomalies as part of desig-
ning the combustion process
Philipp Adomeit, André Brunn, Carsten Dieterich,
Marco Günther, Silja Klier

Abstract
Pre-Ignition at low speed full load of SI-engines is a cause of abnormal
com,bustion, which becomes a limiting factor when it comes to increasing
the level of downsizing. With the onset of pre-ignition strong pressure ri-
ses and extreme knock intensities occur which are a severe risk for engine
damages. Root causes of pre-ignition have been identified, ranging from
hot-spot ignition, ignition at particles or oil droplets to auto-ignition of fuel
vapor or long chained HC which may also originate from the lube oil.

The variety of sources and the stochastic occurrence of pre-ignitions make

it almost impossible to identify their root cause based only on a conventi-
onal thermodynamic analysis. Here a well-targeted application of optical
diagnostics is feasible. For this purpose FEV uses both multi-fiber spark
plugs and endoscopic flame visualization, frequently in combination.

Both diagnostics have proven to be successfully applicable in the develop-

ment process of downsized SI-engines in order to determine the trigge-
ring ignition spot of pre-ignition. In operation the multi-fiber spark plug has
shown in some cases to be invasive by acting as the triggering ignition
spot. In these cases it is most feasible to apply the specifically developed
multiple-image-guide-endoscopy which enables a complete three-dimen-
sional resolution of the ignition spot.

Kurzfassung
Die Vorentflammung bei niedrigen Drehzahlen und hohen Lasten ist eine
Verbrennungsanomalie, die eine Steigerung des Downsizing-Grades bei

159
Ottomotoren einschränkt. Das Auftreten von Vorentflammung führt zu ab-
rupten Druckanstiegen mit extremem Klopfen und kann daher massive
Motorschäden zur Folge haben. Als mögliche Ursachen von Vorentflam-
mungen wurden verschiedene Effekte identifiziert, die von der Oberflä-
chenzündung über partikel- bzw. tröpfcheninduzierte Zündung bis hin zur
Selbstzündung des Kraftstoffs oder langkettiger HC-Komponenten, z.B.
aus dem Motoröl, reichen.

Vor dem Hintergrund der Vielfältigkeit der Ursachen sowie der stochas-
tischen Natur der Vorentflammung ist eine Ermittlung ihrer Ursache aus-
schließlich mit den klassischen Mitteln der thermodynamischen Analyse
schwierig. Daher ist ein zielgerichteter Einsatz optischer Diagnostik sinn-
voll. Dazu kommen bei FEV Mehrfachlichtleiter-Zündkerzen und endosko-
pische Flammenvisualisierung zum Einsatz, häufig auch in Kombination.

Beide Techniken werden bei der Auslegung von aufgeladenen DI-Down-

sizing-Motoren erfolgreich als Analysewerkzeug zur Ermittlung des auslö-
senden Zündorts der Vorentflammung eingesetzt. Im Einsatz hat sich die
Mehrfachlichtleiter-Zündkerze nicht in allen Fällen als nicht-invasiv her-
ausgestellt, da sie als auslösender Zündort fungieren kann. In diesem Fall
ist der Einsatz der speziell entwickelten Mehrfachendoskopie zielführend,
mit der eine dreidimensionale Auflösung der Zündorte erzielt wird.

1. Introduction
For gasoline engines the combination of direct injection and downsizing in
combination with turbo charging has established itself as a universal key
technology for a substantial progress in fuel consumption reduction.

In highly boosted gasoline engines pre-ignition potentially leads to extre-

me knock and severe engine damage. As a consequence of the extreme
mechanical and thermal load of the engine components massive engine
damages can occur. Secondary measures to prevent engine failures can-
not always be implemented without an increase of frictional losses.

Different reasons for this phenomenon were suggested previously, in par-

ticular wall effects including the effect of deposits, particle or droplet-in-
duced ignition and the thermodynamic state and mixture composition in
the gas phase [1-7]. Furthermore different fuel effects on pre-ignition have
been reported.

160
Figure 1 shows an overview of the potential sources for pre-ignition phe-
nomena. Pre-ignition occurs as a consequence of gas phase conditions
which exceed the critical reaction-kinetic auto-ignition limits. Typically a
combination of several root causes leads to the stochastic onset of pre-ig-
nition [1, 4, 8]. Important engine parameters are charge motion intensity,
homogeneity and fuel properties, but also combustion chamber geometry,
material and temperatures.

Figure 1: Overview of the potential sources for pre-ignition phenomena

Especially for highly boosted downsized engines the thermal load and
surface temperatures increase. Reported critical hot spots are spark plug,
exhaust valves, piston and squish edges [4, 9, 10, 11, 12]. The presence
of deposits significantly affects surface temperatures, and may cause hot
spot ignition [13, 14, 15]. Loose deposits and other particulates may survi-
ve a firing cycle and provide hot surfaces for ignition [14, 16].

A further very important source of pre-ignition is the presence of compo-

nents with high ignitability, such as oil or high boiling fuel components.
These components can be introduced into the combustion chamber either
as droplets from different sources, e.g. the crank case ventilation or the
turbocharger lubrication. The entrainment of oil droplets from the cylinder
liner due to oil dilution is regarded as the most common source [8, 14]. But
also fuel evaporating after contact with the liner can cause evaporation of
oil components from the liner oil film. The influence of different base oils

161
and oil additives is examined in detail in [17]. A higher amount of calci-
um detergents in lube oil increases the pre-ignition propensity. However,
oil additives like zinc and molybdenum reduce the pre-ignition sensitivity.
Furthermore, oxidation reactivity is stated to have a more dominant impact
on auto-ignition of oil droplets than oil volatility and evaporation properties.

Several fuel effects on pre-ignition have been reported. A comparison of

conventional gasoline fuels has shown that the pre-ignition propensity
decreases with higher octane number [16]. However, an aged high octa-
ne-rated gasoline has been observed to cause a higher pre-ignition rate
than fresh fuels with lower octane number [13]. The stronger evaporation
cooling of the charge with increased ethanol content in the fuel is ex-
pected to reduce the pre-ignition propensity [16]. Especially for hot spot
triggered pre-ignition a contrary behavior was observed [18] although in
general ethanol containing fuels reveal a reduced ignitability compared
to conventional gasoline. The influence of the fuel molecular structure on
pre-ignition behavior is examined in [19]. Long chain paraffins and olefins
as well as unsaturated naphthenes increase the pre-ignition sensitivity.

The mixture formation process has a major impact on pre-ignition propen-

sity as well. Especially for modern highly boosted gasoline engines the
design of the fuel injector – and the associated charge motion– plays an
essential role in terms of the pre-ignition behavior of the engine [20].

The complex phenomenon pre-ignition requires comprehensive coun-

ter-measures in the engine development process. Therefore, FEV has
developed a chain of development tools for the reduction of pre-ignition
propensity and the analysis of their effectiveness:

1. In the concept phase a fundamental understanding of the pre-ignition

phenomena is required. Here the complex interaction of root-causes
is analysed and potential risks are identified.

2. In the design phase, a pre-optimization tool-chain is used, based on

CAE analysis and optical diagnostics. CFD simulation of in-cylinder
flow and mixture formation is combined with a multi-zone reaction
kinetics analysis to determine the sensitivity to pre-ignition due to gas
phase auto-ignition phenomena [21, 22]. Optical diagnostics are ap-
plied to a rapid prototyping engine set-up to investigate the DI fuel
injection, the fuel liner interaction and mixture formation for the most
critical low speed full load engine operation.

162
 3. Optical diagnostics for pre-ignition investigations are used in the en-
gine thermodynamic development. Here at FEV non-invasive tech-
nologies such as multi-fiber spark plugs and endoscopic combustion
imaging are applied, frequently also in combination.

In the following chapter the above mentioned methodology as well as the

used test engine is described in detail.

4. Experimental set-up and boundary conditions

4.1. Test engine

The test engine is a quantity controlled inline four cylinder engine with
stoichiometric combustion system, direct injection system and conventi-
onal valve train equipped with cam phasers on intake and exhaust side.
The compression ratio of the four valve combustion chamber with lateral
injector position is CR = 10. The intake ports are designed as high tumble
ports in order to achieve a good mixture homogenization and to ensure
combustion of the cylinder charge without any pre-ignition phenomena.
The boosting device is a turbocharger with electrical wastegate actuation.
A summary of the essential technical data of the engine is given in table 1.

Test engine
Bore / mm 76
Stroke / mm 82.6
Stroke/bore ratio / 1 1.087
Number of cylinders 4
Displacement / cm³ 1499
Valves per cylinder 4
Compression ratio / 1 10
Maximum mean effective pressure / bar 22
Cylinder pressure / bar 120
Injector 6-hole multi-hole injector
Maximum fuel pressure / bar 150
Table 1: Technical data of the test engine

163
4.2. Optical measurement techniques
Using an optical spark plug with integrated fiber optics the combustion
radiation of all four cylinders is analyzed. Additionally pre-ignition phen-
omena on the fourth cylinder are detected using multiple high speed en-
doscopy.

4.2.1. Multi-fiber spark plug

Using the different arrival times of light emitted during a pre-ignition at the
single fi-bers, conclusions regarding the origin of a pre-ignition can be
drawn. The detected light is forwarded towards the photomultiplier by total
reflection on the hull of the op-tical fiber. The hull of the fiber is made of
quartz glass with different amounts of dop-ing.

Figure 2: Set-up of used multi-fiber spark plug

Figure 2 left shows a model of the fiber optical spark plug, in the middle
of the figure the arrangement of the fiber viewed from below is presen-
ted. On the right side of figure 2 the relative transmittance of the fibers in
relation to the wavelength of the detected light is depicted. Twelve single
fibers are attached to the spark plug. Four of these are pointing along the
axis of the spark plug, the other eight are tilted by 60 ° from the axis of
the spark plug. Due to the fibers being translucent for ultra violet light (as
well as for visible light), the early stages of pre-ignition can be recorded.
The optical measurements are run continuously at a resolution of 0.1 °CA.
The signals coming from the fibers are converted by photomultipliers and
recorded by the indication system.

The design of the ground electrode and the heat value are taken over
from the base engine configuration. To eliminate interferences of the op-
tical spark plug with the sensitivity for abnormal combustion, comparative
measurements are conducted and statistically evaluated.

164
4.2.2. Multiple high speed endoscopy
The multiple endoscopic system consists of a flexible triple-image-guide
probe, a high speed image intensifier by LaVision and a high speed came-
ra Photron SA1.1. The triple-image-guide probe has a resolution of 10000
pixel at an opening angle of 70 °. Since the image-guides have a diameter
of just 1.2 mm, the additions to the cylinder head for the optical access can
be kept to a minimum.

Figure 3: The triple-image-guide-micro-endoscope and the optical set-up

in the engine test cell

Figure 4 shows the positioning of the optical probes in the cylinder head.
The signals of the image-guides are transmitted via a lens adapter to the
micro channel plates of the image intensifier and are recorded by the
camera. The refresh rate of the camera and the image intensifier has
been adapted to the engine speed in order to realise a resolution of 1 °
to 0.5 °CA. The frame rate is imposed onto the recording system, accor-
dingly. In accordance the spatial resolution of the CMOS chip has to be
reduced to an image range of 512 x 512 pixel. To ensure the best possible
display of the signals of the image-guides on the camera chip an objective
with a focal length of 50 mm is used at an ocular adapter.

165
 Figure 4: Application of the triple-image-guide-micro-endoscope to the
test engine

The high speed camera functions as system master and periodically de-
fines the gate of the image intensifier. Measurements are triggered by
pre-ignition events which are determined from the cylinder pressure si-
gnals by the indication system. Thermal and mechanical isolation of the
image-guide tips is provided by a cooled sensor insert with a window
made of sapphire glass.

5. Methodology of used optical measurement

techniques
For the pre-ignition investigations an engine operating point at minimal
engine speed is chosen to achieve maximum mean effective pressure..
In order to set critical pre-ignition conditions for this operating point on the
one hand an extremely retarded spark timing is chosen. On the other hand
the intake manifold temperature is increased and the injection strategy
is adjusted. The boost pressure is increased until a critical pre-ignition
probability occurs. The boost pressure ranges up to 2000 mbar. In com-
bination with the significantly retarded combustion phasing a brake mean
effective pressure of bmep = 17 bar is reached.

166
5.1. Multi-fiber spark plug
Photomultiplier tubes are used for converting the detected photons into a
recordable signal, figure 5. The arriving photons of combustion radiation
strike the photocathode at which electrons are emitted (photoelectric ef-
fect). Upon striking the anode these electrons flow off to the ground and
cause a voltage drop at the resistor. The voltage is the output signal which
is recorded by an indication system.

Figure 5: Photomultiplier tube and its spectral sensitivity

The photomultiplier tubes respond very fast to incoming photons and con-
vert them into a measurable voltage, even the photons of ultra-violet light
are detected (figure 5).

5.2. Multiple high speed endoscopy

For an exact analysis of pre-ignition onset spots an initial calibration of
the engine-camera system is required before it can be applied for engi-
ne operation. The calibration is done according to the method proposed
by Zhang in [23] using the “Camera Calibration Toolbox for Matlab“ de-
veloped by the California Institute for Technology.

Images of a 2D calibration target are taken from different views. The 3D

Position of the target and the motion between the different views does
not need to be known. From the corresponding 2D/3D pairs of points the
camera parameters are calculated.

167
For the calibration six images of the target are taken with the image-gui-
des positioned outside the engine. An additional image is taken with the
image-guides mounted to their position in the cylinder head. From this
data the intrinsic as well the extrinsic parameters for the different orienta-
tions are estimated for each image-guide individually. Except for the prin-
cipal point, the intrinsic parameters for the three image-guides are the
same since they are all mounted to the same camera. In contrast, the
extrinsic parameters are different due to the different positions in the cylin-
der head. For the calibration the origin of the world coordinate system is
positioned at the center of the piston at top dead center to enable an easy
reconstruction of pre-ignition spots. Exemplary results of the calibration
for one image-guide are shown in figure 6.

Figure 6: Calibration results for one image-guide

Figure 6 left shows a comparison of the original target points (+) with
the back projected points after calibration (o). Even though the image is
strongly distorted, the difference between the points is very small. The
absolute differences are smaller than +/- 1 pixel. The calibration for the
other orientations is similar accurate. On the right of figure 6 the deviations
for all points are shown. With this calibration it is possible to estimate the
3D coordinates, except for a scale factor, of a small light spot recorded by
one of the image-guides. This can be represented by a straight line into
the combustion chamber. If a light spot is seen by two or three image-gui-
des, its 3D coordinates can be determined as the intersection point. Ac-
cordingly, the origin of pre-ignitions recorded right at the start and which
therefore are just a few pixels in size can be exactly calculated using this
method. If the area covered by a pre-ignition at a later point in time is of

168
interest, it can be estimated based on the described methodology using
“Shape from Silhouette”.

This reconstruction requires the image-guides to remain keep their position

on the camera frame during the measurements. So it has to be ensured that
the image-guides do not turn and that there is no relative movement between
the ocular and the camera. Such movement would change the extrinsic pa-
rameters and the entire system would have to be recalibrated. To ensure
this, the positions of the image-guides are checked prior to each measure-
ment and possible shifts are corrected. The orientation of an image-guide is
checked using spot light sources arranged in a defined grid mounted at the
position of the spark plug in the combustion chamber. Thus a clear determi-
nation of the position is possible. For the measurements the spot light source
is again replaced with the spark plug (figure 7).

Figure 7: Raw image of the 3D-calibration image of the combustion

chamber

6. Discussion of different methods for optical combus-

tion diagnostic
Both methods which have been described in detail are useful tools for
comprehensive analysis of abnormal combustion phenomena in early de-
velopment stages. By exact detection of the ignition spot location signifi-
cant conclusions on root causes are possible for the investigated engine.
Multi-fiber spark plug and endoscopic flame visualization methods are
qualified for fundamental analysis as well as for protection from combus-
tion anomalies.

169
Different advantages can be identified for both approaches. First the sys-
tem using the optical spark plug sensor is discussed. Based on these re-
sults requirements for evaluation of multiple high speed endoscopic flame
analysis will be defined.

After a short adaptation time and quick calibration of the photo-multiplier

which converts the light intensity into voltage signals the measurement
data can be recorded with a conventional cylinder pressure indication sys-
tem.

Figure 8 shows a pre-ignition event recorded with twelve light-transmitting

fibers applied to the spark plug.

Figure 8: Pre-ignition measurement with multi-fiber spark plug (full load)

The first signal increase can be detected for fiber #1. After short delay the
combustion is also visible at fiber #3. At fiber #2 which is installed parallel
to the cylinder axis the signal increase occurs with an additional delay of
approximately one degree crank angle. Since the combustion is detected
first at a neighbored pair of fibers and second at the intermediate channel
in cylinder axis direction with a certain delay, it can be concluded that the
ignition spot is located closely to the cylinder liner.

The left half of picture 9 shows all recorded abnormal combustion events
for this operating point compiled with the same algorithm as described
above for one single event. The right sketch relates the mentioned ignition
spots to the combustion chamber roof geometry.

170
 Figure 9: Compilation of all pre-ignition locations (full load)

Even though a slightly higher number of pre-ignitions are located on the

exhaust side, it is clearly visible that there is a random distribution of all
ignition spots without a pronounced single ‘hot spot’. This gives a hint that
the pre-ignition is caused by exceeding limits regarding chemical kinetics
in the gaseous phase of the mixture. Superimposed effects of mixture for-
mation, e.g. homogenization, oil dilution or oil/fuel-interaction at the cylin-
der liner can cause further randomization.

Basically a quick evaluation of pre-ignition characterization could be per-

formed with the optical multi-fiber spark plug on this engine. All ignition
spots could be located very well.

Disadvantages can be expected if a large number of abnormal combusti-

on cycles is caused by the spark plug itself. This can be caused either by
generally unfavorable spark plug layout (heat value, layout of ground elec-
trode or protrusion into combustion chamber) or by changes in the spark
plug behavior due to the installation of the optical fibers. This might lead to
ignition triggering conditions at the spark plug by a reduced heat transfer
or by a changed influence on in-cylinder charge motion. If the spark plug
specification is the fundamental root cause for pre-ignition this will be car-
ried over to the optical spark plug as well. In this case the results of the
measurements are trivial. Figure 10 shows a distribution with all ignition
spots close to the spark plug.

171
Figure 10: Compilation of all pre-ignition locations at full load with adjus-
ted injection parameters

For deeper analysis of such a unique distribution of ignition spots additio-

nal investigation has to be applied. Therefore FEV uses the multi-endos-
copic diagnostic method which requires higher design and adaptation as
well as calibration effort. The original spark plug can be used without mo-
dification to avoid an overlay of effects. Figure 11 displays the raw image
of a triple-endoscopic measurement of pre-ignited combustion caused by
an ignition spot at the spark plug. Analyzing the statistical appearance,
critical areas of the operated spark plug layout can be identified.

Figure 11: Raw image of a pre-ignition initiated at the spark plug under
full load conditions, recorded with multiple image high speed endoscopy

172
Further investigations on a second engine with similar displacement and
combustion system layout have been carried out in terms of spark plug
variation (figure 12). With layout #1 marked by increased combustion
chamber protrusion the central electrode and the directly surrounding gas
phase could be identified as weak point – probably caused by longer dwell
time of hot recirculation gas. The initial optimization step shifts the pre-ig-
nition limit to operation with increased boost pressure by 150 mbar. The
flame visualization using the multi-endoscopic access to the combustion
chamber indicates now the ground electrode, which is oriented towards
the exhaust valves, as primary ignition spot.

Figure 12: Probability density function of pre-ignition distribution for diffe-

rent spark plugs (full load)

7. Summary and Conclusions

Pre-ignition as one of the abnormal combustion phenomena is the most
limiting factor for further downsizing of spark ignition engines. Irregular
fuel conversion before spark timing leads to extreme cylinder pressure
rise and knocking and could damage the engine severely. Miscellaneous
root causes for pre-ignition are published: hot surface ignition, ignition by
glowing particles or oil droplets or even self-ignition of components from
fuel or lubricant. Since pre-ignition occurs stochastically and has multiple
sources even a deeper analysis with conventional measurement based on
cylinder pressure indication will not give the full understanding.

Hence, for the development of downsized and boosted SI engines with

gasoline direct injection additional tools might be required. FEV integrates
multi-fiber spark plugs as well as endoscopic flame visualization in the
development process. Partly both methods will be combined to detect the
exact location in the combustion chamber where pre-ignition starts.

173
Both measurement techniques have been used as a successful tool to de-
termine the initial ignition spot of pre-ignition events. The main advantage
of a multi-fiber spark plug is the easy and fast adaptation on an engine.
However, sometimes the modified spark plug itself has been identified as
ignition spot for abnormal combustion. For high accuracy of ignition spot
identification a large number of light-transmitting fibers is required – lea-
ding partly to different thermal behavior of the spark plug. Endoscopic fla-
me visualization can be used also. However, even with micro-endoscopic
sensors the additional effort for design and adaptation on a multi-cylinder
engine is significantly higher. The additional benefit is an exact determi-
nation of the three-dimensional position of the ignition spot. As discussed
in this paper a change of the ignition spot from center electrode to the
ground electrode could be detected after using a different spark plug lay-
out. In certain cases FEV uses multi-endoscopic combustion analysis for
series production development.

174
Literature
[1] Zaccardi, J.-M..; Duval, L.; Pagot, A.: Development of Specific Tools
for Analysis and Quantification of Pre-ignition in a Boosted SI Engine.
SAE Technical Paper, 2009-01-1795
[2] Zaccardi, J.-M.; Lecompte, M.; Duval, L.; Pagot, A.: Pre-ignition in
Highly Charged Spark Ignition Engines – Visualisation and Analysis.
Motortechnische Zeitschrift, 70:40–47, 2009
[3] [3] Manz, P.W.; Daniel, M.; Jippa, K.-N.; Willand, J.: Pre-ignition in
highly charged turbo-charged engines. Analysis procedure and re-
sults. 8. Internationales Symposium Verbrennungsdiagnostik, Baden
Baden, 2008
[4] Willand, J.; Daniel, M.; Montefrancesco, E.; Geringer, B.; Hofmann,
P.; Kieberger, M.: Grenzen des Downsizing bei Ottomotoren durch
Vorentflammungen. Motortechnische Zeitschrift, 70:422–428, 2009
[5] Dahnz, C.; Spicher, U.: Irregular combustion in supercharged spark
ignition engines- pre-ignition and other phenomena. International
Journal of Engine Research,11:485–498, 2010
[6] Dahnz. C.; Han, K.-M.; Magar, M.: Vorentflammung bei Ottomotoren.
FVV – Vorhaben Nr. 931, Final report, 907, 2010
[7] Rothenberger, P.; Zahdeh, A.; Anbarasu, M.; Göbel, T.; Schäfer,
J.; Schmuck-Soldan, S.: Experimentelle Untersuchungen zur Vor-
entflammung an aufgeladenen Benzinmotoren in Kombination mit
lichtintensivierter Hochgeschwindigkeitskamera und CFD. 8. Inter-
nationales Symposium Verbrennungsdiagnostik, Baden Baden, 2008
[8] Zaccardi, J .M.; Laget, O.; Pagot, A.; Mokhtari, S.; Leone, T: Inves-
tigations on the Effects of In-Cylinder Charge Motion and Injection
Mode on Pre-Ignition in Highly Boosted SI Engines. 19. Aachener
Kolloquium Fahrzeug- und Motorentechnik 2010
[9] Haselhorst M.; Erwig W.: Pre-Ignition Behavior of Alcohol Fuels. SAE
Paper 821210
[10] Spicher U.; Han K.-M.; Rothe M.: Verbrennungsanomalien in Otto-
motoren mit Benzindirekteinspritzung. 7. Tagung der Technik, 2007
[11] Ernst R.; Münzinger S.; Hanula B.; Schnitzler T.; Geiser F.: Das Phä-
nomen Glühzündung – Ursachen, Entstehung und Vermeidung. MTZ
11/2002
[12] Spicher U.; Maas U.; Dahnz C.; Kubach H.; Magar M.; Schießl R.:
Pre-ignition in Supercharged SI Engines. Stuttgarter Symposium
2010

175
[13] Winklhofer E.; Kapus P.; Knorz C.; Moik J.: Ottomotoren im Hochlast-
test. 10. Tagung „Der Arbeitsprozess im Verbrennungsmotor“ Graz
2005
[14] Spicher U.; Dahnz C.; Han K.-M.; Kubach H.; Magar M.; Schießl, R.:
Abschlussbericht über das FVV-Vorhaben Nr. 931, Vorentflammung
bei Ottomotoren – Untersuchung des Auftretens und der Ursache von
Selbstzündungen vor Zündungseinleitung bei aufgeladenen Motoren
mit hohem Verdichtungsverhältnis
[15] Jung H.-S.; Kim S.-H.; Lee S.-M.; Park S.-I.; Whang J-T.: The Effects
of Combustion Chamber Deposits on Octane Requirement Increase
in a Spark Ignition Engine. SAE paper 2008-01-1761
[16] Prochazka G.; Hofmann P.; Geringer B.; Willand J.; Jelitto, C.: Com-
parison of the phenomena of autoignition in highly charged SI Engi-
nes with external and internal mixture formation. FISITA 2006
[17] Takeuchi, K.; Fujimoto, K.; Hirano, S.; Yamashita, M.: Investigation
of Engine Oil Effect on Abnormal Combustion in Turbocharged Direct
Injection – Spark Ignition Engines. SAE Paper 2012-01-1615
[18] Hamilton L.J.; Rostedt M.G.; Caton P.A.; Cowart J.S.: Pre-Ignition
Characteristics of Ethanol and E85 in a Spark Ignition Engine. SAE
paper 2008-01-1774
[19] Nozomi S.; Koichi N.; Katsunori K.; Shunta S.; Manabu W.; Tadahide
S.: The Effect of Fuel Compounds on Pre-ignition under High Tempe-
rature and High Pressure Condition. SAE Paper 2011-01-1984, 2011
[20] Adomeit P.; Dohmen J.; Thewes M.; Ewald J.; Günther M.; Morcin-
kowski B.; Pischinger S.: Einfluss von Kraftstoff und Brennverfahren
auf die Vorentflammung beim aufgeladenen Ottomotor. 34. Internati-
onales Wiener Motorensymposium 2013
[21] Morcinkowski B.; Budde M.; Ewald J.; Schernus C.; Adomeit P.: Si-
mulation von reaktionskinetisch verursachten Verbrennungs-anoma-
lien als Teil der Kanal- und Brennraumauslegung. Klopftagung Berlin
2010
[22] Morcinkowski B.; Budde M.; Ewald J.; Beykirch R.; Adomeit P.: CAE
tool chain for the prediction of pre-ignition risk in gasoline engines.
Comodia Paper EC-1 2012
[23] Zhang Z.: A flexible new technique for camera calibration. IEEE
Transactions on Pattern Analysis and Machine Intelligence,
22(11):1330–1334, 2000.

176
10 Using surface thermocouples and light
conductor measurements to examine the
thermal load on a gasoline engine’s com-
ponents during knocking engine operation
Karsten Michels, Christopher Gessenhardt, Jörg Theobald

Abstract
The reduction of CO2 emissions is an essential goal of the engine develop-
ment. The main focus is the optimization of combustion efficiency which
is often connected to an increased knocking tendency. The evaluation of
these measures requires an exact knowledge of the thermal boundary
conditions in the combustion chamber.

For this reason, the present work is focused on investigations of knocking

combustions of a 4 cylinder gasoline engine by using surface thermocou-
ples and fiber optic measurement techniques in the combustion chamber
as well as a fast detection of hydrocarbons in the corresponding exhaust
port.

Crank-angle resolved measurements of the surface temperature will be

compared to the flame emission signal during different knocking combus-
tion-cycles. In addition to this, the correlation between the hydrocarbon
emissions and the knocking intensity will be considered.

1. Introduction
Actually, one of the essential development goals is the reduction of CO2
emissions, motivated by the climate change and the limited availability of
fossil resources. For this reason, the major concept of gasoline engine at
Volkswagen in the last years is characterized by charging and direct injec-
tion, the TSI technology. The charging of an engine with reduced swept
volume (Downsizing) results in a shifted operation point towards higher
loads which corresponds to increased efficiency. Thereby a knock limited
operation is established more frequently. Similar effects occur due to the
increase of the efficiency due to an increased charging. It is known from
defects caused by combustion anomalies, documented in many cases

177
in the literature, that these change seriously the mechanical and ther-
mal stress of the engine and could finally cause serious damaging of the
IC-engine [1, 2, 3, 4].

Therefore, the knowledge of the mechanical and thermal boundary con-

ditions is of particular interest for the development of new engines. The
understanding of the effects of knocking combustions on the thermal
boundary conditions helps to assess more precisely the risks caused by
gasoline engine combustion anomalies.

Investigations of a 4 cylinder inline engine will be shown and discussed by

using fiber optical measurement technique (Visioknock) and surface ther-
mocouples (OTE) for different load conditions during knocking and regular
combustion. The knocking area will be determined using the optical sig-
nals. Based on crank-angle resolved surface-temperature measurements
the local wall heat flux densities will be calculated for several position of
the surface thermocouples

2. Measurement technology

2.1. Fiber optic measurement technique – Visioknock

Light emission occurs in the end gas zone due to the auto-ignition in this
area during a knocking combustion. A pressure wave moves through the
combustion chamber as a result of this self-ignition. Because this wave is
reflected by the walls of the combustion chamber, a stationary acoustic
wave arises. As the light emission of the flame increases with the density
of the combustion chamber gas, the pressure oscillations of a knocking
combustion correlate with the modulated light signal.

In the context of the present measurements, a detection of this signal

was carried out with the fiber optical measurement technique Visioknock
(AVL). Here, an optically modified spark plug was used which has 35 op-
tical fibers arranged around the circumference of the spark plug body.
These enable the visualization of the combustion chamber area close to
the cylinder head. The design of the spark plug geometry and the location
of the field of view are shown schematically in figure 2.1.1.

178
 Figure 2.1.1: Schematic illustration of the field of view of a Visioknock
fiber optical measurement spark plug [5]

The light signal of the knocking combustion is analyzed using a temporal

high-pass filter regarding the temporal course of the signals of all 35 de-
tection channels. This approach enables the detection of the azimuthal
position of the knocking area as origin of the subsequent pressure wave. A
determination of the radial position of the knocking area isn’t possible due
to the integral detection of the light signals in the detection volumes. Ex-
emplary illustrations of the signals of a knocking and non-knocking com-
bustion cycles are shown in figure 2.2.

179
Figure 2.1.2: Cylinder pressure signal and signal of the flame emission of
a knocking and non-knocking combustion: top: cylinder pressure curve,
center: raw signal of the flame emission, bottom: high-pass filtered flame
emission signal

In figure 2.1.2 cylinder pressure curves of a knocking and non-knocking

combustion are shown in the top row. The diagrams arranged under these
curves show the intensity distributions of the flame emission false–color
coded. Here, the horizontal coordinate is equivalent to the crank angle
and the vertical coordinate is equivalent to the circumferential position in
the field of view according to the sequence: inlet valve (In), clutch, exhaust
valves (Ex), timing belt and inlet valve (In).

The measurement data on the left side of figure 2.1.2 results from a cycle
without knocking. The proceeding combustion can be seen in the curve of
the cylinder pressure as well as in the flame emission. During combustion
stationary acoustic waves occur in the combustion chamber.. These are
obvious in the high-pass filtered flame emission signal as vertical structu-
res with comparatively low amplitude.

180
On the other hand, the knocking combustion (figure 2.1.2, right) causes a
stronger vibration excitation of the combustion chamber due to the higher
burn rate at the time of the knocking. This results in a characteristically
oscillating cylinder pressure curve and considerable distinctive structures
of the high-pass filtered flame emission signal. The last ones occur from a
certain crank angle position for the present measurement and fade away
for the further combustion progress. The initiation of this high-frequency
signal characterizes the temporal and spatial knock onset which was loca-
lized in the area of the exhaust valves in this case (black cross in figure
2.1.2).

During the further progress of the present work, the illustration of the mea-
surement results carried out with the help of the fiber optical measurement
technique is restricted to the high-pass filtered flame emission signal ac-
cording to the illustration in figure 2.1.2.

2.2. Design and application of a surface thermocouple

For the production of surface thermocouples, at first a coated thermo-
couple, here NiCr-Ni (Type K), has been stuck in a bore in the cylinder
head. Afterwards, the thermocouple was cut, fixed, wall-flush machined
and evaporation coated in a vacuum atmosphere. The contact point of the
thermocouple is made by a layer of chrome. Afterwards, a layer of gold
was evaporated for protection against corrosion. The evaporated layer
thickness is about d = 0.5 µm. Figure 2.2.1 contains schematic drawings
of the design and application of a surface thermocouple.

181
 Figure 2.2.1: Construction and adaption of a surface thermocouple

For the local differentiation, twelve surface thermocouples were arranged

in the combustion chamber of cylinder 4 for the measurements. Figure
2.2.2 shows schematically the locations of the OTE and the evaporated
cylinder head with the gold layer for corrosion protection. The surface
thermocouples are arranged with the same distance around the spark-
plug. Furthermore, the engine was equipped with high-pressure indication
on cylinder 4 as well as a low-pressure indication.

Figure 2.2.2: Locations of the OTE (schematically) and evaporated cylin-

der head

182
3. Test engine and test conditions
The presented investigations have been carried out at a turbocharged 4
cylinder gasoline engine (EA 211) of the Volkswagen AG.

Iso-Octan was used as fuel to minimize fouling of the surface thermocou-

ples. The test engine was operated at a speed of n = 4000 min-1 and n =
6000 min-1 for different load condition (IMEP > 11 bar). Knocking combus-
tions was achieved by advanced ignition timing.

4. Principles for the analysis of the surface tempera-

ture curves
The common approach of a Fourier analysis is used to analyse the surfa-
ce temperature measurements [6,7,8,9]. The Fourier differential equation
of the temperature distribution can be reduced to a one-dimensional case
because the thickness of the walls which surround the combustion cham-
ber is considerably higher than the depth of penetration of the temperature
oscillation.

(1)

In the case of an alternating incident heat flux equation 1 can be expres-

sed as follows using a Fourier series.

(2)

(3)

Equation 3 is the consequence of the differentiation of equation 2 accor-

ding to x for x = 0 with the one-dimensional equation of heat transfer ac-
cording to Fourier. Equation 4 is the heat flux density of a measurement
point (OTE) on the wall (x = 0).

183
 (4)

With the introduction of the thermal effusivity b

(5)

It follows from equation 4:

(6)

The thermal effusivity is determined by a submersion test according to

Huber [10] which was also used by other authors [11,12]. Therefore, a
thermocouple will be assembled as a surface thermocouple to an adapter
consisting of the material of the combustion chamber wall and will be sub-
merged into an oil bath. According to the theory of the one-dimensional
heat transfer, the temperature TK which would set for complete tempe-
rature equalization is set on the contact point of two semi-infinite bodies.
This contact temperature is measured during a submersion test and the
thermal effusivity is calculated with the properties of the oil and according
to equation (7).

(7)

The heat transfer coefficient is calculated according to equation 8 by

dividing the instantaneous heat flux density by the difference between the
gas and wall temperature.

(8)

184
Taking into consideration that the heat flux is zero if the difference bet-
ween the gas and wall temperature is zero, the average heat flux can be
determined by using the progress of the heat flux densities.

(9)

5. Test evaluation

5.1. Information from the surface temperature curves

Based on the raw signals of the surface thermocouples several basic in-
formation can be gained as shown in figure 5.1.1. As expected, the surfa-
ce temperature increases with increasing cylinder-pressure during com-
bustion. The surface thermocouples close to the spark-plug (9, 10, 11, 12)
are affected by the flame front first According to this their temperature in-
creases earlier than the temperature of the OTE close to the cylinder wall
which the flame reaches at last. It is also possible to gather the intensity of
the heat-release rate in the area of an OTE from the increase (cf. OTE clo-
se to the spark-plug with OTE on the cylinder edge). The average surface
temperature can be calculated from the temporal average temperature
level. Furthermore, it can be seen a lower fading or even a low increase of
some OTE signals of the elements in the area of the exhaust valves from
140 °KW after ITDC. This results from the increased heat transfer due to
the high flow velocity during the ejection process. The figures show that
the progress of the surface thermocouple signal is different for each cycle
and also reflects the cyclical variations of the gasoline engine combustion
in the instantaneous local surface temperature. The signals of the surface
thermocouples close to the spark-plug indicate the direction of the flame
propagation because it effects the beginning of the steep temperature rise
at the early stage of combustion.

185
Bild 5.1.1: Evolution of cylinder pressure and surface temperature during
a regular combustion

186
5.2. Knocking combustion – Optical fibers and OTE
During the investigations of knocking combustion we wanted to gain detailed
information towards the correlation of specific knocking events and the ther-
mal boundary conditions, particularly with regard to local temperature differen-
ces. Therefore, individual knocking combustion cycles have been analyzed.

Figure 5.2.1: Pressure, heat and temperature increase curves and Visio­
knock signals of a knocking combustion with knock onset at the inlet side

187
The occurring cyclic variations of the in-cylinder flow field cause variations
of the flame expansion direction during the early stage of the combustion.
The resulting cyclical differences of the spatial distribution of the unburned
end gas result again in different knocking areas.

The upper section of figure 5.2.1 shows the temperature increases of the
surface thermocouples 9 to 12 which are arranged around the spark-plug.
It can be seen clearly an early temperature increase of OTE 9 which is fol-
lowed by OTE 12 delayed by about 7 °KW and followed after this by OTE
10 and 11 delayed by additional 5°KW. From this it can be concluded that
the flame burns faster in the outlet direction than in the inlet direction. This
causes an end gas zone on the inlet side where the knocking finally starts
as shown by the analysis of the optical signals.

Prior to knock onset, the signals of the surface thermocouples show spo-
radically weak temperature rises which are assigned to the normal com-
bustion. In comparison, the temperature gradients due to the knocking are
considerably higher. Analog to the analysis of the fiber optical signals, the
first steep temperature rise can be seen at surface thermocouples 5 and 6
in the area of the inlet valves. Afterwards, the higher temperature gradient
appears one to two degrees of crank angle later for the other elements 3,
2, 1 and 8. The surface thermocouple 4 was defective during the investi-
gations of the knocking combustions.

Figure 5.2.2 shows an example for a cycle with the knocking area on the
outlet side. Here, a significant lower temporal difference of the increase is
shown for the signals of the four surface thermocouples around the spark-
plug. This indicates no significant preferred direction of the flame expansi-
on. Consequently, the beginning of the knocking is to be expected on the
outlet side. This tendency could be confirmed by the use of the fiber optic
measurement technique. The knock onset was detected in this case on
the exhaust side, slightly shifted towards the clutch. This correlates with
the signals of the surface thermocouples. These show a significant increa-
se at first for the elements 7 and 8. Afterwards an increased temperature
rise occurs at surface thermocouples 1, 2, 3 and in the other direction for
the sensors 6 and 5.

188
 Figure 5.2.2: Pressure, heat and temperature increase curves and
Visioknock signals of a knocking combustion with knock onset at the
exhaust side

Figure 5.2.3 shows another example of a knocking combustion with the

knocking area between the intake and control side. The central surface

189
thermocouples are the first which show an increase of the signal in the
direction of the exhaust side with a time lag of about 6°KW between OTE
9 and OTE 12 and an additional offset of 3°KW to sensor 10 and 11. It can
be concluded that the flame front reaches the area between the inlet and
control side at last. Because of this an end gas zone is formed there which
serves as a source of knocking. Based on the analysis of the fiber optical
signals the area between intake and control side has been identified as
the knocking area. The surface thermocouples on the edge of the cylinder
reflect this situation. The earliest signal increase can be seen at surface
thermocouple 3 followed counter-clockwise by 2, 1, 8 and 7. In the other
direction, the signals are less pronounced. The signal of surface thermo-
couple 5 shows an increase which could be caused by the knocking at
approximately the same time as the surface thermocouple 3.

190
 Figure 5.2.3: Pressure, heat and temperature increase curves and
Visioknock signals of a knocking combustion with knock onset between
the inlet and control side

In summary for figure 5.2.1 to 5.2.3, the preferred direction of the flame ex-
pansion can be determined with the help of the surface thermocouples loca-
ted in the center of the combustion chamber. Furthermore the knocking area
can be localized based on surface thermocouples at least for a sufficient

191
knocking intensity. In comparison to the temperature gradients prior and af-
ter knock onset, it remains to be stated that higher temperature gradients
and therefore thermal stress appear in the case of a knocking combustion.

5.3. Knocking combustion – Analysis of the heat flux

densities
The maximum relative heat flux densities as a function of the cylinder
pressure are shown in figure 5.3.1. The heat flux densities are calculated
based on surface temperatures for each cycle with the algorithm descri-
bed in section 4. The maximum relative heat flux densities were calculated
from the maximum value for each combustion cycle in regard to the lowest
maximum heat flux densities during an engine operation without knocking
within the series of measurements.

Figure 5.3.1: Maximum heat flux densities as a function of the maximum

cylinder pressure in bar

192
The fundamentally different behavior of the surface thermocouples clo-
se to the cylinder wall in contrast to the surface thermocouples arranged
around the spark-plug is extremely remarkable. The sensors close to the
cylinder wall show a clear response to knocking which is associated with
an increase of the maximum heat flux densities up to more than tenfold.
In contrast to this the central surface thermocouples show only marginally
higher maximum heat flux densities. These effects are caused on the one
hand by the intensive conversion of the chemical energy of the fuel in the
border area of the combustion chamber during a knocking combustion.
At the beginning of the knocking, the area around the spark plug on the
sensors 9 to 12 is already burned. On the other hand, the highest pressure
amplitude occurs at the circumference of the combustion chamber if its
vibration behavior is dominated by circumferential modes. Those modes
have an oscillation node in the center and the highest amplitudes on the
edge of the cylinder, the radial modes produce considerably increased
pressure amplitudes in the center of the combustion chamber due to their
reflection on the cylinder wall [13].

5.4. Knocking combustion – OTE and Fast-FID

To supplement the previous shown measurements a Fast-FID has been
used for further investigations. During strong knocking combustion at high
revolution speed, high wall temperatures can be observed in addition to
partially considerably increased hydrocarbons emissions. Figure 5.4.1
shows the cylinder pressure curves, the relative surface temperatures and
the signal curves of the Fast-FID for several cycles for strong knocking
events at a speed of n=6000 1/min.

193
 Figure 5.4.1: Cylinder pressure curves, relative surface temperatures
and Fast-FID- measurement signal of strong knocking events

Based on these measurements, the question arises whether hydrocar-

bons cause the knocking events or whether they result from the evapo-
ration of the lubricating film due to the high wall temperatures or other
possible HC accumulations due to e.g. an insufficient mixing formation.
Therefore, a trigger setup to initiate knocking was used which enables a
purposefully early ignition of one combustion cycle by electrical ignition
control. Here, defined advanced ignition timing takes place during a limi-
ted number of combustion cycles. This setup wasn’t available at the time
of the measurements shown in figure 5.4.1. Therefore, the investigations
were conducted with a purposefully early ignition on an engine without
surface thermocouples later. Figure 5.4.2 shows the possibility to observe
increased HC emissions due to a strong knocking for this engine.

194
 Figure 5.4.2: Cylinder pressure curves, trigger and Fast-FID measure-
ment signal of a strong knocking event at the test engine without surface
thermocouples

In contrast to the measurements shown in figure 5.4.1, a lower HC emis-

sions level was reached for this engine until the end of the measurement
range of the Fast-FID.

Figure 5.4.3 shows cylinder pressure curves, trigger and Fast-FID signal
for early ignited knocking combustions.

195
 Figure 5.4.3: Cylinder pressure curves, trigger and Fast-FID signal of
early ignited knocking combustions at the test engine without surface
thermocouples

In contrast to the sporadically appearing strong knocking, the Fast-FID

signal has no significant increase of the amplitudes in the case of a purpo-
sefully early ignition and the resulting strong knocking combustions.

Therefore it can be concluded that the high HC emissions shown in figure

5.4.1 and 5.4.2 are not caused by the higher evaporation of the lubrica-
tion film or fuel residuals due to higher wall temperatures after knocking
events.

It is most likely that hydrocarbons can cause extreme or mega knocking

and can be detected afterwards by Fast-FID. Similar mechanisms are also
discussed in the context of the triggering of preignition in the literature
[14,15].

6. Summary and outlook

In the context of the present work, knocking combustions of a 4 cylinder
gasoline engine were investigated with regard to the changes of the ther-

196
mal boundary conditions in the combustion chamber and the occurrence
of increased HC emissions. Here, a cylinder head equipped with surface
thermocouples, the fiber optic measurement technique Visioknock and a
Fast-FID were used.

A clear correlation between der preferred direction of the flame expansi-

on, the knocking area and the local increase of the surface temperature
was determined during different knocking combustion cycles. Here, the
knocking areas and timings determined by the fiber optic measurement
technique correlate well with the significant increases of the surface tem-
perature gradients. The characterization of the thermal stress increase of
the test vehicle was based on the relative heat flux densities. These in-
crease partially tenfold during a knocking combustion in the area close to
the cylinder wall. In contrast, a considerably lower change of the tempera-
ture boundary conditions was determined in the center of the combustion
chamber.

In the context of these investigations, it could be shown that the sporadi-

cally occurring very strong knockings are connected with increased HC
emissions. Based on a systematic variation of the operating boundary
conditions, an increased occurrence of hydrocarbons was identified as
trigger of these very strong knockings.

197
7. References
[1] Messung verschiedener Einflussgrößen bei klopfender Verbrennung
auf die Kolbenbelastung, Schelling, H. Freier, R., Geisselbrecht, M.,
Stuttgarter Symposium Kraftfahrwesen und Verbrennungsmotoren,
1997
[2] Theoretical and Experimental Investigation of Knock Induced Surface
Destruction, Maly, R. R., Klein, R.,Peters, N., König, G., SAE 900025,
1990
[3] Das Phänomen Glühzündung, Münzinger, St., Hanula, B., Schnitzler,
Th., Geiser, F., MTZ 11/2002
[4] Extremklopfer, Spicher, U., Rothe, M., FVV Abschlussbericht, Heft
836, Vorhaben Nr. 816, 2007
[5] Visioknock-Gerätehandbuch, AVL List GmbH, 2004
[6] Instationäre thermische und mechanische Motorbelastung, FVV-Ab-
schlussbericht Vorhaben 750, 2002
[7] Beitrag zur experimentellen Bestimmung des instationären, gassei-
tigen Wärmeübergangskoeffizienten in Dieselmotoren, K. Sihling,
Diss. TU Braunschweig, 1976
[8] Experimentelle Untersuchungen zum Wärmeübergang bei normaler
und klopfender Verbrennung im Ottomotor, G. Woschni, J. Fieger,
MTZ 2/1982
[9] Ein Gleichungsansatz zur Berechnung der instationären Wandwär-
meverluste im Hochdruckteil von Ottomotoren, M. Bargende, Diss.
TH Darmstadt, 1990
[10] Der Wärmeübergang schnelllaufender, direkteinspritzender Diesel-
motoren, K. Huber, Diss. TU München, 1990
[11] Modellierung der Verbrennung und des Wärmeübergangs in Ottomo-
toren mit homogen kompressionsgezündeter Verbrennung, S. Hen-
sel, Diss. Universität Karlsruhe, 2009
[12] Thermische Randbedingungen I, FVV-Abschlussbericht Vorhaben
505, 1993
[13] Druckindizierung bei klopfender Verbrennung, Bertola, A., Stadler, J.,
Walter, T., Wolfer, P., Gossweiler, Chr., Rothe, M., 7. Internationales
Symposium für Verbrennungsdiagnostik, 2006
[14] Lubricant Reactivity effects on Gasoline Spark Ignition Engine Knock,
Amann, M., Alger, T., SAE 2012-01-1140

198
[15] Investigation of Engine Oil Effect on Abnormal Combustion in Tur-
bocharged Direct Injection - Spark Ignition Engines, Radwan, M. S.,
Elfeky, S., Abu-Elyazed, O., SAE 2012-01-1599

199
11 Comparative analysis of low-speed
pre-ignition phenomena in SI gasoline
and dual fuel diesel-methane engines
Jean-Marc Zaccardi, David Serrano

Abstract
Low speed pre-ignition (LSPI) in downsized spark-ignition engines has
been studied for more than a decade but no definitive explanation has
been found regarding the exact sources of auto-ignition. No single me-
chanism explains all the occurrences of LSPI and each engine has to be
considered as a particular case supporting different conditions for auto-ig-
nition.

In a different context, dual fuel Diesel-methane engines are now conside-

red for being applied to passenger cars since low CO2 emissions can be
reached by combining conventional compression ignition engines charac-
terized by high compression ratios with methane‘s high knocking resistan-
ce and low carbon content. As for SI engines, LSPI remains, however, one
of the main limitations of low-end torque for dual fuel engines.

This article aims at highlighting the similarities and discrepancies between

LSPI phenomena in SI gasoline and dual fuel Diesel-methane engines. To
this end, experimental results recorded on single and multi-cylinder dual
fuel engines are compared with previously obtained results on SI engines.

Similar results to those obtained in SI engines have been obtained in Dual

Fuel mode regarding the overall LSPI combustion process, and regarding
the positive effects of EGR and combustion timing on the LSPI resistan-
ce. Besides, through the use of a purely gaseous mixture in very specific
conditions (in terms of charge motion, compression ratio and ignition pro-
cess), dual fuel engines offer novel perspectives on some proposed LSPI
mechanisms in SI engines.

This comparative analysis shows that the main source of LSPI in dual fuel
engines is related to a spontaneous gas phase auto-ignition within the air
/ methane mixture, this auto-ignition being triggered by stochastic effects
locally inducing higher reactivities. The impact of residual burned gases is

200
notably highlighted via parametric variations on different engine hardware
configurations.

1. Introduction
Nowadays, Diesel Compression Ignited (CI) engines are widely used for
transportation in Europe because they offer higher fuel efficiency and lo-
wer CO2 emissions than Spark Ignited (SI) engines at equal rated pow-
er. However, stringent emission regulations on NOx and PM are making
after-treatment systems for Diesel CI engines expensive and more and
more complex. That is why alternative paths are explored like the dual
fuel concept.

The general principle consists in initiating the combustion of a main fuel

with a pilot injection of a secondary fuel characterized by a high ceta-
ne number like Diesel fuel). The main fuel should have therefore a high
octane number (like gasoline, ethanol or methane) to maintain the high
compression ratio of CI engines, synonym of high thermal efficiency. Such
concepts have already been well investigated by IFPEN with the combina-
tions of Diesel / gasoline and Diesel / methane fuels [1][2][3].

The concept of dual fuel Diesel-methane is well known for example in he-
avy duty applications as the high octane number of methane allows con-
verting easily existing CI engines without varying compression ratio, thus
reducing adaptation costs. However, only limited Diesel substitution ratios
between 30 to 70 % [4][5] are used on this type of heavy duty applications
while recent studies have shown that dual fuel combustion is possible
over a large operating range with high Diesel fuel substitution ratios, up to
99 % [2][3][6] allowing high CO2 savings. In this perspective, IFPEN has
applied the dual fuel concept to a passenger car multi-cylinder engine [3].
From mid to high loads, dual fuel combustion is operated at stoichiome-
try with huge CO2 benefits beyond theoretical diesel full substitution by
methane. From low to mid loads, dual fuel combustion is operated at lean
burn with EGR.

However, dual fuel combustion suffers from major combustion issues that
need to be addressed. At low loads, significant unburned methane emissi-
ons can result from both combustion chamber geometry and combustion
conditions. Indeed, dual fuel engines are generally derived from Diesel
engines that use pistons with high crevices that can trap port-injected met-

201
hane (especially high top land heights that are necessary because of steel
inserts and cooling oil galleries [7]). Emissions and efficiency also depend
on flow motion inside the combustion chamber: Carlucci et al. [8] have
shown that swirl seems to be adapted to a fast combustion development.
Besides, if the methane / air mixture is too lean, meaning out of flam-
mability limits, combustion misfiring and flame quenching both on walls
and bulk can occur, causing high unburned HC emissions. Ishiyama et al.
[9] and Königsson et al. [6][7][10] gave lean limits values for equivalence
ratio around 0.45 for avoiding misfires and around 0.55 for reasonable
unburned HC emissions. It seems that clean dual fuel combustion is a
challenging task. Many scientists have tried to list the effects of all engine
parameters like the Diesel pilot Start Of Injection (SOI), Diesel rail pressu-
re or Exhaust Gas Recirculation (EGR) rate [4][6][9].

Other indirect issues exist: some recent studies have investigated the Die-
sel injector fouling issue caused by high temperatures reached by the in-
jector nozzle [10]. Indeed, the reduction in Diesel flow through the nozzle
decreases heat dissipation, thus inducing high temperatures and causing
the formation of deposits inside the nozzle. This item remains a key issue
for future industrialization of dual fuel solutions.

At high loads, maximal performance can be reduced by the occurrence of

knock [11] but also of Low-Speed Pre-Ignition (LSPI) [3][12][13], limiting
thus the compression ratios that can be used compared to the Diesel ba-
seline engines.

LSPI has been widely observed on highly charged SI gasoline engines

over the last 10 years. It has sometimes been referred to as Super-Knock
[14][15] or Mega-Knock [16] but more often as LSPI [17][18][19][20][21]. In
fact, “mega-knock” pressure oscillations appear because of the premature
and uncontrolled ignition.

The first occurrences of this abnormal combustion dates back to the very
beginning of years 2000 in Europe but surprisingly, there is still no con-
sensus on the exact causes of the irregular auto-ignition. Considering
the great variety of engines, fuels and operating conditions used for the
reported analyses of LSPI, it is however not surprising that different hy-
potheses emerged. In fact, the authors are deeply convinced that the-
re has never been only one cause to LSPI and that each engine should
rather be considered as a particular case of the more general theory of
auto-ignition [22]. The increase in temperature and pressure levels with

202
load explains indeed the highly unstable state of the mixture at the end
of the compression stroke. In these conditions, spatial, cycle-to-cycle and
cylinder-to-cylinder fluctuations can easily unbalance the system and lead
to uncontrolled auto-ignitions.

The experimental study of this last form of irregular combustion is compli-

cated by its variable occurrence mode. It is generally a rare and occasio-
nal event but it can also appear on a regular basis and even sometimes
alternatively with normal combustions [23][24]. Defining relevant test and
quantification methodologies and performing reliable analyses are thus all
the more challenging that LSPI events are usually very violent and cannot
be controlled as simply as knocking combustion. Moreover, due to its in-
tensity in terms of maximal cylinder pressure, LSPI is now widely seen as
a crucial issue and a strong limit for the downsizing of SI engines [25][26].

This article aims at highlighting the similarities and discrepancies between

LSPI phenomena in SI gasoline and dual fuel Diesel-methane engines.
Indeed, basic considerations on hardware configurations, mixture forma-
tion processes and ignition in both types of engines can first enhance
our understanding of uncontrolled auto-ignition at high load. Then, the
detailed comparison of events recorded in both cases can bring a novel
perspective on some proposed LSPI mechanisms and help in understan-
ding possible conflicting theories.

To this end, experimental results recorded on single and multi-cylinder

dual fuel engines will be compared with previously obtained results on SI
engines. Engines and fuels used for this analysis are first introduced in
section 2. Then, section 3 gives an overview of the combustion proces-
ses followed during normal operation and pre-ignition in Diesel-methane
fuelled engines. Section 4 is dedicated to a comparative analysis of LSPI
events recorded on SI and dual fuel engines during parametric and hard-
ware variations. This comparative analysis will also be completed with a
literature review discussing the main LSPI mechanisms. Section 5 finally
points out the most probable LSPI mechanism associated with dual fuel
engines.

203
2. Experimental apparatus

2.1. Engines
Results presented in this article have been obtained on three different
engines: a multi-cylinder SI engine and two Diesel engines used for dual
fuel operation (see Table 1).

As for dual fuel engines, the tests have been carried out on a single cylin-
der engine (displacement of 499 cm3) and on 4-cylinder PSA 1560 cm3
DV6 Diesel engine (80 kW Euro 6 variant using a variable nozzle turbine),
both being originally designed for Diesel fuel operation. Methane could be
injected directly [27] but in our case port fuel injection was preferred so
as to keep possible full Diesel operation. That way, only minor changes
were necessary on the baseline engines to allow dual fuel operation. Gas
injection systems were notably implemented without modifying the glo-
bal engines packaging. In both cases, the gas is injected into the intake
duct of each cylinder with Start Of Injection (SOI) set at 360 CAD before
combustion Top Dead Centre (TDC), just upstream of the inlet valves.
Thus, methane is assumed to be homogeneously distributed in each of
the cylinders.

For all the dual fuel tests, the Diesel injection strategy was simplified by
keeping only one sole injection. The injection duration was kept as low as
possible in order to maintain the lowest and stable Diesel pilot quantity.
The injection pressure was set to 400 bar for all the tests. The Diesel mass
fraction was then rather low (around 1 to 6 % except for very low loads).
The equivalence ratio was regulated via the measurement of a wideband
lambda sensor and by adjusting only the methane injection duration to
maintain the selected equivalence ratio.

Several configurations were also tested on each of these engines (in

terms of compression ratios or swirl motions).

204
 4-cylinder PSA Single cylinder 4-cylinder PSA
DV6d IFPEN EP6Dt
Combustion mode CI & Dual fuel CI & Dual fuel SI
Displacement 1560 cm3 499 cm3 1598 cm3
Bore x Stroke 75 x 88.3 mm 85 x 88 mm 77 x 85.8 mm
Nominal comp. ratio 16:1 17:1 10.5:1
Connecting rod 136.80 mm 145.00 mm 138.54 mm
length
Number of valves 2 4 4
IVO @ 1 mm lift 17 CAD ATDC 33 CAD ATDC variable timing
IVC @ 1 mm lift 10 CAD ABDC 30 CAD ABDC variable timing
EVO @ 1 mm lift 16 CAD BBDC 23 CAD BBDC 22.5 CAD BBDC
EVC @ 1 mm lift 10 CAD BBDC 17 CAD BTDC 7.5 CAD BTDC
Diesel FIS (DI) Bosch CRI 2.2 Bosch CRI 3.1
(400 bar) (400 bar)
Gas FIS (PFI) Bosch NGI Landi Renzo
(9 bar abs.) (3 bar abs.)
Gasoline FIS (DI) Bosch HDEV 5
(up to 200 bar)
EGR Cooled high pres- Cooled high pres- Cooled low pressure
sure sure

Table 1: Dual fuel and SI engines specifications

Regarding SI engines, results presented in this article have been recorded

on a serial production turbocharged engine (displacement 1600 cm3, see
Table 1) being run most of the time with direct injection in lateral position.

Different configurations have been tested in terms of injection strategy,

charge motion (variable tumble level) and dilution (cooled low pressure
EGR). For all of these configurations, optimal injection and ignition set-
tings were determined to minimize exhaust emissions and to optimize the
combustion timing.

No after-treatment system was used on these engines. Therefore, to be

representative of full production exhaust lines, the exhaust back-pressure
was adjusted with a dedicated throttle (the same approach was followed
for the single cylinder engine fed by an external sonic neck system).

205
Safety and stability limits were also set to protect the engine integrity
during the tests (at least during normal operation at full load, see Table
2). Regarding dual fuel engines, these constraints are not different from
Diesel operation, as no major modification of engine hardware has been
made.

Dual fuel engines SI engine

Average maximum ≤ 160 bar ≤ 120 bar
cylinder pressure (for
each cylinder)
CoV of IMEP ≤ 3% ≤ 3%
Upstream turbine ≤ 810 °C ≤ 950 °C
temperature
Downstream turbine ≤ 700 °C ≤ 850 °C
temperature

Table 2: Limits of engine operation for multi-cylinder engines (exhaust

temperature limited at 750 °C for the single cylinder engine)

The temperature limits were not reached with the multi-cylinder SI engi-
ne at low engine speeds, even for loads greater than the baseline serial
production engine. However, regarding dual fuel engines, the Diesel SOI
is a critical parameter and must be adjusted to respect both the maximal
pressure limit (reached for early injections) and the temperature limit (re-
ached for late injections).

2.2. Fuels
The multi-cylinder SI engine was run with a commercially available E5
fuel (RON 97) and dual fuel engines were run with an association of pure
methane and a standard EN590 European Diesel fuel (B5 and Cetane
Number > 51).

As shown by Figures 1 and 2, auto-ignition delay times i of methane are

more than ten times greater than those of gasoline (represented here by
a ternary reference fuel consisting of n-heptane, iso-octane and toluene).
The higher auto-ignition de-lay and octane number of methane (~130)
allow maintaining knock-free operation over a broader range of engine

206
speeds and loads than with gasoline. Higher com-pression ratios can also
be used while reaching higher maximal brake torque and keeping optimal
combustion timings at high loads.

Figur1: Calculated auto-ignition Figur2: Calculated auto-ignition

delays times i for = 0.7 and xdil = delays times i for = 1.0 and xdil =
0 (Chemkin calculations) 0 (Chemkin calculations)

It can also be noted that methane presents great advantages compared

to other high octane number fuels. Methane is notably the lowest carbon
content fossil fuel, and reduces significantly CO2 emissions compared to
Diesel fuel (~26 % theoretically and more if it comes from a renewable
source like biogas).

2.3. Measurements and control

Conventional measurements of temperature and pressure were made on
the cool-ant, oil, fuel, intake and exhaust paths. Based on these measure-
ments, and depending on the engine used, safety and stability limits were
set to protect the engine integrity during the tests.

Fast pressure transducers were also used to record the instantaneous

intake and exhaust pressures. Cylinder pressure has been recorded with
different angular resolutions: from 0.1 CAD for analyzing knocking com-
bustion, up to 1 CAD when statistical analysis over several minutes was
required.

207
The engines were controlled with IFPEN in-house management system
devices. These fully by-passed systems allowed flexible engine control of
all engine devices and notably considering that all of the engines have to
use two injection systems (methane and Diesel for dual fuel engines, PFI
and DI for SI engines).

3. General overview of the combustion process

3.1. Regular dual fuel combustion

The dual fuel concept has been applied by minimizing the Diesel fuel
mass fraction (pilot injection size) and thus maximizing methane proporti-
on for the whole engine operating range. In these conditions, the dual fuel
combustion process can be described by four different steps:

(1) The homogeneous carburetted mixture of air and methane (and

sometimes with EGR too) is compressed but should not auto-igni-
te because of its high auto-ignition delay.

(2) The combustion is then initiated by the direct injection of a Diesel

fuel micro pilot quantity which auto-ignites during the end of the
compression stroke. Diesel fuel is then only used as a high ener-
gy chemical ignition device and Diesel SOI is the trigger of the
dual fuel combustion (excluding Diesel fuel auto-ignition).

(3) Diesel combustion occurs (which can be composed by premixed

and diffusion phases) with part of the mixed methane due to the
mixture entrainment into the Diesel spray

(4) Methane combustion around the Diesel fuel spray [2] occurs. This
last step involves complex mechanisms (including multi-propaga-
tion flames) that are currently under investigations at IFPEN.

208
3.2. SPI in dual fuel engines

3.2.1. Critical operating range

Similarly to SI gasoline engines, LSPI generally appears on dual fuel en-
gines between 1500 and 2500 rpm and for IMEP above 20 bar. Lower en-
gine speeds may also be concerned if high low-end torques are targeted.

Figure 3 gives an illustrative example of full load curve obtained on a

specific configuration of the 4-cylinder PSA DV6d engine running on dual
fuel operation. Black-filled symbols indicate the operating points on which
LSPI could be recorded, and occurrence frequencies are also mentioned.

As already observed in SI engines, no specific pattern of occurrence can

be identified in dual fuel engines. Indeed, LSPI can occur within sequen-
ces alternating with regular combustions, or as isolated events.

It can also be mentioned that, until now, no cycle-to-cycle determinism has

been identified in the occurrence of LSPI which remains thus a pure stoch-
astic event. As it will be shown in next sections, causal relations can be
found between mean operating conditions and LSPI occurrence frequen-
cy both in SI and dual fuel engines, but no direct determinism between
successive cycles can explain the mixture premature auto-ignition [28].

209
3.2.2. LSPI combustion process
The LSPI combustion process in dual fuel engines is quite similar to the
one that can be observed in SI engines. Indeed, a three steps combustion
process can be identified as illustrated by the example in Figures 5 and 6.

(1) First, the beginning of combustion can be observed before the

Diesel pilot injection (as it starts before the spark in SI engines).
It means that some heterogeneities, or hot spots, within the mix-
ture have sufficient energy to trigger an auto-ignition. This first
step represents the critical characteristic of LSPI since, as for SI
engines, several mechanisms can be assumed to explain how the
mixture‘s reactivity can be enhanced to the point of being able to
trigger an auto-ignition.

(2) After the first auto-ignition of fresh gases, a second phase similar
to a classical flame front propagation can be observed. The heat
release rate associated with this second step is indeed compara-
ble to the one obtained during normal com-bustion.

(3) Finally, a third phase can be observed in the form of a bulk au-
to-ignition. The fresh gases remaining in the combustion cham-
ber are indeed highly com-pressed and heated up and, therefore,
a second knocking-like auto-ignition occurs. Depending on the
combustion timing and on the load, this auto-ignition can lead to
severe engine damages. As a remark, it can also be mentioned
that these knock oscillation frequencies do not seem to be related
to the engine bore as for SI engines but more to the piston bowl
diameter.

This combustion process in SI engines has already been described by

several authors (see [17][23][26] for example). First analyses were based
on cylinder pressure and on heat release calculations but advanced dia-
gnostics are required to confirm these approaches. That is why endosco-
pic visualizations, and in some cases 3D CFD, have been performed to
further investigate the LSPI process.

210
Considering the auto-ignition delay times of air / CH4 mixtures (see Figu-
res 1 and 2 for example), the temporal auto-ignition onset can be estima-
ted by the Livengood-Wu integral [29].

(1)

This approach shows, however, that the homogeneous gas phase auto-ig-
nition of air / CH4 mixture is theoretically impossible before the Diesel pilot
injection. The auto-ignition delay times associated with typical operating
conditions (mean pressure and temperature) in dual fuel engines are too
high for auto-ignition to be expected and this despite the use of high com-
pression ratios. It is the case for the example illustrated in Figures 5 and
6. When normal combustion cycles are considered, the Livengood-Wu
integral reaches unity around 380 CAD, which means that knocking com-
bustion is not supposed to occur before the end of normal combustion. It
also means that there is absolutely no chance that an homogeneous gas
phase auto-ignition can explain the pre-ignition cycle starting at ~ 350
CAD.

Consequently, and coherently with similar observations made in SI en-

gines fuelled with gasoline, only local heterogeneities can explain the

211
premature auto-ignition of the charge, these heterogeneities being more
reactive than the surrounding mean mixture.

It is also noteworthy that the combustion stability obtained with dual fuel
operation is far better than that usually obtained in SI engines, especially
at high loads with delayed combustion timings (see exemplary results in
Figure 7). Critical pre-ignition events characterized by very early combus-
tion timings are thus all the more easily identifiable (see exemplary results
in Figure 8).

4. Results

4.1. Impact of engine settings

4.1.1. Diesel start of injection

Regarding SI engines, it is generally observed that delaying the spark
advance leads to higher LSPI frequencies [30]. Among others, Adomeit
et al. [31] have notably used extremely retarded spark timings to favour
critical pre-ignition conditions during their tests. It has also been shown by
Rothenberger et al. [32] that reducing the spark advance from its knock
limit could lead to a significant increase in pre-ignition frequency. In all ca-
ses, the effect on LSPI is not due to the spark itself but to the longer time
available for preliminary auto-ignition reactions to occur within the fresh

212
gases, and to side effects of late spark advance and associated combus-
tion phasing on exhaust temperatures.

Coherently, the Diesel fuel pilot injection is certainly the most critical set-
ting impacting LSPI in dual fuel engines. This parameter is, to a certain
extent, the equivalent of spark advance in SI engines for which it has
already been shown that delayed spark timings lead to more LSPI events.

Figure 9 illustrate the normalized LSPI occurrence frequency as a function

of the Diesel fuel SOI settings for the multi-cylinder engine (similar results
have been ob-tained with the single cylinder engine). Normalized frequen-
cies have been illustrated here because absolute values are too different
to be directly compared.

These results show that delaying the Diesel fuel SOI has the same detri-
mental effect as delaying the spark advance in SI engines. This injection
setting is even more critical than the spark advance since one crank angle
degree is enough to significantly decrease the LSPI frequency.

Similarly to SI engines, two effects can be hypothesized to explain this

result. Firstly, delaying the Diesel fuel SOI gives more time for chemical
reactions to occur within the air / methane mixture. Therefore, heteroge-
neities with sufficient reactivity may auto-ignite more easily. Secondly, de-
laying the combustion timing also leads to higher exhaust temperatures,
and consequently to higher temperatures of residual burned gases.

The exhaust temperatures (Texh) and burned gases rates (xbg) for each
operating point are mentioned in Figure 9. A correlation between the LSPI
frequency and this temperature can be observed. It can thus be hypothe-
sized that gas phase auto-ignitions enhanced by higher temperatures of
burned gases are responsible for LSPI events.

213
As a remark, it can be noted that this effect of Diesel SOI timing is not
related to any kind of direct determinism between successive cycles but
only reflects a mean effect.

4.1.2. Exhaust Gas Recirculation

Dilution with cooled EGR has already proved to be an efficient solution
in order to enhance the auto-ignition resistance of air / fuel mixture in SI
gasoline engines (both for knocking combustion and for LSPI [17][33][34]).
Figure 10 illustrates for example the decrease in pre-ignition frequency
allowed by the addition of EGR (results obtained on the multi-cylinder SI
engine with two intake temperatures Ti). It can be shown that this decrea-
se is directly linked to the earlier combustion timings allowed by EGR, and
by the chemical and thermal effects of dilution.

Several effects must, however, be accounted for. Firstly, cooled EGR has
indeed a significant positive effect on knocking combustion even at mo-
derate loads, i.e. before LSPI appears. The earlier propagation of normal
combustion is thus beneficial to consume the carburetted mixture before it
reaches a critical state for auto-ignition.

EGR favourably modifies as well the mixture heat capacity, and therefore
its temperature during compression. The dilutant composition itself can
also affect the chemical processes leading to auto-ignition (notably NO

214
and C2H2 within the burned gases). That is why different behaviours can
be observed with and without catalytic converter in the EGR path [35].

Besides, it has to be noted that the mean cylinder pressure during com-
pression necessarily increases with the EGR rate so as to maintain the
desired air flow rate and torque. This rise in pressure is, however, coun-
terbalanced by the increase in trapped mass inside the cylinder. Conse-
quently, the in-cylinder temperature at TDC is nearly the same whatever
the EGR rate. Considering the combination of all these effects, the additi-
on of EGR has finally a positive impact on LSPI.

Tests made on dual fuel engines, with cooled high pressure EGR systems
and in various conditions, have also shown that the higher the EGR rate
is, the lower the LSPI occurrence frequency is. Figure 11 illustrates for
example the decrease in LSPI frequencies as the EGR rate increases for
three different engine configurations (different hardware configurations,
loads and settings in terms of methane / Diesel ratios, air / fuel ratios
and SOI, see Table 3). Absolute occurrence frequencies obtained on each
configuration are too different to be directly compared. However, normali-
zed frequencies illustrated here show that EGR can help in avoiding LSPI
without in-creasing cycle-to-cycle fluctuations (IMEP stability is kept in-
deed below 2 % and even below 1 % for most of these operating points).

215
 Table 3: Operating conditions for results illustrated in Figure 11

For the first two configurations ( = 0.8 - Baseline piston and = 1 - Piston
A), the decrease in LSPI frequency is due to the increase in EGR rate, but
also to SOI set-tings. Indeed, the Diesel pilot injection is set earlier as the
EGR rate increases which helps in limiting LSPI events (as indicated in
previous section, see Figure 9). The Diesel pilot injection is, however, kept
constant for the third configuration ( = 1 - Pis-ton B). The decrease in LSPI
frequency is thus only due to dilution with EGR.

Similarly to what has been observed with SI engines, dilution with EGR
helps in controlling the mixture auto-ignition delay during compression.
Despite the increase in intake pressure when EGR rate increases, the
mixture temperature is kept roughly constant because of the trapped
mass inside the cylinder also increases. It is therefore possible to improve
the auto-ignition resistance thanks to the chemical effect of dilution.

216
Further investigations should be performed to evaluate the interest of
using a three-way catalyst, whether in the same high-pressure EGR sys-
tem, or in a cooled low pressure system.

4.1.3. Fuel / air equivalence ratio

The impact of Fuel / Air Equivalence Ratio (FAER) on LSPI in gasoline
SI engines is not straightforward. Regarding perfectly premixed homoge-
neous mixtures, the auto-ignition delay depends on the equivalence ratio
(see Figures 1 and 2): the leaner the mixture is, the higher the auto-igni-
tion delays are. However, in SI engines, this expected link between the
mean fuel / air equivalence ratio and LSPI can not always be observed
because auto-ignition is locally triggered on hot-spot(s) having a higher
reactivity than the surrounding mean mixture.

Besides, other side effects must also be considered. For instance, com-
bustion chamber fouling can be favoured by rich mixtures. LSPI events
originating from heated deposits can thus appear, independently from the
global fuel / air equivalence ratio. It is also quite difficult to characterize
the effects of lean mixtures because cycle-to-cycle fluctuations can sig-
nificantly increase for FAER below 0.9. In these conditions, LSPI events
may not be solely dependent on the mean FAER, but also for instance on
cycle-to-cycle fluctuations of burned gases rate, temperature and chemi-
cal composition.

Dual fuel engines benefit from a better combustion stability than SI engi-
nes, especially when lean mixtures are used. It is indeed possible to reach
FAER around 0.6 with very low cycle fluctuations (CoV of IMEP below 1.5
%). Consequently, it is easier to analyze the evolution of LSPI as a functi-
on of the global FAER.

Figure 12 illustrates the impact of FAER on LSPI events recorded in the

single cylinder dual fuel engine at 2000 rpm, 21 bar IMEP. An increase in
the pre-ignition frequency can first be observed as the FAER decreases
from the reference case = 1 to the case = 0.75. It can also be seen in this
figure that the same SOI setting (5 CAD BTDC) was used for these four
operating points ( = 1, 0.9, 0.85 and 0.75). For homogeneous premixed
mixtures, a decrease in FAER should result in an increase in auto-ignition
delay times (see Figure 13) which seem to be contradictory to what can be
observed in Figure 12. However, the comparison illustrated in this figure

217
is not only based on different FAER, but also on different pressure levels
inside the combustion chamber. Indeed, when FAER decreases, the tar-
geted load is maintained by increasing the air flow rate and the intake
pressure (from 2.03 bar @ = 1, to 2.46 bar @ = 0.75 and 2.93 bar @ =
0.6, see Figure 14). Consequently, the in-cylinder pressure is significantly
increased during compression but it can also be noted that, in the same
time, the mixture temperature is kept roughly unchanged thanks to the
increase in trapped mass (the increase of temperature before TDC is only
~5°K, not illustrated here). The positive effect of FAER on auto-ignition
delay is, however, counterbalanced by the increase in pressure. Indeed,
it can be seen in Figure 13 that a 40 bar increase in pressure has more
effect on the auto-ignition de-lay times than a decrease in FAER from 1
to 0.7.

Below = 0.75, a decrease in LSPI occurrence frequency can be observed

in Figure 12. This decrease can be explained by three effects.

(1) The use of earlier SOI settings, which are necessary to maintain
optimal combustion timings and conventional combustion stabili-
ties in diluted mixtures;

(2) The decrease in FAER: the negative impact of pressure on au-

to-ignition delays can be partly counterbalanced by the positive
impact of FAER below 0.7;

(3) The continuous decrease in the residual burned gases rates and
temperatures (see Figure 15).

Consequently, despite the increase pressure, all these positive effects re-
sult in a better auto-ignition resistance and consequently in fewer LSPI
events.

Dual fuel engines are well adapted to the study of FAER effects because
of the nearly homogeneous air / CH4 mixtures and the good combustion
stability compared to SI engines. This FAER sweep is a good example of
all the interactions that have to be considered when analyzing auto-igniti-
on in practical engine conditions. It is not sufficient, and even misleading,
to consider only one parameter at once. Several effects must be analyzed
simultaneously to explain the unstable state of the mixture.

218
 Figure 12: Impact of FAER on Figure 13: Calculated auto-igni-
LSPI frequency (single cylinder tion delays times i for air / CH4
engine, CR = 17:1, 2000 rpm, 21 mixtures (xdil = 0 %, Chemkin
bar IMEP) calculations)

Figure 14: Pressures at the intake Figure 15: Residual burned gases
and befor TDC during the FAER rate and temperature as a functi-
sweep on of FAER

This analysis should be completed with rich mixtures so as to confirm the

chemical effects of FAER on LSPI. Increasing the FAER should result
indeed in a decrease in auto-ignition delay times.

219
4.1.4. Diesel fuel mass fraction
The last parameter variation performed on the single cylinder engine con-
cerns the Diesel fuel mass fraction. In terms of efficiency, CO2 emissions
and pollutants, the lowest possible Diesel fuel quantity should be used on
the whole engine operating range. Even if there is no interest in increasing
the Diesel fuel mass fraction, simple parametric variations were perfor-
med to analyze its effect on LSPI.

Results at full load show that low Diesel fuel mass fractions are also com-
patible with very low LSPI occurrence frequencies. This tendency can be
observed for instance in Figures 16 and 17 corresponding to results ob-
tained on the single cylinder engine at 2000 rpm and for FAER of 1 and
0.8. In the first case ( = 1, see Figure 16), the same SOI setting was used
with Diesel fuel mass fractions from 2 up to 15 %. The minimal Diesel fuel
mass fraction was chosen here as the reference configuration since it
represents the optimal setting in terms of efficiency.

As long as the Diesel SOI can be kept constant, it can be observed in both
cases that the LSPI frequency increases when the Diesel fuel mass frac-
tion increases. However, operating points with Diesel fuel mass fractions
above 15 % in the first case ( = 1) and above 5 % in the second case ( =
0.8) were impossible to reach with the same settings of Diesel SOI. There-
fore, coherently with results presented in previous sections, the only way
to improve the resistance to LSPI and to run the engine with higher Diesel
full mass fractions was to advance the start of Diesel injection. In the first
case for instance ( = 1), increasing the SOI from 5 to 7 CAD is enough to
avoid LSPI and to increase the Diesel fuel mass fraction from 15 to 25 %.
These are, however, clearly not the optimal settings for dual fuel combus-
tion since the indicated efficiency is downgraded by the use of high Diesel
fuel mass fractions.

220
 Figure 16: Impact of Diesel fuel
mass fraction on LSPI frequency
(single cylinder engine, CR = 17:1,
2000 rpm, 21 bar IMEP, = 1)
Figure 17: Impact of Diesel fuel
mass fraction on LSPI frequency
(single cylinder engine, CR =
17:1, 2000 rpm, 21 bar IMEP, =
0.8)

The decrease in efficiency as the Diesel fuel mass fraction increases in-
duces an increase in intake pressure in order to maintain the targeted
load. This increase is, however, limited (~5 bar at TDC between the extre-
me settings with Diesel fuel mass fractions of 2 and 25 %), and the tem-
perature of fresh gases during combustion remains unaffected because of
the increase in trapped mass.

It can also be noted that the air / fuel ratio of the premixed mixture (air and
methane) decreases when the Diesel fuel mass fraction increases. This
should lead therefore to a higher resistance to auto-ignition, which is not
observed here. It means that a stronger effect than that of the air / fuel
ratio is induced by the increase in Diesel fuel mass fraction.

In fact, it seems that LSPI is only correlated here to the residual burned
gases temperatures (see Figure 18). As long as the Diesel SOI is kept un-
changed, the exhaust temperature increases as the Diesel fuel mass frac-
tion increases. Indeed, the methane combustion is progressively slowed
down when the Diesel mass fraction in-creases, which results in higher
exhaust temperatures. Then, when Diesel SOI is advanced, the residual
burned gases temperature can be decreased and the combustion timing
improved, limiting thus highly reactive hot spots and giving more chance
to consume the fresh gases mixture before auto-ignition occurs.

221
The exhaust temperature variations may seem low but the hot residual
burned gases act as highly reactive hot spots within the fresh mixture
and their temperature increases during compression. Their effect on the
auto-ignition resistance might thus be significant (especially here with a
high compression ratio of 17:1). 3D CFD tools could be useful to simulate
the gas exchange and mixture preparation processes, and to better analy-
ze the formation of mixture heterogeneities. Besides, cylinder-to-cylinder
fluctuations should also be taken into account since these spatial tem-
perature variations can be amplified by an unequal distribution of fresh
gases (air and EGR) resulting from the intake manifold shape.

The impact of the Diesel fuel to methane ratio needs anyway further inves-
tigations. The combustion process in dual fuel engines is still not exactly
known and the way CH4 and Diesel fuel are consumed by diffusion fla-
mes, flame propagation and auto-ignition evolves when the different fuels
mass fractions change. IFPEN is currently carrying out further investiga-
tions with advanced tools (like optical diagnostics) in order to understand
these complex mechanisms involved in dual fuel combustion process.

222
4.2. Impact of engine hardware configuration

4.2.1. Compression ratio

In the case of gasoline SI engines, the compression ratio has a significant
and direct effect on auto-ignition. The rise in temperature and pressure le-
vels during compression as the compression ratio increases, results more
easily in knocking combustion at middle loads. Then, at high loads, these
high pressure and temperature levels, associated with delayed combusti-
on timings, lead to more LSPI events. Parameter variations have then the
same effects on LSPI but at different absolute levels.

Interestingly, this critical relationship between the compression ratio and

LSPI has not been observed in dual fuel engines (at least not on the tested
configurations). On the contrary, a decrease in LSPI frequency has occur-
red when the compression ratio of the multi-cylinder engine was raised
from 16:1 to 17:1. This remarkable result is illustrated in Figure 19 at 1750
and 2000 rpm full load. Two sets of data are illustrated in this figure for the
case CR = 17:1. The first one corresponds to results obtained with opti-
mal SOI settings (early injections allowing the cylinder pressure to reach
its maximal limit at 160 bar), and the second one corresponds to results
obtained with delayed SOI settings.

In accordance with previous results, the negative effect of late Diesel SOI
on LSPI can be observed. It can also be observed that higher pre-ignition
frequencies have been obtained with a compression ratio of 16:1. This
offers interesting perspectives in terms of trade-off between higher effi-
ciency and higher resistance to LSPI.

As expected, the cylinder pressure during compression is higher with a

compression ratio of 17:1 than with a compression ratio of 16:1. However,
because of the different trapped masses and volumes in these two cases,
the fresh gases temperature during compression does not significantly
increase. There is thus no clear correlation between LSPI, and pressure
and temperature levels inside the combustion chamber.

On the other hand, it can be shown that the LSPI frequency is correlated
with the residual burned gases rates and temperatures. Figure 20 shows
that the lowest LSPI frequency obtained with a compression ratio of 17:1
and optimal SOI settings also correspond to the lowest exhaust tempera-
ture and to the lowest residual burned gases rate.

223
Figure 19: Impact of compression ratio on LSPI frequency (4-cylinder en-
gine, = 1, EGR rate 0 %, methane mass fraction from 96.5 to 97.5 %)
Figure 20: LSPI frequency as a function of the residual burned gases rate
at 1750 rpm

It also has to be noted that changing the compression ratio on a mul-

ti-cylinder engine with given turbocharger and air cooler may result in dif-
ferent air flow rates, back pressures and intake temperatures for the same
operating load. These side effects should also be taken into consideration
when analyzing LSPI.

4.2.2. Swirl motion

The impact of charge motion has been previously studied in SI engines
and it has been found that intense charge motion can help in limiting LSPI
events (see example in Figure 21). Tumble motions can have a positive
effect on the mixture preparation process and enhance the mixture homo-
geneity, limiting thus the formation of hot spots with higher reactivities (in
terms of chemical composition or temperature). Besides, the control of the
interactions between the air flow and the injection spray is also of great
importance to limit the formation of fuel droplets and fuel impacts on the
walls, and consequently to avoid oil / fuel interactions.

In most cases, only tumble motions have been investigated but combina-
tions of swirl and tumble motions have also been reported [36].

224
The effect of charge motion on LSPI is not straightforward, whether it be
in SI or in dual fuel engines. The mixture preparation process strongly
depends on the charge motion in SI engines, especially in direct injection
engines. The mixture‘s reactivity (in terms of equivalence ratio, tempera-
ture, or burned gases content) can thus be significantly altered. Several
critical parameters are impacted by the air motion:

(1) The global air / fuel mixture homogeneity;

(2) The interactions between the fuel spray and the air flow, and con-
sequently the formation of liquid fuel films on the walls;

(3) The mixing of fresh with residual burned gases;

(4) The convective heat exchanges at the walls;

(5) The pressure during compression since high levels of charge mo-
tion are generally associated with low flow capabilities (and the-
refore higher intake pressure is needed to reach a given level of
torque).

A „high“ swirl configuration has been tested on the 4-cylinder dual fuel
engine. In the Tippelmann framework, this swirl level is ~0.46 (while the
serial production baseline configuration is ~0.32).

225
As illustrated by Figure 22, the lowest LSPI occurrence frequencies have
been ob-tained with the baseline configuration, while the high swirl confi-
guration has lead to the highest LSPI frequencies. It can also be noticed in
Table 4 that maximal IMEP reached with this latter configuration are lower
than those reached with the baseline one (and this despite the use of 5 %
EGR with the high swirl configuration).

Ns = 0.32 (baseline) Ns = 0.46

EGR (%) 0 5
22.6 @ 1750 rpm 20.6 @ 1750 rpm
IMEP (bar) 23.2 @ 2000 rpm 22.4 @ 2000 rpm
22.4 @ 2500 rpm 22.0 @ 2500 rpm

Table 4: Operating conditions for results illustrated in Figure 22

These two cylinder heads being characterized by different trade-offs bet-

ween air flow capabilities and swirl levels, different cylinder pressure and
temperatures levels are reached inside the combustion chamber. It has
notably been noticed at 2500 rpm (where IMEP are quite comparable)
that the high swirl configuration leads to the lowest cylinder pressure and
temperature during compression, while it is associated with the highest
LSPI frequency.

226
Methane being port injected, the mixture homogeneity can be supposed to
be very good. Besides, contrary to results presented in previous sections,
neither the burned gases rates nor their temperatures are correlated with
the LSPI frequencies (these rates are very similar between the three con-
figurations, from 2.3 up to 2.4 %).

Besides, it has also been observed that the high swirl configuration led to
higher heat losses than the baseline configuration. For example, for com-
parable indicated works at 2500 rpm, calculated heat balances show that
the heat loss to the walls increases by 2.3 % with the high swirl configurati-
on (see Figure 23). Therefore, based on these observations and conside-
rations, a thermal related mechanism can be hypothesized to explain the
higher LSPI frequency obtained with the high swirl motion. By enhancing
the combustion speed and increasing the convective heat losses to the
walls, this configuration can indeed lead to higher walls temperature. Con-
sequently, the fresh gases mixture may be more heated during the intake
and compression strokes and lead to more reactive hot spots.

Figure 23: Impact of swirl motion on heat balances (4-cylinder engine, CR

= 16:1, = 1, 2500 rpm, 22.4 bar IMEP for the baseline configuration and
22.0 bar for the high swirl configuration, methane mass fraction 97 %)

Another hypothesis which was not analyzed here concern indirect effects
of swirl motion. Altering the swirl level can notably impact the Diesel fuel
auto-ignition. Because of different flow capabilities and different convecti-
ve heat exchanges, different pressure and temperature levels are reached
inside the combustion chamber. The interactions between the charge mo-
tion and the Diesel spray also lead to different local air / fuel ratios and
therefore to different auto-ignition delay times for the Diesel fuel. Conse-
quently, the whole dual fuel combustion process can be impacted (com-

227
bustion timing, heat losses, residual burned gases characteristics, etc.) as
well as the resistance to LSPI.

Despite the supposed good mixture homogeneity in terms of air / fuel

ratio, temperature stratifications and mixing of fresh and residual burned
gases are certainly very different in both cases illustrated here. All these
effects should be investigated more deeply with 3D CFD as it has already
been done in SI engines [36].

5. Discussion
LSPI events in dual fuel engines have the same critical feature as those
observed in SI engines: they depend on several interacting phenomena.
Besides, the lack of concrete control on some of these phenomena further
complicates their analysis.

Only a few observations of LSPI in dual fuel engines can be found in

the literature [12]. However, based on the results reported here, the main
pre-ignition mechanism can be identified.

In terms of settings, the Diesel start of injection is of crucial importance.

Advancing or delaying this injection by one or two crank angle degree(s)
can radically change the LSPI resistance and it has been shown that this
effect can be linked on one side to the combustion timing, and on the other
to the exhaust temperatures, and consequently to the temperature of the
residual burned gases (see section 4.1.1).

It has also been shown later that this Diesel injection timing could com-
pensate for non-optimal air / fuel equivalence ratio (see section 4.1.3)
or Diesel fuel mass fraction (see 4.1.4). In all cases, the temperature
and pressure history during compression can not fully explain LSPI. The
burned gases rate and temperature should be taken into account as well
since they account for the mixture‘s reactivity.

In terms of engine hardware configuration, it has also been shown that

higher compression ratios are not inevitably associated with higher LSPI
occurrence frequencies (at least for the tested compression ratios) con-
trary to what can be observed with SI engines. Coherently with results
obtained during settings variations, the residual burned gases rate has
proved to be a critical parameter.

228
Even if further investigations are still needed, the swirl motion variation
has also pointed out that the convective heat exchanges at the walls, and
therefore the mixture‘s temperature, could also be directly responsible for
some LSPI events.

At this stage, to complete the analysis, it is also possible to make a com-

parative analysis of the other main reported sources of LSPI in SI engines
[31] with what has been observed in dual fuel engines.

(1) Critical hot spots: the spark plug, the exhaust valves and the piston
head are often cited as potential hot spots in the combustion chamber,
and thus as potential trigger of auto-ignition. It can be recalled first that
dual fuel engines used here are derived from Diesel serial production
engines. Therefore, they do not use any spark plug and original glow
plugs have been replaced by cooled pressure sensors. The potential
mechanisms related to the spark plug or to the spark itself can thus
be eliminated. It can also be mentioned that no run-away pre-ignition
phenomena has been observed on the tested dual fuel engines, neit-
her at low nor at high engine speeds. Hot surfaces are thus unlikely to
be the source of LSPI in dual fuel engines (as it is very often the case
in SI engines).

(2) Deposits: the presence of deposits on the piston or the valves is also
often cited as a critical factor enhancing LSPI. In this work, as in most
of the reported studies in the literature, specific test methodologies
were followed to avoid as much as possible the combustion chamber
fouling and to ensure repeatable results. Moreover, methane is natu-
rally not a soot precursor. This mechanism can thus not be the main
path of triggering LSPI, especially as this kind of LSPI events can be
identified even at middle loads as punctual events while LSPI reported
here is quite „stable“.

(3) Liquid fuel films: in direct injection SI engines, the fuel can be directly
injected towards the piston head or the liner if the fuel spray penetrati-
on is too important. These fuel impacts are detrimental to the mixture
preparation but also to LSPI. Indeed, these spray-walls interactions
can lead to the formation of liquid fuel films, and therefore to the for-
mation of very rich mixtures associated with lower auto-ignition delay
times. In some cases, this effect can also be enhanced by the lower
mixture temperature obtained near the walls if this temperature cor-
re-sponds to the negative temperature coefficient zone.

229
Dual fuel engines represent, however, an ideal case with the port injection
of a pure gaseous fuel. This mechanism can thus be eliminated as well.

(4) Oil / fuel interactions: this mechanism is very often cited as the
main source of LSPI in SI engines. This process consists in inter-
actions between the fuel and the lubricating oil on the liner (espe-
cially when lateral injectors promoting fuel impacts are used), then
in a release in the combustion chamber of fuel and oil droplets
from the top land, and finally in the auto-ignition of the air / fuel
mixture whose reactivity has been enhanced by high ignitability
components of oil.

The case of dual fuel engines is, however, very special and interesting
since the gaseous fuel is port-injected and the Diesel fuel is injected only
around top dead centre. The Diesel injection timings notably ensure that
the injection takes place only in the piston bowl. Besides, given the low
flow rates used, it can be mentioned as well that no fuel impacts occur in
the piston bowl.

LSPI events due to oil auto-ignition in large gas engines have recently
been mentioned by Yasueda et al. [37][38] but for different operating con-
ditions after extended and sometimes abnormal operation with high oil
consumption, and without taking into account the interactions between
gas and oil (the proposed mechanism concerning only the oil vaporization
and oxidation).

In our work, oil / fuel interactions can hardly be considered as the main
source of LSPI but it should be confirmed by endoscopic visualizations
which could help in localizing the spatial onsets of auto-ignition.

Consequently, considering all these observations, it seems that the source

of LSPI in dual fuel engines is more related to a spontaneous gas phase
auto-ignition within the mixture. Of course, this assumption should be con-
firmed by further investigations, notably on different engine configurations,
but the comparative analysis with all the studies made for SI engines tend
to confirm this potential LSPI mechanism.

As a concluding remark concerning the role of the Diesel injector on LSPI, it

can also be mentioned that the Diesel fuel injection does not seem to have
any effect on combustion once pre-ignition has started (in the same way, the
spark has no particular effect on combustion once pre-ignition has started in

230
SI engines). The way the Diesel fuel is consumed and interacts with pre-igni-
tion needs, however, to be further investigated (notably by optical diagnostic
tools).

It also has to be mentioned that the Diesel injector fouling could be responsib-
le for some LSPI events. This issue is particularly critical for dual fuel engines
since the reduction in Diesel fuel flow through the nozzle decreases heat dissi-
pation, inducing thus high temperatures and causing the formation of deposits
on the nozzle [10]. Consequently, these deposits can be heated up cycles af-
ter cycles and potentially act as hot spots and trigger pre-ignition either on the
Diesel injector, or within the mixture if they are stripped away by combustion.

Unfortunately, as it is the case for SI engines, these LSPI events are very dif-
ficult to identify and to analyze since they remove themselves the mechanism
that caused them by burning or expelling the critical deposits. In this context,
the numerous investigations currently being carried out on the formation of
deposits on and in Diesel injectors should also be applied to dual fuel com-
bustion.

6. Conclusion
This article gives a new point of view concerning LSPI by comparing the
occurrence of this complex phenomenon in dual fuel Diesel-methane CI
engines and in gaso-line SI engines. Dual fuel concept has recently re-
born with the renewal of interest towards gas applications. Indeed, low
cost and low soot potential as well as low carbon content make natural
gas a good solution for future highly efficient engines. IFPEN has been
working in order to improve the dual fuel concept by minimizing the Diesel
pilot injection size. The results show that dual fuel operation is possible
over a wide engine operating range with huge CO2 savings compared to
Diesel operation. However, dual fuel combustion can suffer from LSPI and
this could be a limiting factor for performance and compression ratios.
Dual fuel engines are commonly derived from conventional CI engines
that are designed in order to withstand higher thermomechanical cons-
traints than SI engines. Despite this higher hardware resistance, dual fuel
engines must, however, prevent LSPI occurrence at high loads.

In terms of critical settings, LSPI occurrence in dual fuel engines can be

linked to Diesel pilot SOI which variation can sufficiently affect the tempe-
rature of both exhaust gases and in-cylinder trapped burned gases.

231
In terms of engine hardware configuration, the increase of compression
ratio in dual fuel engines does not necessarily lead to higher LSPI oc-
currence: there‘s no need to downgrade compression ratio thus engine
efficiency to operate in dual fuel conditions.

In all cases, the pressure and temperature history during compression

stroke only are not sufficient to explain LSPI. The residual burned gases
rate and temperature within the cylinder should be considered as well for
the analysis.

The causes of LSPI can be multiple and complex to identify. For the typical
full load performance targeted in modern engines (SI or dual fuel engi-
nes), the mean mixture‘s characteristics are not critical enough to trigger
auto-ignition. Only spatial and temporal fluctuations can explain that hot
spots with higher reactivities than the surrounding mean mixture can be
formed. In this perspective, the comparison between LSPI occurrence in
SI and dual fuel engines gives a new point of view on the possible causes
of pre-ignition. Indeed, many scientists point out mechanisms related to
overheated spark plugs, liquid fuel films or fuel / oil interactions. But the
combustion process in dual fuel engines is radically different from that
in SI engines in terms of ignition (no spark plug) and mixture preparati-
on (methane is port-injected and supposed to be perfectly mixed with air
inside the cylinder). Besides, the Diesel pilot in-jection conditions (fuel
quantity, timing, spray targeting, etc.) do not allow any fuel impact on the
cylinder walls or on the piston.

Thus, the mechanisms mentioned in the literature for explaining LSPI in

SI engines seem to be no longer adapted to dual fuel engines. The main
source of LSPI seems to be more related to a spontaneous gas phase
auto-ignition within the mixture (coming from temperature heterogeneity
or residual burned gases). To go further, it seems implicitly that among all
LSPI occurrences in SI engines, some pre-ignitions should be caused as
well by this mechanism and not only by liquid films or fuel / oil interactions.

A deeper analysis of the different combustion modes (auto-ignition, flame

propagation) involved in the dual fuel combustion process should help to
clarify the causes of LSPI occurrence. IFPEN is currently working on this
topic and results will be presented in an upcoming paper.

Moreover, the properties of methane as a fuel should be taken into account

to complete the analysis of the three-steps LSPI combustion process. Be-

232
sides methane, H2 and ethanol are also considered for association with
Diesel in dual fuel engines. If backfire problems are already known with
H2, LPSI phenomena have been observed with ethanol. Further investi-
gations on the fuel sensitivity are thus also required.

233
Literature
[1] Duffour, F.; Ternel, C.; Pagot, A.; IFP Energies nouvelles Approach for
Dual Fuel Diesel-Gasoline Engines. SAE Technical Paper 2011-24-
0065, 2011, doi:10.4271/2011-24-0065.
[2] Serrano, D.; Lecointe, B.; Exploring the Potential of Dual Fuel Die-
sel-CNG Combustion for Passenger Car Engine. Fisita World Au-
tomotive Congress, F2012-B01-025, 2012, doi:10.1007/978-3-642-
33777-2.
[3] Serrano, D.; Obiols, J.; Lecointe, B.; Optimization of Dual Fuel Die-
sel-Methane Operation on a Production Passenger Car Engine –
Thermodynamic Analysis. SAE Technical Paper 2013-01-2505,
2013, doi:10.4271/2013-01-2505.
[4] Sahoo, B.B.; Sahoo, N.; Saha, U.K.; Effect of Engine Parameters
and Type of Gaseous Fuel on the Performance of Dual Fuel Gas
Diesel Engines – A Critical Review. Renewable and Sustainab-
le Energy Reviews 13 (2009) pp. 1151–1184, 2009, doi:10.1016/j.
bbr.2011.03.031.
[5] Fritz, S.G.; Egbuonu, R.I.; Emissions form Heavy-Duty Trucks Con-
verted to Compressed Natural Gas. SAE Technical paper 932950,
1993, doi:10.4271/932950.
[6] Königsson, F.; Stalhammar, P.; Characterization and Potential of Dual
Fuel Combustion in a Modern Diesel Engine. SAE Technical Paper
2011-01-2223, 2011, doi:10.4271/2011-01-2223.
[7] Königsson, F.; Kuyper, J.; Stalhammar, P.; Angstrom, HE.; The Influ-
ence of Crevices on Hydrocarbon Emissions from a Diesel-Metha-
ne Dual Fuel Engine. SAE Technical Paper 2013-01-0848, 2013,
doi:10.4271/2013-01-0848.
[8] Carlucci, A.P.; Laforgia, D.; Saracino, R.; Toto, G.; Combustion and
Emissions Control in Diesel-Methane Dual Fuel Engines: the Ef-
fects of Methane Supply Method Combined with Variable In-Cylin-
der Charge Bulk Motion. Energy Conversion and Management 52
(2011) 3004–3017, 2011, doi:10.1016/S0196-8904(03)00150-X .
[9] Ishiyama, T.; Kang, J.; Ozawa, Y.; Improvement of Performance and
Reduction of Exhaust Emissions by Pilot-Fuel-Injection Control in a
Lean-Burning Natural-Gas Dual Fuel Engine. SAE Technical Paper
2011-01-1963, 2011, doi:10.4271/2011-01-1963.

234
[10] Königsson, F.; Stalhammar, P.; Controlling the Injector Tip Tempera-
ture in a Diesel Dual Fuel Engine. SAE Technical Paper 2012-01-
0826, 2012, doi:10.4271/2012-01-0826.
[11] Selim, M.Y.E.; Sensitivity of Dual Fuel Engine Combustion and
Knocking Limits to Gaseous Fuel Combustion. Energy Conversion
And Management 45(3), pp. 411-425, 2004, doi:10.1016/S0196-
8904(03)00150-X.
[12] Königsson, F.; Advancing the Limits of Dual Fuel Combustion. Licenti-
ate Thesis Report, KTH Royal Institute of Technology, 2012.
[13] Eichlseder, H.; Grabner, P.; Hadl, K.; Hepp, C.; Luef, R.; Dual Fuel
Concepts for Mobile Application. 34. Internationales Wiener Moto-
rensymposium, 2013.
[14] Inoue, T.; Inoue, Y.; Ishikawa, M.; Abnormal Combustion in a Highly
Boosted SI Engine - The Occurrence of Super Knock. SAE Techni-
cal Paper 2012-01-1141, 2012, doi:10.4271/2012-01-1141.
[15] Kalghatgi, G.T.; Bradley, D.; Andrae, J.; Harrison, A.J.; The nature of
“super-knock” and its origins in SI engines. Institution of Mechanical
Engineers - Internal Combustion Engines: Performance, Fuel Eco-
nomy and Emissions, pp.259-269,2009.
[16] Smallbone, A.; Bhave, A.; Advanced Simulations of „Knocking“ and
„Mega-Knocking“ Combustion Events in Highly Boosted DISI En-
gines. 10th International Symposium on Combustion Diagnostics,
Baden-Baden, 2012.
[17] Amann, M.; Alger, T.; Mehta, D.; The Effect of EGR on Low-Speed
Pre-Ignition in Boosted SI Engines. SAE Int. J. Engines 4(1):235-
245, 2011, doi:10.4271/2011-01-0339.
[18] Amann, M.; Mehta, D.; Alger, T.; Engine Operating Condition and
Gasoline Fuel Composition Effects on Low-Speed Pre-Ignition in
High-Performance Spark Ignited Gasoline Engines. SAE Int. J. En-
gines 4(1):274-285, 2011, doi:10.4271/2011-01-0342.
[19] Amann, M.; Alger, T.; Westmoreland, B.; Rothmaier, A.; The Effects
of Piston Crevices and Injection Strategy on Low-Speed Pre-Ignition
in Boosted SI Engines. SAE Int. J. Engines 5(3):1216-1228, 2012,
doi:10.4271/2012-01-1148.
[20] Sasaki, N.; Nakata, K.; Effect of Fuel Components on Engine Ab-
normal Combustion. SAE Technical Paper 2012-01-1276, 2012,
doi:10.4271/2012-01-1276.

235
[21] Takeuchi, K.; Fujimoto, K.; Hirano, S.; Yamashita, M.; Investigation of
Engine Oil Effect on Abnormal Combustion in Turbocharged Direct
Injection - Spark Ignition Engines. SAE Int. J. Fuels Lubr. 5(3):1017-
1024, 2012, doi:10.4271/2012-01-1615.
[22] Zaccardi, J.-M.; Escudié, D.; Overview of The Main Mechanisms Trig-
gering Low-Speed Pre-Ignition in Spark-Ignition Engines. To be pub-
lished, submitted to Int. J. Engine Res.
[23] Dahnz, C.; Spicher, U.; Irregular combustion in supercharged spark
ignition engines – pre-ignition and other phenomena. Int. J. Engine
Res. Vol. 11, pp. 485-498, 2010.
[24] Zahdeh, A.; Rothenberger, P.; Nguyen, W.; Anbarasu, M.;
Schmuck-Soldan, S.; Schäfer, J.; Goebel, Th.; Fundamental Appro-
ach to Investigate Pre-Ignition in Boosted SI Engines. SAE Int. J.
Engines 4(1):246-273, 2011, doi:10.4271/2011-01-0340.
[25] Willand, J.; Daniel, M.; Montefrancesco, E.; Geringer, B.; Hofmann,
P.; Kieberger, M.; Limits on Downsizing in Spark Ignition Engines
due to Pre-ignition. MTZ Motortechnische Zeitschrift 70(5), pp. 422-
428, 2009.
[26] Spicher, U.; History and Diagnostics of Irregular Combustion in Spark
Ignition Engines. The 8th International Conference on Modeling and
Diagnostics for Advanced Engine Systems, 2012.
[27] Brown, B.; Rogak, S.; Munshi, S.; Multiple Injection Strategy in a Di-
rect-Injection Natural Gas Engine with Entrained Diesel. SAE Int. J.
Fuels Lubr. 2(1):953-965, 2009, doi:10.4271/2009-01-1954.
[28] Zaccardi, J.-M.; Pagot, A.; Statistical observations on pre-ignition in
SI engines. 3rd Conference on Knocking at Gasoline Engines – Irre-
gular Combustion, Berlin, 2010.
[29] Livengood, J.C.; Wu, P.C.; Correlation of Autoignition Phenomena in
Internal Combustion Engines and Rapid Compression Machines.
Proceedings of the Fifth International Symposium on Combustion,
pp. 347-356, 1955.
[30] Vangraefschepe, F.; Zaccardi, J.-M.; Analysis of destructive abnormal
combustions appearing at high load and low engine speed on high
performance gasoline engines. SIA Conference : The Spark Ignition
Engine of the Future, 2007.
[31] Adomeit, P.; Thewes, M.; Ewald, J.; Günther, M.; Morcinkowski, B.,
Pischinger, S.; Effect of Fuel and Combustion System on the Pre-Ig-

236
 nition of Boosted SI Engines. 34th International Vienna Motor Sym-
posium, 2013.
[32] Rothenberger, P.; Zahdeh, A.; Anbarasu, M.; Göbel, T.; Simon, J.S.;
Soldan, S.S.; Experimental Pre-Ignition Investigations of Boosted
Spark Ignition Engines in Combination with Light-Intensified High
Speed Camera and CFD. IAV Conference Knocking at Gasoline En-
gines – Irregular Combustion, 2010.
[33] Hofmann, P.; Kieberger, M.; Geringer, B.; Willand, J.; Jelitto, C.; Aus-
lösemechanismen und Einflussfaktoren auf Vorentflammungsphä-
nomene an hoch aufgeladenen Ottomotoren. 17. Aachener Kollo-
quium Fahrzeug- und Motorentechnik, 2008.
[34] Freeland, P.; Taylor, J.; Oudenijeweme, D.; Stansfield, P.; Warth, M.;
Advanced technologies for downsized direct injection gasoline engi-
nes. THIESEL Conference on Thermo- and Fluid Dynamic Proces-
ses in Direct Injection Engines, 2012.
[35] Hoffmeyer, H.; Montefrancesco, E.; Beck, L.; Willand, J.; Ziebart, F.;
Mauss, F.; CARE – CAtalytic Reformated Exhaust Gases in Tur-
bocharged DISI-Engines. SAE Int. J. Fuels Lubr. 2(1):139-148,
doi:10.4271/2009-01-0503, 2009.
[36] Zaccardi, J.-M.; Laget, O.; Duval, L.; Pagot, A.; Mokhtari, S.; Leone,
T.; Investigations on the effects of in-cylinder charge motion and in-
jection mode on pre-ignition in highly boosted spark ignition engines.
19. Aachener Kolloquium Fahrzeug- und Motorentechnik, 2010.
[37] Yasueda, S.; Takasaki, K.; Tajima, H.; The Abnormal Combustion
caused by Lubricating Oil on High BMEP Gas Engines. Der Arbeits-
prozess des Verbrennungsmotors, 2011
[38] Yasueda, S.; Takasaki, K.; Tajima, H.; Abnormal Combustion caused
by Lubricating Oil in High BMEP Gas Engines. MTZ Motortechni-
sche Zeitschrift Industrial (MTZ Special Edition), pp. 34-39, 2013.

237
Acknowledgments
The authors would like to thank PSA Peugeot-Citroen for providing both
dual fuel and SI multi-cylinder engines used for these tests.

Notations and abbreviations

BDC: Bottom Dead Centre (ABDC / BBDC: After and Before BDC)
BMEP: Brake Mean Effective Pressure
CI: Compression Ignition
CoV: Coefficient of Variation
CR: Compression Ratio
DI: Direct Injection
EGR: Exhaust Gas Recirculation
EVC: Exhaust Valve Closing
EVO: Exhaust Valve Opening
FAER: Fuel / Air Equivalence Ratio
FIS: Fuel Injection System
IMEP: Indicated Mean Effective Pressure
IVO: Inlet Valve Opening
IVC: Inlet Valve Closing
LSPI: Low-Speed Pre-Ignition
MFB: Mass Fraction Burned
PFI: Port Fuel Injection
RON: Research Octane Index
SA: Spark Advance
SI: Spark Ignition
SOI: Start Of Injection
TDC: Top Dead Centre (ATDC / BTDC: After and Before TDC)
: Fuel / Air Equivalence Ratio
: Auto-ignition delay time
xdil: Dilution rate

238
12 LEC-GPN – a new Index for assessing
the knock behavior of gaseous fuels for
large engines
Andreas Wimmer, Franz Chmela, Martin Kirsten,
Gerhard Pirker, Peter Christiner, Christian Trapp,
Herbert Schaumberger

Abstract
While conventional natural gas is still the standard source of fuel for large
gas engines, fuels with higher hydrocarbons and fuels with added inert ga-
ses such as N2 and CO2 are becoming increasingly important in the ope-
ration of these engines. When engines suitable for use with these fuels
are developed, the occurrence of knocking combustion is the limiting fac-
tor in optimizing power density and efficiency. Knowledge of the knocking
tendency of the fuel is required when designing the combustion process.
It has been shown that the methane number normally used in combination
with the above-mentioned mixed gases is not a useful criterion.

This paper presents a new knock number with which the achievable level
of indicated mean effective pressure is predicted based on the gas com-
position. The process of determining the value referred to as LEC‑GPN
(LEC Gas Performance Number) is similar to that for determining the con-
ventional methane number; both are measured on the engine test bed.
In contrast to the methane number, knocking behavior is evaluated while
taking full account of the combustion that can in fact be expected with the
gaseous fuel. The load level obtained is measured at substantially the
same position of combustion and intake temperature as with natural gas.
A measuring plan was developed using DoE to deal with the large number
of possible gas compositions. First the gas mixtures to be investigated
were defined and then the LEC‑GPN values were calculated from mea-
surements taken on a single-cylinder research engine. From this data, a
functional was derived with which the LEC‑GPN value of any gas mixture
can be established.

239
Kurzfassung
Neben dem konventionellen Kraftstoff Erdgas werden für Großmotoren
Kraftstoffe mit höheren Kohlenwasserstoffen und mit Zumischung von In-
ertgasen wie N2 und CO2 immer wichtiger. Bei der Entwicklung dafür ge-
eigneter Motoren ist auch hier das Auftreten klopfender Verbrennung der
limitierende Faktor bei der Optimierung der Leistungsdichte und des Wir-
kungsgrads. Bei der Auslegung des Brennverfahrens ist daher die Kennt-
nis der Klopfneigung des Kraftstoffs erforderlich. Es hat sich gezeigt, dass
die bisher verwendete Methanzahl in Verbindung mit den oben genannten
Mischgasen kein brauchbares Beurteilungskriterium darstellt.

In diesem Beitrag wird eine neue Klopfkennzahl vorgestellt, mit der das er-
reichbare Mitteldruckniveau abhängig von der jeweiligen Gaszusammenset-
zung prognostiziert werden kann. Die Bestimmung der als LEC‑GPN (LEC
Gas Performance Number) bezeichneten Kennzahl erfolgt ähnlich wie bei der
konventionellen Methanzahl über Messungen am Motorenprüfstand. Im Ge-
gensatz zur Methanzahl wird aber die Bewertung des Klopfverhaltens unter
weitgehender Berücksichtigung des tatsächlich mit dem jeweiligen Brenngas
zu erwartenden Brennverhaltens durchgeführt, wobei dazu im Wesentlichen
bei gleicher Verbrennungslage und Einlasstemperatur das erreichbare Last-
niveau im Vergleich zum Erdgasbetrieb erfasst wird. Um die große Zahl mög-
licher Gaszusammensetzungen abdecken zu können, wurden zunächst auf
Basis eines DoE‑Messplans die zu untersuchenden Gasmischungen definiert
und in weiterer Folge am Einzylinder‑Forschungsmotor die entsprechenden
LEC-GPN-Werte ermittelt. Daraus wurde ein Funktional abgeleitet, mit dem
die LEC-GPN‑Werte für beliebige Gasgemische bestimmt werden können.

1. Introduction
Since the invention of the gasoline engine by Nicolaus August Otto, knock
has always gone hand in hand with combustion. In 1923, Ricardo deduced
from the acoustic perception of the knock phenomenon that it is a sponta-
neous ignition of the end gas [1]. Pressure pulsations as well as thermal
stresses caused by this abnormal combustion affect engine durability and
can lead to severe engine failure. Hence, knock is the limiting factor in
increasing both the power density and efficiency of gasoline engines. Fac-
tors that influence knock tendency are engine load, excess air ratio, intake
air temperature, compression ratio, combustion chamber design, charge
motion and of course the gas properties of the gaseous fuel.

240
Raising efficiency, increasing power density and reducing emissions are
the main development objectives in the field of large stationary gas engi-
nes. These requirements have been met repeatedly in recent decades,
and the left-hand graph in Figure 1 shows how load has increased over the
past two decades for a GE Jenbacher Type 6 engine. Recently, attempts
to raise power density have led from a strategy based on single-stage
turbocharged (SSTC) engines two-stage turbocharged (TSTC) engines.

Figure 1: Large gas engines - area of conflict

Load increase leads to a reduction in the safety margin between the ope-
rating point and the knock limit as shown in the right-hand diagram [2].
This diagram depicts the knock and misfire limits versus excess air ratio. It
can be seen that increasing IMEP narrows the gap between the knock and
misfire limits and increasing the compression ratio narrows it even more,
measures favorable to increasing efficiency.

The knock tendency of the gaseous fuel is one of the most important
boundary conditions for large stationary gas engines in particular, which
are operated mainly at full load. Knocking combustion limits the achievab-
le engine load; therefore, the compression ratio is decisive for the engine
layout. Thus it is important to be aware of the knock properties of the fuel
gas in the design stage.

The most common index for characterizing fuel gases has been methane
number (MN). This number indicates the volumetric mixture ratio of met-

241
hane and hydrogen in a reference gas with the same knock intensity on
a single-cylinder CFR test engine as the gas being tested. A MN of 100
indicates knock-resistant pure methane with no hydrogen, a MN of zero
100% “easy-knocking” hydrogen. Since the test procedure is performed at
a stoichiometric excess air ratio, the conventional MN does not appear to
be well suited for lean burn engines operated at excess air ratios higher
than two.

Besides measuring the MN on a CFR test engine, which precisely de-

termines MN, it can also be calculated as described in FVV report 37
[3]. This report introduced an approximation method using an extended
mixture approach. The calculation is performed by dividing the mixture
into groups of three components. This is done under the premises that the
MN of these mixtures is known from experiments and that the highest MN
differs by five units from each of the others [3]. This paper makes use of
this calculation method, referring to it as MNFVV.

Figure 2: MNFVV of various gas mixtures and inadequacy of MNFVV

The left-hand diagram in Figure 2 provides an example of the calculated

MN of methane/propane mixtures. The MN of mixtures with significant
ratios of inert gas (carbon dioxide and nitrogen) were also calculated. The
MNFVV decreases as the amount of propane increases and increases as
the amount of carbon dioxide and nitrogen increases.

242
When the MN is high, however, the calculated MNFVV no longer seems
to be suitable for predicting knock-limited IMEP. This is illustrated in the
right-hand diagram, which plots the load levels that can be achieved with
different gas compositions versus the calculated MN of these mixtures.
Contrary to the expectation that gas mixtures with similar MN’s have si-
milar load levels, two different mixtures with a MNFVV of 110 have two
considerably different knock-limited IMEPs. Therefore, a new method for
determinating achievable engine load appears to be necessary.

Another index commonly used to characterize engine knock is the Wau-

kesha Knock Index (WKITM). This index is an extension of the test pro-
cedure for MN. Focusing on mixtures with higher hydrocarbons and their
isomers, it considers the knock resistance of inert gases such as carbon
dioxide and nitrogen. The test bed measurement data serves as the input
data for a computer program. The method can also be used to determine
the maximum obtainable power and ignition timing [4, 5].

Another way of characterizing the knock resistance of fuel gases was in-
troduced by DNV KEMA. A method to compute the autoignition process
during the compression and power phases was derived from the physical
and chemical combustion properties of the fuel mixtures. The autoignition
delay time of the end gas is used as the primary yardstick for ranking the
knock resistance of fuel gases [6, 7].

The need for a reliable tool to predict knock-limited IMEP also arises from
the large variety of fuel gases from various sources, e.g. biogas, digester
gas, mine gas, and landfill gas as shown in Table 1 [8].

Table 1: Variety of gas compositions for large gas engines

This paper introduces a new index referred to as LEC-GPN which was de-
veloped for lean burn engines in particular to predict the achievable IMEP
for a broad variety of fuel gases with various gas compositions.

243
2. Test procedure

2.1. LEC-GPN – Gas Performance Number

This assessment procedure was developed in order to determine the
knock-limited IMEP of a specific fuel gas under certain operating condi-
tions and to compare it with test results obtained using natural gas with a
MNFVV of 92. The reference IMEP refers to the full scale engine nominal
load level. Excess air ratio and spark timing were optimized for best effi-
ciency at a NOx level of 500 mg/Nm³. The intake air temperature was set
for a knock frequency of 10%. The cylinder pressure trace was measured
and the apparent rate of heat release curve was calculated and normali-
zed with the rate of heat release integral. The normalized rate of heat re-
lease curve served as reference for the later assessment of the LEC-GPN
of other fuel gases.

The normalized rate of heat release curve of the gas being tested was
matched with the reference curve by adjusting spark timing and excess air
ratio while maintaining the baseline intake temperature and the baseline
knock frequency of 10%.

The equality of the normalized rate of heat release curves ensures an

equal history of the unburned zone for each measurement point with a
specific engine configuration. From this, the temperature of the unburned
zone can be regarded as a knock indicator.

The LEC-GPN was calculated from the IMEP obtained from tests using
Equation 1.

(1)

244
Figure 3: Apparent rates of heat release and cylinder pressure traces for
natural gas (MNFVV = 92) and a mixture of 80 vol% NG and 20 vol%
C3H8

Figure 3 provides an example of the method as applied to a fuel gas mix-

ture of 80 vol% natural gas with a MNFVV of 92 and 20 vol% propane. The
left-hand diagram depicts the normalized apparent rate of heat release
curves for the baseline natural gas and the gas being evaluated. These
curves clearly match very well.

The right-hand diagram showing the cylinder pressure traces reveal a

considerable difference in IMEP with a value of about 9.5 bar for the natu-
ral gas/propane mixture. Equation 1 yields a LEC-GPN of 60% while the
calculated MNFVV amounts to 53.4.

One significant advantage of this index is its independence from engine

configuration, making this method universally applicable as no specially
equipped research engine is required. Furthermore, this index depicts the
actual load which can be achieved with a certain fuel gas, a factor of great
importance to the customer.

2.2. Engine
All the experimental work was done on a three liter, single-cylinder spark
ignited gas engine at 1500 min-1. The schematic diagram shown in Figu-
re 2 depicts the low-pressure gas supply provided at the SCE test bed.

245
A compressor is used to provide boost pressure. Further degrees of free-
dom are the temperature and the humidity of the intake air.

A gas mixing station capable of providing fuels containing up to six diffe-

rent components was used to prepare different gas compositions. Each
individual fraction was measured with a separate mass flow controller
(MFC). Fuels with no more than four components were investigated to
determine the LEC-GPN.

A Venturi mixer generated the air/fuel mixture. The piping system was long
enough to achieve a homogenous charge before it reaches the engine.

Figure 4: Test bed low pressure gas supply

2.3. Gas mixtures tested

Table 2 provides an overview of the range of gas blends needed to set
up the LEC-GPN. The mixtures tested included natural gas (MNFVV 92),
propane, carbon dioxide and nitrogen. It was important both to cover a
broad variety of fuel compositions in order to assess the mixtures as well
as to minimize the time on the test bed.

246
 Table 2: Concentration ranges of the gas components

A sophisticated measurement plan was created using DoE and a D-Op-

timal model approach. The input parameters were the concentrations of
the four gases. This measurement plan included a grid of 43 different gas
compositions with mixtures consisting of 1 to 4 components.

2.4. Knock detection

Knock recognition involves separating knocking cycles from non-knocking
cycles using simple algorithms. The method should be capable of determi-
ning precisely when knock starts. The measured cylinder pressure curve
is the basic physical parameter for knock recognition because it is usually
superimposed by high frequency pressure oscillations as a result of knock.

The knocking criterion [9, 10, 11] is basically a combination of several cri-
teria. To increase sensitivity by differentiation, the apparent heat release
rate is calculated from the pressure curve using the first law of thermo-
dynamics. The high frequency part is extracted using a high-pass filter.
Then, a critical threshold value is determined that is two thirds of the ma-
ximum amplitude. Second, the amplitude that first exceeds this threshold
value is sought. The zero value before this amplitude is the start of kno-
cking, see Figure 5.

To ascertain that this is in fact the start of knock, the high frequency pres-
sure oscillations in the area of several degrees crank angle before and
after the start of knock are integrated and set in relation to each other. This
ratio must also exceed a threshold value. In addition to the start of knock,
knock intensity is another parameter used to characterize knock behavior
and can be determined from the high-frequency part of the cylinder pres-
sure curve.

247
 Figure 5: High-pass filtered heat release and cylinder pressure

3. Discussion

3.1. Basic results

It is well known that adding propane to the fuel gas lowers achievable
IMEP significantly. On the other hand, inert gas has a positive effect on
the knock tendency of fuel gases. Figure 6 shows this effect with three
different natural gas MNFVV 92/propane mixtures. The methane number
of the combustible components in the fuel gas is calculated, and then
the inert gases nitrogen and carbon dioxide are added to the fuel gas in
different ratios.

The first two curves in Figure 6 show the results with natural gas (MNF-
VV 92) as the only combustible component followed by two natural gas
(MNFVV 92)/propane mixtures with methane numbers of 70 and 50, re-
spectively. Three statements can be made based on the measurements.
First, there is a steady increase in the achievable load when the inert gas

248
ratio is increased. Second, a gradient can be observed which is almost in-
dependent of the methane/propane mixture ratio with a slight tendency to
increase towards higher methane numbers of the fuel gas. Third, carbon
dioxide has a greater impact on reducing knock tendency than nitrogen.

Figure 6: LEC-GPN for various gas mixtures

The left-hand graph in Figure 7 uses the data points shown in Figure 2
once again. IMEP is now plotted in relation to the LEC-GPN, indicating the
relationship between the baseline point (natural gas MNFVV 92, repre-
sented by a grey triangle) and the IMEP reached with individual gas mix-
tures. The achievable load of the fuel gas can be read from this illustration.

249
 Figure 7: LEC-GPN data versus achievable IMEP and versus MNFVV

The right-hand graph in Figure 7 plots the LEC-GPN in relation to calcu-

lated methane number. A linear relationship exists between the achievab-
le load and the methane number with natural gas (MNFVV 92)/propane
mixtures. The measurement results also show that the knock resistance
of fuel gases with high concentrations of inert gas is overestimated by the
calculated methane number. Therefore, the methane number cannot be
used to reliably predict achievable load at least for this kind of mixtures.

3.2. Model and model quality

The main goal was to derive a functional from the measurements to pre-
dict the achievable load under the boundary conditions of a specific fuel
gas composition. Figure 8 shows the work flow to establish this functional.
The first step involved generating an extensive database containing a bro-
ad variety of different gas compositions. The Design of Experiments (DoE)
method was used to come up with a measurement plan which included 43
different fuel gas compositions. The measurement results were then used
to derive a four-dimensional functional which can predict achievable load
with any fuel gas composition containing up to four fractions (methane,
propane, nitrogen and carbon dioxide). The last step in the work flow was
to verify the functional using additional measuring points not part of the
previous modeling process.

250
 Figure 8: Work flow for deriving the functional

Figure 9 shows the correlation between the measured IMEP and the IMEP
predicted by the model. Both axes are scaled in the same way. A satisfac-
tory agreement between the measured data and the calculated data can
be seen. The largest deviations are in the range of about 1 bar.

Figure 9: Model quality

3.3. Applying different combustion processes to prove

independence
Since the method must be independent from the used engine configu-
ration, two different concepts were investigated. Tests were carried out
with both a direct ignition combustion system as well as a pre-chamber
combustion system.

251
 Figure 10: Comparison between a direct ignition combustion concept
and a pre-chamber combustion concept

Figure 10 compares these two configurations with three different gas com-
positions. The black curve illustrates the measurements derived from the
direct ignition concept and the dotted curve those from the pre-chamber
concept.

The graphs in Figure 6 reveal a slight disadvantage in terms of the level

of achievable load. The reason for this difference is the behavior of these
two concepts at the knock limit. The direct ignition combustion concept
enables a steady rise in the knock frequency. In contrast, the operation of
a pre-chamber is far more sensitive at the knock limit and tends to lead
to spontaneous runaway combustion. Consequently, achievable load is
slightly lower with the pre-chamber concept, which has a lower knock fre-
quency than the direction ignition concept (5% and >10%, respectively).

4. Conclusions
Since the composition of fuel gases for large gas engines varies widely
depending on their source (wells, sewage, landfills or biogas plants), it is
important to have a reliable tool which can predict the highest possible
load as well as engine efficiency.

252
This paper introduces a new index called LEC-GPN which is derived from
measure-ments taken on a SCE. It sets the load achievable with a certain
fuel gas in relation to a baseline point (natural gas).

This approach is subject to the boundary conditions of:

• a constant intake temperature,
• an equal normalized apparent rate of heat release and
• 10 % knocking cycles.

This work concentrates on the operation with four different gases, where
natural gas (methane) and propane were used as fuel gas in addition to
the inert gases nitrogen and carbon dioxide. Due to the broad variety of
gas compositions DoE was used to define a measuring plan.

Up to now, investigations have been made with two different engine configu-
rations, a direct ignition concept and a pre-chamber concept. A comparison of
these concepts reveals that the approach used with LEC-GPN is independent
of the combustion concept used. Further steps will be necessary to verify this
method on other engines with different engine concepts and configurations.

Abbreviations
AHRR Apparent heat release rate
CFR Cooperative fuel research
DoE Design of Experiments
GPN Gas Performance Number
IMEP Indicated mean effective pressure
LEC Large Engines Competence Center
MFC Mass flow controller
MN Methane number
NG Natural gas
SSTC Single-staged turbocharged
SOC Start of combustion
TSTC Two-stage turbocharged
WKI Waukesha Knock Index

253
References
[1] RICARDO, H.: The Internal Combustion Engine, Band 2, Blackie
and Son Lim., Glasgow and Bombay
[2] Wimmer, A.: Combustion concept development for the next genera-
tion of GE Jenbacher gas engines, 7th Dessauer Gasmotoren-Kon-
ferenz, 2011
[3] Untersuchungen der Übertragbarkeit der am CFR-Motor gefunde-
nen Ergebnisse auf andere Motoren – Gültigkeitsbereich der Met-
hanzahl, Forschungbericht Heft 120, FVV 1971
[4] Callahan, T.: Engine knock rating of natural gasses – expanding the
methane number database, ASME International, 1996
[5] Selberg, J.: Application of the Waukesha Knock IndexTM number for
rating gaseous fuels, CIMAC, 1998
[6] Van Dijk, G.H.J: Ranking the knock resistance of gaseous fuel by
their physical and chemical properties, CIMAC, 2013
[7] Van Dijk, G.H.J: Two-Zone Model to predict cylinder pressure history
and knock propensity in a gas engine, 14. Tagung Der Arbeitspro-
zess des Verbrennungsmotors, 2013
[8] Albus, R.: Biogaserzeugung- Verfahren, Gasbestandteile und Po-
tenziale, 14 DVV Kolloquium, Wien, 2004
[9] Worret, R.: Zylinderdruckbasierte Detektion und Simulation der
Klopfgrenze mit einem verbesserten thermodynamischen Ansatz.
Dissertation, Universität Karlsruhe, 2002.
[10] Dimitrov, D.; Strasser, Ch.; Chmela, F.; Wimmer, A.: Vorhersage
des Klopfverhaltens für Groß-Gasmotoren mit Direktzündung oder
Vorkammer. 2. Tagung Klopfregelung für Ottomotoren – Trends für
Serienentwickler, Berlin, 2006.
[11] Dimitrov, D.; Chmela, F.; Wimmer, A.: Eine Methode zur Vorausbe-
rechnung des Klopfverhaltens von Gasmotoren. 4. Dessauer Gas-
motoren Konferenz, Dessau, 2005

254
13 A statistical modeling approach with
detailed chemical kinetics for use in
3D-CFD engine knock predictions
Dirk Linse, Andreas Kleemann, Christian Hasse

Abstract
In this work, a statistical modeling approach in the context of Rey-
nolds-averaged Navier-Stokes (RANS) equations is presented. It was ap-
plied to simulate knocking combustion in a turbocharged spark ignition
engine with central direct injection. This model solves transport equations
for the Favre mean and variance of an ignition progress variable that re-
presents the level of pre-reactions. For the closure of the mean chemical
source terms in these equations a clipped Gaussian probability density
function (PDF) for temperature and a -PDF for mixture fraction are em-
ployed, which is combined with the use of tabulated detailed chemistry.
Hence, the impact of mixture fraction and temperature fluctuations on the
autoignition process is accounted for. Statistical information is introduced
by employing a presumed probability density function approach for the ig-
nition progress variable. The shape of the function is defined by assuming
a clipped Gaussian distribution, which is determined by the first and se-
cond moments of the ignition progress. Based on the PDF the probability
of autoignition can be computed at each location in the combustion cham-
ber. Moreover, the knock model provides distinct criteria to determine the
mean knock onset and the number of knocking cycles. For validation pur-
poses a spark timing sweep was performed on the engine test bench for
a full-load operating point at n=2000 rpm. 3D-CFD simulations including
gas exchange, mixture formation, combustion, and knock were carried out
and the results are compared with experimental results. It is shown that
the proposed knock model is able to accurately predict the mean knock
onset and the trend of the number of knocking cycles with respect to a
spark timing sweep. Moreover, the model provides additional information
to determine critical regions in the combustion chamber that are characte-
rized by a high ignition probability.

255
1. Introduction
Turbocharged direct injection spark ignition (DISI) engines represent a
major step in the evolution of gasoline engines. DISI engines have be-
come increasingly popular due to the potential to significantly improve
fuel efficiency and emission levels while the performance of the engine
is maintained or even enhanced. In turbocharged DISI engines both the
compression ratio and spark timing are limited by the occurrence of knock.
Hence, it limits the thermodynamic efficiency and the performance of the
engine. The optimization of the combustion process in spark ignition (SI)
engines with respect to knock is therefore of significant importance in the
engine development process. A more complete understanding of engi-
ne knock can be obtained by combining experimental and numerical de-
velopment tools. Especially the numerical simulation makes it possible to
improve and to optimize the combustion process in internal combustion
(IC) engines. Although the 3D-CFD simulation of in-cylinder flow including
gas exchange, mixture formation, and combustion has become well esta-
blished in the engine development process, the prediction of engine knock
is still a major challenge. However, in order to reduce the costs and time
effort during the engine development process, there is an increasing de-
mand for efficient and reliable computational tools that are able to predict
the occurrence/probability of knocking combustion or can at least predict
tendencies when comparing engine concepts.

Engine knock is a very complex phenomenon involving chemical proces-

ses in the unburnt mixture, turbulent flame propagation, and detonation
waves. Especially the erratic behavior and the impact of cycle-by-cycle
variations makes the modeling of engine knock a challenging task. Ge-
nerally all 3D-CFD codes used in industrial applications are based on the
Reynolds-averaged Navier-Stokes (RANS) equations, which provide only
averaged quantities. Hence, cyclic variations are not cannot be described
by this approach and the interpretation of the CFD results with respect to
knock onset, intensity, and number of knocking cycles (all being the result
of cycle-to-cycle variations) is rather difficult.

256
Figure 1: Measured pressure traces for 1000 consecutive cycles where
270 are regarded as knocking cycles. The measurements were taken for
a full-load engine operating point at 5000 rpm. The solid black line repre-
sents the averaged pressure history.

The effect of cyclic variations on these criteria is illustrated in Figure 1,

where the measured pressure traces for 1000 consecutive cycles for a
full-load engine operating point at 5000 rpm are shown. Additionally, the
corresponding averaged pressure trace is depicted and is represented by
the solid black line. The cycle-by-cycle variations are clearly visible. The-
se variations lead to different peak pressures, changes in heat release,
and to irregular combustion phenomena. From these 1000 consecutive
cycles 270 were identified as knocking cycles. A simulation of this engine
operating point with a RANS CFD code would yield a pressure trace very
similar to the solid black line in Figure 1. It is obvious that an evaluation of
knock criteria solely based on averaged quantities is quite limited or even
impossible. The discussion shows that a RANS 3D-CFD knock model that
is solely based on mean quantities is unable to predict the number of kno-
cking cycles or the onset of knock with reasonable accuracy. More strictly
speaking, it is not possible to deduce the impact of knocking combustion
for the investigated engine operating point from the mean pressure history

257
in Figure 1. Hence, for a reliable prediction of knocking combustion it is
crucial to provide statistical information regarding the formation of engine
knock. Moreover, for a comprehensive knock model it is also important
to incorporate detailed chemical kinetics and to account for the impact of
turbulence chemistry interaction on the autoignition process in the unburnt
fuel-air mixture. Linse et. al. [1] and Linse [2] have proposed a new stati-
stical modeling approach which addresses these demands and accounts
for the stochastic nature of engine knock. This model can be interpreted
as a generalized knock integral method (gKIM) and is briefly summarized
in Section 2. It was applied to simulate the impact of a spark timing sweep
on the knock characteristics of a turbocharged DISI engine for a full-load
engine operating point at n = 2000 rpm. Based on a reference engine ope-
rating point, which was operated close to the allowable knock limit on the
engine test bench, the spark timing was advanced in order to increase the
number of knocking cycles. The in-cylinder flow of the investigated engine
was simulated including gas exchange, mixture formation, combustion,
and knock. An overview of the experimental and numerical setup is given
in Section 3. In Section 4 the simulation results are presented and analy-
zed, showing that the knock model is able to predict the mean knock onset
with reasonable accuracy and that the impact of a spark timing sweep on
the number of knocking cycles is well captured. Finally, the main conclusi-
ons are summarized in Section 5.

2. Generalized Knock Integral Method

Most of the current modeling approaches do not provide statistical infor-
mation regarding the formation of engine knock and the impact of turbul-
ent fluctuations of temperature and mixture composition on the autoigni-
tion process is in general not accounted for. Hence, they do not take into
account the inherent feature of engine knock, namely that it is a stochastic
phenomenon. For the modeling of autoignition and knock it is common
to introduce a pseudo precursor to define the onset of ignition. This pre-
cursor generally represents a critical concentration of a fictitious species,
and in case of the knock integral method (KIM) [3], its rate of change is
directly related to the inverse of the ignition delay. The generalized knock
integral method does not only provide the mean value of the precursor it
also provides its statistical distribution. The basic concept of this model is
outlined in the following.

258
2.1. Ignition Progress Variable
In order to quantify the autoignition process in the unburnt fuel-air mixture
an ignition progress variable is introduced. It can be viewed as a nor-
malized mass fraction that characterizes the reactive state of the fuel-air
mixture and is defined as

(1)

represents the mass fraction of the characteristic species and is

the maximum value at the instant of autoignition of the fuel-air mixture,
which corresponds to a value of =1. For modeling engine knock in SI
engines it is necessary to account for the propagating flame front, which
divides the mixture into an unburnt and burnt region. Hence it is feasible
to split the ignition progress variable into an unburnt and burnt
part [1, 2]

(2)

where is the reaction progress variable. Linse et. al. [1] have proposed a
balance equation for which reads

(3)

The corresponding source term is defined as

(4)

where is the ignition delay time conditioned on unburnt gases and

is the source term of the reaction progress variable. As it is evident,
the chemical source term is split into two different contributions. The first
term on the right-hand side of Eq. (4) accounts for the formation of the
ignition progress. The second term accounts for the chemical conversion
of the ignition progress variable due to combustion. It therefore acts as a
sink term and has to be provided by the combustion model. In case of a

259
homogeneous mixture, quiescent flow, and the absence of a flame front,
Eqs. (3) and (4) represent the classical KIM approach.

2.2. Turbulence Modeling of the Ignition Progress Vari-

able
The transport equations for the Favre mean and variance of the ignition
progress variable are given by [1, 2]

(5)

(6)

The difficulty lies in the closure of the source terms and ̃ since
is a reactive scalar, and therefore it is essential to account for turbulence
chemistry interaction. Linse et. al. [1] have proposed a closure based on
the presumed PDF approach, where a clipped Gaussian PDF for tempe-
rature and a -PDF for mixture fraction are employed. Hence, the impact
of temperature and mixture fraction fluctuations on the ignition progress
variable is accounted for. The closure of the chemical source term reads
[1, 2]

(7)

and

(8)

Finally, the mean inverse ignition delay time needs to be specified. It is

computed according to

(9)

260
The computation of the mean inverse ignition delay time (Eq. (9)) was
performed a priori and tabulated as a function of pressure, unburnt tempe-
rature, mixture fraction, residual gas mass fraction, mixture fraction vari-
ance, and unburnt temperature variance. The mean inverse ignition delay
was then linearly interpolated in six-dimensional phase space during the
simulation run. Detailed chemical kinetics has been incorporated by em-
ploying the complex iso-octane mechanism proposed by Curran et al. [4].

2.3. Ignition Probability

The probability distribution of the ignition progress variable is determined
by employing a presumed PDF approach in which the shape of the PDF
is defined by assuming a clipped Gaussian distribution. The advantage of
this approach is that the PDF can be constructed by the first and second
moments of , which are determined by Eqs. (5) and (6). Moreover, ba-
sed on the PDF of the ignition progress variable it is possible to derive a
mathematically consistent approach to evaluate the ignition probability at
each location in the computational domain. It is computed by integrating
the PDF from a user-defined lower limit (here chosen as 0.95) up to the
maximum value of [1, 2)

(10)

Hence, the ignition probability determines the probability of finding valu-

es of >0.95. In order to identify regions in the computational domain at
which the ignition probability is greater than 1%, the set is introduced

(11)

Another useful quantity to assess the knock characteristics of an engine is

the volume defined by

(12)

It represents the total volume of regions in the combustion chamber where

the ignition probability is greater than 1%.

261
3. Experimental and Numerical Setup

3.1. Experimental Setup

Engine type Gasoline engine 4-stroke

Cylinders/valves per cyl. 6/4
Compression ratio 10.2
Displacement volume 2979 cm3
Bore 84 mm
Stroke 89.6 mm
Connecting rod length 145 mm
Injection system Solenoid valve fuel injection
Table 1: Engine specifications.

The experimental investigations were performed on a BMW turbochar-

ged 6-cylinder engine with direct injection and homogeneous charge com-
bustion. The engine is equipped with the third-generation Valvetronic that
allows for variable valve timing and valve lift. At BMW, this technological
combination is referred to as TwinPower Turbo. The engine specifications
are presented in Table 1. For the assessment of the generalized knock
integral method a spark timing sweep was performed on the engine test
bench for a full-load operating point at n = 2000 rpm. The timings of igni-
tion were spark = 2.4° CA aTDC, spark = 1.2° CA aTDC, and spark
= 0.0° CA aTDC, which are denoted in the following as Case 1, Case 2,
and Case 3, respectively. The first spark timing spark = 2.4° CA aTDC
was used as a reference. It was operated close to the acceptable knock
limit. The spark timing was then advanced by two consecutive steps of
= 1.2° CA in order to increase the number of knocking cycles. It should be
noted that for the spark timing sweep the valve timings, injection timing,
injected fuel mass, and trapped mass were kept constant - only the spark
timing was adjusted. This ensures that the mixture formation and charge
motion are essentially the same for all three cases. For the identification
of a knocking cycle and to determine the corresponding knock onset the
algorithm according to Worret and Spicher [5] was employed.

262
3.2. Numerical Setup
For the three dimensional simulation of the in-cylinder flow the commerci-
al CFD code ANSYS CFX was employed. The simulation of Case 1 was
started at exhaust valves opening (EVO) with a valve lift of 0.3 mm. In the
computational domain the initial velocity was set to zero. The temperature
and pressure in both the combustion chamber and the exhaust port were
initialized according to the results of the 1D simulation. In the entire com-
putational domain residual exhaust gas was initialized according to the
engine out equivalence ratio of the corresponding operating point. The
results of the 1D gas exchange simulation were used to define the inflow
and outflow conditions at the intake and exhaust ports.

The 3D-CFD simulation was at first performed for Case 1. The spark
timing sweep was then performed by restarts at TDC based on the results
of Case 1. This is a reasonable approach, since the boundary conditions
were essentially the same due to the operating strategy. Hence, it was not
necessary to simulate a full cycle for each of these three cases. For the si-
mulation of engine knock the generalized knock integral method was em-
ployed. It was initialized after the liquid fuel was completely evaporated,
which corresponds to 630 CAD for the investigated operating points. Due
to the geometrical symmetry of the combustion chamber and the valve lift
profile, only half of the cylinder was simulated.

4. Results
At first, the simulation results are evaluated at TDC, which corresponds
to the spark timing of Case 3. The basic idea is to analyze the conditions
at start of combustion. Analysis of the residual gas distribution has shown
that the concentration in the combustion chamber varies only in range bet-
ween 1% and 1.1%, indicating that the residual gas is thoroughly mixed
with the surrounding fuel-air mixture. In Figure 2 the distributions of mix-
ture fraction, temperature, and ignition progress evaluated on a horizontal
cut plane are shown for half the combustion chamber (symmetric model).
It can be seen that the fuel-air mixture is not thoroughly mixed and that a
fuel-rich region can be found in the area beneath the exhaust valve and in
the squish area. The temperature distribution is much more homogeneous
and shows a slightly higher temperature in the center of the combustion
chamber. More interesting is the distribution of the mean ignition progress
variable. It shows two distinct areas where the ignition progress is close

263
to its current peak value: the region beneath the exhaust valve and in the
vicinity of the spark plug.

Figure 2: Distribution of the mean mixture fraction (left), temperature

(center), and ignition progress variable (right) evaluated at TDC.

The spots in the vicinity of the spark plug are expected to be of minor im-
portance for engine knock. The initial flame front will grow fast enough to
consume the fuel in this area before the occurrence of autoignition. More
critical is the region beneath the exhaust valve, since in this area the high
temperature is accompanied by a fuel-rich mixture composition. Moreo-
ver, this region is relatively far from the spark plug such that it is likely that
self-ignition may occur before the flame front finally arrives. However, the-
se are just mean quantities and do not provide information how large the
fluctuations around the mean value really are. Now, in order to compare
the turbulence fluctuation intensity of different quantities, it is convenient
to normalize the corresponding standard deviation by its mean value. This
leads to the definition of the fluctuation intensity of mixture fraction, tem-
perature, and ignition progress

(13)

Since the residual gas is thoroughly mixed and the fluctuations of tempe-
rature are low, it can be concluded that the fluctuations of are mainly
induced by fluctuations of the mixture fraction.

264
Figure 3: Distribution of the turbulence fluctuation intensity (Eq. (13)) of
mixture fraction (left), temperature (center), and ignition progress variab-
le (right) evaluated at TDC.

Hence, improving the homogenization of the fuel-air mixture might be a

measure to reduce the knock tendency in this particular case. Next, the
impact of combustion on the ignition progress is at first discussed for Case
3.

Figure 4 shows the distribution of the mean ignition progress evaluated

on a horizontal cut plane at 745 CAD and the probability distributions at
the locations P1 and P2. The mean flame front is depicted as well. It can
be seen that the maximum value is close to at point P1 and zero
far behind the flame front. The gKIM model takes into account that the
probability of autoignition in the burnt mixture is zero. Since the depicted
flame front is just the mean flame front, there is also a high probability that
self-ignition may occur directly behind this front. It depends on how the
instantaneous flame front fluctuates around the mean value.

265
Figure 4: Distribution of the mean ignition progress variable (left) and the
PDF of the ignition progress variable at P1 and P2 (right) evaluated at
745 CAD for Case 3.

The PDF of the ignition progress at the location P2 is close to a delta peak
located at = 0.25, which simply states that the fluctuations around the
mean value are very low. In contrast, the PDF at location P1 is very broad,
and although the mean value is close to = 0.7, there is a probability of
finding values of close to unity. This illustrates that it is very difficult to
predict the onset of autoignition or knock solely based on mean quantities.
With the definition of the ignition probability (Eq. (10)) the probability of
autoignition can be computed at location P1, and it follows that = 0.04
or = 4%.

Figure 5 shows the areas of self-ignition ( ) for Case 1, Case 2, and

Case 3 evaluated at different instants. These areas are colored by the
ignition probability . By comparing the three cases it can be seen that the
locations of self-ignition are nearly identical. This is due to the operating
strategy of the spark timing sweep. The charge motion, mixture formation,
and turbulence level are the same for all three cases. Hence, the increase
of knocking cycles is more likely to be caused by an increase of pressure
and temperature in the unburnt mixture rather than by differences in the
local mixture composition. For Case 1 the region where it may lead to
autoignition is very small and it appears late ( ≥ 751 CAD) during the
combustion process. This is in accordance with the engine settings of
this case. It was operated close to the allowable knock limit, and hence
the number of knocking cycles is close to zero. By advancing the spark

266
timing, the regions of self-ignition and the probability increase. It can also
be observed that the instant of autoignition appears earlier. There are two
regions where autoignition may occur. Figure 5 indicates that the spots
of self-ignition beneath the exhaust valve are at first established, and as
these spots vanish, the region of self-ignition in the pocket valve grows.
The existence of two different areas that are characterized by a high pro-
bability to autoignite gives rise to the question which region is the most
critical with respect to knock. It can be argued that the spots of self-ignition
beneath the exhaust valve are the major cause for engine knock.

Figure 5: Identification of areas with an ignition probability greater than

1% (Eq. (11)) evaluated at different instants for Case 1 (left), Case 2
(center), and Case 3 (right).

267
These spots appear first and they have a high probability to autoignite.
Moreover, the flame front is still far enough from the combustion chamber
walls. The regions of self-ignition in the pocket valve appear at a later
stage of combustion. In this case, the flame front is closer to the walls and
most of the unburnt fuel-air mixture is already consumed.

Figure 6 shows the quantity (Eq. (12)) as a function of crank angle for
all three cases of the spark timing sweep. The volume increases until it
reaches its peak value, and then it gradually decreases with the advance-
ment of the combustion process.

Figure 6: Total volume (Eq. (12)) of cells that satisfy the following
condition:

This behavior is the same for all three cases. They differ only in the value
of the maximum volume and in the time when the peak volume is
reached. It can be seen that by advancing the spark timing increa-
ses and that the peak value is shifted towards TDC. From Case 1 to Case
3 the maximum volume is increased by a factor of six. Hence, the total
area where autoignition may occur grows considerably with the advance-
ment of the spark timing. This characteristic is used to define the mean
knock onset.

268
Figure 7: Histogram of the ignition probability distribution for Case 1
(left), Case 2 (center), and Case 3 (right) evaluated on the set . The
distributions are evaluated at the instant of their corresponding mean
knock onset.

It is assumed that the crank angle α max at which ( max) = is

reached is equal to the mean knock onset. The intrinsic idea of this appro-
ach is that a large volume increases the probability of autoignition at
different locations within .

Figure 7 shows the histograms of the ignition probability evaluated so-

lely on the set for all three cases of the spark timing sweep. The his-
tograms are determined at the instant of their corresponding mean knock
onset. It can be seen that for Case 1 the maximum ignition probability is
less than 7%. With the advancement of the spark timing the frequency of
finding higher values of increases considerably. For Case 3 the maximum
values of are close to 40%. It should be noted that the values of the
frequency are only relative to the volume , which corresponds to
the volume at the instant at which the histogram is evaluated. Hence, the
absolute values of the frequency cannot be directly compared. However,
the maximum values of can be used as a rough approximation for the
number of knocking cycles.

269
Figure 8: Comparison of the simulated mean knock onset (left) and num-
ber of knocking cycles (right) with experimental data for a spark timing
variation.

The simulated mean knock onset of the spark timing sweep and the cor-
responding number of knocking cycles are compared with experimental
data in Figure 8. The number of knocking cycles is determined based on
the maximum ignition probability at the instant of the mean knock onset.
It can be seen that the simulated mean knock onsets of the spark timing
sweep are in good agreement with the experimental data and the trend
is well captured. The absolute values of the mean knock onset are only
slightly overpredicted by the simulation as it is evident from Figure 8 (left).
The largest deviation between simulation and experiment are observed
for Case 1 with a spark timing of spark = 2.4° CA aTDC. This is not sur-
prising, since the number of knocking cycles in the experiment was less
than 0.3%. Hence, the computation of a mean knock onset which is based
only on three knocking cycles is associated with statistical inaccuracy.
Figure 8 (right) shows that the number of knocking cycles is clearly over-
predicted by the applied knock model. It has to be recognized that the ge-
neralized knock model provides only the probability of autoignition and not
of engine knock. Although autoignition is a necessary condition for engine
knock to occur, it is only one among other conditions for the development
of knocking combustion. Another reason for the overprediction is the use
of a lower integration limit of = 0.95 for the evaluation of the ignition
probability (Eq. (10)). Higher values for the lower bound would result in
a decrease of the ignition probability and, therefore, in a reduction of the
number of knocking cycles. It has to be recognized that the number of

270
knocking cycles will always be overpredicted by the gKIM model. Howe-
ver, the trend is well captured - with the advancement of the spark timing
the number of knocking cycles increases.

5. Conclusions
The generalized knock integral was applied to assess the impact of a
spark timing variation on the knock behavior of a turbocharged DISI engi-
ne. Special emphasis was put on an analysis of the impact of residual gas,
mixture fraction, and temperature distribution on engine knock. Based on
the statistical information provided by the knock model, the ignition proba-
bility at each location within the combustion chamber was determined.
Two distinct regions were identified with increased probability of engine
knock occurrence: in the gap between the bottom part of the exhaust val-
ve disk and the combustion chamber walls as well as in the exhaust valve
pocket, respectively. These regions are characterized by high tempera-
ture and fuel-rich mixture composition, which considerably decreases the
local ignition delay time. Moreover, these areas are also characterized
by relatively large turbulent fluctuations of mixture fraction and moderate
fluctuations of temperature. As a result, large fluctuations of the ignition
progress variable are induced. This shows that it is of significant import-
ance to account for the impact of turbulent fluctuations of mixture fraction
and temperature on the autoignition process. It was shown that the model
is able to predict engine knock with reasonable accuracy. The simulated
mean knock onsets were in good agreement with experimental data. Mo-
reover, the trend of the numbers of knocking cycles with respect to a spark
timing variation was well captured. This can be considered as the maxi-
mum information that can be gained from a RANS simulation. An impro-
ved description with higher quantitative predictability can only be obtained
with Large Eddy Simulations (LES), which can capture the cycle-to-cycle
variations of the underlying flow and mixing field. However, given the sig-
nificant increase of computational requirements for LES, knock prediction
in the industrial context will depend on RANS-based models such as the
one presented here.

271
Literature
[1] Linse, D.; Kleemann, A.; Hasse, C.: Probability Density Function
Approach Coupled with Detailed Chemical Kinetics for the Prediction
of Knock in Turbocharged Direct Injection Spark Ignition Engines, in
press, Combustion and Flame, 2013
[2] Linse, D.: Simulation and Modeling of Knock and Nitric Oxide Emis-
sions in Turbocharged Direct Injection Spark Ignition Engines, PhD
thesis, Technische Universität Bergakademie Freiberg, 2013
[3] Livengood, J. C.; Wu, P. C.: Correlation of Autoignition Phenomena
in Internal Combustion Engines and Rapid Compression Machines.
In: 5th Symposium on Combustion, pp. 347–356, The Combustion
Institute, 1955.
[4] Curran, H. J.; Gaffuri, P.; Pitz, W. J.; Westbrook, C. K.: A Compre-
hensive Modeling Study of Iso-Octane Oxidation. In: Combustion and
Flame, vol. 129, pp. 253, 2002.
[5] Worret, R.; Spicher, U.: Klopfkriterium. In: Heft 741-2002, FVV-Ab-
schlußbericht, Vorhaben Nr. 700, Forschungsvereinigung Verbren-
nungskraftmaschinen e.V., 2002.

272
14 Investigation on knocking
combustion with reaction kinetics
for a turbocharged SIDI engine
Werner Holly, Michael Heiss, Nikola Bobicic, Thomas
Lauer, Stefan Pritze

Abstract
For highly boosted gasoline engines with direct injection (DI) the high-
load operating conditions with lowest fuel consumption are restricted by
irregular combustion like knocking. Therefore, the analysis of the initiation
mechanism for knocking was the subject of this research work.

Cylinder pressures and burn rates were measured at the knock limit for a
turbocharged DI test engine. A 1D multi-cycle three-pressure-analysis was
applied to investigate the cycle-to-cycle variations. The results showed
that important mean charge properties such as the charge mass or the
composition were within a narrow range after inlet valve closing. Further-
more, it could be shown that the duration until 2 % mass fraction burned
is a clear indicator if autoignition occurs for a given cycle. All cycles with
an accelerated early flame development showed an irregular heat release
during the combustion phase. A CFD-simulation was carried out in order
to study the mixture properties and turbulence close to the spark plug.

The self-ignition of the unburned mixture was simulated with a stochastic

reactor model. The mixture properties were modelled with probability den-
sity functions. A toluene reference fuel was used to describe the chemistry
of gasoline with a detailed chemical mechanism. With this approach, the
onset of the timing was calculated in good agreement with the measure-
ments. The formation of typical intermediate species during the autoigniti-
on could be identified. A critical mass fraction of HO2 in the range of 10-4
was derived for the knocking cycles.

Kurzfassung
Die Effizienz von aufgeladenen Ottomotoren ist bei hohen Lasten durch
die klopfende Verbrennung limitiert. Deshalb war das Ziel dieser Arbeit,

273
die Entstehungsmechanismen die zur Selbstzündung im Endgas führen,
zu untersuchen.

Zylinderdrücke und Brennraten wurden für einen Turbomotor mit Direkt-

einspritzung vermessen, um das Betriebsverhalten an der Klopfgrenze zu
untersuchen. Mittels einer 1D Multi-Zyklen Three-Pressure-Analyse wur-
den die zyklischen Schwankungen untersucht und statistisch ausgewer-
tet. Es zeigte sich, dass die gemittelten Zylindergrößen wie Zylindermas-
se oder Restgasgehalt nach Einlass-Schließen innerhalb eines engen
Wertebereichs liegen. Dagegen konnte ein eindeutiger Zusammenhang
zwischen der Brenndauer bis 2 % Massenumsatz und dem Auftreten von
Selbstzündungen festgestellt werden. Alle Zyklen mit schneller erster
Flammenausbreitung wiesen während der Verbrennung eine irreguläre
Wärmefreisetzung auf. Die Gemischeigenschaften und die Turbulenz in
der Nähe der Zündkerze wurden mit der CFD-Simulation ausgewertet.

Die Selbstzündung des unverbrannten Gemisches wurde mit einem stoch-

astischen Reaktormodell modelliert. Die Gemischeigenschaften wurden
durch Wahrscheinlichkeitsdichtefunktionen abgebildet. Die chemischen
Eigenschaften des Kraftstoffes wurden mit einem detaillierten chemi-
schen Mechanismus eines abgestimmten Modellkraftstoffes abgebildet.
Es konnten so die Klopfzeitpunkte repräsentativer klopfender Zyklen in
guter Übereinstimmung mit den Messungen berechnet werden. Wichti-
ge Zwischenspezies und Radikale, die während des Selbstzündungsvor-
gangs gebildet werden, konnten ermittelt und als Indikator für den Reakti-
onsfortschritt bewertet werden. Ein kritischer HO2 Massenanteil von 10-4
wurde für die klopfenden Zyklen beobachtet.

1. Introduction
The future need for efficient gasoline engines with lowest CO2 emissions
suggests combustion concepts with high boost pressures and high po-
wer density. For such engines, the combustion is restricted by irregular
combustion like knocking with critical pressure peaks and oscillations [1].
Common remedies to avoid knocking combustion like a delayed spark
timing or fuel enrichment are inevitably associated with an efficiency loss
[2]. The presented paper deals with experimental and numerical investi-
gations on knocking combustion with the goal to understand the mecha-
nisms behind autoignition and to identify characteristic properties of the
combustion that typically lead to a knocking combustion.

274
2. Impact of cycle-to-cycle variations on knocking
combustion

2.1. Experimental investigation and multi-cycle

­analysis
A turbocharged 4-cylinder DI test engine was set up on a test bench and
was equipped with high- and low-pressure indication in order to investiga-
te the knocking combustion. The measurements were carried out at two
operating points with spark timing at the knock limit, see Table 1.

Engine Equivalen- Spark

Operating- IMEP
speed ce ratio timing
point [bar]
[rpm] [-] [°CAbTDC]
1 2,000 18.4 1 3.6

2 4,300 21 1.28 9.4

Table 1: Investigated operating points

The knock limit for the test bench measurements was defined as 2-3 de-
tected knock events per 100 cycles. In order to determine the bulk proper-
ties at inlet valve closing for the knocking cycles, a three pressure analysis
(TPA) was carried out with GT-Power for 150 consecutive measured cy-
cles. The measured pressure signals for intake, exhaust and the cylinder
were applied as model boundaries to reproduce the cycle-to-cycle varia-
tions of the charge properties and the burn rates.
In a first step, the bulk properties of the cylinder charge that are general-
ly regarded as knock critical like the pressure, temperature and residual
gas fraction were investigated at inlet valve closing and spark timing. The
simulation results showed that the fluctuations of these parameters lay wi-
thin a narrow range without any noticeable outliers that would indicate the
knocking cycles. In contrast, the burn rate varied considerably at the very
beginning of the flame propagation. Fig. 1 shows the 0-2 % burn duration
of 150 consecutive cycles at 2,000 rpm. Although the applied knock sen-
sor on the test engine detected knocking only for the two cycles that are
marked with black circles, the multi-cycle simulation showed an irregular
heat release for all fast burning cycles with a 0-2 % burn duration of 10

275
°CA or below (circles with dotted lines). These cycles showed exothermic
pre-reactions that are precursors of knocking combustions. Furthermore,
it becomes obvious from Fig. 1 that the very first phase of the combustion
is a clear indicator whether knocking combustion occurs for a given cycle
or not. A sharp threshold for the 0-2 % burn duration could be observed
between knocking and non-knocking cycles, see Fig. 1.

Figure 1: 0-2 % combustion duration of 150 consecutive cycles at 2,000

rpm

In contrast, there was no correlation between the 10-90 % combustion

duration and the cycles with autoignition, see Fig. 2. This means that the
flame development immediately after the spark ignition and therefore the
mixture properties in the vicinity of the spark plug have the highest impact
on the cycle-to-cycle variations. Fast burning cycles with favourable char-
ge properties at the spark plug support a quick development of the initial
flame. Therefore, the temperatures in the unburned zone are high enough
to reach the threshold of autoignition before the regular flame burns the
mixture.

276
 Figure 2: 10-90 % combustion duration of 150 consecutive cycles at
2,000 rpm

To get a better understanding how the cyclic variations of the combustion

effect the autoignition process, the 0-2 % and 10-90 % burn duration (BD)
are plotted against the point of 50 % mass fraction burned (MFB) in Fig. 3.

Figure 3: Correlation between 50 % MFB, 0-2% BD and 10-90 % BD of

150 consecutive cycles at 2,000 rpm

277
A distinct statistical correlation between the 0-2 % burn duration and 50 %
MFB is visible. This confirms that already the very first combustion phase
has a major influence on the duration of the entire combustion process
and therefore the knock tendency. In contrast, the 10-90 % burn duration
shows a less distinct correlation with 50 % MFB. This behaviour was si-
milar for OP 2.

2.2. Investigation of the spatial charge properties

As previously discussed, the very first combustion phase determines the
entire combustion duration. This suggests that inhomogeneities of the
cylinder charge and particularly the charge properties close to the spark
plug have a high impact on the knocking combustion [3, 4, 11]. To inves-
tigate the spatial distribution of local mixture properties like the tempera-
ture, equivalence ratio and turbulent kinetic energy in the vicinity of the
spark plug, CFD-simulations were carried out for both engine speeds. A
surrogate fuel consisting of seven hydrocarbons with different boiling tem-
peratures was used for an accurate representation of the real fuel evapo-
ration and mixture cooling [5, 6]. The CFD wall temperature boundary
conditions for each engine speed were pre-calculated with a 1D cycle
simulation including a simple FE-based wall temperature solver.
Fig. 4 shows contour plots of the local lambda value and the turbulent
kinetic energy close to the spark plug. The results reveal considerable
inhomogeneities for both properties. Particularly for the turbulence, the
spark plug acts as an obstacle that disturbs the turbulent flow field resul-
ting in steep gradients close to the spark gap. CFD-simulations based on
a RANS-approach only solve for ensemble averaged properties what is a
well known limitation when discussing cyclic variations. However, the gra-
dients suggest that there will be significant differences of the local mixture
properties from cycle-to-cycle and hence the conditions for the early flame
development will vary considerably [7].
Therefore it can be stated that the variations of the very first combustion
phase are mainly determined by the turbulent flow field close to the spark
plug in combination with the spark development. Both have a consider-
able influence on the early flame development, the direction of the flame
propagation and therefore the knock probability.
In order to initialise the stochastic reactor model accurately, see the follo-
wing chapter, the temperature distribution in the whole combustion cham-
ber was analysed at inlet valve closing and at ignition timing. Fig. 5 shows
the results of the CFD-analysis for the two engine speeds.

278
 Figure 4: Lambda- (top) and turbulence distribution (bottom) at ignition
timing in orthogonal sectional views through the spark plug at 2,000 rpm

Figure 5: Temperature distributions at IVC for 2,000 rpm and 4,300 rpm

It becomes obvious that at 4,300 rpm there is a higher temperature inho-

mogeneity compared to 2,000 rpm. This can be explained with the mixture
formation and the resulting mixture cooling, which is illustrated in Fig. 6.

279
At 2,000 rpm, the spray cone becomes less distorted by the momentum
of the incoming air than for the higher speed (Fig. 6). In contrast, for the
high engine speed the spray deflection separates cooler regions with fuel
evaporation and regions with increased charge temperature. This sepa-
ration remains stable until spark timing due to the high tumble flow. Com-
bined with a shorter mixing time this results in a more inhomogeneous
distribution of the fuel and consequently in higher temperature differences
in the combustion chamber.
The temperature distribution at IVC was used to initialise the stochastic
reactor model. This will be described in the following section.

Figure 6: Simulated spray penetration into the cylinder and temperature

distributions at end-of-injection (EOI) for 2,000 rpm (left) and 4,300 rpm
(right)

3. Analysis of knock initiation with reaction kinetics

3.1. Setup of the reactor model

The autoignition of the unburned mixture was simulated with a 0D reactor
model and detailed chemistry [8, 15]. The inhomogeneities in the cylin-
der were modelled with probability density functions (PDF) and discreti-
zed with 100 stochastic particles. For each time-step, the chemistry, heat
transfer and turbulent mixing processes were solved for each stochastic
particle using a turbulent mixing model, which will be described in the fol-
lowing section. The heat release of the regular combustion was modelled
using the Wiebe-approach [14]. The reaction kinetics are very sensitive

280
with respect to temperature. In order to initiate the cylinder temperature
as accurate as possible, the PDF of the temperature from the CFD-simu-
lation at intake valve closing was assigned to the stochastic particles, see
Fig. 5.
The reaction mechanism of a Toluene-Reference-Fuel (TRF) was used
that consists of toluene, isooctane and n‑heptane [9]. It contains the reac-
tion mechanisms of the three components as well as their cross reactions
for low and high temperature combustion (137 species, 633 reactions).
The appropriate volume fractions of the three components were deter-
mined according to the RON/MON mapping described by Morgan et al.
[10] to satisfy a RON of 95 and a MON of 85, see the components’ volume
fractions in Table 2 next to the pure components.

Vol-% of Vol-% of Vol-% of

Fuel RON MON
Isooctane n-Heptane Toluene
Isooctane 100.0 0.0 0.0 100.0 100.0

n-Heptane 0.0 100.0 0.0 0.0 0.0

Toluene 0.0 0.0 100.0 120.0 109.0

PRF 95 95.0 5.0 0.0 95.0 95.0

TRF 20.64 18.82 60.54 95.0 85.0

Table 2: Octane numbers of different fuels

To study the influence of the reference fuel components, Fig. 7 shows the
calculated ignition delay times in a constant volume reactor for the fuel
mixtures of Table 2. The calculations were carried out for an initial pressu-
re of 60 bar and a stoichiometric air/fuel ratio.

281
 Figure 7: Self-ignition delay time at constant volume for different hy-
drocarbon mixtures

Higher alkanes like isooctane or n-heptane show the well documented two
stage ignition at low temperatures that is related to the negative tempera-
ture coefficient (NTC) at about 900 to 950 K, e.g. [12]. This is in contrast
to most non-alkanes. As an example the self-ignition time of the aromatic
toluene is illustrated in Fig. 7. Due to the content of non-alkanes in gaso-
line, surrogate fuels that are purely based on alkanes and iso-alkanes
show a different low temperature chemistry than gasoline. Particularly, the
load sensitivity and the difference between RON and MON will not be
predicted [10].
For demonstration, a primary reference fuel (PRF) consisting of isooctane
and n-heptane is added in Table 2 and Fig. 7. The difference between the
TRF and PRF fuel is clearly noticeable.

3.2. Modelling of the turbulent mixing processes

During the compression stroke, the spatial distribution of the mixture pro-
perties is affected by the in-cylinder flow, the turbulent mixing and the heat
transfer to the walls. In order to model these mixing processes with the
PDFs a turbulent mixing time (TMT) is used that models the spatial energy
and species transport. It is proportional to the ratio of the turbulent kinetic
energy k and the dissipation rate ε [15], see Eq. 1.

282
 (1)

C is a constant. The value of k/ε was determined for both operating points
with the CFD-simulation using a 2-equation k-ε turbulence model. Its
mean value in dependency of the crank angle is shown in Fig. 8.

Figure 8: Values of TMT (k/ε) for 2,000 rpm and 4,300 rpm and the ratio
between both values

Additionally, the ratio of the turbulent mixing time between both operating
points is plotted on the secondary y-axis. It is noticeable that the TMT sca-
les with the mean piston speed. The constant C had the same value for
both engine speeds. Therefore, this parameter must be determined only
once for a given combustion process.
Fig. 9 shows for the mean cycle of both operating points a comparison of
the temperature distributions in the unburned zone between the stocha-
stic reactor model (SRM) and the CFD at 50 % MFB. The evolution of the
temperature distribution is well described by the TMT model compared to
the CFD-results. The marked areas with high temperature indicate par-
ticles, where first exothermic pre-reactions have started but did not lead to
knocking combustion because the reactions were not fast enough for the
mean cycle. The hottest areas do not become visible in the CFD simula-
tions because no knock model was used and therefore no pre-reactions
were computed.

283
Figure 9: Temperature distributions at 50 % MFB for 2,000 rpm (top) and
4,300 rpm (bottom)

3.3. Comparison of calculated knock timing with test

bench data
For both operating points a representative knocking cycle was chosen
from the measurements to reproduce the autoignition with the stochastic
reactor model. The cylinder pressures and burn rates are depicted as
dotted grey lines and the simulation as solid black lines in Fig.  10. The
results of the calculated cylinder pressures and heat releases in Fig. 10
reveal that the timing of the irregular heat release due to autoignition of
the endgas (arrows) is in good agreement with the measurements for both
operating points.

284
 Figure 10: Comparison of the cylinder pressure and heat release rate
between measurement and SRM for 2,000 rpm (top) and 4,300 rpm (bot-
tom)

In Fig. 11 the calculated temperatures for all particles are shown over
crank angle at 2,000  rpm. All particles with regular combustion are de-
picted as thin grey lines. They are characterised by an almost vertical
increase to a high temperature level above 2,500 K which means that the
particle has been entrained into the hot flame. In contrast to this, self-ig-
niting particles show a progressive rise in temperature until self-ignition,
followed by a fast oxidation of the remaining fuel (thick black curves). The
first autoignition occurs at 28 °CAaTDC and denotes the onset of the kno-
cking combustion.

285
Figure 11: Calculated temperature of all 100 particles for a representative
knocking cycle at 2,000 rpm

3.4. Formation of chemical radicals during knock initia-

tion
As discussed in many reference books [1, 9, 12, 13] and already sugge-
sted in chapter 3.1, the self-ignition of hydrocarbons is characterised by
several temperature regimes with the formation and decay of specific in-
termediate species. At low temperature, the peroxy chemistry leads to the
chain-branching steps. An important reaction which initiates the peroxy
chemistry is given in Reaction 1:

R + O2 RO2 (R1)

The alkyl radical R is oxidised in a reversible reaction to form the peroxy

radical RO2. After several isomerisation and oxidation reactions via RO2
highly reactive hydroxyl radicals OH are formed, leading to a first tempera-
ture rise. At the resulting elevated temperature, the equilibrium of Reacti-
on 1 shifts increasingly to the left and chain-branching via RO2 is inhibited.
Reactions involving the alkyl radical R, according to Reaction 2, become
important for further branching:

286
R + O2 HO2 + alkene (R2)
RH + HO2 R + H2O2 (R3)
H2O2 + M OH + OH + M (R4)

This denotes the intermediate regime, where reactions of HO2 over H2O2
(R3) finally lead to the hot ignition with branching into two OH radicals
(R4). RH and M denote the hydrocarbon fuel or other intermediates, res-
pectively. Consequently, the formation of HO2 as an important intermedi-
ate was analysed in the following. Fig. 12 shows the mass fraction of HO2
(lower section) and OH (upper section) for all particles over crank angle
for a knocking cycle at 2,000 rpm. The thick, solid lines denote the mass
fractions in the self-igniting particles and the dotted lines the mean value
for the unburned region.

Figure 12: Calculated OH- (top) and HO2- (bottom) mass fractions of all
100 particles for a representative knocking cycle at 2,000 rpm

A first increase of HO2 and OH becomes visible between 5 and 20 °CAaT-

DC (Fig. 12). This is due to the relatively slow temperature increase via low
temperature branching that increasingly shifts the equilibrium in reaction 2
to the left with subsequent HO2 formation. At about 20 °CAaTDC the HO2
formation stagnates what indicates the first stage of ignition together with

287
the shift to the intermediate temperature regime, where the reaction rate
slows down. With increasing unburned zone temperature the HO2 forma-
tion proceeds until ignition with H2O2 decomposing into two OH radicals.
This can be seen by means of the steep increase of OH radicals in Fig.
12. The excess of OH-radicals immediately leads to a steep temperature
gradient at 28 °CAaTDC and an almost instantaneous consumption of the
remaining unburned mixture.
To compare the self-ignition process at different engine speeds, the mean
HO2mass fractions in the unburned zone for the investigated cycles are
shown in the lower section of Fig. 13. Additionally, the mean temperature
of the unburned zone is depicted in the upper section of Fig. 13. With in-
creasing engine speed the pre-reactions have less time to form radicals
and therefore the knock resistance increases with the engine speed. As a
consequence, the HO2 fraction between -30 and 10°CAaTDC is lower for
OP 2 than for OP 1, despite both operating points have the same unburned
temperature during compression. Due to an earlier spark timing and point
of 50 % MFB at the knock limit the unburned zone has a higher tempera-
ture and there are higher reaction rates during combustion. Consequently,
similar concentrations of HO2 occur for both mean cycles as well as for both
knocking cycles at the end of the combustion. A critical mass fraction of HO2
in the range of 10-4 can be derived for both knocking cycles.

Figure 13: Calculated mean temperatures (top) and HO2 mass fractions
(bottom) for four different cycles

288
The stagnation of the HO2 formation for all four cycles in Fig. 13 makes
clear that the intermediate regime and thus the first stage of ignition has
been reached even for the mean cycles what is typical for operating points
that are close to the knock limit. Nevertheless, the unburned zone tempe-
ratures of the mean cycles are not sufficient to ignite the mixture before it
is burned by the regular flame.

4. Summary and Conclusion

Experiments for two operating points at the knock limit for a 4-cylinder DI
test engine were the basis for an analysis of the chemical mechanisms
of irregular combustion. A multi-cycle analysis of the gas exchange and
combustion showed that bulk properties like the charge temperature at
inlet valve closing or the internal residual gas fraction were within a narrow
range over all cycles. Furthermore, it could be shown that a fast 0-2% burn
duration is a strong indicator for knocking cycles. This suggested that the
development of the flame kernel and the cycle-to-cycle variations mainly
depend on the charge properties and turbulence in close proximity to the
spark plug, what could be explained with CFD simulations.
In order to analyse the autoignition of the unburned mixture during a kno-
cking cycle, a detailed modelling of the reaction kinetics was carried out
using a stochastic reactor model that included the full chemistry of low and
high temperature combustion. A toluene reference fuel reaction mecha-
nism was used to simulate a fuel with a RON and MON according to the
fuel used at the test bench.
The prediction of the onset of autoignition was in good agreement with
the measurements for both operating points. The crank angle resolved
progress of typical intermediate species occurring during self-ignition, like
OH and HO2 was investigated in order to explain the mechanism of knock
initiation. With an analysis of the radical formation it could be shown that
the first stage of ignition and hence the transition to the intermediate tem-
perature chemistry already occurred for the mean cycle. However, due
to the lower unburned zone temperatures, the reaction rate slowed down
and self-ignition was inhibited. A critical threshold for the HO2 mass fracti-
on of 10-4 was determined for both investigated operating points.
For the analysis of knocking combustion, the combination of engine cycle
simulation, CFD-simulation and chemical reaction kinetics proved to be
an appropriate combination to investigate the complex mechanisms that
initiate knocking combustion. Nevertheless, further investigations on cy-
clic variations are necessary. Thereby, the development of a combustion

289
model that predicts the cycle-to-cycle variations with respect to the charge
properties and the turbulence intensity in the combustion chamber could
be a next step. The development of further mechanisms will allow exten-
ding this approach on different fuel qualities and fuel chemistry. Finally,
large eddy simulations have the potential to calculate the knock onset with
respect to the local properties of the unburned mixture for consecutive
cycles. However, this approach is still too expensive in a numeric sense
but will become more important with increasing computer power during
the next years.

290
References
[1] Heywood, John B.: Internal Combustion Engine Fundamentals. New
York: McGraw-Hill 1988. ISBN 0-07-100499-8
[2] Eichlseder, H. et al.; List, H. (Ed): Grundlagen und Technologien des
Ottomotors - Der Fahrzeugantrieb. Wien: Springer Verlag 2008. ISBN
978-3-211-25774-6
[3] Heiss, M.; Bobicic, N.; Lauer, T.; Geringer, B.; Schmuck-Soldan, S.: A
detailed analysis of the initiation of abnormal combustion with reacti-
on kinetics and multi-cycle simulation, Fisita 2012-World Automotive
Congress, F2012-A06-021
[4] Schießl, R.; Maas, U.: Analysis of endgas temperature fluctuations in
an SIengine by laser-induced fluorescence, Combustion and Flame
133 (2003) 19-27
[5] Batteh, J. J.; Curtis, E. W.: Modeling Transient Fuel Effects with Alter-
native Fuels, SAE Paper 2005-01-1127, 2005
[6] Heiss, M.; Lauer, T.: Simulation of the Mixture Preparation for an
SI Engine using Multi-Component Fuels, STAR Global Conference
2012, Amsterdam: 2012.
[7] Holmström, K.; Denbratt, I.: Cyclic Variation in an SI Engine Due to
the Random Motion of the Flame Kernel, SAE Paper 961152, 1996
[8] Gogan, A.; Sundén, B.; Lehtiniemi, H.: Mauss, F.: Stochastic Model
for the Investigation of the Influence of Turbulent Mixing on Engine
Knock, SAE Paper 2004-01-2999, 2004
[9] Andrae, J.C.G. et al.: HCCI experiments with toluene reference fuels
modeled by a semidetailed chemical kinetic model, Combustion and
Flame 155 (4) (2008) 696–712
[10] Morgan, N. et al.: Mapping surrogate gasoline compositions into
RON/MON space, Combustion and Flame 157 (2010) 1122–1131
[11] Pöschl, M.; Sattelmayer T.; Influence of temperature inhomogeneities
on knocking combustion, Combustion and Flame 153 (2008) 562–573
[12] Warnatz, J.: Maas, U.; Dibble, R. W.: Verbrennung: Physikalisch-Che-
mische Grundlagen, Modellierung und Simulation, Experimente,
Schadstoffentstehung. Berlin: Springer, 2001
[13] Compton, R.G.; Hancok, G.: Low-Temperature Combustion and Au-
toignition. Comprehensive Chemical Kinetics, Volume 35. Amster-
dam: Elsevier 1997

291
[14] Heiss, M.; Lauer, T.; Geringer, B.: A detailed analysis of engine knock
initiation by means of a stochastic reactor model. ECCOMAS, Vienna
2012
[15] cmcl Innovations: SRM Suite User Manual, Version 7.2.1. Cambrid-
ge: 2012.

292
15 Knocking simulation at
Mercedes-Benz – application in series
production d
­ evelopment
Arnold Kaden, Alfred Frommelt, Helmut Gildein,
Eberhard Kraus, Uwe Schaupp

Abstract
The ongoing trend towards increasing power density while at the same
time improving fuel consumption objectives places ever greater emphasis
on refining the knocking characteristics associated with developing the
combustion system of gasoline-powered engines.

This article discusses the use of CFD simulation techniques for develo-
ping the gasoline combustion system at Mercedes-Benz with a focus on
engine knocking.

To this end, a turbocharged engine with direct injection is used to illustrate

the application of a combustion model together with an active knocking mo-
del. CFD simulation facilitates a detailed analysis of flame propagation and
the operative conditions in the end gas zones. Coupling the spontaneous re-
lease of heat in these end gas zones with the normal combustion cycle leads
to a deeper understanding of how knocking progresses with respect to time
and space. The results obtained are then compared to quantitative measure-
ments, and the sensitivity of the simulation approach covered is characterized
by varying typical engine parameters such as intake and surface temperatures.

Attention will also be paid to pre-ignition and its correlation with specific
engine oil properties as investigated in a project conducted by the FVV.
A CFD simulation exercise likewise highlights the possibilities afforded by
the piezo injection technology developed by Mercedes-Benz for optimizing
the pre-ignition performance of an engine currently in series production.

1. Introduction
Development activities surrounding the combustion system of gaso-
line-powered engines have been characterized by two predominant trends
for some time. The first trend is downsizing combined with ever-higher

293
turbocharging pressures and an operative displacement of engine torque
toward the lower speed range (low-end torque) in an effort to increasing
driving pleasure while making an active contribution to achieving target
fuel consumption objectives. The second trend is elevating compression
as far as possible to further approach these consumption objectives in
particular. Doing this, however, makes gasoline engines more susceptible
to the effects of knocking, which is why even more emphasis must be pla-
ced on this aspect when developing the combustion system.

The engine development program at Mercedes-Benz is heavily influenced

by computer-aided simulation techniques, which are integrated in the con-
cept phase early on (front loading) and are closely correlated with different
simulation techniques and systematic performance summaries for the di-
gital assembly stage with release approval coordination. [1]
Viewed in this context, CFD simulation forms an essential design compo-
nent of combustion system development. Different aspects of this simula-
tion application have been reported on in the last few years. [2]

2. Knocking Simulation
This article explains the methodology behind knocking simulation and il-
lustrates different application scenarios.

2.1. Knocking Simulation Methodology

Knocking simulation is based on a realistic representation of the com-
bustion process and maps the processes gas exchange, injection, mix-
ture formation, ignition and combustion (see Figure 1). A well-tuned spray
model that characterizes the properties of the fuel injector in question is
critical in addition to detailed geometric mapping of the combustion cham-
ber and port geometry in the calculation mesh used. For this purpose, an
internally developed model is used that simulates the characteristics of
the fuel injector with outward-opening nozzle needle, the resulting dro-
plet spectrum, and the dynamics of the piezo actuators. Simulation of the
injection and combustion process is based on the fundamental design of
the well-known ECFM model [3], which was enhanced to include several
submodels.
All of these measures are critical in realizing an accurate knocking simula-
tion exercise. The knocking model itself uses a correlation for the ignition
delay time with respect to the local lambda (oxygen sensor), pressure,

294
end gas temperature, EGR rate and octane number of the fuel. As soon as
operative conditions allow for self-ignition, the spontaneous, local release
of heat is coupled into that of the normal combustion process. The model
is therefore an active model [4].

Figure 1: Combustion process

3. Application Example
The following illustrates the application of the aforementioned methodo-
logy on a typical Mercedes-Benz engine. The four-cylinder engine ana-
lyzed integrates all components of the Mercedes-Benz combustion sys-
tem: four-valve design, turbocharging, piezo injector in central combustion
chamber position and arrangement of the spark plug between the exhaust
valves.
An operating point of 1,500 rpm, high turbocharging pressure, and λ = 1
operation are observed for each of the analyses below.

3.1. Basis
The indicator evaluation and CFD simulation are first compared with each
other to establish a basis for the following parameter variation. To this end,

295
a CFD simulation is carried out using all engine configuration parameters
(valve timing, intake manifold pressure and temperature, injecting timing,
and ignition point).
Figure 2 compares the rate of cylinder pressure during the high-pressure
stage in this simulation to that of the indicator measurement for the four
cylinders of the engine. A good correlation can be observed.

Figure 2: Comparison of rate of cylinder pressure in CFD simulation to

that of indicator measurement

This operative state is consistent with engine operation at the knock limit.
To better represent dependencies in the simulation so that more accurate
statements can be made by also leveraging a cycle-averaged RANS si-
mulation, pronounced knocking operation will now be examined.

3.2. Variations in Intake Manifold Temperature

The intake manifold temperature is increased by 20 °C with respect to the
aforementioned application state. The intake manifold pressure is also
increased at this time until the same intake quantity of fresh air is realized
so that the identical amount of fuel can be injected. All other engine ope-
rating parameters remain unchanged. The average combustion chamber

296
compression temperature at tdc thereby increases by 20 kelvin and leads
to pronounced knocking. Figure 3 compares the cylinder pressure and
heat release rate for the base operating state and with increased intake
temperature.

Figure 3: Cylinder pressure and combustion curve

In the scenario with the increased intake temperature, one sees a steep
pressure gradient at a crank angle (KW) of approximately 750 degrees,
which can be interpreted as knocking. The heat release comparison (Fi-
gure 3, right) reveals a harmonious progression in the base operating sta-
te that releases only a minimal quantity of spontaneous heat at a crank
angle of approximately 755 degrees. In the operative scenario involving
the increased intake temperature, at a crank angle of approximately 750
degrees, there is a strong deviation from the heat release progression
associated with the harmonious base operating state, which is indicative
of pronounced knocking. A typical two-stage progression offers a more
detailed analysis of the knocking occurrence. Self-ignition first takes place
with a lower intensity, which is followed by a much more intense ignition
sequence at a further crank angle of 2 degrees that ultimately is respon-
sible for the steeper cylinder pressure gradient.
Figure 4 shows the propagation of the regular flame front (brown) and the
local distribution of the areas where self-ignition is recognize (cyan). The
regular flame front starts to propagate at the spark plug. Propagation is
influenced by the local conditions for charge movement, turbulence, tem-

297
perature, residual gas and lambda distribution. Self-ignition pockets below
the hot exhaust valves first form at a crank angle of 750 degrees. This
correlates with the initial, smaller rise in heat release. At a further crank
angle of two degrees, more voluminous self-ignition pockets form throug-
hout the entire combustion chamber and particularly on the intake side,
where larger end gas zones are present at this time. These self-ignition
sequences trigger the second, higher rise in heat release (Figure 3, right).

Figure 4: Flame propagation and self-ignition pockets

A more in-depth analysis of the progressional transitions between the

crank angles of 750 and 752 degrees reveals that the first self-ignitions
produce a pressure wave that causes additional end gas zones to self
ignite.
This pressure wave is depicted in the graphic at the right in Figure 5. Whi-
le the cylinder pressure distribution in the combustion chamber is nearly
constant during knock-free combustion (Figure 5, left), as soon as the
crank angle reaches 750 degrees and the first self-ignition sequence oc-
curs below the hot exhaust valves, an intense pressure wave is generated
in the combustion chamber that coincides with amplitudes of over 10 bar.
This is where the active knocking model comes in, which couples in the
spontaneous release of heat as a result of local knocking events.

298
Figure 5: Cylinder pressure distribution at a crank angle of 750 degrees

3.3. VisioKnock measurement

Figure 6 shows the outcome of a VisioKnock analysis for the engine in
question to validate the above results. The actual engine obviously could
not be transitioned to quantitative knocking operation in the extreme man-
ner that was possible during simulation.
It is also important to note that the majority of knocking regions exist on
the intake side and not on the exhaust side. The most intense events thus
occur on the intake side because this is where the largest quantity of end
gas resides. Due to the angled installation position of the spark plugs, the
knocking regions cannot be determined as accurately as would be the
case with a central spark plug arrangement, where the spark plug would
be in a vertical arrangement. The exhaust-side sensors point toward the
piston and the intake-side sensors toward the combustion chamber roof;
the pressure waves generated along the plane of symmetry can therefore
only be registered with a delay, and the point of origin is also somewhat
uncertain.
Making a direct comparison with the simulation results is difficult because
only a cylce-averaged RANS simulation was done. If one interprets the
knocking region distribution calculated as an individual cycle that behaves
in line with the mean value, this will reflect the most probable knocking re-
gion. It is in this context that the simulation and quantitative measurement
can be stated as coinciding very well.

299
 Figure 6: VisioKnock analysis of knocking region

3.4. Variations in Piston Temperature

Variations in the surface temperature of the combustion chamber are ex-
amined as additional factors that influence knocking as experienced in
gasoline engines. Starting with the knocking state described in section
3.2, the influence of a reduction in the piston temperature is analyzed by
e.g. spraying the bottom side of the piston with oil or integrating a different
piston design. This measure reduces the average combustion chamber
compression temperature at tdc by 4 kelvin. Figure 7 illustrates the effect
on the release heat rate. A considerable reduction in knocking can be
observed.

300
 Figure 7: Combustion curve for a reduction in piston temperature

Measurements taken using such a piston revealed a similarly improved

knock limit during trial testing and thus confirm these simulation results.

3.5. Variations in Exhaust Valve Temperature

Starting with the example referenced in 3.2 with an increased intake tem-
perature, the temperature of the exhaust valve is reduced by 50 kelvin. All
other parameters remain unchanged. Although the average combustion
chamber compression temperature at tdc drops by just 1 kelvin, the local
drop in temperature below the exhaust valves is much more pronounced.
Figure 8 illustrates the effect on the release of heat.
It becomes evident that the knocking events also occur at a much later
point and are less intense. The initial knocking event below the exhaust
valves in particular occurs later than in section 3.4; the second knocking
event on the intake side, however, takes place sooner and with a higher
intensity as the end gas temperature there was barely influenced by this
scenario.

301
Figure 8: Combustion curve for a reduction in exhaust valve temperature

Trial testing carried out using correspondingly modified exhaust valves

with an improved sodium fill level that is also realized in the valve disk
area demonstrate more favorable operative performance in the designa-
ted quantity.

These examples underscore the significance of a sufficiently dimensioned

intercooler system and relatively low component temperatures in the com-
bustion chamber for highly turbocharged engines.

4. Pre-Ignition
The results shown thus far pertain to the knocking events that always lead
to a spontaneous release of heat as already triggered by normally induced
ignition combustion and the associated rise in pressure in the end gas
region. These knocking events can therefore be easily influenced by the
ignition timing.

Other phenomena are also observed, however, especially in the case of

highly turbocharged engines, in which a significant amount of heat is al-
ready released far before ignition occurs at the spark plug. This phenome-
non is known as pre-ignition (VE) and poses a considerable risk because
extremely high peak pressures are generated at this time and the change
of ignition timing is useless.

302
4.1. Phenomenology of Pre-Ignition
Many analytical investigations have revealed a wide variety of sources
of pre-ignition [5]. These causes generally fall into one of two categories:
excessively high local component temperatures, or hot spots, and the pre-
sence of “ignition improvers”. Hot spots can be the spark plug or exhaust
valves, for example. “Ignition improvers” can refer to particulate matter
that dislodges from deposits in the combustion chamber and becomes
very hot or oil droplets that accumulate in the blow-by gas return circuit or
enter the combustion chamber through leaky valve stem seals.

Figure 9 depicts important factors that influence the occurrence of pre-ig-

nition.

Figure 9: Factors that influence pre-ignition

An additional factor has also emerged in the recent past and relates to the
use of specific samples of lubricating oil in conjunction with pre-ignition oc-
currences, particularly where engines with direct injection are concerned.
This correlation was also focused on a FVV project entitled “Vorentflam-
mung bei Ottomotoren 2” (pre-ignition in gasoline engines 2), which was
overseen by Daimler AG.
Additional internal testing has, among other things, revealed that the wet-
ting of the liner surface area by the injected fuel leads to the formation of

303
a fuel-oil mixture in the top land of the piston that is randomly flung throug-
hout the combustion chamber and causes highly spontaneous pre-igniti-
on occurrences when modern, low-friction oils containing specific additive
packages are used.

Figure 10: Pre-ignition frequency as a function of coolant temperature

and oil samples

Of particular interest is the correlation between the increasing pre-ignition

frequency and lower coolant temperature level (Figure 10). This negati-
ve temperature coefficient ultimately provides verification of the fact that
pre-ignition cannot occur due to gas-induced self-ignition, but is the result
of secondary ignition mechanisms that enter the combustion chamber and
cannot yet be captured directly by the simulation process.

4.2. Optimizing Pre-Ignition

Armed with this knowledge, the circumstances surrounding combustion
chamber wetting – particularly with respect to the liner surface – become
even more important. The fuel injector used by Mercedes-Benz therefore
offers exceptional operative possibilities thanks to its ability to inject small
quantities of fuel at intermittent intervals.
Although pre-ignition cannot be directly described in a CFD simulation,
the mechanism that causes it (i.e. wetting of the cylinder barrel liner) is
very well represented and accounted for. It is against this background that
applying CFD simulation techniques makes it possible to understand the

304
interaction between the injected fuel spray, the flow of the gas exchan-
ge and combustion chamber wetting so that the injection timing system
can be optimized. Figure 11 charts typical results that are achieved. By
integrating an optimized, triple-point injection, able to minimize wetting of
the cylinder liner with fuel by a factor of 5 as compared to a single-point
injection. During testing of the engine selected, which was operated using
a pre-ignition-critical oil sample, this measure was sufficient in eliminating
almost all pre-ignition occurrences.

Figure 11: Comparison of liner wetting calculated with pre-ignition fre-

quency measured

This optimized injection timing was further enhanced by injecting a small

quantity of fuel directly before the ignition timing, which considerably in-
creased turbulence in the vicinity of the spark plug (see Figure 12), there-
by improving combustion stability at the same time.

Figure 12: Comparison of turbulence distribution before ignition using a

single-point and a triple-point injection

305
5. Summary
The methodology for simulating engine knocking as applied by Merce-
des-Benz was described and explained using relevant examples. The re-
sults of the simulation exercises were also validated and confirmed via
optical combustion analysis and indicator measurements. Key influencing
factors such as the intake air temperature and the surface temperatures
of various components were examined. The different effects of global and
local temperature reduction on the operative knocking event were shown.
The pre-ignition phenomenon and the different response to specific en-
gine operating parameters that was observed in certain scenarios have
been explained. The significance of avoiding or preventing liner wetting
to optimize pre-ignition behavior through the use of a piezo injector was
illustrated.

306
Bibliography
[1] Kaden, A.; Gildein, H.; Liang, Z.; Schaupp U.; Sundheim D.: Einlass-
kanalauslegung in einem modernen Serienentwicklungsprozess. 4th
MTZ symposium entitled, „Ladungswechsel im Verbrennungsmotor“,
2011
[2] Kaden, A.; Altenschmidt, F.; Gildein, H.; Kraus, E.; Schaupp, U.; Pri-
lop H.: Analysis of soot formation by optical measurement and CFD
simulation on a turbocharged gasoline engine with direct injection.
9th International Symposium on Combustion Diagnostics, 2010
[3] Colin, O.; Benkenida, A.; Angelberger C.: 3D Modelling of Mixing, Ig-
nition and Combustion Phenomena in Highly Stratified Gasoline En-
gines. Oil & Gas Science and Technology – Rev. IFP, Vol. 58, 2003
[4] Lafossas, F.A; Castagne, M.; Dumas, J.P.; Henriot S.: Development
and Validiaton of a Knock Model in Spark Ignition Engines Using a
CFD code. SAE 2002-01-2701, 2002
[5] Spicher, U.: Untersuchung des Auftretens und der Ursachen von
Selbstzündungen vor Zündungseinleitung bei aufgeladenen Motoren
mit hohem Verdichtungsverhältnis. Research Association for Com-
bustion Engines fall conference, final report entitled, „Vorentflam-
mung bei Ottomotoren“, 2009

307
16 The DELTA knocking control – the
necessary paradigm shift for engines
with high power density
Alexander Stahr, Philipp Langfritz, Michael Guenther,
Matthias Kratzsch

Abstract
This article describes the necessity of introducing a new characteristic
value for controlling the knocking intensity. The structure-borne sound
information from the knocking sensor is used to determine the current
knocking intensity in a quality that is comparable with that of pressure indi-
cation. The article describes an adaptation of a measurement chain requi-
red to guarantee absolute accuracy of the knocking intensity value. The
definition of statistical parameters enables simplification in the evaluation
of the knocking sensor signal and a pronounced enhancement in robust-
ness, also toward interference noise. The continuity of the new parameter
permits the application of a continuous controller to enable adjustment
of the ignition timing as the needs dictate. This cuts adjustments of the
ignition timing to a minimum. The article describes system properties that
allow the control to implement early adjustment of ignition in the event that
the cylinders have not been pre-set ideally.

1. Introduction / Motivation
The calls by the EU Commission to reduce CO2 emissions to 95 g CO2/
km require substantial undertakings to increase the efficiency of combus-
tion engines. One method is to shift the duty point toward high mean pres-
sure with contemporaneous reduction in stroke volume – known as down-
sizing. With this in mind, current combustion methods permit compression
ratios of 1:10 and more. But the knock threshold prevents operation at
ideal ignition, especially at low speeds. Knocking control is used in order
to guarantee the engine’s good working order. However, as the degree of
downsizing increases, the weaknesses inherent to the standard version
are becoming ever more apparent.

308
Firstly, the optimization of the combustion method reduces the combus-
tion dispersion and hence also the dispersion of knocking events. This
goes hand in hand with the reduction of the ratio between knocking and
non-knocking events, which is important for the detection of knocking. This
cuts the quality and stability of standard knocking detection and knocking
control. It follows therefore that stabilizing the combustion processes in a
gasoline engine as an objective in engine development run directly cont-
rary to the requirements of standard knocking detection/knocking control
(fig. 1).

Figure 1: Conflicting objectives in current engine development

It must be stated furthermore that the rising number of systems designed

to optimize combustion has also caused an increase in mechanical struc-
ture-borne sound emission. This also produces difficulties to compensate
the reduction in the useable diference between structure-borne knocking
signal and the basic engine noise.

Additionally, the ignition timing sequence which typically follows knocking

detection prompts a disproportionate reduction in the engine torque. The
later the ignition timing lies relative to the optimum knock threshold, the
more the torque will fall in the subsequent positions. Engines with high
power density in particular may therefore experience noticeable impact on
drivability and fuel consumption.

309
The availability of high-octane fuels and the demands of the market for
multiple fuel compatibilities present a further challenge to knocking con-
trol, as the knocking intensity exerts substantial influence on the knock
distribution and hence on the difference between knocking and non-kno-
cking.

Reasons of engine safety and engine efficiency necessitate a modification

in the strategy of knock detection and the resulting ignition timing measu-
res to suit current engine and market circumstances.

2. Repercussions of combustion optimization on

knock detection

2.1. Knocking description through pressure indication

2.1.1. The standard counting method

The absolute maximum of band pass-filtered combustion pressure has
become established in the determination of knocking intensity for individu-
al combustions. The number of events above a defined pressure value is
most commonly determined in order to describe knocking at a given duty
point. In this example, there are 7 events of 1,000 that are located above
a threshold of 3 bar (figure 2). For a long time this was deemed sufficiently
precise.

Figure 2: Knocking events in a standard engine

310
One primary objective in the development of combustion methods is to
reduce the cyclic dispersion of the combustion process. This means that
considerably more light knocking burnings appear. The strength of these
individual events themself is uncritical. Figure 3 illustrates that an engine
with an optimized burning procedure shows a considerably higher number
of mean and small knocking events compared at an equal conventional
numerator criterion (see figure 2).

Figure 3: Knocking events in a modern engine

Accordingly, discrete description of knocking by using the standard coun-

ting method must be classified inadequate to cope even with the evaluati-
on of the pressure gauge as a reference signal.

2.1.2. The statistical method

IAV tackled this problem as far back as the nineties, developing a stati-
stical assessment of the knocking intensity. To do this, the relative accu-
mulated distribution (figure 4, right) is determined for the example used in
2.1.1 (figure 4, left).

311
Figure 4: Statistical calculation of knocking intensity values

In this the sample is sorted according to the size of the elements, and the
number of elements is standardized to 100 %. The strongest knocking
incident describes the sample to 100 %. Hence this permits the conside-
ration of any number of sample sections (quantiles).

At IAV, the 85 % quantile has proven most effective for assessing change
knocking states like those found in test stand monitoring and automatic ig-
nition timing optimization. It yields a stable value, even with a sample size
of merely 240 cycles. Accordingly, the pressure value at the 85 % quantile
in the sample selected is 0.8 bar.

If one applies the criterion found in 2.1.1 – threshold value of 3 bar – the
quantile value for this distribution is 99.3 %. This means 0.7 % – in this
case that seven events – are greater than 3 bar. 993 elements or 99.3 %
are therefore equivalent to, or lower. Given that the pressure threshold re-
mains constant, the quantile value in this method is variable. But this does
not permit any statement on the mean knocking level.

Figure 5 presents the distribution for the example of an optimized combus-

tion process, as seen in 2.1.1.

312
Figure 5: Statistical calculation of knocking intensity values

In this, the pressure value of the 85% quantile is 2.2 bar. This value re-
flects, to a far greater extent, the optical impression in the presentation of
all cycles.

Given that the number of cycles with pressure vertex above 3 bar was
held constant at 7 elements, the resulting quantile also remains constant
at 99.3 %.

If one compares the position of the 85% quantile with the quantiles (cons-
tant here) determined at 3 bar, it is to observe a pronounced convergence
between these two criteria in the optimized method. This reduced distance
emphasizes the fact that a substantial reduction in difference between
knocking and non-knocking can be ascertained in the reference signal
already.

2.2. Impact of the altered distribution function on

­traditional knock detection and control
The problems associated with a reduced ratio between knocking and
non-knocking are exacerbated once more in the evaluation of the struc-
ture-borne sound. The additive introduction of the engine’s basic noise
alone effects a linear, upward shift of the knocking signals, hence inducing
a lower ratio between the two criteria.

This relevant distance is used, however, as a knock detection criterion at most

control systems. A consequence for those systems is a strongly reduced selec-
tivity between knocking and No knocking. The selectivity is also reduced con-
siderably at control systems which evaluate the spread of the knocking events
to themselves. As represented in 2.1.2, the distribution the knocking events

313
approaches more and more the normal distribution. Therewith the difference
between the standard deviations of knocking and No knocking also will sink.

The increasing effect of the firing angle on the mean knocking strength (e.g.
85% quantile at the pressure and reference level at the handling noise) also
seems negative to the detection and control accuracy. Every intervention of
the control – here the retarding of the firing angle – leads close to the kno-
cking limit to a significant reduction of the knocking strength and therefore
to a sinking noise. Due to the changed knock distribution the magnitude of
hard knocking events does not drop much stronger then the reference any
more, Therefore Signal- Noise Ratio does not drop clearly enough. Now also
significantly lower knocking events are interpreted as knocking and respon-
ded by another firing angle retarding which in turn lets down the reference or
noise drop again. The control is in a stable state of false detection (figure 6).

The opposite can happen in the equal operation point too. By advancing the
firing angle the mean knocking strength and therefore the noise increases.
Inevitably significantly higher knocking is not recognized anymore and no fi-
ring angle retarding is carried out. The control is in a state of NON-detection.

Figure 6: Weaknesses of standard knock detection/control

The stability and quality of knock detection and control continue to suffer
if the combustion dispersion falls further. Hence, the goals of engine de-
velopment run directly contrary to the requirements of the current knock
detection and control strategy.

314
3. Parameterization for engine control

3.1. Ignition timing control at the IAV test stand using

pressure gauging
The knocking intensity parameter described in 2.1.2 has been used suc-
cessfully in test stand automation for many years at IAV. Firstly, it permits se-
cure monitoring of the test sample on the basis of archived threshold values
and response measures.

Secondly, the continuity and currency of the parameter permits control of

the ignition timing.

The sequence of operations of an automatic firing angle adjustment repre-

sented is in the figure 7. Beginning with a sure firing angle the knocking
strength target or limit is approached by linear advance of the firing angle
(1).

A first estimation of the distance to the target value is carried out with the
progress of 85% quantile (here dp85). By checking the distance to the tar-
get and boundary conditions an acceleration of the spark advancing can
speed up the target achievement (2). In a safe distance to the target value
the advancing of firing angle is reduced again (3). At transgression of the
set point a retard is carried out on the last knocking free firing angle value
(4). The optimum is reached and the measuring is executed (5).

Figure 7: Automatic ignition timing control on the IAV test stand

315
This optimization can be combined with additional threshold and target
values, for instance the location of the center of combustion.

Monitoring and optimizing the knocking intensity on the basis of a statistic

knocking parameter has proven effective at IAV. The task at hand is now
to transfer the principle to engine control.

3.2. Parameterization on the basis of structure-borne

sound signals

3.2.1. Correlation between pressure and structure-

borne sound signals
The principal requirement for the applicability of statistical evaluation is the
proof of informational equivalence between pressure and structure-borne
sound signals. This correspondence can be proven on the basis of an ig-
nition graph presented in figure 8. It is clearly recognizable that the quan-
tile value of pressure and also the level of structure-borne sound change
exponentially with the ignition timing.

Figure 8: Statistical pressure and structure-borne sound criteria

316
In terms of the structure-borne sound, the explanation for this is that, up to
a certain level, it consists solely of the normal noise emitted during engine
operation. This includes mechanical noises, noises created by the electri-
cal systems and the combustion itself.

But the structure-borne sound level rises sharply as soon as the early
ignition creates knocking. This shows the separation of basic noise and
the signal level generated by knocking. This furthermore proves that any
knocking intensity identified by means of pressure indication can be as-
signed unambiguously to a structure-borne sound level. The demand for
informational equivalence is satisfied in this 100 % correlation.

3.2.2. Optimization of signal recording by simplification

The better the structure-borne knocking information rises from the basic
noise, the earlier and with greater the knocking intensity can be deter-
mined. This is achieved by optimizing the evaluation of the structure-born
noise in a temporal and also spectral optimal form. Therefore signal filte-
ring is simply reduced to a minimum.

A broad and rigid measurement window (figure 9) means that the location
of combustion, and hence of knocking, is immaterial to the signal energy
recorded. It follows that the measurement window does not have to move
together with the ignition. Any and all interference is therefore recorded
identically at all times, thus leading to stable basic noise, independent of
ignition.

317
Figure 9: Broad, rigid measurement window

Filtering is expanded to include the entire spectrum of useful signals in the

frequency range also (figure 10). This records the entire oscillation energy
of the knocking events. Given that only the location of combustion will ch-
ange in an ignition sweep with stable duty point, the basic noise remains
constant, and only the knocking noise rises.

Figure 10: Broad signal filtering

A welcome side-effect of this simplification is the substantial reduction

in effort involved in recording and evaluating the signals. The elaborate

318
identification of filtering and windowing, dependent on the cylinder, speed,
load, and fuel, are entirely unnecessary.

3.2.3. The TARGET value model

This context described in 3.2.1 applies to every duty point. The absolute
value of target knocking level, however, is dependent on the duty point
and must therefore be archived accordingly. This requires knowledge of
the respective target knocking intensity, and it must be suitably set; for
instance through the application of automatic ignition timing adjustment
as defined under 3.1. This produces an engine map covering load and
speed, in which the target level is archived (figure 11).

Figure 11: Target knocking level engine map

In modern engines, however, a speed-load point can be achieved by me-

ans of systems (camshaft adjustment, valve stroke, boost level, number
of cylinders, injection strategy, air- fuel- ratio, etc.), activated or positioned
in different ways. Furthermore, these systems are driven, which may lead
to interfering signals.

The target value is also exposed to the ambient thermal conditions of the
engine. As soon as a change in one of the system impacts on the basic

319
noise, it will also influence the target knocking level. It follows therefore
that the changes in parameter relevant to noise must be known and the
reference target value requires correction equivalent to these changes. To
do this, it is necessary to create a model, consisting of five modules.

Mechanical noise

This incorporates alterations in the noise emitted by all moving compo-

nents.

However, it only relates to components that can be (de)activated or pos-

sess an adjustment range, where they leave from and return to the mea-
surement window, for instance the noises of inlet and outlet valves closing.

Combustion noise

Combustion impacts very substantially on knocking distribution (see

2.1.2). Increasingly lean fuel, for instance, will raise knock dispersion, me-
aning that individual events will occur that should be classified as critical,
even if the knocking level remains constant. Accordingly, modifications of
carburetion must lead to an adjustment in the target level.

If the charge movement is variable, its activation will lead to an improved

mix preparation and hence to a reduction in knocking dispersion. At the
same time, the rise in charge cycle work increases the final compression
temperature, which in itself prompts an increase in knocking susceptibili-
ty. The accumulating effect on the knocking sound must be available for
target correction.

The individual values are monitored at the same time as defining the kno-
cking level. Any extremely distinct, individual events that occur, indicate
an anomaly compared with the known/archived statistical distribution of
knocking and hence in the process of combustion. Given that the inci-
dence of these extreme knocks is a response to changes in the ignition,
the target level for the relevant engine map range is reduced significantly.
Nevertheless, this system is not in the position to fulfill an independent
pre-ignition detection.

Electronic noise

The principle here is the same as with the mechanics. The systems that
can enter and depart the measurement window with their electrical distur-

320
bance must be identified. Significant repercussions on the basic noise are
archived in a corrective value.

Octane number

The anti-knock properties of the fuel used have a very strong influence
on the knock distribution. Accordingly, determining the octane number is
a correct element of modeling the target value. Details will be presented
in 3.2.4.

Measurement chain

The knocking signal is exposed to a wide range of damping and conversi-

on on its way from the oscillating gas in the combustion chamber. Via the
engine structure, the knock sensor, the jack connection, the cable, and the
input circuit of the control device until it finally reaches into the analogue/
digital converter. All elements in this chain are subject to exemplary dis-
persion and thermal influence, and some are also susceptible to ageing.

The thermal effects are summarized in the knocking sensor temperature

model.

The exemplary dispersion of the individual elements and their ageing can
only be assessed in their current assembly. Details are described in 3.2.4.

3.2.4. Noise adaptation

The target knocking level determined in 3.2.3 only applies to this specific
engine used. Any stable control on the basis of the absolute target value
requires an absolutely accurate target value for each individual engine in
operation under all prevalent circumstances. This adaptation of the exem-
plary dispersion takes place by means of noise adaptation. It is conducted
selectively for each cylinder and is virtually unaffected by driving behavior.
Evaluation and classification of the measuring chain requiring adaptation
is possible after merely one successful sequence. Approximately 5 suc-
cessful adaptations are necessary in order to obtain sufficient precision
for each cylinder.

As described in 3.2.1, the knocking intensity or the knocking level chan-

ges exponentially by advancing the ignition timing. The knocking intensity
is often extraordinarily sensitive to any adjustment in the ignition timing,

321
especially when approaching the target knocking intensity. This effect can
be represented mathematically in the first derivative of the level curve
through determination of the alteration gradient. Given that the level curve
is a polynomial of at least the 3rd degree, the gradient produced will be a
polynomial of at least the 2nd degree. Accordingly, one single level gra-
dient can be assigned to each ACTUAL level. A target gradient belongs to
the target level (figure 12).

Figure 12: Level and gradient curve

The aim is to align the target value of the current measurement chain to
the reference value by the application of linearization formula, in which

y = ax + b (1).

By applying the first derivative a curve shift towards the zero point was
produced.

If the gradient curve of the system requiring adaptation (y) is equal to the
reference system (x) applied in 3.2.3, the difference in level curves shall
be equivalent to the displacement in zero point or the offset (b).

322
Figure 13: Impact of signal offset in measuring chain

The sensitivity of a measurement chain, that is the ratio between the po-
tential created with given acceleration, is substantially influenced by the
knocking sensor itself and the impact of applied impedance. It corres-
ponds with the linear part.

If the comparison of the reference and actual gradient reveals a proportio-

nate change, this represents the gain between the two level curves (a), i.e.
the difference in sensitivity between the measurement chains.

Figure 14: Impact of signal gain in measuring chain

Use of a low-octane fuel will produce a substantially steeper level and

therefore gradient curve. Dependence between the current curve and the
reference would be cubic. We shall now precede to the TARGET model for
this gradient curve/this octane number using a switch threshold.

323
Figure 15: Impact of octane number

This evaluation of the gradient – the (delta) – in knocking intensity pro-

vided the name of this control strategy.

3.2.5. Updating the knocking level

The statistical reliability is dependent on the dispersion within the sample
and the number of elements considered. This applies to calculating the
mean and also the quantile. A trend is apparent, however: the statistical
parameter becomes increasingly stable or robust as the number of ele-
ments rises. In control, this is equivalent to high damping or a low filter
factor.

When the working point is stable, the ambient conditions of very many
events are identical. This is why a large number of elements can be
evaluated here. The description of the state is extremely precise.

Figure 16: Dynamism-dependent compromise between currency and

statistical security

324
The change in any parameter described in 3.2.3 produces signal dyna-
mism. And currency comes increasingly to the forefront as the signal dy-
namism rises. To do this, the sample size, i.e. the level damping, is re-
duced in dependence on the signal dynamism. This reduction takes place
in a compromise with the consequent decrease in statistical security of
the knocking level. This decrease is permissible, as the dynamism case
is limited in time and hence possesses barely any statistical relevance.
Further, the gain in quality due to enhanced updating speed more com-
pensates than the quality drops for sample reduction.

4. Knocking control with the engine control

4.1. Control structure

The structure of the DELTA knocking control is equivalent to the standard
structure. In it, the actual level described under 3.2.5 replaces the control-
led variable “ratio between knocking and non-knocking”.

The standard common control factor “threshold for ratio between knocking
and non-knocking” is replaced by the TARGET level as defined under
3.2.3 and 3.2.4.

Figure 17: Comparison of standard versus DELTA control structure

325
A PID controller is used instead of the standard offset in ignition timing
with following advance sequence. This is described in greater detail in 4.2.

4.2. Control strategy

The provision of a continuous control factor (current knocking level, see
3.2.5) and the derivation of controlled variable (target knocking level, see
3.2.3) permit the use of a continuous controller. Application of a PID con-
troller means that the quasi predictive properties of the D part and the
precise short and long-term reactions of the P and I parts can now also be
exploited in knocking control.

This permits an extremely high increase of control quality.

Figure 18 uses a real measurement to provide an example of control with

gradual, stage by stage increase in the target knocking level in volts and
therefore the actual knocking intensity as dp85 in bar.

Figure 18: Gradual increase in the target knocking intensity

The control sets a very low deviation between TARGET and ACTUAL level
with barely substantial alteration of ignition timing.

Progress of the target level now appears linear due to the temporal varia-
tion. This sequence in knocking intensity identified by pressure indication

326
after the 85 % quantile (dp85) also progresses quasi linear and hence
proportional to the target resp. the actual noise level.

The mean correction of ignition timing falls each noise set point increase
as the knocking intensity increases. The curve of ignition correction is
hyperbolic.

Presentation of the correlation between the ignition timing and knocking

level or knocking intensity over time confirms the illustrations using the
ignition timing variation as in figure 19, for example.

4.3. Control security

Use of a PID controller means that the ignition timing is set to early when
values fall below the target. Two functions ensure that this early adjust-
ment does not produce any hazardous knocking intensity.

Limitation in the adjustment factor range

During adaptation (see 3.2.4), more than just the target level is deter-
mined. Polynomial calculation permits extrapolation of level and gradient
curve beyond the target value, hence enabling identification of the ignition
timing at which the control parameters no longer guarantee the requisite
security and quality (see “Limitation in ignition timing dynamism”). When
the controller is in operation, this ignition timing serves as early limitation.

Limitation in ignition timing dynamism

In the case that the optimum control range – in other words, the target
level – is exceeded, the pronounced non-linearity of the controlled varia-
ble (significant change in the rise of knocking level by early adjustment of
the ignition angle) provokes substantial controller activity, that is ignition
timing fluctuation (see top right in figure 15). This is a secure indication
that the target knocking threshold has been exceeded.

The controller activity resp. the ignition timing dynamism is determined

and monitored. In case of transgressing of the target knocking intensity
caused by premature ignition timing the controller activity exceeds its limit.
As necessary consequence the target level is reduced to a secure target
value.

327
4.4. Control stability and quality
As presented in figure 19, the steep gradient of the controlled variable
has markedly stabilizing effects. A deviation of ± 1° KW from the optimum
ignition timing is equivalent in this example to a change of approximately
± 25 % from the target level.

The extended signal registration (see 3.2.2) and the adaptation (see 3.2.4)
prevent any such fault. It is therefore fair to assume a control stability and
precision of less than ± 1° KW.

Figure 19: Control precision and maximum early adjustment

4.5. Robustness towards interference

Sporadic interference

Interference signals that do not follow any systematic pattern may be in-
troduced sporadically into the structure of the engine. Provided the energy
is sufficient, standard knocking control will detect these signals as kno-
cking events and will respond by retarding the ignition timing. Statistical
evaluation incorporates them in the overall sample, and their significance
is thereafter extremely subordinate. Their impact on the control deviation
and hence on changes in ignition timing is negligibly slight.

328
Constant interference

The control will set the target level by late adjustment of the ignition timing
if the knocking level rises due to continuous interference, caused, for ex-
ample, by a defect component. If this falls below an archived threshold, the
control stroke is classified implausible and a noise adaptation is initiated.

If the mechanical defect is repaired, the limitation in ignition timing dyna-

mism according to 4.3 shall remain in force until the next cyclic adaptation.

5. Summary and Outlook

The statistical approach to calculate the actual knocking level delivers a
reliable criterion to establish secure controls, especially in engines with
reduced knocking stochastic.

The DELTA knocking control enables for the first time the definition of an
actual target knocking level to control knocking.

The level adaptation introduced guarantees a high degree of signal quality

and the simple and secure adjustment of units within an engine series,
along with a broad spread in octane number and across the entire life
cycle.

Simplification of signal recording and modeling of the engine noise repre-

sent a significant reduction in the effort related to adaptation of knocking
control for the stage of build, i.e. engine derivatives.

329
Figure 20: Solution of the target conflict

The PID controller and the correct target level cut the ignition timing re-
duction to a minimum. Level adaptation in cylinders with substantial po-
tential in pre-control of the ignition timing permits its optimization. The
consequent ignition timing effectiveness with DELTA knocking control is
therefore improved, which may in turn have positive repercussions on fuel
consumption.

The minimized impact range of ignition timing reduces the stress of follo-
wing acting systems e.g. boost pressure control or drive ability.

The DELTA control method solves the conflicting targets between the on-
going optimization of combustion procedure and the demands of current
knocking detection. This shall permit secure exploitation of additional be-
nefits in terms of further downsizing and combustion stabilization.

330
17 Artificial Intelligence for knock detection
Matthias Biehl, Elvira Perless, Robert Sloboda

Abstract
This article deals with the identification of abnormal combustion phenome-
na in gasoline engines by reconstructing physical values, which cannot be
detected directly or only at high costs. Therefore methods from the field of
Artificial Intelligence, Pattern Recognition and Statistics are combined in
this approach to create a basis for important Electrical Control Unit (ECU)
algorithms. The influence of databases with an extremely uneven data
distribution and data spectrums of high-variance is investigated. Standard
measurements from various engines are processed to ensure a high ge-
neralizability of the model and the comparability to the results of the ECU.
A noticeable increase of robustness against interferences resulting in an
improved quality and a reduced parameterization effort is achieved by
using this model.

Kurzfassung
Diese Veröffentlichung beschäftigt sich mit der Identifikation anormaler
Verbrennungsphänomenen in Ottomotoren mit Hilfe der Nachbildung phy-
sikalischer Größen, die nicht direkt oder nur unter hohem Kostenaufwand
bestimmt werden können. Daher werden Methoden aus dem Bereich der
Künstlichen Intelligenz, der Mustererkennung und der Statistik in diesem
Ansatz kombiniert, um eine Grundlage für wichtige Steuergeräte Algorith-
men zu bilden. Der Einfluss von Datenbasen mit extrem ungleich verteil-
ten Daten und Datenspektren mit hoher Streuung wird dabei untersucht.
Um eine hohe Generalisierungsfähigkeit des Models und die Vergleich-
barkeit zu den Steuergerätergebnissen zu gewährleisten werden Stan-
dardmessungen verschiedener Motoren verarbeitet. Eine deutlich verbes-
serte Robustheit bezogen auf Störungen, was weiterhin zu einer erhöhten
Güte und einem verringerten Parametrisierungsaufwand führt, wird durch
das Model erreicht.

331
1. Introduction
The ever-growing economical and ecological demands made on modern
combustion engines, in particular with regard to the improvement in effi-
ciency and emission performance standards, require permanent new and
further development of components, methods, and functions. A vast num-
ber of these developments results in increased complexity and in side
effects amongst each other, which necessitates adequate continued pro-
cessing of the calibration.

For the knock detection that is based on a structure-borne noise sensor

this adds up to a negative influence due to a rising number of mechanical
and electrical interferences on the sensor or on the sensor signal. The
cause for these disturbances is usually a result of the growing diversity
of engine components and new strategies for the mixture formation and
combustion process. The current reference value method for the knock
detection, in particular when based on a multiple-filter strategy, provides a
solution for this, although the calibration effort compared to the single-filter
strategy is distinctly increased and can hardly be implemented without the
support of a tool chain.

1.1. State of technology

The maximum value of the bandpass-filtered cylinder pressure signal
(Peak) has proven to be the most representative value for the evaluation
of the knock intensity. This value is used to verify the knock detection qua-
lity of different knock detection strategies. To date, the pressure sensor
was not able to establish itself for the series production application, due to
the comparatively high costs and the short service life.

The reference value method currently used in series production for the
detection of knocking, calculates cylinder-individually the knock integral
(IKR) on the basis of the structure-borne noise signal that has been pro-
cessed with one or several adjustable bandpass filters. For this, the sen-
sor signal is recorded in a limited time period after the ignition, which is
specified by the adjustable measurement window (beginning and length).
This integral is divided by a permanently updated mean value that is refer-
red to as reference level (RKR) and that is based on the IKR. The result
(RKI) is compared with an adjustable knock detection threshold (KEK),
and when the threshold is reached a knock occurrence is detected.

332
Sporadically occurring interferences in the typical useful frequency range
of the structure-borne noise sensor (5-25 kHz) may result in misdetec-
tions. Continuous influence of interferences, on the other hand, denotes
an increase of the reference level, which in turn results in knocking com-
bustions not being detected reliably. Thorough parameterization of the fil-
ters, of the position of the measurement window, and of the KEK during
the calibration process helps to reduce the influence of inter-ferences.

The increase of disturbing noise is caused by modern engine concepts,

such as variable camshafts, multiple fuel injections, bi-fuel concepts, con-
solidated designs, downsizing, and an ever growing number of mechanical
and electrical engine components. Parallel to the technical trends, econo-
mic considerations lead to cost savings. Besides cutting the calibration
costs, only a limited number of experimental vehicles is available, which
makes representative investigations of series-production tolerances more
difficult. In addition, the number of knock sensors is minimized and this
usually impairs the signal quality of those cylinders that are located further
away from the knock sensor.

Strategies to master this challenge lead to rising model and function com-
plexity of the ECU software. An increase of the calibration effort can be
prevented by means of a knock control tool chain supporting an efficient
calibration. In extreme cases, merely sufficient knock detection quality can
still be ensured.

1.2. State of art

Within the framework of a research project Robert Bosch GmbH in co-
operation with the Technical University of Kaiserslautern investigated the
adequacy of Support Vector Machines and of Neural Networks for the
knock detection. The results obtained in the process show an increased
knock detection quality and robustness against known interferences. The
computational effort required for this by far exceeds the resources of cur-
rent ECUs and of those of the next generation.

333
2. Future-oriented knock detection
For a future-oriented knock detection strategy, on the basis of the struc-
ture-borne noise signal, the specialized calibration group for knock detec-
tion and control of Robert Bosch GmbH is presently developing an alter-
native approach that is based on methods of Artificial Intelligence, Pattern
Recognition and Statistics. The core of this procedure is the Regression
Analysis. Thereby the maximum value of the band-pass-filtered cylinder
pressure signal (Peak) is calculated on the basis of the structure-borne
noise signal. This computed value is denoted by Peak*.

2.1. Signal processing and evaluation

Signal processing and signal evaluation is required to assess the com-
bustions with regard to their knock intensity. The complexity required for
this and its distribution to the two parts depend on the necessary model
quality and on the available computing capacity. In consequence of the
performance of current control units, the computing capacity is the decisi-
ve determining factor, since the best possible model quality is considered
a prerequisite.

A detailed analysis of the knock sensor signal and of the pressure sensor
signal in the time and frequency domain is required for the extraction of
features determining the knock characteristics. In the process, the exact
position is identified in those time and frequency ranges, where the useful
signal is subjected to as little interference as possible. Thereby, characte-
ristic features are obtained by means of which the knock intensity can be
determined.

334
 Figure 1: Example for time-frequency analysis

Figure 1 shows an illustration of the described analyses applied to the

knock sensor signal (left) and the pressure signal (right). The core of this
evaluation is the presentation of the time-frequency analysis by the scalo-
gram, where blue stands for low and red for high intensity. The frequency
spectrum is plotted in blue as result of an Fast Fourier Transformation
(FFT) and in green as sum of the time-frequency analysis over the time
period to the left of the respective scalogram. Directly underneath, the
energy distribution over the time is illustrated for every frequency band
individually. Below this diagram, the respective signals are plotted against
the crank angle. In case of the knock sensor this corresponds to the mea-
sured sensor signal; for the pressure sensor the signal that has been pro-
cessed by filtering is depicted.

2.2. Comparison of different models

For the determination of the explicit knock intensity, on the basis of the
features obtained during the signal processing, a regression model is
used that depends on the operating point. The combination of linear and
nonlinear model part allows for the computation effort of the knock inten-
sity to be reduced, and for an increased robustness and ability of genera-
lization compared to the purely nonlinear model. Additionally the training

335
effort is reduced as well as the necessity for complex safety functions.
This combined model provides the additional advantage that more com-
plex system behaviour can be reproduced and consequently that a higher
model quality can be achieved compared to the purely linear model. The
accomplishable quality of the different models is depicted schematically
in figure 2 by means of a radar diagram about the decisive determining
factors for the knock detection. The rating for each criterion ranges from
zero indicating high effort or poor quality up to ten which stands for mini-
mum effort or maximum quality. The area enclosed by each of the three
illustrated models shows the overall model rating.

Figure 2: Radar diagram to compare linear, nonlinear and combined mo-

del

2.3. Strategy – training and application

The objective of an efficient calibration in combination with an improved
signal analysis is achieved by using a modified calibration method. For
a closer consideration of the implementation a division into two phases
takes place. The offline training phase combines the procedure that more
closely describes the training of the individual components “offline” rela-
ting to the ECU. Based on this, the online application phase comprises

336
the implementation of the “online” calculation in the ECU. Objective of
this two-phase division is to reduce the computing effort in the ECU that
is required to detect knocking combustions to the greatest possible ex-
tent. At the same time, this implementation strategy provides the option of
resorting offline to complex evaluation and adaptation methods and thus
to achieve the best possible knock detection quality subject to bounds of
limited ECU resources.

2.3.1. Training process

An overview of the implemented model that is called Virtual Artificial Sen-
sor (VAS) is shown in figure 3. For the extraction of features for the deter-
mination of knock characteristics, complex methods of signal evaluation in
the time and frequency domain are applied to the knock sensor signal and
to the cylinder pressure signal. Amongst these are the Fourier Transfor-
mation to identify characteristic frequency bands for knocking and interfe-
rences and the Wavelet Transformation to determine the temporal energy
distribution within these frequency bands. This allows for an identification
of knock features in the time and frequency range that are subjected to
minimal interferences.

After the complex determination of features specific to time and frequency,

the calculation of comparable features takes place by using methods that
are suitable for application in the ECU. In a first implementation step, the
established methods – limitation of time range by means of measurement
window and limitation of frequency range by means of filtering – are used.
In addition to this, the measurement window is adjusted for every frequen-
cy band and may show discontinuity. Thereby, targeted suppression of
interferences is achieved. In figure 3 this step is called ‘signal-processing’.

The thus determined measurement windows and filters are applied to the
structure-borne noise signal and the result is subsequently integrated ran-
ge by range. These obtained features are the input for the model-based
calculation of the Peak value of the bandpass-filtered cylinder pressure
signal. The number of features depends on the complexity of the engine
behaviour and on the position and intensity of the interferences.

The variance of the knock characteristics, which is also known from the
state of the technology and which is primarily caused by aging of the en-
gine and by production tolerances, requires the basic characteristics of

337
the engine noise to be considered in the model. The model part called
feature-adaption calculates the basic characteristics of the engine. This
allows for adequate knock detection across all operating ranges and for
the entire service life of the engine.

Figure 3: Overview of the model

338
The features determined in this way are used in equal measure for the
training of the linear and of the nonlinear model part. The Peak value of
the bandpass-filtered cylinder pressure signal is specified as target value
in the training of the linear model. Based on the result, the nonlinear part
is adapted with the deviation of the linear model.

The parameters of the linear model are determined by a main-component

analysis on the basis of all data available for the training. This procedure
forms a new feature space by transformation. The orthogonal coordinate
axes of this feature space are the main directions of propagation of the
input data in the original feature space. This strategy rules out overfitting
of the model related to the behaviour pattern of outliers and an uneven
distribution of the data basis up to this part of the training method.

Mapping of specific behaviour within the data base is implemented by me-

ans of the subsequent nonlinear model. The training is done by means of
the combustions that cannot be mapped in sufficient quality by the linear
model.

This model consists of an Artificial Neural Network (ANN) with Radial Ba-
sis Functions (RBF) neurons and limited complexity. In order to consider
the ECU resources and minimize the risk of overfitting, the amount of used
neurons is reduced to such an extent that the quality will not be negatively
affected.

Figure 4 shows the output of the linear model (left side) in comparison to
the output of the combined model (right side). The most distinctive outlier
of the linear model is marked in red. By applying the non-linear model this
deviation is compensated and the correlation between measured Peak
pressure and Peak* pressure calculated by the combined model is impro-
ved.

339
Figure 4: Output of the linear model with outlier identification (left) – out-
put of the combined model (right)

The possibly extreme output behaviour of nonlinear models for unknown

input features is restricted in this implementation by limiting the influence
on the overall result. In the process, a positive and a negative limit are de-
fined on the basis of the training data. Figure 5 gives a general impression
of the implemented limiter block. In the predefined range resembling the
model deviation from –PeakError to PeakError the corrective measures
are zero. From –Peakerror to Peakneg and from Peakerror to Peakpos
the output of the nonlinear model is not restricted. As safety precaution
against unknown input data higher or lower corrections by the non-linear
model are not allowed.

Figure 5: Limiter with positive and negative limits

340
2.3.2. Application
For the ECU implementation, the calculation of the signal evaluation and
of the regression model, on the basis of the parameters that were deter-
mined in the training process, takes place in real-time. The output value of
this calculation Peak*, based on the structure-borne noise signal, subse-
quently serves as basis for the evaluation of knocking combustions.

Two different detection strategies are pursued when detecting knocking

combustions, both of which are based on the physical compulsory th-
reshold for the Peak* value of the bandpass-filtered cylinder pressure si-
gnal. The main distinguishing feature is the detection threshold, which is
firmly specified dependent on the operating point for one of the methods.
For the second method the detection threshold adapts self-reliantly within
specified limits dependent on the operating point and the engine behavi-
our, which allows more sensitive knock detection. Figure 6 shows how the
effective threshold may vary within the two defined thresholds when the
second detection method is applied under steady-state conditions.

Figure 6: Self-reliant knock detection threshold adaptation

341
3. Results
Compared to the results of the current knock detection and to those of
[2] and [3], the output of the combined model, given as VAS, shows an
enhanced correlation (see fig. 7).

Figure 7: Correlation graph for the various investigated knock detection

methods at different engine speeds

In figure 8 the output of linear model and of combined model is compared

at different operating points. The given examples illustrate that the addi-
tional non-linear model part reduces deviations, which results in a higher
model quality.

342
Figure 8: Output Peak* of the linear model (left) and of the combined
model (right) at different operating points – with regression line (solid)
and standard deviation (dashed)

343
A particularly challenging project on the part of the knock detection im-
plemented in the ECU is illustrated in figure 9. On the left, the results are
shown that were obtained with the standard ECU at an early stage of the
calibration. Due to interferences the knock detection tends towards mis-
detections, which can be reduced by immense calibration effort. Applying
the same data to the combined model leads to the results illustrated on the
right side. The influence of interferences can be reduced by appropriate
selection of features. By combining the linear and nonlinear model a good
correlation between calculated Peak* and measured Peak is achieved,
which in turn increases the overall knock detection quality.

Figure 9: Comparison between ECU results and combined model results

4. Conclusion
The results obtained by the new method, compared to the state of tech-
nology, feature an increased knock detection quality and a distinctly more
robust behaviour as regards unknown interferences. Known systematic
and sporadic interferences are just about completely compensated by this
strategy.

Compared to the state of art, the combination of linear and nonlinear mo-
del conserves drastically more resources, which allows for an ECU imple-
mentation. In addition to this essential advantage, this new training stra-
tegy results in mapping of the knock characteristics that is more capable
of generalization, whereby more robust model behaviour and increased
knock detection quality are achieved.

344
The combination of signal evaluation of the structure-borne noise signal,
shared model structure, and “offline” training strategy that is adjusted to
the general requirements of the knock detection, achieves results that de-
monstrate the potential of this strategy for the use in future ECUs.

5. Prospect
The achieved results implicate prompt continuance of the investigations.
Projects with major interferences serve as basis here to derive possibly
required further developments on the basis of the modular structure.

Furthermore, integration of the algorithms in the ECU software is intended

to verify the maximum performance of the functions in the vehicle inclu-
ding all actual deter-mining factors, as well as for endurance testing. In a
further development step provision is made for functions that serve the
adaptation of signs of aging and series-production tolerances. Moreover,
the transferability of this strategy to further fields of application, such as
detection of preignition and misfiring, will be investigated more closely in
the course of other scientific research projects.

The increased-power rated ECU generation MDG1 (Motronic Diesel Gaso-

line 1) by Robert Bosch GmbH provides the hardware requirement for a
realistic series-production application of the new knock detection method.

345
Literatur
[1] Biehl, M.; Perless, E.: Virtual Artificial Sensor to identify abnormal
combustion phenomena in gasoline engines. In: AMA Conferences
2013 - Sensor. pp. 340–345, 2013
[2] Biehl, M.; Kempf, S.; König, A.: Messtechnische Hardwareplattform
zur Entwicklung neuer Motorsteuergeräte-Algorithmen am Beispiel
der Klopferkennung. In: XXV. Messtechnisches Symposium. pp. 39-
50, 2011
[3] Iswandy, K.; Kempf, S.; König, A.; Sloboda, R.: Robustheitsuntersu-
chung eines SVM-basierten Klopferkennungsverfahrens. In: Motor-
technische Zeitschrift. pp. 486-491, 2010

346
18 Knock detection strategies based on
engine acoustic emission analysis
Filippo Cavanna, Nicola Garagnani, Riccardo Lanzoni,
Stefano Sgatti, Massimo Zanotti

Abstract
Among the various known methods to detect knock events and measure
their entity, we tried to identify a relatively new path, that may be conside-
red, at a first glance, as quite “strange”.

Since several years, Magneti Marelli is investigating the opportunity of ex-

tracting as more information as possible from the acoustic signal acquired
by the ECU through a microphone. Both the combustion process and the
engine status may be monitored by real-time processing acoustic data:
beside knock control, we can cite misfire and/or misfuel detection, valve
closing events closed-loop phase recognition, engine and turbocharger
speed reconstruction, in-cylinder pressure rise rate identification, in-use-
fuel characteristics definition and so on.

The paper describes the development of a real-time knock detection al-

gorithm and the followed critical approach to set a first demo-system up.
Several aspects in particular have been taken into account, regarding
both the hardware and the software. We tried to get an overall idea about
the best type of microphone to be used, with a comparison between the
performances of various common low-cost commercial unidirectional and
omnidirectional devices respect to a high-quality laboratory equipment.
Even several installation issues have been taken into account in order
to understand if it would be possible to think of a direct connection of the
microphone with the ECU case, thus sparing costs in cables and vehic-
le-assembling.

The performances of a couple of processing algorithms have been com-

pared and the impact of the introduction of new strategies for acoustic-ba-
sed knock control on actual ECU software has been roughly estimated.

An experimental campaign has been carried out in a test-cell environ-

ment. In order to evaluate the signal processing system and algorithm per-
formances, knocking events have been externally induced and monitored.

347
1. Introduction
The idea of monitoring the combustion process by means of investiga-
ting its acoustic emission comes both from the awareness of the huge
quantity of information hidden inside that signal (humans and animals are
benefiting from the hearing sense since millions of years…) and from the
quite economic availability of an ad-hoc displacement to be set together
with an ECU system. Recent studies performed by our company together
with some University research centres ([1], [2], [3]) allowed us to investi-
gate the potential benefit of registering the acoustic emission of an engine
in order to get information about several parameters describing both the
combustion process and the working conditions of the engine itself. Se-
veral industries –among various fields– have already tested the use of
vibro-acoustic analysis for diagnostic goals ([4]) and demonstrated how
this technique may be easily applied to accurately prevent and/or detect
punctual failures even inside very complex installations.

Our activity started from the consideration that the pressure rising rate and
the disturbs due to knocking phoenomena must influence in a similar way
every signal correlated with the state of the pressure field in the engine
region. Obviously the best observer for the combustion should be a pres-
sure sensor directly displaced inside the cylinder, but even all the gauges
set to measure the different parameters which are influenced by irregular
engine vibrations must be able to register signals showing that a “troubled”
combustion has taken place. The transmission of the pressure oscillation
-first along the crankcase and then directly in the air– must reach, just few
instants after the normally-displaced vibrometer, also microphones set not
so far from the engine. And even the experience of several senior mecha-
nician suggested us to search for the information of knocking inside the
sound emitted by the engine…

Finally we decided to perform a comparison between the “truth”, i.e. the

signal coming from a pressure sensor directly mounted into the cylinder,
and two “reconstructed” information, coming from a set of accelerometers
and microphones.

Three types of influence have been taken into account: the type of used
sensor and its position, the structure of the algorithm adopted to detect
knock and the calibration parameters together with the specific working
point we were working in.

348
2. Experimental activity
We took into consideration a 1.4L gasoline PFI engine, equipped with the
dedicated sensors showed in the following scheme:

Figure 1: Scheme of the experimental layout

The following picture may allow to better understand the real displace-
ment of the engine test bench:

349
 Figure 2: Engine bench layout

As we wrote before, one of the aim of our activity was to investigate the
impact of both the type (omnidirectional or monodirectional) and the posi-
tion of the adopted microphones. We took into consideration back electret
condenser microphones, already used for mass production and thus very
cheap but, at the same time, significantly small and resistant to vibrations
and atmospheric agents.

Beside the choice of laboratory-type cylinder pressure sensors (as they

were used as reference gauges, no cost issues were considered), the
remaining item was the discussion about the position of the two additional
standard production accelerometers. The idea was to be able to compare
the influence of the positioning of the vibrometers and the microphones.
Which sensor would be more influenced by mounting instances?

Since years from experience in production-system knock control, we know

engine working point characteristics have got a not negligible influence
onto the performances of the different algorithms; we hence took into
consideration different speed and load breakpoints. The most interesting
ones have been:

350
1. Engine speed: 2000rpm, Intake efficiency: 0.7

2. Engine speed: 2000rpm, Full load condition

3. Engine speed: 3000rpm, Intake efficiency: 0.7

4. Engine speed: 3000rpm, Full load condition

5. Engine speed: 3500rpm, Intake efficiency: 0.8

6. Engine speed: 3500rpm, Full load condition

7. Engine speed: 4000rpm, Intake efficiency: 0.8

8. Engine speed: 4000rpm, Full load condition

9. Engine speed: 5000rpm, Full load condition

Knock phoenomena have been induced by means of forcing a proper

spark advance increment until the peak-to-peak value of the in-cylinder
pressure oscillation reaches a satisfiable level.

Several types of indexes for the estimation of the knock level –starting
from the signal acquired from the different sensors– have been taken into
account ([5], [6]): two of them –as expected– proved to be very helpful for
our study.

MAPO represents the maximum amplitude of the absolute value of the

band-pass filtered signal, windowed around a defined knocking region.
SUM represents the energy of the signal associated to the given frequen-
cy band, again considering only an angular portion of the signal around
the typical knocking region.

351
3. Results
The outcoming for our activity basically confirmed what we expected. First
of all we considered the impact of the positioning onto the accelerometer
signal. The following images show that –in general– knocking phoenome-
na are more evident the nearer the vibrometer is respect to the cylinder
where the event took place.

Figure 3: Cylinder pressure and vibrometers signals

At a first glance, the knocking information is completely absent from any

microphone acquisition (please see the following image as an example).

352
 Figure 4: Cylinder pressure and microphones signals

The post-processing of the data allows to discover something different:

even the information inside the microphone signals appear accurate
enough to perform further considerations. We calculated the value of the
indexes described in the previous sections in all the registered cycles for
all the acquired signals. We estimated the best frequency and angular
windows to be considered for every signal in the different engine working
points, by means of adopting a dedicate numeric optimization algorithm.

As expected, the optimal angular window for acoustic signal is far delayed
respect to the correspondent accelrometric one:

353
     Intake efficiency

    0.7 0.8 FL
Pressure: 30…50 Pressure: 35…55
Central Vib: Central Vib:
2000 30…60 – 30…60
Omni Mic: Omni Mic:
50…130 70…120
Pressure: 30…60 Pressure: 30…55
Central Vib: Central Vib:
3000 30…50 – 30…40
Omni Mic: Omni Mic:
50…150 80…160
Pressure: Pressure:
45…100 25…120
Engine speed

Central Vib: Central Vib:

3500 –
40…60 30…60
Omni Mic: Omni Mic:
40…180 80…180
Pressure: 20…65 Pressure: 10…70
Central Vib: Central Vib:
4000 – 20…30 -30…30
Omni Mic: Omni Mic:
50…170 60…170
Pressure:
45…105
Central Vib:
5000 – –
40…50
Omni Mic:
120…170
Table 1: Optimal angular windows [CA° aTdc] (Pressure vs Central Vib
vs Omni Mic)

354
The obtained correlation level between all the considered energetic in-
dexes calculated for the pressure signal and the ones coming from the
microphones is absolutely comparable with the correspondent one achie-
ved taking into consideration the accelerometers outcome.

Figure 5: Correlation level comparison between pressure and vib vs mic

The following figures show a global comparison between the correlation

level registered both for the accelerometric signals and for the acoustic
ones (respect to the pressure acquisitions) in all the considered tests. The
“SUM” energetic index has been first taken into account.

Figure 6: Correlation between pressure and accelerometer signal ener-

355
 getic indexes

Figure 7: Correlation between pressure and acoustic signal energetic

indexes

The just shown results are very encouraging: we demonstrated that a de-
dicated calculation chain allows to obtain an acoustic index strictly corre-
lated with the pressure oscillation inside the cylinder, due to knock events.

The additional CPU load to be taken into account for an eventual integ-
ration of a similar algorithm into the engine control architecture is almost
negligible.

Just to outline the importance of the choice for the better algorithm, we
would like to show a comparison between the results availabile after a dif-
ferent treatment of the signal. Independently on the engine working point
and on the considered sensor, for example, MAPO index revealed itself
as a worse indicator of knocking (both for accelerometer signals and for
acoustic ones).

356
357
 Figure 8: MAPO vs SUM correlation comparison

Finally, through the numeric optimization algorithm, we may get a further

confirmation of the accuracy of the signal treatment (hw+sw) that we de-
cide to follow. In fact, while considering a defined frequency window for
the filter, we have to obtain a level of correlation coherent with what we
expect, starting from the known content of the signals themselves inside
that considered window.

Figure 9: Correlation level function of frequency filtering shape (Test 2,

Sensor 4)

358
Literatur
[1] N. Cavina, D. Moro, M. De Cesare, G. Serra, Exhaust Gas Turboch-
arger Speed Measurement via Acoustic Emission Analysis, SAE
Technical Paper 2008-01-1007, 2008.
[2] N. Cavina, S. Sgatti, F. Cavanna, G. Bisanti, Combustion Monitoring
Based on Engine Acoustic Emission Signal Processing, SAE Techni-
cal Paper 2009-01-1024, 2009.
[3] N. Cavina, D. Moro, S. Sgatti, F. Cavanna, Ethanol to Gasoline Ratio
Detection via Time-Frequency Analysis of Engine Acoustic Emission,
SAE Technical Paper 2012-01-1629, 2012.
[4] P.D. McFadden, Examination of a technique for the early detection of
failure in gears by signal processing of the time domain average of
the meshing vibration, “Mechanical systems and signal processing”,
1987
[5] O. Boubal, J. Oksman, Knock Acoustic Signal Estimation Using Para-
metric Inversion, IEEE 0018-9456, IEEE Transactions On Instrumen-
tation And Measurement, Vol. 49, No. 4, August 2000.
[6] Millo, F., Ferraro, C., Knock in S.I. Engines: A Comparison between
Different Techniques for Detection and Control, SAE Technical Paper
982477, 1998

359
19 Continental’s pre-ignition and glow
­ignition function – detection and
­avoidance of irregular combustions
Markus Kieberger, Dominik Biehl

Abstract
For realization of regular fuel consumption targets, OEMs equip passen-
ger cars with downsized engines. For this state of the art engines direct
injection systems and supercharging systems are combined. Due to an
operation point shift up to higher mean effective pressures, thermal strain
is increasing during driving operation.

Linked to a high thermal strain is an occurring of irregular combustions,

which highly stress the engine. In highly charged engines in addition to
knocking combustion also pre-ignition and glow ignition occur.

Pre-ignitions normally occur in the low-end-torque range at low engine

speed and high load. These irregular combustions are characterized by
stochastic occurrence and by an alternate order of regular combustion
and pre ignition. The occurrence of glow ignition is mostly linked to hot
spots in the combustion chambers at high engine speed and load. These
hot spots are responsible for an early ignition of the air-fuel-mixture.

Equal for all above mentioned irregular combustion phenomena is loss of

control of the combustion process. This leads to the need of avoidance of
these phenomena to protect the engine of heavy damage. By reason of
the occurrence of these phenomena in the whole engine operation map
an according observation is needed.

The Continental pre-ignition and glow ignition function is able to detect

irregular combustion phenomena in the whole operation map and to ac-
tivate system reactions to reduce the strain to a minimum. Splitting the
function into several simultaneous active detection path guarantees a de-
finite detection during operation. Apart from a good detection avoidance of
wrong detections by analyzing driving conditions can be guaranteed with
the function.

360
The paper highlights the pre-ignition and glow ignition function and the
different detection paths. Content of this paper are the detection algorithm
and the signal processing in the different paths. Furthermore the system
reactions of the function are highlighted, which especially should prevent
the continuing respectively the formation of above mentioned irregular
combustions.

1. Introduction
During the last years a trend to turbocharged SI engines with small dis-
placements has established worldwide. These so called downsizing con-
cepts can be calibrated very efficiently and with low fuel consumption.
These engines are an effective measure to reach the regular CO2 targets.
The enhancements in charging technology provide very high mean effec-
tive pressure levels over a large engine speed range and a strong driving
performance also. By reason of high low-end-torque values a modification
of axle ratio and an application of downspeeding are possible. The stress
for the engine during driving conditions, by reason of an operation point
shift caused by downsizing/speeding, is increasing and leads to higher
occurrence of irregular combustion phenomena. The spectrum of irregular
combustions contains knocking combustion, pre-ignitions and glow igni-
tions.

By reason of different characters of irregular combustions, an avoidan-

ce of continuing by retarding the ignition timing is not possible. Equal for
pre ignition and glow ignition is loss of control of combustion initialization.
The air-fuel-mixture is inflamed by a self ignition process before the spark
plug ignition. The early ignition causes a quick energy conversion and
therefore an extreme pressure and temperature rise in the combustion
chamber. The combustion pressure is similar to a knocking combustion
heavily oscillating. Already after few events a heavy damage can happen
due to the resulting mechanic and thermal stress. The avoidance of these
combustion phenomena or the limitation to some events only is the target
during engine operation.
Corresponding to different occurring combustion phenomena, Continen-
tal´s pre-ignition and glow ignition function was developed with two in-
dependent detection paths. In addition to the detection a functionality to
control system reactions to avoid further occurrence is implemented.

361
2. Irregular combustion

2.1. Knocking combustion

This irregular combustion occurs at low load up to full load over the com-
plete engine speed range of the engine. Typical for knocking combus-
tion is the ignition of the air-fuel-mixture by a spark of the spark plug.
The propagating flame front started at the spark causes an increase of
pressure and temperature level in the tail gas and leads to critical condi-
tions also. Induced by these critical conditions in the unburned mixture a
spontaneous self ignition with energy conversion happens which causes
a further ignition of unburned parts of the mixture [1]. The quick energy
conversion during the burning process of the mixture causes high cylinder
pressure peaks with heavy overpressures (figure 1) as well as thermal
stress. With a rising quantity of unburned mixture the pressure rise and
also the overpressure increase.
Retarding the ignition timing is an easy way to avoid knocking combustion.
The detection of knock is done by a knock sensor and by the analysis of
knock sensor signal. The knock sensor records the cylinder overpressure
during knock and transfers it to the ECU. In the ECU the received signal
is analyzed by the knock function in a calibrated crank angle range after
IGTDC.

Figure 1: Cylinder pressure curve at knocking combustion

362
2.2. Pre-ignition
This irregular combustion occurs at low engine speed and high load con-
ditions up to full load. This combustion is initiated by a self ignition of the
air fuel mixture. The air-fuel-mixture is self ignited after reaching energetic
critical conditions before the initialization of combustion by the spark of the
spark plug. This combustion type is characterized by a stochastic occur-
rence of single events and also of a sequence of events. A direct progres-
sion of pre-ignitions is normally not observed and the pre-ignitions events
are interrupted by one or two regular combustions [2].
Caused by the early ignition of the mixture before TDC during compressi-
on and therefore in the upward motion pressure peaks up to 250bar can
occur. Below following figure 3 shows the cylinder pressure curve of a
pre-ignition. Typical for a pre-ignition is a heavily oscillating cylinder pres-
sure located after a steeply rise at beginning.
In comparison to a knocking combustion an avoidance of pre-ignition
events by shifting the ignition timing is not possible.

Figure 2:Cylinder pressure curve by pre-igntion

363
2.3. Glow ignition
This irregular combustion type occurs at higher load up to full load over
the whole engine speed rang predominant at high engine speed. This
combustion is also started by a self ignition of the air-fuel-mixture, induced
by hot or overheated parts in the combustion chamber [3]. Critical thermal
temperatures of parts are released by knocking combustion cycles. By
reason of a quick heat release on knocking combustions an increase of
heat introduction in the combustion chamber is released. If the combus-
tion chamber increases a critical temperature limit caused by an ongoing
knocking, it will lead to an earlier self ignition of the air-fuel-mixture. De-
pending on the released energy a self-sustaining or self-energizing com-
bustion can result [4].

The early self ignition timing leads to a changed combustion and therefore
to a different cylinder pressure curve, see figure 2. Starting from a regular
combustion without knocking with a smooth cylinder pressure curve the
cylinder pressure gets sharper and also the overpressure higher with in-
creasing number of cycles and earlier self ignition timing. For very early
self ignition timings, more than 50°CAbTDC, the shape of the pressure
curve turns back to smooth without oscillation.

Figure 3: Typical cylinder pressure curve by glow igntion

364
3. Pre-Ignition- and Glow Ignition-Function

3.1. Structure of the function

According to several occurring irregular combustions in SI engines dif-
ferent detection algorithms for a faultless detection were realized in the
function. All known irregular combustions were compared and existing dif-
ferences were carved out.
As pictured in the figures of chapter 2 pre-ignitions and glow ignitions
show a heavy pressure oscillation as well at and after cylinder pressure
peak. This pressure oscillation is formed strongest at ignition timing or
rather before of a regular combustion. By reason of this characteristic a
new segment for detection was introduced.

Corresponding to an already existing crank angle segment for detection

of knocking combustion in a first step an additional crank angle seg-
ment was implemented. The new crank angle segment was positioned
in front of the segment for knock detection. Naming of the new crank
angle segment was given as a result of the position of the segment as
“pre-window”. Naming of the already existing segment was chosen as
“main window”.

Due to the implementation of the pre-window combustion cycles with early

ignition can be detected. A further advantage of the new window is distin-
ction between knocking combustions and irregular combustions with self
ignition before the spark. In case of pre-ignition high knock values in both
windows are detected, which therefore give the opportunity to distinguish
from a knocking combustion with detection in main window only.

3.1.1. Main window

The crank angle segment is normally used for detection of knocking combus-
tion. Position of the segment depends on the operation point but the begin is
always positioned next to the maximum of the cylinder pressure. For knock
observation the knock sensor signal is recorded and a knock value calcula-
ted. By comparing knock value and threshold a knocking combustion can be
detected.

Similar to the strategy for detection of knocking combustions is the proce-

dure for detection of irregular combustions with high cylinder overpressu-

365
res. Corresponding to the rise of cylinder overpressure in the combustion
chamber the calculated knock value is increasing by irregular combustions.

For detection of irregular combustion an additional threshold was imple-

mented. Calculation of this new threshold is based on the knock detection
threshold. By the reference on the knock threshold an accurate classifica-
tion of knock and combustion abnormity is possible.

An exceeding of threshold by the knock value therefore leads to detecti-

on of irregular combustion. Combined with an exceeding of the detection
threshold is an overpressure in the cylinder and for this reason stress on
the engine. A slight difference between both detection thresholds offers
the opportunity to reduce stress and to protect engine of serious dama-
ge. Therefore mechanical characteristics of the engine can be taken into
account by the calibration. Also an individual calibration fitted on every
particular engine is possible.

As mentioned before another observation segment is needed for distincti-

on between hard knock and pre-ignition.

3.1.2. Pre-window
The window was introduced for detection of irregular combustions with
self ignition of the air fuel mixture before ignition timing. In a crank angle
segment free to calibrate and positioned in front of the main window the
knock sensor signal is recorded and a knock value is calculated. By ana-
lysis of the knock value a detection of irregular combustion and therefore
a distinction to regular combustion are possible. The detection of an irre-
gular combustion is done by comparing the knock value with a calibrated
detection threshold. Combustions with high cylinder overpressure are de-
tected in case of knock value exceeds value of threshold. Corresponding
to the setting of the detection threshold a sensitive or robust detection can
be executed. Setting of the detection threshold is therefore essential for
resulting stress on engine during operation.

Calculation of detection threshold in pre-window is based on an average

noise level. Due to this fact the engine noise is recorded by the knock sen-
sor in the pre-window. Based on the measured engine noise an averaged
noise level in the function is calculated. The calibration of the detection th-
reshold is done under operating conditions with irregular combustions by

366
comparison of knock values and cylinder overpressures of the indicating
measurement system. Differences between noise level and knock value
can be chosen relatively high, like measurements have shown. Heavy
overpressures of pre ignitions, which occur in the range of the pre window
and thereby cause high knock values, are responsible. A second advanta-
ge of high differences is a robust classification of fault detections caused
by disturbing noise of the engine.

Pre-ignitions, remaining heavy knocking and beginning of glow ignitions

are easily to detect by analyzing knock sensor signals in both windows.
Problematical for detection with knock sensor are combustion cycles with
a smooth cylinder pressure curve without cylinder overpressure, as they
occur at glow ignition. Due to this behavior detection by single use of
knock sensor cannot be performed.

3.1.3. Observation of engine roughness

For detection of combustion cycles of a definitive glow ignition with smoo-
th cylinder pressure curve a further functionality for observation was im-
plemented.
Ignition of air fuel mixture occurs very early at glow ignition about 50°CAbT-
DC. This behavior is shown in figure 3, where the cylinder pressure curves
over number of cycles are pictured. Caused by the early self ignition is the
combination of energy conversion and upward motion of the piston. Both
are responsible for a steeply rising pressure in the combustion chamber.
Corresponding to the rising pressure volumetric work is rising. At that time
it leads to a higher torque requirement at the crank shaft and therefore to
an influence on engine roughness. The higher torque requirement causes
an extension of cycle duration, which is analyzed in the engine roughness
signal. Influence on engine roughness of the engine is rising with an in-
creasing shift of self ignition towards BDC.

By using already existing signals of misfire functionality evaluation of influ-

ence on engine roughness could be done.
For detection of combustion cycles with smooth pressure curve a new th-
reshold was implemented. Calculation of the threshold is based on misfire
detection threshold. The logic for calculation of engine roughness for each
cylinder in misfire function was not changed. Detection of glow ignition
is done by comparison of current engine roughness value and detection
threshold. In case of threshold undercut by engine roughness an irregular

367
combustion is detected. Despite analysis of the same signals a faultless
detection and distinction between misfire events and irregular combus-
tions could be represented. Due to a stronger influence on engine rough-
ness at misfire distinction to glow ignition is possible.

3.2. Detection paths in the function

Measurements on several engines showed two different behaviors con-
cerning transition speed from regular to irregular combustion. On one
hand a very fast transition from regular combustion to regular combustion
with smooth cylinder pressure curve was observed and on the other hand
a very slow but constant transition from regular to uncontrolled self ignition
with high overpressure in the end was observed.
The first group of irregular combustion includes all combustion with slow
transition. These combustions normally start with knock followed by heavy
knock and in the end by self ignition before the spark. Corresponding to
this behavior detection is moving from main to both windows and in the
end to pre window. Detection of all these irregular combustion by knock
sensor only is therefore possible.
The second group of irregular combustions includes all combustions with
fast transition. Most of the time regular combustion and definitive irregular
combustion are separated by few knocking cycles. Detection by knock
sensor only is therefore not possible. For detection of these irregular com-
bustions an additional observation of engine roughness is necessary.
Corresponding to different behaviors of transition two detections path
were implemented in the function. Detection of irregular combustions in
first group are done in “SLOW”-path and of combustions in second group
in “FAST”-path.

3.2.1. Slow path

Slow path of function is used for detection of irregular combustions, which
can be identified by single analysis of knock sensor signal. Typical types of
combustion detected in slow path are single pre-ignition events or a row of
pre-ignition events and continuing hard knock.
Knock sensor signals of pre-window and main window are used for de-
tection of combustion phenomena in this path. Irregular combustions
are detected and counted if the of knock value exceeds the detection
threshold.

368
Figure 4 shows all steps of slow path from first detection to injection de-
activation.

Figure 4: Steps in slow-path from first detection to cylinder shut-off

First step in the slow path is detection of some irregular combustion

events in pre and main window. In the function included is a separate
identification of both windows. Events in both windows are counted and
additionally a summation of both windows is calculated. In a next step
several system reactions are activated after reaching a preset number
of detections. If these system reactions cannot stop continuing irregu-
lar combustions in a further step injection of cylinder is switched off.
Cylinder shut off is conducted after exceeding a permissible number of
detections. Another enabling of injection during the current driving cycle
depends on the calibration. Reactivation of injection is done after elapse
of calibrated shut off time.

System reaction

In slow path of function an activation of system reaction is possible. Sys-

tem reactions in function were implemented to avoid continuity of irregular
combustions by adjustments in the ECU. Activation and deactivation of
the different system reactions are completely independent of each other.
In function implemented system reactions are:

• Fuel enrichment

• Adjustment of camshaft position

• Load reduction

Fuel enrichment

At activation of system reaction fuel enrichment global reduction of lam-

bda in the engine to calibrated value is conducted. Therefore the engine
is operated with fuel enrichment for a preset duration. At end of duration

369
reset of lambda is done continuously without a jump. Afterwards engine is
operated with same lambda value as before the activation.
Measurements have shown that this system reaction is the most effective.
Continuing of irregular combustion can be easily avoided by this system
reaction. A further advantage is a not existing influence on driveability.
Torque- and power level of the vehicle are maintaining completely unch-
anged for the driver. Dangerous and unpredictable situations for the driver
could be easily avoided.

Adjustment of camshaft position

When activating the system reaction a relative adjustment of intake cams-

haft position is conducted. Based on the current setpoint a shift to earlier
or later opening timing is possible. Adjustment of valve timing after first
activation is executed in a jump. Duration of activation for the system re-
action is completely free to calibrate. At end of duration reset is done back
to setpoint continuously.
This system reaction causes a strong influence on charge of cylinder and
on torque of the engine. A significant shift of intake valve timing is most
of the time linked to a strong torque reduction which can cause an uncer-
tainty of the driver. Impact on driveability has to be taken into account at
calibration of this system reaction.

For state of the art internal combustion engines operated with scavenging
in the low end torque range combination of enrichment and shift of inta-
ke camshaft setpoint is necessary to protect catalyst of thermal damage.
This fact shows that the function is adapted to state of the art operation
modes of an internal combustion engine.

Torque reduction

This system reaction causes a reduction of engine torque. In the function

two different kinds of reduction are available. First opportunity is reduction
of full load torque curve of the engine by a factor. During active system
reaction only a limited engine torque map is available for the driver. Re-
duction of driver input is therefore only done between full load and torque
limitation.
Second opportunity is the application of a reduction factor on the current
driver input. Therefore reduction is active over the whole operation map
of the engine. By reason of no occurrence of irregular combustions in low
load condition a minimal threshold for torque reduction was implemented.

370
In case of driver input falls below the threshold reduction factor is set in-
active.
To maximize protection of the engine torque reduction is conducted by a
jump at beginning. Corresponding to the applied reduction factor influence
on driveability can be very high. This fact should be taken into account
during calibration of this system reaction. Thus maximum protection of the
engine with high reduction factor causes maximum impact on driveability
with uncertainty of the driver. In such a case it is necessary to inform the
driver. At the end of duration reset of torque value is done continuously to
reduce impact on driveability.

Activation of each system reaction starts after exceeding a preset value of

detected events in both windows. Dependent on irregular combustion acti-
vation therefore happens after the first or the second event. At pre-ignition
event two detections in pre-window and main window are counted. Same
result is given after two hard knocking cycles in main window.

To protect engine of damage an early activation, e.g.: after one pre-igniti-

on, is necessary. As previously mentioned fuel enrichment offers the big-
gest influence to avoid irregular combustion. Due to this behavior general
fuel enrichment independent of combination with other system reactions
is always meaningful.

If system reactions are not sufficient to avoid consistency of irregular com-

bustions in a next step injection of the affected cylinder is switched off
to protect engine of damage. Threshold values for injection deactivation
are separately calibrated for each detection window. Additional to these
threshold values a summation threshold is checked. For this reason all
events in both windows are summed up. In case one threshold is excee-
ded a request for cylinder shut-off is generated.

3.2.2. Fast path

Fast path of function is used for detection of irregular combustions, which
cannot be identified by single analysis of knock sensor signal. Typical ty-
pes of combustion detected in fast path are glow ignitions.

Figure 5 shows all steps of fast path from first detection via knock sensor
to observation of engine roughness and in the end to injection deactiva-
tion.

371
 Figure 5: Steps in fast-path from first detection to cylinder shut-off

In this path for detection of combustion phenomena on one hand knock

sensor signals in pre-window and main window and on the other hand
engine roughness signals are used. An analysis of knock signal, calcu-
lation of a knock value and comparison with an individual threshold is
carried out for both crank angle segments. The detection thresholds in
fast path are completely independent to the detection thresholds in slow
path. Calculation of identification thresholds is done based on same si-
gnals as in slow path. As mentioned above calculation of threshold in
main window is based on the knock detection threshold. In pre-window
calculation is based on an average noise level of the engine in fired
operation conditions.

As a result of a mostly short transition duration from regular combustion to

glow ignition a faultless detection of knock to enabling observation of engine
roughness has to be guaranteed. By trend values for the thresholds in both
windows in fast path are normally smaller than in slow path. Furthermore
a couple of detected events are sufficient to activate observation of engine
roughness. Unlike in slow path a recurring detection of knock does not lead
to cylinder shut-off.

Starting with detections in pre-window or main window and confirmation

of a first detection condition observation of engine roughness is activated.
As mentioned before identification of glow ignition is done by comparison
of current engine roughness value and calculated detection threshold. In
case of an undercut of the threshold a glow ignition combustion cycle is
detected. A confirmed status of identification is reached after exceeding
a permissible number of cycles with glow ignition. Linked to the second
confirmation is the setting of an injection deactivation request.
Observation of engine roughness is conducted only for a calibrated durati-
on after confirmation of detection in pre or main window via knock sensor.
Each time confirmation is refreshed duration of observation is reset to
a calibrated value. After elapsed observations time all activation condi-
tions and counted detections are cleared. So if during observation time

372
the number of permitted glow ignition cycles is not reached and cylinder
shut-off request is not set everything is cleared at the end. Due to the facts
some experience for a robust and faultless calibration are necessary.

3.3. Cylinder shut-off

To protect engine of damage an injection deactivation on the affected
cylinder after confirmation of irregular combustion can be requested. The
function enables two different kinds of injection deactivation.

Execution of injection deactivation can be adjusted reversible or irrever-

sible in the driving cycle. As the name suggests the irreversible cylinder
shut-off is active for the remaining part of the driving cycle. Influence of
the strategy on driveability is extreme, but the reachable protection of the
engine is the most efficent.
Second strategy is a reversible injection deactivation. In case of a rever-
sible shut-off request calibrated minimum time duration has to elapse be-
fore injection reactivation. After elapsed shut-off time and end of driving
condition trailing throttle injection is switch on again. Therefore reactivati-
on of injection can be done torque neutral. The repeatability per driving cy-
cle can be calibrated. In case of reaching the number of applied reversible
shut-offs per driving cycle injection of the affected cylinder is switched off
irreversibly.

4. Validation of the pre-ignition and glow ignition de-

tection
In the following the result of glow ignition detection validation is discussed.
The validation was conducted on a state of the art engine equipped with
turbocharger and direct injection on the engine test bench at 2000rpm
engine speed. For validation measured values of the ECU and signals of
the indication system have been compared and analyzed. The measured
signals of the ECU from the beginning to the end of glow ignition with
injection deactivation are pictured in figure 6. Signals pictured in the figu-
re are the knock values of pre-window and main window plus detection
thresholds, engine roughness values of the affected and of the next fired
cylinder plus detection threshold.

373
For the validation of glow ignition detection slow path functionality to-
gether with system reactions were enabled. Thereby cylinder shut-off by
slow path functionality was not possible.
In below inserted figure heavy knocking events in main window at be-
ginning of glow ignition are visible. At this point of time these events are
interrupted by regular combustions. During measurement number of hard
knocking cycles is increasing and events are becoming more intensive.
After reaching an applied number of detected cycles in main window ob-
servation of engine roughness is started.

Figure 6: Validation of glow ignition detection

During measurement knock is becoming harder and combustion initializ-

ation is shifting earlier. The shift of combustion causes a transition of de-
tection from main window to both windows. Responsible for the transition
is an occurring hot spot in the combustion chamber. The hot spot has a
self-energizing and self-sustaining behavior and causes the transition of
combustion. In case of a simultaneous confirmation of detection in pre-
window and main window engine roughness threshold is shifted closer to
the engine roughness values.
With increasing measurement duration knock values in both windows are
becoming higher until a maximum is reached. Figure 8 shows a combus-
tion cycle at the time of maximum knock values. The exact position of the
cycle is shown in figure 7 and is marked with “cycle 1”.

374
 Figure 7: measured signals at validation of glow ignition detection

The pictured combustion cycle shows a typical cylinder pressure curve

with strong oscillation as it occurs at pre ignition. Inflammation of the air
fuel mixture starts at the end of the compression stroke and causes a hea-
vy pressure rise. After the first pressure peak next to TDC an oscillation of
cylinder pressure occurs. Optimal positions for the detection windows are
therefore in the crank angle range of heavy oscillation. As figure 8 shows
the position of pre and main window are calibrated ideally. As a result of
the calibration knock values correlate to the measured cylinder overpres-
sure and show high values, too. Furthermore the figure shows how the
detection of a pre-ignition event in the function is conducted.

After the maximum knock values in both windows decrease very fast and
reach noise levels of a regular combustion at last. Responsible for the
behavior is the continuing movement of combustion initialization towards
early. The smooth cylinder pressure curve of the very early combustion
causes reduced knock values at a level of regular combustion. Figure 9
shows a combustion cycle of glow ignition at the time. The exact position
of the cycle is shown in figure 7 and is marked with “cycle 2”. The absence
of a cylinder overpressure is the reason for the low knock values. From
this point of time slow path would be without any further detection and
could not protect the engine of damage.

375
 Figure 8: combustion with overpressure at cycle 1

Figure 9: combustion with smooth cylinder pressure at cycle 2

After the decrease of knock values an influence on engine roughness

can be observed. During the following combustion cycles the influence
on engine roughness is becoming stronger and leads to an undercut of
the detection threshold. As mentioned before calculation of the detection
threshold is based on the misfire identification threshold. After reaching a
permissible number of detected glow ignition cycles an injection deactiva-
tion request is generated.

376
5. Conclusion
For realization of regular fuel consumption targets, passenger cars are
equipped with downsized engines. For this state of the art engines direct
injection systems and turbo charging systems are combined. Due to an
operation point shift up to higher mean effective pressures of these engi-
nes, thermal strain is increasing during driving operation. Linked to a high
thermal strain is an occurring of irregular combustions, which highly stress
the engine and may lead to engine damage.

Continental´s pre-ignition and glow ignition function offers the opportunity

to protect the engine of damages caused by irregular combustions. Based
on knocking functionality of Continental a new function for detection of dif-
ferent kinds of irregular combustions was developed. At low engine speed
main task is to detect pre-ignitions and to avoid a sequence of several
events in a row. At mid to high engine speed glow ignitions are becoming
more interesting. At these engine speeds a faultless detection of glow
ignitions is the main task.

Due to these different characters of irregular combustions two detection

paths were implemented in the function. For detection of pre-ignition and
hard knock a path with single analysis of knock sensor signal is used. This
first path, called slow path, includes also an activation of system reactions
to avoid a continuing of irregular combustion. For detection of irregular
combustions with a smooth cylinder pressure curve especially glow ig-
nitions a second path is used. This second path, called fast path, uses a
combination of knock sensor signal and engine roughness analysis.

Both paths are discussed in the paper and execution of detection is de-
scribed. Furthermore the system reactions and the cylinder shut-off for
maximum protect of the engine in case of irregular combustions are exp-
lained.

Based on a measurement the paper shows the detection of glow ignition

combustion in fast path of the function. Pictured in the measurement are
all steps from beginning with detection of knock followed by the detection
of influence on engine roughness until the end with cylinder shut off.

Continental´s pre and glow ignition function supports the costumer to con-
tinue trend of downsizing to reach regular fuel consumption targets with
maximum protect of the engine at occurrence of irregular combustions.

377
Based on the experience of several costumer projects and by reason of
a continuing development Continentals pre and glow ignition function is
prepared for future challenges concerning charged SI engines.

378
Literature
[1] Willand, J.; Nieberding, R.; Vent, G.; Enderle, C.: The Knocking Syn-
drome - Its Cure and Its Potential. SAE 982483, 1998.
[2] Palaveev, St.; Dahnz, Ch.: Selbstzündungsphänomene in Ottomoto-
ren mit Direkteinspritzung. Direkteinspritzung im Ottomotor VII. Haus
der Technik Fachbuch, Band 111, S.80-98, 2010
[3] Ernst, R.; Münzinger, St.; Hanula, B.; Schnitzler, T.; Geiser, F.: Das
Phänomen Glühzündung Ursachen, Entstehung und Vermeidung.
MTZ 63, Nr. 11, S. 916-924, 2002
[4] Dahnz, Ch.; Palaveev, St.; Schwenger Ch.; Spicher, U.: Faseropti-
sche Untersuchungsmethoden zur Analyse von Verbrennungsano-
malien in Ottomotoren. Ottomotorisches Klopfen - Irreguläre Verbren-
nung. Berlin : Expert Verlag, ISBN: 978-3-8169-3047-1, 2010.

379
20 Pre-ignition analysis on a turbocharged
gasoline engine with direct injection
Erik Schünemann, Andreas Witt, Markus Selder,
Christian Schwarz

Abstract
The consistent implementation of additional functional potentials in the
combustion engine, which will have the largest share of the worldwide
drivetrain portfolio despite rising rates of electrification, is a major part of
the BMW EfficientDynamics strategy. The BMW TwinPower Turbo techno-
logy package, containing turbocharging, fully variable valvetrain with VAL-
VETRONIC and Bi-VANOS, and direct fuel injection, is used in the petrol
engines of the BMW Group. Besides typical BMW driving dynamics and
a simultaneously significant reduction in fuel consumption irrespective of
the locally available fuel quality, using BMW TwinPower Turbo technolo-
gy permits compliance with all current world-wide exhaust gas emission
legislations. Additional improvements can be achieved by further raising
specific torque and power and possibly increasing the compression ratio
at the same time. Therefore, it is necessary to understand current limita-
tions that are given by the occurrence of irregular combustion phenomena
at high engine loads, e.g. pre-ignitions. Pre-ignition analysis showed that
blowby liquid containing engine oil as well as hot deposits/particles inside
the combustion chamber during the compression stroke are significant
sources of pre-ignitions. Hence, best possible separation efficiency of the
crankcase ventilation system as well as prevention of deposit formation
are required. Pre-ignition analysis is greatly supported by a robust cylinder
pressure indication and by the application of “keyhole” optical measure-
ment techniques e.g. high-speed endoscopy.

Kurzfassung
Die konsequente Umsetzung weiterer funktionaler Potenziale am Ver-
brennungsmotor, der auch bei steigendem Elektrifizierungsgrad den
größten Anteil am weltweiten Antriebsportfolio haben wird, ist wesentli-
cher Bestandteil der BMW EfficientDynamics Strategie. Dazu kommt in
den Otto- und Dieselmotoren der BMW Group das Technologiepaket

380
BMW TwinPower Turbo zum Einsatz, das für Ottomotoren aus den Ele-
menten Abgasturboaufladung, vollvariabler Ventiltrieb mit VALVETRONIC
und Direkteinspritzung besteht. Neben BMW typischer Fahrdynamik und
gleichzeitig signifikanter Absenkung des Kraftstoffverbrauchs unabhängig
von der lokal vorliegenden Kraftstoffqualität erlaubt die BMW TwinPower
Turbo Technologie die Erfüllung aller aktuellen weltweiten Abgasgesetzge-
bungen. Zusätzliche Potenziale können durch die weitere Anhebung von
spezifischem Drehmoment und spezifischer Leistung bei gleichzeitig weiter
erhöhtem Verdichtungsverhältnis gehoben werden. Voraussetzung dafür
ist ein umfassendes Verständnis der bei hohen spezifischen Lasten auftre-
tenden irregulären Verbrennungsphänomene, wie z.B. Vorentflammungen,
die aktuell begrenzend für Aufladegrad und Verdichtungsverhältnis von
Turbo DI Ottomotoren sind. Die in diesem Beitrag beispielhaft vorgestell-
ten Ergebnisse von Vorentflammungsanalysen zeigen, dass sowohl nicht
abgeschiedene Blowby-Flüssigkeit, die Motoröl enthält, als auch heiße
Ablagerungen/Partikel, die sich während der Kompression im Brennraum
befinden, signifikante Quellen für Vorentflammungen sind. Deshalb sind so-
wohl bestmöglicher Abscheidegrad der Kurbelgehäuseentlüftung als auch
die weitestgehende Vermeidung von Ablagerungen zwingende Vorausset-
zungen für die Absenkung des Vorentflammungsrisikos. Die Analyse von
Vorentflammungen wird wesentlich unterstützt durch den Einsatz robus-
ter Indiziermeßtechnik und die Anwendung von optischen „Schlüsselloch“
Meßverfahren wie z.B. Hochgeschwindigkeits-Endoskopie.

1. Introduction
The BMW Group is convinced that the need for individual mobility of their
customers is closely linked to the desire for sustainability. Rising fuel prices
and the tax benefits provided for low-emission vehicles also mean that lower
fuel consumption and low exhaust gas emissions are becoming increasingly
important decision criteria when a vehicle is purchased. Legislation in the
worldwide automobile markets also demands further significant reductions
in CO2 and pollutant emissions. In contrast to this is the customer wish for
more driving dynamics, comfort and safety. The strategic approach of the
BMW Group to resolve this conflict of objectives is BMW EfficientDynamics.
Besides measures involving aerodynamics and weight reduction, as well as
an overall energy management, the consistent further development of the
combustion engine, which will also have the largest share of BMW Group´s
worldwide powertrain portfolio despite rising rates of electrification, will be a
major factor for success in the foreseeable future.

381
2. BMW TwinPower Turbo Technology
The BMW TwinPower Turbo technology package is used in current engi-
nes of the BMW Group and will also be used in the next engine generati-
on. For gasoline engines, it consists of the elements (Fig. 1):

• Turbocharging
• Fully variable valvetrain with VALVETRONIC and Bi-VANOS
• High Precision direct fuel injection

Fig. 1: Components of the BMW TwinPower Turbo Technology for Petrol

Engines

The combination of Turbocharging with VALVETRONIC, Bi-VANOS and

High Precision DI injection provides high specific torque and power cou-
pled with significant response advantages in comparison with turbochar-
ged petrol DI engines with throttle valve load control. Injection pressures
of up to 200 bar and the degrees of freedom resulting from the spray and
flow configuration for each individual hole in the multi-hole injector ensure
good mixture homogenization and minimize the amount of fuel wetting
the walls. Making use of the valvetrain variabilities of VALVETRONIC and
Bi-VANOS enables fuel consumption advantages in broad ranges of the
characteristic engine map by largely dethrottling the gas exchange in par-
tial load with high internal EGR rates. Additional measures that increa-
se the charge motion level over the entire valve lift curve, provide the
necessary combustion stability at low loads as well as rapid and complete

382
combustion in the upper load range. This results in low brake specific fuel
consumption figures compared with the competition, shown in Fig. 2 for
the best point in the characteristic engine map and for the part load re-
feence point n = 2000 rpm; we = 0.2 kJ/l.

Fig. 2: Minimum BSFC and BSFC @ n=2000 rpm / we=0.2 kJ/l

for BMW TwinPower Turbo engines vs. competition

3. Classification of Irregular Combustion Phenomena

in Turbocharged DI Petrol Engines
Contributing to these low fuel consumption figures are relatively high
compression ratios that can be achieved with the BMW TwinPower Turbo
technology package. Current BMW TwinPower Turbo petrol engines are
already reaching a specific torque of more than 200 Nm/l and a specific

383
power of more than 100 kW/l while still maintaining a compression ratio
of CR = 9.5. However, additional potentials regarding increases of both
specific engine loads and engine efficiency i.e. compression ratio can be
reached. Therefore, it is necessary to understand the limitations that are
currently given by the occurrence of irregular combustion (Fig. 3). The-
se phenomena can be divided into irregular combustion events that start
temporally after or before the regular spark angle. Irregular combustion
with an ignition after the regular spark angle is typically related to spark
knock appearing either as “classical” knock or “super knock”. In case of
an ignition before the regular spark angle the event can be classified into
a glow ignition or a pre-ignition.

Glow ignition is caused by hot surfaces of the combustion chamber, e.g.

exhaust valves and can occur at high engine loads over a wide engine
speed range. At steady-state operation, the corresponding early combus-
tion leads to a further elevated temperature level of the hot combustion
chamber surfaces which again shifts the consecutive glow ignition and
combustion of the respective cylinder towards earlier crank angles. Thus,
glow ignition is generally self-amplifying and if started once, it occurs in
every combustion cycle of the affected cylinder.

Pre-ignition events show a different behavior. They occur at low to medi-

um engine speeds and very high engine loads. Pre-ignition can appear as
a single event or as a series of events, where either consecutive pre-igni-
tions occur or pre-ignitions and regular ignitions alternate.

384
 Fig. 3: Irregular combustion at turbocharged DI petrol engines

4. Pre-Ignition Sources at Turbocharged DI Petrol Engi-

nes
Understanding pre-ignition is the subject of many recent studies. Accor-
ding to [1], it can be categorized as shown in Fig. 4. Due to the high com-
plexity of the pre-ignition phenomena, the relevant pre-ignition paths can
be highly dependent on the examined engine configuration, operating pa-
rameters and boundary conditions. This article focuses on an exemplary
study of two possible sources:

• the influence of blowby fed into the engine by the crankcase ventilation
system in an early development status
• the influence of deposits

385
 Fig. 4: Pre-ignition sources at turbocharged DI petrol engines [1]

The examined engine configuration is shown in Fig. 5. The TwinPower

Turbo multi-cylinder engine has a displacement of 0.5 l/cylinder, a central
DI injector position with transversal arrangement of injector and spark plug
with the injector on the intake side and the spark plug on the exhaust side.

Fig. 5: Engine configuration

386
4.1. Influence of Blowby on Pre-Ignitions
For the measurements, the engine was equipped with an additional filter
in the charge air path before the throttle valve. The main task of this additi-
onal filter was an efficient filtration of oil droplets at low additional pressure
losses. The crankcase ventilation system had one “low-load” blowby path
for naturally aspirated operation, where the blowby after separation of the
liquid phase was fed into the intake ports. A second “high-load” path was
used for turbocharged operation, where the blowby after separation of the
liquid phase was fed into the intake of the compressor. It must be noted
that the system was in an early development status with a separation effi-
ciency that proved to be not sufficient.

All pre-ignition measurements for this have been performed at a steady

state high-load operating point with a duration of several hours under de-
fined boundary conditions. Before the start of every measurement, the
engine was conditioned and an oil change was conducted. Pre-ignitions
have been detected by cylinder pressure measurement through compari-
son of the start of combustion with the spark ignition angle. Fig. 6 summa-
rizes the results. Increased pre-ignition numbers can be observed when
the “high-load” path of the crankcase ventilation system or the complete
crankcase ventilation system is active, i.e. blowby is fed back into the
engine. If the crankcase ventilation system is not active, i.e. blowby is not
fed back into the engine or the additional filter blocks the liquid phase,
very few pre-ignitions are measured. This leads to the conclusion that in
this configuration the liquid phase of the blowby contains components that
force the occurrence of pre-ignition because the blowby gas phase is not
blocked by the additional filter.

387
Fig. 6: Influence of blowby fed into the engine by the crankcase ventilati-
on system (early development status) on pre-ignitions

Chemical analysis of the blowby gas phase showed no critical compo-

nents and hence confirmed this observation. However, the blowby liquid
phase contained a significant amount of engine oil as expected due to the
insufficient oil separation. To confirm the blowby liquid phase as source of
the detected pre-ignitions, measurements of the ignition delay have been
performed with several gasoline qualities as well as with the blowby liquid
phase and with original engine oil at the Ignition Quality Tester [2]. For en-
gine oil and blowby liquid phase, ignition delays are shortened by a factor
of more than 3 compared with gasoline (Fig. 7). Hence, the best possible
separation efficiency of the crankcase ventilation system is a robust mea-
sure to reduce the occurrence of pre-ignitions.

388
 Fig. 7: Ignition delays of gasoline, blowby liquid phase and engine oil
at the Ignition Quality Tester

4.2. Influence of Deposits on Pre-Ignitions

In order to examine the possible role of deposits as a source of pre-igni-
tions, the aforementioned engine was equipped with an optical access
for the application of high-speed endoscopy at one cylinder (Fig. 8). The
two bores were used for endoscopic access and for illumination of the
combustion chamber as well as the injected fuel spray. The endoscopic
detection path was transmitting in the visible region of the spectral range.
Endoscopic measurements and cylinder pressure indication were syn-
chronized. All data was written into ring buffers containing approximately
50 cycles. When a pre-ignition was detected by the indication system, the
respective cycle and 50 cycles prior to the pre-ignition cycle were saved.

389
 Fig. 8: Optical access to the combustion chamber for
high-speed endoscopy

Fig. 9 shows that deposits can be separated from the combustion cham-
ber walls e.g. by the intake air motion during the compression stroke. Al-
though not shown here, it is expected that a similar process can also occur
at the surfaces of intake ports and intake valves. Separation of deposits
from these surfaces can also result from vibrations caused by knocking
combustion. The separated deposits become air-borne as particles and
can take part in the regular combustion process. If these particles are not
completely consumed by the end of the combustion process and survive
the exhaust stroke, they should be visible as light-emitting, i.e. glowing
particles in the following cycle and could act as a pre-ignition source. Fig.
10 shows such an air-borne glowing particle at the early compression stro-
ke of a pre-ignition cycle. In this image also the spray of the fuel injection
is present.

Fig. 9: Deposits separate from the combustion chamber wall

390
 Fig. 10: Glowing air-borne particle at the early part of the compression
stroke of a pre-ignition cycle

Fig. 11 shows a more detailed example. At the start of the compression

stroke (177 °CA BTDC) together with the fuel spray a glowing air-borne
particle is visible for the first time in this cycle. The particle is present also
in the later phase of the compression at 156 °CA and 24 °CA BTDC. At 9.6
°CA a premixed flame front is visible in the vicinity of the particle, which
pre-ignites the compressed charge before the regular spark angle. It
should be noted that for the examined engine configuration and boundary
conditions approximately 40 to 50 percent of the detected pre-ignitions
were connected with glowing particles visible in the compression stroke.
Consequently, the prevention of deposit formation is one key to reduce the
risk of pre-ignitions.

391
 Fig. 11: Glowing air-borne particle during the compression stroke
pre-ignites the charge

5. Summary and Conclusions

The combination of turbocharging, a variable valvetrain with VALVETRO-
NIC + Bi-VANOS and High Precision direct injection is the technology of
choice for current and future BMW Group petrol engines. BMW TwinPow-
er Turbo petrol engines deliver significantly reduced fuel consumption as
well as BMW typical driving dynamics while at the same time complying
with the most stringent emission legislations and world-wide fuel quali-
ties. Additional improvements can be achieved by further raising specific
torque and power, and by increasing the compression ratio. Therefore, it
is necessary to understand current limitations that are given by the oc-
currence of irregular combustion phenomena, e.g. pre-ignitions. Pre-ig-
nition analysis showed that blowby liquid containing engine oil that is fed
into the engine as well as hot deposits/particles are significant sources of
pre-ignitions. Hence, best possible separation efficiency of the crankcase
ventilation system as well as prevention of deposit formation are requi-
red. Pre-ignition analysis is supported greatly by a robust cylinder pres-
sure indication and by the application of “keyhole” optical measurement
techniques e.g. high-speed endoscopy.

392
References
[1] S. Palaveev, H. Kubach, M. Magar, R. Schiessl:
FVV Vorhaben Nr. 1051 Vorentflammungen bei Ottomotoren II;
Zwischenbericht, FVV Herbsttagung, Dortmund, 2012
[2] http://www.aet.ca

393
21 Knock and irregular combustion –
­challenges for the new turbocharged,
highperformance four-cylinder AMG
­engine
Benjamin Kraus, Sebastian Wolf, Jürgen Fischer

Abstract
The brand-new AMG A-Class model “A45 AMG” with its 2.0l high-perfor-
mance engine defines new standards in terms of performance and effi-
ciency. The M133, a transversely mounted four cylinder engine has the
geometrical basis from the known M270 from Mercedes-Benz. Important
components like the high pressure injection system with the central positi-
on of the piezo injector, the cylinder head and the complete auxiliary drive
system have been taken over from the original series engine. To achieve
the desired high performance profound modifications were necessary for
the base engine, i.e. a complete new developed crankcase and the coo-
ling circuit. Also the inlet and outlet systems for the gas exchange had to
be rede-signed entirely.

The “A45 AMG” marks the beginning of a new era of high-performance

engines from AMG. Based on the modular Mercedes construction kit of
the M270/274 family the M133 has an output of 265kW / 360PS from
a displacement of two liters and a fuel efficiency of only 6,9l for 100km
based on the NEDC. The specific power output of 133kW/l sets a new
benchmark for engines built in large-volume production and is in the range
of race engines. The knock behavior has to be known very precisely to
control this high level of turbocharging in series. Additionally any irregular
combustion has to be avoided wherever possible.

In the first part of the article the challenges to control conventional knock
in this engine are explained. Specific influences of compression ratio,
back pressure, combustion chamber temperature and fuel quality with
their influence on the knock limit are analyzed in relation to the applied
knock control.

The second part explains two critical operating states of the highly tur-
bocharged engine with the possible occurrence of pre-ignitions. One ope-

394
rating stage is the steady-state low-end–torque which has been extensi-
vely analyzed. The possible levers like scavenging, back pressure, charge
air cooling, pressure gradient over inlet and outlet and compression ra-
tio are explored in detail and are described within this article. The other
operating stage is the very high dynamic full load operation as it occurs
while accelerating the car in combination with torque intervention due to
gear shifts. Within this dynamic change of engine speed and engine load
irregular combustions can occur under specific circumstances. This phe-
nomenon is explained in detail and reasons and measures are discussed
and described.

Kurzfassung
Für die neue AMG A-Klasse „A45 AMG“ wurde auf der geometrischen
Basis des bereits bekannten, quer eingebauten Vierzylindermotors M270
von Mercedes-Benz ein 2,0l Hochleistungsaggregat entwickelt, das in
seiner Klasse neue Maßstäbe in Bezug auf Performance und Effizienz
setzt. Neben tiefgreifenden Modifikationen am Grundmotor, wie z.B. der
Neukonstruktion des Kurbelgehäuses und des Kühlkreislaufes, war eine
komplette Neuentwicklung der ein- und auslassseitigen Ladungswechsel-
systeme erforderlich. Aus dem Serienkonzept konnten die in zahlreichen
Serienaggregaten bewährten Komponenten der Hochdruckeinspritzanla-
ge mit zentraler Piezo-Injektorlage, der Zylinderkopf und der komplette
Nebenaggregatetrieb übernommen werden.

Mit dem A45 AMG beginnt bei AMG eine neue Ära von Hochleistungsmoto-
ren. Auf Basis des modularen Mercedes-Baukastens der M270/M274-Mo-
torenfamilie wurde mit dem M133 ein Aggregat dargestellt, das aus zwei
Litern Hubraum eine Leistung von 265 KW/360 PS entwickelt und dabei
nur 6,9 Liter je 100 Kilometer im NEFZ benötigt. Die Literleistung von 133
kW/l ist dabei Benchmark für in Großserie gefertigte Motoren und rangiert
auf Rennsport-Niveau. Um diesen hohen spezifischen Aufladegrad in Se-
rie sicher beherrschen zu können, musste das Klopfverhalten des Motors
sehr gut bekannt und irreguläre Verbrennungen weitestgehend vermieden
werden.

Der Beitrag befasst sich im ersten Teil mit den Herausforderungen des
konventionellen Klopfens an diesem Aggregat. Hierzu werden die spezi-
fischen Einflüsse von Verdichtung, Abgasgendruck und Brennraumtem-

395
peratur sowie der Kraftstoffqualität auf die Klopfgrenze analysiert und in
Verbindung mit der applizierten Klopferkennung und -regelung betrachtet.

Im zweiten Teil wird aufgrund des hohen Aufladegrads besonderes Au-

genmerk auf zwei kritische Betriebszustände, welche irreguläre Verbren-
nungen aufweisen können, gerichtet. Zum einen der quasistationäre
Low-End-Torque-Bereich, welcher intensiv untersucht und analysiert wur-
de. Stellhebel wie Scavenging, das Gegendruckniveau, Ladeluftkühlung
sowie die Verdichtung werden dabei im Detail betrachtet und im Beitrag
ausführlich aufgeführt. Zum anderen wird der hochdynamische Volllastbe-
reich, wie er z.B. bei Volllastbeschleunigungen in Verbindung mit Momen-
teneingriffen durch Getriebeschaltungen auftritt, beleuchtet. Innerhalb
dieser Last- und Drehzahldynamik kann es unter bestimmten Randbedin-
gungen ebenfalls zu irregulären Verbrennungen kommen. Dieses Phäno-
men wird detailliert betrachtet und die Ursachen sowie die Maßnahmen
eingehend diskutiert und beschrieben.

1. Introduction
The global increasing restriction on CO2-emissions forces the automoti-
ve OEMs to reduce the fuel consumption of the complete fleet. The fuel
saving measures like direct injection and adjustable valve timing are not
sufficient anymore. Engine downsizing by means of reducing the displa-
cement and implementing a supercharging device becomes more import-
ant. Even the high-performance engines of Mercedes-AMG are affected
by this downsizing trend. The fierce competition in the high-performance
segment leads to fast-increasing requirements concerning e.g. power out-
put and dynamic engine response. These performance demands make
downsizing even more challenging.

Especially for the compact car models additional restrictions come from
the available packaging space. To comply with this complete set of requi-
rements each component must be state-of-the-art in its own performance
field. It is mandatory to optimize all components taking into account their
interdependencies in an integral development process.

One of the main challenges for an effective and efficient downsized gaso-
line engines is the knock limit. A parallel increase of power and reduc-
tion of fuel consumption can be accomplished by well-tuned shifting of
the knock boundaries. An effective shift can only be obtained by simulta-

396
neously designing and calibrating the charge motion, cooling, exhaust gas
properties and the high pressure temperature. Durable operation of the
engine can be assured by understanding the process of pre-ignition and
the implementation of counteracting mechanisms.

2. The compact high-performance engine M133

From the very beginning the goals for the new compact Mercedes-AMG
engine, the M133, were very challenging. Besides compliance with the
Euro 6 directives also the packaging of the engine in the small spaced
NCCG („New Compact Car Generation“) had to be accomplished. The
M133 is based on the M270, which is the Mercedes-Benz compact car
production engine [1, 2, 3]. But to be able to deliver the increase power
output target several important components of the engine had to be adap-
ted. The crankcase has been redesigned and is now completely manufac-
tured out of aluminum. Also the structure itself has been modified to allow
for an increased cylinder pressure of 150 bar.

The original cooling circuit had several components exchanged. The in-
direct cooling circuit for the charge air has been optimized for maximal
efficiency, especially in the lower temperature ranges. The cooling of the
pistons, crankcase, valves and cylinder head has been modified to be
able to shift the knock boundary as much as possible to prevent material
fatigue.

The camshafts for the intake and exhaust valves have been redesigned
for a maximum engine torque of 450 Nm and a maximum power output
of 265 kW (figure 1). The fuel system has been equipped with a four-cam
high pressure pump to be able to supply an injected fuel amount of up
to150 mg/stroke.

The intake manifolds have been designed for a minimum pressure loss.
This included a short intake system length and a dissipative muffler. An
optimal layout of the exhaust gas manifold has been developed together
with a twin-scroll turbocharger. The exhaust system and the catalyst have
optimized large cross sections to minimize the back pressure and provide
an efficient exhaust gas exchange [4].

397
 Figure 1: M133 basic engine data

The M133 is benchmark for 2013 with a power density of 133 kW/liter and
represents a complete novelty amongst series-production engines. This
is impressively demonstrated by a glance at the FEV scatter. The range
must be significantly expanded upwards from 2013 for to these specific
power densities for gasoline engines (figure 2).

398
 Figure 2: FEV scatter – power density for combustion engines (Source:
FEV)

3. The knock behavior

The main factors that influence the knock behavior of an engine are:

• Compression ratio
• Exhaust gas back pressure and the gas exchange
• Combustion chamber temperature
• Fuel quality
• Charge air pressure
• Valve timing and air-fuel mixing
• Charge motion
• Oil droplets in the combustion chamber
In the next chapters more details will be given related to the first four items.

3.1. Compression ratio

The dimensioning of the compression ratio largely affects the knocking
behavior of an engine. High compression ratios lead to the best possible
efficiency for part load operations. At the same time the compression ratio
is limited by the reduced efficiency in full load. However, experience pro-
ved that also lower compression ratios in full load operation are bounded.
This can be explained as a result of the unfavorable gas exchange and the
overall cycle efficiency decrease.

399
In general a reduction of the compression ratio also reduces the end
temperature of the compression stroke, which in turn benefits the knock
boundary. The resulting larger volume leads to a higher internal exhaust
gas recirculation. This rises the cylinder temperature and adversely af-
fects the knock boundary.

The engine speed has an amplifying effect on the above described phe-
nomena. The best engine set-up was therefore adjusted for maximum ef-
fective power output operating condition and the corresponding available
efficiency (figure 3).

Figure 3: Full load efficiencies for ε = 8.6 / 7.8 / 7.0, respectively

Meeting all the desired targets the best full load efficiency was reached for
a compression ratio of 8.6. Lower ratios allowed significantly more spark
advance for engine speeds in the range of 6000 rpm. Nevertheless this
did not result in better combustion efficiency due to the higher remaining
exhaust gas ratio. Primarily in the engine speed range of 2500 – 4500
rpm the lower compression ratios still achieve a high efficiency. For the-
se conditions the knock boundary could be shifted and thereby allowing
earlier fuel conversion. With a compression ratio of 8.6 the M133 engine
has the best possible part load efficiency and simultaneously attains a fuel
consumption of 350 g/kWh or better under full load condition. Therefore
this compression ratio has been implemented in the series production.

400
3.2. Exhaust gas back pressure
Minimizing the internal exhaust gas recirculation is an important measure
to advance the knock margin. For this mechanical redesign of the cams-
hafts as well as a re-adjustment of the valve timing is required.

The exhaust gas back pressure itself also has a large influence on the
knock margin. The exhaust system has been redesigned with a sing-
le-flow exhaust system as well as a tandem-catalyst (5.66“). The complete
exhaust system has achieved a back pressure level of less than 500 mbar.
To obtain the required charging efficiency the twin-scroll turbocharger has
been optimized in combination with the design of the flow separated mani-
fold, both mechanically and in terms of calibration adjustments. The cross
sections have been defined to obtain maximum power output and best
possible response simultaneously.

All the modifications combined decreased the back pressure to an abso-

lute value of 2.8 bar, which is an important contributing factor to the high
specific power output (figure 4).

Figure 4: Exhaust gas back pressure before and after the optimization

Through the reduction of the back pressure also the exhaust gas exchan-
ge and the knock margin have been improved. This improves the overall
engine efficiency. The complete charging optimization process leaded to
an upward cycle: due to the in-creased efficiency, less air and fuel are re-

401
quired to provide the same performance. This as such results in a reduced
back pressure and exhaust gas ratio.

3.3. Combustion chamber temperature

To reduce the susceptibility of knocking the cylinder temperatures before
and during the combustion must be lowered. The Blue Direct combustion
process with its Piezo direct-injection system and a corresponding injecti-
on strategy support this required temperature decrease. An effective wa-
ter-air cooling system lowers the charge air temperature. Even for full load
operation a maximum charge air inlet temperature of only 25 K more than
the environmental temperature can be obtained.

The cooled cylinder surface temperatures are a main contributing factor

to a knock occurrence reduction. Because of this large impact a well-de-
signed layout of the cooling circuit around the cylinder was an important
aspect in the development process. Figure 5 shows the results of the de-
sign process. The redesigned water jacket and improved heat dissipati-
on resulted in an optimized temperature distribution in the cylinder head.
These active cooling measures lead to a lower combustion chamber tem-
perature.

The chosen position of the intake and exhaust valves, the injector and the
spark plug also contribute to a knock beneficial temperature distribution.
The hottest temperature spots within the cylinder occur at the exhaust val-
ves and around the spark plug. Because of the ignition of the air-fuel mix-
ture in this area, knock is not expected to occur in this area. The cylinder
surface in regions far away from the spark location reaches much lower
temperatures. This again favors the knock limitation.

402
 Figure 5: The surface temparature distribution of the cylinder head

3.4. Fuel quality and knock control

Like all AMG high-performance engines also the M133 runs best on super
unleaded gasoline (RON 98). Naturally also fuels with lower RON have
been tested and validated with respect to the knock control. All regular fuel
types down to RON 91 have been assessed.

Two knock sensors are applied for the knock control management. They
have been installed between cylinder 1 and 2 and between cylinder 3
and 4. The sensor signals are filtered by means of an antialiasing filter
after which they are amplified and converted to digital signals. The next
steps of signal processing are performed with specific programmed filters.
These filters are modified and adapted for each engine model separately.
Furthermore the filters can be calibrated for each cylinder individually. The
knock window and its margin can independently be calibrated and stored
for each cylinder as well.

The knock control function can adjust the knock margin for each stationary
operating point by means a feed-forward control that uses the stored engi-
ne map data. As an extension an adaptive learning algorithm is available

403
on the ECU. This function can analyze and process knock occurrences
for any deviation from the original engine speed and load maps. As such it
can create new maps that contain adjusted values for this engine specific
knock margin.

From figure 6 it can be derived that lower RON fuels have the largest effect
on the M133 knock limit in full load operation. The ignition delay causes a
decrease in combustion efficiency, which can only partly be compensated
by increasing the charge air pressure.

A secondary knock control function that can identify low RON fuels has
been implemented in the ECU. Based on this data map different ignition
timings can be activated. Despite the ignition retarding the well-designed
gas exchange and the compression ratio of 8.6 provide for an acceptable
performance for RON91 fuels. A maximum torque of 450 Nm at 3500 rpm
and about 90% of the nominal performance is reached.

Figure 6: Spark retarding in full load for RON 98, 95 und 91

4. Pre-ignition
The extended knock identification and control measures largely reduce
the amount and impact of knock events. If knock does occur it usually
results in no damages on the engine. Pre-ignition events on the other
hand might result in severe damages [5]. By definition a pre-ignition is an

404
irregular, not spark-initiated combustion. Even a single pre-ignition event
usually implies extreme pressure gradients and peaks as well as a rapid
temperature rise. This can result in larger damages of the piston, valves,
spark plug and subsequent damage of other components. The probability
of pre-ignition occurrence increases with higher charging pressures. Also
the thermal layout of an engine has a clear effect on pre-ignitions.

4.1. Steady state pre-ignition

For turbocharged engines the most sensitive operating condition for pre-ig-
nition to occur is the low-end torque region [6, 7]. This is the lower engine
speed range with high engine torque. The low engine speed results in a
small charge motion and an accordingly low turbulence level. The resi-
dence time of the air-fuel mixture is longer for low speed operation. The
high charging pressure causes a high cylinder pressure. In addition, the
exhaust gas tends to stagnate at the turbine entrance, causing an increa-
sed back pressure. This in turn results in a higher residual gas fraction.
All these factors negatively contribute to provide the boundary conditions
required for pre-ignition.

It was possible to significantly reduce the occurrence of pre-ignition in

low-end torque operation by means of a continuous performance oriented
engine design. Also the impact of pre-ignition and the damage levels could
be reduced. A significant contributing factor in full load operation is the
optimized compression ratio. This causes a temperature reduction during
the compression stroke, which results in a lower reaction speed of the
air-fuel mixture as well as a lower sensitivity for the spontaneous ignition
caused by particles or oil residuals.

An optimized design of the high-performance turbocharging concept also

attains a positive effect on the pre-ignition occurrence. Despite the fact
that the one-stage turbocharger design focus has been on the maximum
power range, the maximum torque of 450 Nm can be obtained beginning
at an engine speed of 2250 rpm. The most critical operating range for
pre-ignition, which is below 2000 rpm, is thereby being avoided (figure 7).

405
 Figure 7: M133 Power and Torque

The high-performance turbocharging experiences a low pressure loss

over the entire engine speed range. Also the optimized design of the in-
take and exhaust cross sections contributes to this low pressure drop. Up
to an engine speed of about 5000 rpm the intake pressure is higher than
the exhaust manifold pressure. This positively contributes to an optimal
cylinder charging and low residual gas fraction (figure 8).

Figure 8: Pressure drop (Intake manifold - Exhaust manifold)

406
The splitted gas flow to the twin-scroll turbocharger and exhaust manifold
attain a very low residual gas fraction. This fraction is smaller than 5%
in full load operation. Flushing out the exhaust gas in low-end torque is
called scavenging. By means of scavenging the torque characteristics are
improved and it is possible to reduce the residual gas fraction to even less
than 1% for the critical low speed operations.

The efficient charge air cooling benefits the compression stroke tempera-
tures and also contributes to appropriate turbocharger response charac-
teristics. The Piezo-injector and the multiple injection strategy provide a
good mixture preparation and avoid fuel wall wetting and oil dilution.

These arrangements contribute to a reduction of pre-ignition occurrences

and thereby assure durable operation. The engine has been intensively
tested for durability on test benches, public roads and test tracks. The
engine successfully sustained and passed all applied tests.

4.2. Non-steady state pre-ignition

Not only does pre-ignition occur in (quasi-) stationary conditions. Even
more critical situations arise in and can even be caused by high dynamic
and transient non-steady state operations. The requests for torque reduc-
tion coming from the gear box ECU or ABS/ESP ECU cause a high ther-
mal loading due to the high dynamic behavior of the engine. During such
a procedure the sensors and signals processors that support the ECU for
engine monitoring also approach their capacity limits. In worst case this
results to incorrect measurements and a false amount of fuel supply.

The M133 is equipped with a cylinder cut-off system for high load condi-
tions. If the boundary conditions for such an operation apply the fuel supply
for e.g. two of the four cylinders is cut off and these cylinders accordingly
do not deliver any power output. An increase of irregular pre-ignitions has
been observed during operation conditions of a requested torque reduc-
tion in combination with a fuel cut-off. The gear changing process cause
high engine speed gradients (up to 15.000 rpm/s) in addition to the low tor-
que request. This dynamic procedure also requires a very accurate data
acquisition quality in order for the ECU to calculate the correct injection
and ignition timing.

407
4.2.1. Full load gear shifting sequences
The perceived vehicle dynamics can be improved by shorter gear shifting
response times. This is especially noticeable for acceleration upshifting
with cylinder cut-off or for deceleration downshifting with throttle activation.

For upshifting during acceleration the required torque output can be rea-
ched by means of ignition retarding or injection deactivation. The availa-
ble energy through the air mass flow remains constant during the shifting
event. This also enables slow responsive turbocharged engines a faster
annexation to the demanded (dynamic) torque output.

The M133 showed pre-ignitions during shifting events in combination with

both ignition retarding and injection deactivation. It is expected that an
extreme ignition retard results in misfire and consequently in pre-ignition.
Nevertheless this expected cause could not be confirmed. During the ex-
ternal triggered shifting events without injection deactivation no irregular
combustion events have been detected.

The highly dynamic gear shifting sequences could only be analyzed du-
ring in-vehicle measurements. For this purpose special pressure induced
spark plugs and a mobile cylinder pressure indication system have been
installed in the car. Combined with the signals from the ECU a better mo-
nitoring and therefore understanding of the combustion processes could
be obtained. An example of the pressure curve during a shifting event with
pre-ignition can be found in figure 9.

408
 Figure 9: Pressure curve for gear shifting events

Striking is the pressure gradient in the gas exchange stroke, see also figu-
re 10. To gain a better understanding of these phenomena additional sen-
sors have been installed in the vehicle. A low pressure indication system
has been integrated in the intake manifold. Also clear pressure gradients
could be observed in the intake system. During the gas exchange stroke
intermediate compression levels up to over 12 bar have been measured
in the cylinder. These pressure gradients only appear after a delayed com-
bustion event. A delayed combustion causes an according delay in the
heat release process. A complete heat release is not completed before the
gas exchange stroke starts. This may lead to very early pre-ignition in the
succeeding induction stroke.

409
 Figure 10: Detailed view of shifting event

4.2.2. Prevention of pre-ignition

The M133 ECU software contains a pre-ignition detection function. This
function analyses the knock detection sensor signals in a time frame well
before actual knock events might occur. Pre-ignition events can be detec-
ted by their strong high-frequent sound level. Especially for steady-state
operations this detection functions enables initiation of counteractive mea-
sures. The appropriate counteractions depend on the measured signal
and the derived characteristic of the expected event. The measures can
in many cases prevent engine damage. Possible counter measures are a
temporary load reduction or a cylinder cut-off.

In non-steady and transient operation these reactive measures do not suf-

fice. The succeeding combustions are quasi-steady-state and outside the
critical low-end torque range. Usually the subsequent combustion events
are not inherently irregular.

This knowledge was implemented in a ‘pre-ignition occurrence probability

analysis method’ especially developed for the M133. The method incor-
porates two approaches. One approach increases the flame propagation
velocity. The other approach adapts the ignition sequence.

410
Research has shown that the pre-ignition probability during cylinder cut-
off decreases if the fired cylinders contain a rich air-fuel mixture. This air-
fuel ratio increases the flame propagation velocity. The fastest combustion
velocity is reached at lambda of 0.90. The pre-ignition occurrence is re-
duced by 80%.

The second approach, the adjusted ignition sequence, reduces the con-
sequences of a delayed combustion event. If a delayed combustion is
detected, the ECU changes to a special cylinder cut-off strategy. For this
strategy the cylinders that are cut-off use a special pattern. This approach
ensures that the cylinder that contains the delayed combustion event is
not followed by a fired combustion.

Both approaches combined accomplished a significant reduction of pre-ig-

nition in transient conditions. Therefore the engine can operate well within
its loading design limits regardless of the operating conditions.

5. Conclusion
With the new A45 AMG model Mercedes-AMG started a new era in the
compact high-performance engine segment. The new AMG engine, the
M133, is based on the Mercedes-Benz modular engine series M270/
M274. Despite its merely two-liter displacement the engine accomplishes
a power output of 265 kW / 360 bhp at a fuel consumption of only 6.9
l/100km (NEDC). With the specific volumetric power output of 133kW/l the
engine performs in a range of race engines and is the new bench-mark
for the large-volume production engines. The new M133 shows evidently
that knock and pre-ignition can be successfully controlled even for highly
turbocharged engines.

The most influencing factor that can reduce the knock margin is the con-
sequent and simultaneous optimization of all relevant engine compo-
nents. The combination of a proper selected compression ratio, an effi-
cient charge cooling, a lower back pressure and a reduced combustion
chamber temperature contribute to a knock resistant combustion process.
A deep understanding of the interaction of the residual gas fraction, cylin-
der pressure and temperature, injection timing, oil consumption and air
turbulence is required to optimally control the knock margin. The measu-
res that positively affect the knock margin also benefit the prevention of
stationary pre-ignition. The understanding of non-stead state pre-ignition

411
phenomena, especially in the higher engine load and speed range, has
been gained through the thorough analysis of single combustions cycles.
Based on the results of this analysis appropriate countermeasures in the
area of injection and ignition strategies could be implemented.

Through the consistent design focus on its very high power density the
M133 shows that it is possible to achieve a high-performing, a durab-
le and a very responsive engine simultaneous. This was achieved by a
precise calibration and optimization under full load condition. To meet all
requirements a detailed analysis of very specific operating conditions was
necessary. But all for the best: The new AMG M133 has become the revo-
lutionary benchmark engine in the sportive compact car segment.

412
Literature/References
[1] Hart M., Gindele J., Ramsteiner T., Thater G., Tschamon B., Karres
M., Keiner B., Fischer J.; Der neue Hochleistungsvierzylindermotor
mit Turboaufladung von AMG; 34. Wiener Motorensymposium; Wien;
2013
[2] Merdes, N.; Enderle, C.; Vent, G.; Kreitmann, F.; Weller, R.; The new
turbocharged 4-cylinder inline gasoline engine by Mercedes-Benz;
20. Aachener Kolloquium Motoren- und Fahrzeugtechnik, Aachen
2011
[3] Mürwald, M.; Kemmler, R.; Waltner, A.; Kreitmann, F.: Die neuen
Vierzylinder-Ottomotoren von Mercedes-Benz, Motortechnische Zeit-
schrift, Ausgabe 11/2013
[4] Gindele J.; Ramsteiner, T.; Fischer, J.; Tschamon, B.: Der neue
2,0-l-Hochleistungs-Vierzylindermotor von Mercedes-AMG, Motor-
technische Zeitschrift, Ausgabe 09/2013
[5] Zaccardi, J.; Lecompte, M.; Duval, L; Pagot, A.: Vorentflammung an
hoch aufgeladenen Ottomotoren, Motortechnische Zeitschrift, Ausga-
be 12/2009
[6] Willand, J.; Daniel, M.; Montefranceso, E.; Geringer, B.; Hofmann,
P.; Kieberger, M.: Grenzen des Downsizing bei Ottomotoren durch
Vorentflammungen , Motortechnische Zeitschrift, Ausgabe 05/2009
[7] Zahdeh, A., Rothenberger, P., Nguyen, W., Anbarasu, M. et al., „Fun-
damental Approach to Investigate Pre-Ignition in Boosted SI Engi-
nes,“ SAE Int. J. En-gines, SAE 2011-01-0340, 2011

413
22 Simulations and experimental investiga-
tions of intermittent pre-ignition series
in a turbocharged DISI engine
Stefan Palaveev, Max Magar, Christian Disch,
Robert Schießl, Heiko Kubach, Ulrich Spicher,
Ulrich Maas, Thomas Koch

Abstract
In this work, a particular phenomenon associated with pre-ignition in SI en-
gines is studied, namely its occurrence as alternating sequence of pre-ig-
niting (PI) and regularly burning engine cycles. While several researchers
report on this phenomenon, a conclusive explanation is still missing. A
major reason for the lack of knowledge is the sporadic and stochastic
occurrence of pre-ignition events which make systematic studies difficult.

This paper introduces a method for more systematically studying the

pre-ignition series. In the experiment, the spark timing of one single engine
cycle is advanced well before the knock limit. This initiation cycle features
a pressure trace that is very similar to a “real” PI-cycle. It is also often fol-
lowed by an alternating sequence of PI and regular combustion cycles (all
at regular spark timing). Experiments are conducted in a series-production
engine with optically accessible combustion chamber, using high-speed
crank-angle-resolved imaging and analysis of in-cylinder luminescence
phenomena to study the pre-ignition series phenomenon. The artificially
induced PI events are also compared to natural PI events.

The observations indicate that hot particles in the combustion chamber

play an essential role: In all natural PI cycles, and in all PI cycles from an
artificially triggered series (except for the first cycle), a few particle-like
luminescent spots could be observed already early in the compression
stroke. In contrast, for regular cycles, such spots were never detected.
In the expansion stroke of cycles following the first (natural or artificial)
PI-cycle, numerous luminescent spots were visible in the compression
stroke; the number of those spots always decreased within subsequent
cycles a PI-series.

414
The observations are consistent with the release of particles of combustib-
le material in the combustion chamber that are gradually heated by con-
tact to exhaust gas, until they self-ignite. If these particles remain inside
the engine for a sufficiently long period, they can keep burning throughout
multiple cycles, causing the gas phase to ignite in later cycles. Exemplary
modeling studies of carbon particles using detailed models for chemistry
and molecular transport corroborate this scenario.

1. Introduction
Reducing fuel consumption as well as tailpipe emissions has become a
major challenge in SI engine development in recent years. Actual effi-
ciency requirements are promoting new engine concepts. In particular,
operation under low load conditions bears a high potential for optimiz-
ation. In addition to stratified lean combustion, the downsizing strategy
represents a promising approach. Reducing the displacement provides a
decrease in throttling losses during the gas exchange cycle. Furthermore
low power operation is shifted towards higher loads and lower effective
fuel consumption. Both measures benefit overall combustion efficiency.
Supercharging is applied to meet high output requirements [9].

However further downsizing increases the likelihood of abnormal combus-

tion phenomena under full load conditions due to critical thermodynamic
states in the compressed gas. Knocking combustion is most frequently
observed. “Regular” knocking combustion can be avoided by implemen-
ting an antiknock control although in general at the expense of efficien-
cy. In addition, events of extreme knocking combustion and of low speed
pre-ignition (PI) highly threaten operation reliability. The occurrence of
these combustion anomalies is completely incidental and uncontrollable.
In comparison to “regular” engine knock extremely high peak pressures
are observed provoking severe engine damage. Modern antiknock con-
trols cannot capture the dynamics of these phenomena and thus fail to
prevent them.

Premature auto-ignition has been observed predominantly in highly su-

percharged spark ignition engines at low engine speeds and high load
conditions. Figure 1 displays a comparison of engine cycles featuring a
regular combustion and a premature auto-ignition with subsequent heavy
knocking combustion.

415
Figure 1: Comparison of an engine cycle with regular combustion and a
cycle with pre-ignition

Previous studies revealed the sporadic and stochastic nature of low speed
pre-ignition and report a frequent occurrence of series of several subse-
quent pre-ignition events. A characteristic feature of these series is the
intermittence between two consecutive events. In general a pre-ignition is
followed by one or even several regular combustion cycles before the next
pre-ignition is triggered. Figure 2 depicts a characteristic series of several
pre-ignitions. The number of events per series as well as the length of the
intermittence is subject to variations.

Figure 2: Exemplary series of multiple pre-ignition events

416
Figure 3 presents an overview of the main properties of these pre-ignition
series. The presented data was gathered in previous investigations at KIT.

Figure 3: Properties of pre-ignition series. Left figure: Number of events

per series. Right figure: Length of the intermittence between consecutive
events

The sudden and unpredictable incidence of pre-ignition as well as the long

gaps of regular combustion cycles between consecutive series supports
the hypothesis of a reservoir that is build up during regular operation and
depleted again during the series.

Many investigations focusing on the identification of the triggering mecha-

nisms of low speed pre-ignition have been performed recently. Willand et
al. [23] concluded from experiments and numerical simulations that hot
spots in the bulk gas, deposits on the combustion chamber walls as well
as residual gas represent promising candidates for pre-ignition sources.
The experiments conducted by Zadeh et al. [24] revealed an interrelation
of the pre-ignition frequency and spray impingement. Günther et al. [11].
deduced an approach to estimate the probability of pre-ignition based on
the enthalpy of the mixture at the end of the compression. Haenel et al.
[12] suggested an experimental procedure providing a means of genera-
ting high pre-ignition rates systematically. Takeuchi et al [22] investigated
the effect of lubricant oil composition on the pre-ignition rates. The results
indicate that both the type of base oil as well as the kind of additives exhi-
bits a significant effect. Moreover Amann et al. [1] successfully reduced
the pre-ignition rates and intensity applying exhaust gas recirculation in
an experiment. Dahnz et al. [5] and Palaveev et al. [17] have identified

417
numerous potential triggers of low speed pre-ignition using a systematical
approach.

The combination of experimental investigations and numerical simulations

provided an assessment of the effectiveness of the considered mecha-
nisms. The authors concluded that multiphase processes, in particular
droplets originating from liquid films and hot particles represent probable
sources of pre-ignition. The studies of Bradley et al. [4] and Peters et al.
[18] concentrate on the extreme knock observed in conjunction with low
speed pre-ignition. The majority of the studies report the existence of a
correlation between the pre-ignition rates, the wall wetting and the pro-
perties of the lubricant oil [2], [5], [17], [22]. Furthermore Amann et al. [2]
achieved a mitigation of the pre-ignition by adapting the geometry of the
piston top land gap. These results strengthen the aforementioned reser-
voir theory.

The accumulation of lubricant oil and liquid fuel in the top land gap forms
a liquid film representing a potential reservoir. Such a reservoir can have
different effects. Inertial forces caused by the non-uniform piston motion
or reverse blow-by may extract droplets from the film that could trigger
auto-ignition. Numerical simulations of the auto-ignition around droplets
composed of highly ignitable n-alkanes (e.g. n-heptane, n-decane and
n-hexadecane) under engine relevant conditions yield that such droplets
are rather ineffective as pre-ignition promoter [17].

In addition to auto-ignition around droplets released from the liquid film,

the emerging vapor may ignite due to the high temperatures in the ex-
pansion stroke similar to pool fires. The rich mixture in conjunction with
quenching in the top land gap enhances soot formation. The prevailing
conditions in the expansion, the rich mixture and the low oxygen content
in the exhaust gas, favor the formation of large soot particles [20].

Moreover deposits can form as a consequence of incomplete combus-

tion and pyrolysis of the liquid film in the top land gap. These deposits
can store unburned and partially oxidized hydrocarbons [19]. A release of
these substances may alter the ignitability of the mixture locally promoting
auto-ignition. Besides particles may be a result of the flaking of deposits.

The contact with the flame front and the hot exhaust gases in conjunction
with exothermal reactions on the surface lead to high particle tempera-
tures. In addition the surface reactions compensate part of the heat loss

418
of the particles during the charge cycle and thus prevent a quick cooling
of the particles. If particles remain in the cylinder after the scavenging is
completed they could cause auto-ignition in the subsequent compression
stroke.

Although the majority of the researchers report the observation of pre-ig-

nition series, to our knowledge a systematic analysis of this behavior has
not yet been published. In this article an experimental approach based
on the systematic generation of sequences of pre-ignition is presented.
This procedure provides a high number of pre-ignition events in a shor-
ter time compared to regular experiments. In addition optical diagnostics
were applied to study the processes during a sequence. A major goal of
the study is to understand the conditions and the processes that provoke
the occurrence of series.

The recorded engine cycles indicate that solid particles are involved in the
triggering of pre-ignition events within a series. In order to obtain a deeper
insight in the processes governing particle and gas phase interactions
numerical simulations are performed.

2. Methodology
The presented work is based on optical investigations of the combustion
process in a modified serial production engine as well as numerical simu-
lations of the auto-ignition around solid particles with detailed physical and
chemical models. The following chapters describe the experimental setup
and the numerical model.

2.1. Engine setup

Super-knock events cause enormous loads on the first piston ring groo-
ve, so investigations of their occurrence require a test engine with serial
pistons strengthened by a castin steel ring carrier. The experiments pre-
sented in this paper as well as in previous investigation at KIT [5], [6], [7],
[17] were performed on a modified four-cylinder turbocharged DISI engine
from the EA113 series of Audi.

In order to increase the maximum low-end torque, the serial exhaust

camshaft was replaced with a customized one. The modified cam profile
is also used in the EA888 engine generation [10]. A water-cooled pressure

419
transducer, type 6061B by Kistler, was installed in each of the four cylin-
ders. The camshaft drive was modified to free space for an optical access
to the combustion chamber of the first cylinder. The access is designed
as an endoscopic sleeve with integrated sapphire window for endoscopes
with a diameter up to 8 mm. The endoscope tip is cooled by pressurized
air, which is introduced trough channels in the sleeve’s wall. The sleeve is
screwed into the access bore and can be removed for window cleaning.

Basic engine data is listed in Table 1. The application of an “open” engi-

ne control unit allows the adjustment of all engine calibration parameters
and maps through an interface. In order to achieve fixed spark-timing the
knock control system was deactivated and the ignition maps were mani-
pulated.

Table 1: Engine specifications

Manufacturer Audi
Engine Type EA113; 2.0 l TFSI
Displaced volume 1983 cm³
Stroke 92.8 mm
Bore 82.5 mm
Connecting Rod 143 mm
Compression ratio 10.5:1
Fuel injection system 6-hole-Injectors, inj. fuel pressure= 15 MPa

The investigations are focused on initiation of PI-series by a single su-

per-knock cycle. This could be achieved by advancing the spark-timing to
a crank angle well before the knock limit. For this purpose an electronic
single cycle spark-timing switch was developed and installed at the optical
cylinder. The principle of operation of the dedicated ignition system is illus-
trated in Figure 4. A single cylinder engine control unit (ECU) by IFKM was
operated in addition to the Bosch Motronic 9.1 ECU. An angle encoder
mounted to the crankshaft provides degree marks and trigger signal to
the IFKM-ECU. The ignition signal of the Bosch-ECU (terminal nr. 1) is
disconnected from the ignition coil and connected as input to a switch as
well as the ignition signal from the IFKM-ECU. The switch output is then
connected to the ignition coil. With this arrangement a spark-timing well
before the knock limit can be adjusted in the IFKM-ECU. By pushing a

420
button the switch electronics activates the IFKM-ECU-input for one cycle,
what causes a single cycle with super-knock. A resistor at the Motronic
output prevents error inputs to the onboard diagnostic system.

Figure 4: Principle of operation of the single cycle spark-timing switch

Cylinder pressure data was recorded against time with a sampling rate of
500 kHz. Due to the high temporal resolution of the signals, determination
of peak pressures and pressure gradients is more precise compared to
angle resolved data acquisition. Frequency domain signal processing is
also possible. Furthermore, both tracks of the angular encoder, CDM and
Trigger, are connected to the data recorder that allows thermodynamic
analysis of the combustion process.

Pre-ignitions are detected by event triggering on raw cylinder pressure

data with a threshold value of 120 bar, which is possible when using a
pressure amplifier with drift compensation. Data of all channels was re-
corded for 30 consecutive combustion cycles including 10 cycles before
PI-event. Since the peak pressure during regular combustion cycles does
not exceed 100 bars, only cycles with heavy knocking combustion can
trigger the system. Pushing the switch button also triggers the acquisitions
system with the same cycle configuration.

421
Besides the high-frequency data acquisition, the test engine was equip-
ped with various low-frequency temperature and pressure sensors to mo-
nitor operating parameters. Data from these sensors was logged with 1
Hz. Information about the timestamp of each experiment was saved in an
additional channel.

2.2. Optical setup

Previous optical studies of PI occurrence were focused on the spatial dis-
tribution of PI-onsets and the early phase of flame development. Thus, an
UV-sensitive high-speed camera system based on photomultipliers and
optical fibers was used to record the chemiluminescence from radicals in
the flame front, such OH* at 306.4 nm, CH at 431.5 nm and C2 at 516.5
nm [5], [6], [7], [8], [13], [14], [17]. A spatial reconstruction of the PI-onsets
was achieved by the complementary use of the camera and a fiber-op-
tic-spark plug (FOSP). Both camera system and the FOSP-photomultiplier
arrangement have an excellent sensitivity in the UV-range and reaction
time below 1 ns, which allows a maximum acquisition frequencies of up
to 200 kHz. However, the sensitivity in the visible range is limited to <
650 nm. In our previous work [5], [6], [7], a systematic categorization of
potential PI sources was carried out, resulting in a tree structure including
particles and deposits. However, an optical detection of incandescence
from hot particles in the combustion chamber with the photomultiplier-ba-
sed-systems is nearly impossible because of the radiation spectrum (λ >
600 nm). Thus, the visualization of hot particles requires a different optical
setup which can detect light sources in the entire visible light spectrum
with a sufficient temporal and spatial resolution.

The optical arrangement used for the presented investigations consists of

an engine-mounted CMOS high-speed camera system (LaVision, HSS6)
and a rigid boroscope (Karl Storz, diameter = 6.5 mm). In order to ensu-
re precise correlation between frame timing and engine crank angle, the
camera-setup is extended by a high-speed controller (HSCv2), which is
connected to the crank angle encoder. The socalled Hypersampling sys-
tem used with DaVis8 (software) is a time-based high-speed setup able
to deal with a fluctuating engine speed and offering a batch processing
function to reconstruct images for a pre-defined crank angle resolution.
Considering the intensity of soot incandescence from test measurements,
the temporary resolution of the visualization is configured to 12.5 kHz,
i.e. an angular resolution of 1 CA at 1750 rpm when using the described

422
batch processing function. The visualization is performed in a CA-range
between 180 CA before top dead center (bTDC) and 180 CA after top
dead center (aTDC).

2.3. Modeling
Engine combustion is governed by the coactions of physical and chemical
processes and their interactions as well as by a turbulent flow field. The
complexity of these processes requires a detailed investigation of the en-
gaged sub-processes in order to understand the overall process. Deep
insights into the different sub-processes can be provided by means of
numerical simulations.

A detailed combustion model is applied to investigate multiphase proces-

ses in the combustion chamber in particular the impact of hot, solid par-
ticles. The model is based on the assumption of spherical particles loca-
ted in a quiescent fuel/air mixture enabling the use of a one-dimensional
approach. The reduction of the problem to one-dimensional geometries
allows the use of very detailed models for the physical and chemical pro-
cesses [15], [20]. In the model particles are considered to be composed
of pure carbon featuring exothermal chemical reactions on the surface
leading to a burn-off. The surface reactions are modeled using the mecha-
nism of Bradley et al. [3]. Particles are either formed in fuel rich parts of a
stratified mixture or due to burn-up of liquid films in the power stroke [21].
The exposition to the flame front and the hot exhaust gases in conjunction
with the exothermal surface reactions provoke very high particle tempe-
ratures. Therefore, it is necessary to account for radiative heat loss in the
combustion model. The unsteady thermodynamic conditions prevailing in
engines are modeled including a transient pressure profile. The pressure
profile is determined by the geometrical properties and the operation point
of the engine.

The numerical studies presented in this article were performed to determi-

ne particle properties (size and temperature) enabling high temperatures
at the end of the charge cycle. It is assumed that particles formed during
the combustion in the power stroke feature temperatures in the range of
1500 K < T0 < 2000 K and sizes ranging from 50 μm < r0 < 150 μm.

These particles are exposed to conditions similar to those prevailing du-

ring the expansion, the exhaust and the intake stroke. Expansion starts at

423
a pressure of 100 bar and a gas phase temperature equal to the particle
temperature. The exhaust stroke is modeled assuming constant pressure.
Initial conditions of pressure and temperature correspond to the values
obtained at the end of the power stroke. During the intake stroke pressure
is assumed to be constant as well. Initial conditions correspond to the
conditions of the air (TAir = 300 K, pAir = 2 bar). Radiative heat loss is
computed based on a wall temperature of TWall = 450 K.

3. Results and Discussion

3.1. Initiation of PI-Series

The approach of determination the mechanism leading to PI-Series in this
experiments can be compared on an abstract level with the determination
of a transfer function in the control theory. An artificially generated PI can
be used as an input signal to the system (engine) in order to measure the
output signal (PI-series / no PI-series) and analyze the transfer function
(observations of the processes in the com-bustion chamber). Therefore
the first step consists of evaluating a characteristic heat release of natural
preignitions in order to recreate it.

The optical and thermodynamically investigations within the project

“Pre-Ignition in SI-engines” funded by the German FVV-association sho-
wed that pre-ignitions typi-cally start from a local kernel with a consequent
flame front propagation. The mean PI-timing detected in the performed
measurement at a break mean effective pressure (BMEP) of 22.6 bar (360
Nm) and engine speed of 1750 rpm was 12 CA bTDC, i.e. 18 CA before
knock limit. The advanced begin of the heat release consequently causes
a rise of pressure and temperature in the end-gas, resulting in detona-
tions and propagation of high amplitude, high frequency pressure waves
trough the com-bustion chamber. This phenomenon is also referred to as
“super-knock” [4].

The described phenomenon can be basically compared with a combus-

tion process caused by spark-ignition well before the knock-limit. There
are two differences to be taken into account: fixed starting position of the
flame propagation (spark gap) and the ignition delay. Experiments with
advanced spark-timing reproduce therefor a particular case of PIs at the
spark plug, what should be the case with lowest knock intensity (lowest

424
input signal). Thus, the experimental results (system output) can be trans-
ferred to all PI-onsets.

A second step consists in the implementation of an ignition system, which

is able to change spark-timing for a single cycle. The hardware setup of
the system used in our investigations has been already described in sec-
tion 2.1.

A first experiment was performed at n = 2000 min-1 and BMEP = 18.85

bar, i.e. a torque of 300 Nm. Spark timing was adjusted at the knock limit
at 3 CA aTDC. Within 2 min. we undertook 5 consecutive attempts to
initiate a PI-series by advancing the spark-timing by 9 CA. This caused
knocking combustion with peak pressures up to 120 bars, but no PI-se-
quences. Compared to previous investigations [6], [7], [17] the achieved
knock intensity was below values of „natural“ super-knock, what could be
a reason for non-active PI-sequence-mechanism.

In a next attempt the „jump“ in spark-timing was extended by another 5 CA

to 11°CA bTDC. Within the next 5 attempts 3 PI-sequences were success-
fully initiated. The pressure trace of one of these series is shown in Figure
5 and will be discussed in the optic section.

Figure 5: PI-sequence initiated by a single „super-knocking“ cycle

The experimental results make evident, that the entire PI-sequences was
induced by the first, artificially initiated PI. The result can be explained if

425
we assume a reservoir of a PI-promoter which has been activated by the
first super-knocking cycle and causing following PI-events. The self-sus-
taining mechanism of PI-initiation lasts until the reservoir content is gone.
It is remarkable that the typical intermittency of cycles with regular com-
bustion and PI was also observed, which indicates that the conclusions of
experiments with initiated PI-series can be transferred to regular PI-series.

The influence of the engine load, which also represents the thermodyna-
mic state in the combustion chamber, on the rate of initiated PI-sequences
was investigated in a next experiment. In a variation of three BMEP-set-
tings at n = 1750 min-1 the spark advance to the knock limit was changed
in three steps as illustrated in Table 2. In order to reduce the probability
of potential engine damages the number of attempts to initiate PI-series
was reduced to only forty, eleven of which succeeded. As expected, both
larger advance of the initiating spark timing and higher engine load increa-
sed the knock intensity of the initiated PI-cycle. However, with increasing
number of experiments we observed a weak trend of decreasing number
of PI-sequences, despite of the higher engine loads. Thus, the occurrence
of PI-sequences is less controlled by the thermodynamic state and chemi-
cal kinetics, but much more by the probability of existence of PI-promoter
in the compression stroke. With growing number of super-knocking cycles
the reservoir of PI-promoter is gradually reduced which explains the zero
number of initiated PI-series at highest load and spark advance.

426
 Table 2: Variation engine load and spark advance of the initiating PI-
ignition

Load Md = 300 Nm Md = 315 Nm Md = 330 Nm

(BMEP = 18.9 bar) (BMEP = 19.8 bar) (BMEP = 20.7 bar)
spark

PI-sequences = 1 PI-sequences = 0 PI-sequences = 1

Sequence-length = 2 PI Sequence-length = 2 PI
10 CA pmax = 140 bar pmax = 148 bar pmax = 205 bar
= 254 = 603 = 729
bar/°CA bar/°CA bar/°CA

PI-sequences = 2 PI-sequences = 1 PI-sequences = 1

1. Sequence-length = 4 PI Sequence-length = 4 PI Sequence-length = 4 PI
15 CA 2. Sequence-length = 4 PI
pmax = 215 bar pmax = 214 bar pmax = 247 bar
= 985 = 762 = 1020
bar/°CA bar/°CA bar/°CA

PI-sequences = 2 PI-sequences = 3 PI-sequences = 0

1. Sequence-length = 2 PI 1. Sequence-length = 2 PI
2. Sequence-length = 2 PI 2. Sequence-length = 2 PI
20 CA 3. Sequence-length = 2 PI
pmax = 197 bar pmax = 250 bar pmax = 251 bar
= 755 = 1130 = 1680
bar/°CA bar/°CA bar/°CA

The described trend of decreasing number of PI-sequences with growing

number of attempts to induce them became much more significant in the
following optical investigations, as shown in Figure 6. The black bars mark
the initiated PI-series. The height of the bars provides information of the
PI-number in a sequence including the introducing PI. The balloons inform
about initial operating conditions and changes. Altogether 155 attempts to
provoke a PI-sequences were made, from which only 9 succeeded. The
most effective measure to provoke PI-sequences turned to be the retar-
ded injection timing, which causes more impingement of the cylinder liner
with fuel.

427
 Figure 6: Experimental timeline

The processes in the combustion chamber within the initiated PI-series

in Figure 6 were recorded by the high-speed-camera. The results of the
visualization are presented and discussed in the following chapter.

3.2. Optical studies of PI-sequences

3.2.1. Image processing

In order to facilitate the analysis of the recorded images from the com-
bustion chamber, a raw data processing is required. To ensure that the
camera noise isn’t taken into account an intensity threshold of 20 counts
was set to every pixel. All images are visualized in a false color illustration
where the pixels with the highest in-tensities are shown in dark colors. The
batch processing was extended with a further batch function where every
pixel was processed with a non-linear filter. The filter compares the neigh-
boring pixels of every individual pixel and subtracts the intensity average
of them. This means that the principle of operation corresponds to a high-
pass filter. As a result, regions with steep rise of their intensity, such as
particles and burning droplets, can be better distinguished.

Figure 7 shows the result of the described approach of background sub-

traction. A raw image from the combustion process at 60 CA aTDC is
compared to its filtered version on the right side. Some particles in the left
image can be easily separated from the background as the one marked
by , but there are also regions, where particles are less distinguished

428
 , because of the high intensity of combustion radiation. In contrast, the
filtered image on the right side shows a good segmentation of all particles.

Figure 7: Image filtering: raw image (left) and corresponding processed

image (right)

3.2.2. Results
The filtered images are analyzed and discussed in this section. Figure 8
shows an cycle-to-cycle frame sequence at 35 CA aTDC, which was taken
during the PI-sequence in Figure 5. The images from combustion cycles
with PI are marked by a bold border line. The transient sequence includes
one artificially initiated PI-cycle (2) followed by regular combustion cycles
and two “regular” PI-events (4 and 7) in an interchanging order.

429
Figure 8: Processed images from consequent combustion cycles in a PI-
sequence. Power stroke at 35 CA

The first frame illustrates the regular combustion cycle just before the in-
itiated PI (2) and a typical scene from the cylinder without knocking com-
bustion. Most of the image area appears in a white color – an evidence of
continuous intensity distribution and pre-mixed combustion. Only a small
number of hot particles / burning fuel drop-lets can be observed at the
right (intake) side, close to the injector tip. They probably result from some

430
fuel dropping out of the nozzle orifices. However, these droplets burn out
before the end of the power stroke and the soot radiation disappears.

The hatched areas in the middle of (2) and (7) represent an overloaded
chip area caused by the knocking combustion and the high level of light
emission. Unfortunately a detailed analysis of the in-cylinder process in
these cycles is not possible.

The regular combustion cycle (3) after the introduced PI, introduces re-
markable characteristics. A large number of hot particles fills the entire
field of view of the camera. In the following cycle (4) the combustion starts
with a “natural” PI, which is discussed in Figure 9. The PI-cycle (4) itself
is also characterized by a high density of glowing particles. In the follo-
wing cycles (5)+(6) the observed particle density continuously decreases.
Combustion in the ensuing cycle (7) starts again with a “normal” PI, which
is discussed in Figure 10, followed by cycle (8) with an increased num-
ber of particles. In the following combustion cycles no more PIs occurred
and the particles nearly disappeared. The characteristics of the recorded
images in the regular combustion cycles (9)-(12) are similar to the scene
in cycle (1) before the PI-sequence.

Obviously each PI-event in the showed cycle-to-cycle sequence is follo-

wed by a regular combustion cycle with a high number of particles. In the
consequent regular combustion cycles particle density quickly decreases
and reaches the observed initial level before the PI-sequence. Compari-
son of cycles (3), (5) and (8) also shows that number of released particles
after each PI-cycle decreases as well.

3.2.3. Discussion
Discussion of the PI in cycle (4)
The cycle-to-cycle sequence in 3.2.1 shows the consequences of PI in
the power stroke, but does not explain how PI occurs. The high-speed-
imaging during the com-pression stroke of cycle (4) reveals a first glowing
particle already at 152 CA bTDC. In order to depict its movement, the
frames in the relevant crank angle section are summed up, showing in a
multilayer illustration of all scenes. The left image in Figure 9 shows the
result in a range between 152 CA and 18 CA bTDC. Altogether five par-
ticles can be distinguished by their traces.

431
Particle appears first at 152 CA moving bottom-up and disappears
from the view field at 139 CA, probably in the deepening area of the roof
at the injector tip. Shortly after that at 120 CA, close to the place where
the trace of was lost, a new trace , probably from the same particle
starts. The trace goes from right to left following the shape of the roof (see
arrows). A similar particle movement was also observed in cycle (7) (see
Figure 10, left) indicating that trace and belong to the same par-
ticle. The particle finally disappears at the left border of the viewing field
at 43 CA bTDC. Another particle was observed in the range between
41 CA and 31 CA bTDC. It enters the view field at the right side and mo-
ves to the left. The big size of this particle and the relatively high velocity
both indicate that it was very close to the optical access. Another particle
is traced by . It is characterized by a weak incan-descence during its
movement between 57 CA and 19 CA bTDC. All traces 1 to 4 were lost,
so there is no conclusive proof of the involvement of these particles in the
mechanism of preignition. In contrast, the images in the middle and on the
right show that particle initiates in the center of the combustion cham-
ber one of many flame kernels. This particle appears at 38 CA and moves
from left to right. At the same time, at 13 CA many further ignition kernels
entered the view field of the camera causing fast flame development.

Figure 9: Pre-ignition mechanism in cycle 4.

left: multilayer image of the frames between 152 and 18 CA bTDC.
middle and right: flame initiation caused by particles Summarized, in the
compression stroke of the PI-cycle (4) incandescence of different moving
particle has been detected. It became evident that at least one of them
causes premature flame initiation.
Discussion of the PI in cycle (7)

432
The phenomena during the compression stroke of cycle (7) are similar to
cycle (4). A single hot particle was detected well before TDC entering the
field of view of the camera and following the gas movement in the com-
bustion chamber. The left image in Figure 10 shows again a sum-up of
all frames in the range between 77 and 12 CA. The trace illustrates very
clear the movement of the particle. At 2 CA aTDC the particle initiates a
flame propagation starting behind the spark plug (Figure 7, middle). The
right image shows that at spark-timing the flame propagation has already
developed.

Figure 10: Pre-ignition mechanism in cycle 7

Both examples show that following PIs are caused by glowing particles,
which are generated in the preceding cycles. Furthermore, a detailed ana-
lyses of all 8 recorded PI-sequences revealed, that all of the following PIs
start from a hot particle in the combustion chamber. There is certainly no
conclusive proof of a correlation between high particle density and PI in
the consecutive cycle, but optical data from all sequences supports such
theory. Furthermore, only few of the attempts to start a PI-sequence suc-
ceeded, indicating that certain conditions in the combustion chamber must
be given in order to develop single preignition to a sequence.

Particles or Droplets?
Due to missing illumination of the combustion chamber it is not possible
to distinguish hot particles from burning droplets. However, observations
indicate that following pre-ignitions were induced by solid particles. One
example is shown in Figure 11. The trace in the left image is generated
by the sum of every second frame in the range between 138 and 50 CA
bTDC. The object enters the viewing area of the camera at and moves
bottom-up. The direction is twice deflected at + and finally at

433
the object leaves the observation field. It is remarkable that shape and
size nearly do not change, even when the object obviously hits the intake
valve and changes the moving direction. Assuming that this object is a
burning droplet, it is very unlikely that it can “survive” the preceding gas
exchange, and can move through a wide range of the combustion cham-
ber without atomization keeping shape and size. Since all of the observed
objects showed similar properties, we assume that they are hot particle,
probably from combustion deposits.

Figure 11: PI at the border of the field of view initiated by particle

Interestingly the flame front caused by a PI enters the camera field at the
same position, where the particle has left it. An explicit correlation bet-
ween this particle and the occurrence of PI is unfortunately not possible,
while the PI-onset is not in the observation field.

Origin of the particles

Looking back to Figure 8, regular combustion cycles consecutive to PI-cy-
cles showed a large number of particles during the combustion, but not
a single hot particle in the compression stroke. It is unlikely that all these
particles are released from the cylinder walls by the super-knock and re-
mained in the combustion chamber after the exhaust stroke. Thus, ano-
ther explanation of their origin must be found.

By impingement of the cylinder liner with liquid fuel, a mix of fuel and
oil accumulates in the top land. The optical investigations showed, that
super-knock is often followed by intense combustion radiation from the
piston top land (see Figure 12).

434
 Figure 12: Combustion in the top land area

As a result of the detonating combustion by an initiating PI, the flame

can reach regions close to the boundary layer and propagate into the top
land area forming particles and deposits. Because of the high in-cylin-
der pressure during the power stroke these particles cannot leave the
top land area. Therefore in the following exhaust stroke, it is likely that
deposits keep remain in the combustion chamber. However, because of
the heat flux to cylinder liner, piston ring and piston the temperature of the
particles decreases. In the intake stroke the piston moves down and a
significant number of the particles can leave the top land and mix with the
air-fuel-mixture. Pre-ignition in the following compression stroke is yet un-
likely, since the temperature of the particles is too low. During the regular
combustion the particles get heated-up to a level close to the combustion
temperature. The optical investigations make evident that some of these
hot particles survive the following exhaust stroke at a temperature, which
is significant to initiate a pre-ignition.

3.3. Results of the numerical simulation

High particle temperatures may promote auto-ignition in the surrounding
fuel/air mixture. The assumption that hot particles represent a potential
trigger of pre-ignition requires high particle temperatures at the start of
the compression stroke. The existence of hot particles in the compression
implies that particles survive the scavenging. During the intake cycle par-
ticles mix with the fresh air and the injected fuel provoking a decrease of
particle temperatures.

435
Figure 13 presents the transient evolution of the particle temperature du-
ring the power, the exhaust and the intake stroke for different particle sizes
and initial temperatures.

Figure 13: Transient evolution of particle temperature during the charge

cycle for different particle sizes and initial temperatures

The results yield that temperature decreases much faster for small par-
ticles compared to the large ones. Thus the size represents a decisive
parameter for the evolution of the particle temperature. Moreover the com-
parison of two particles of the same size and different initial temperatures
shows that the particle exhibiting the higher temperature is cooling down
faster. As a consequence simulations were performed with different par-
ticle sizes to determine the minimum particle temperature at the start of
the compression required to observe auto-ignition before the compression
stroke is completed. The results are displayed in Figure 14. The different
symbols represent the different initial particle temperatures before the gas
exchange. The particles within the dashed loop successfully triggered an
auto-ignition before the end of the compression stroke.

436
Figure 14: Ignition limit for auto-ignition around particles of different sizes

The results indicate that a particle temperature of approximately 1300 K at

the beginning of the compression is a prerequisite for premature auto-igni-
tion independent from the particle size. In addition no auto-ignition can be
observed for particles featuring sizes of less than r0 < 85.

The results of these studies show that premature auto-ignition can be trig-
gered by solid particles if they exhibit temperatures exceeding 1300 K at
the start of the compression. In these investigations a potential effect of
the turbulent flow field was not taken into account. Convection is likely to
enhance the cooling of the particles. Although considering the low mass of
the particle it can be assumed that hey get carried away easily by the flow
and thus the effect of convection will not play a major role. Nevertheless
critical particle temperatures are expected to be marginally higher in the
real engine.

4. Summary
This work investigates alternating series of cycles with pre-ignition (PI)
and regular combustion in SI engines. Experimental investigations were
performed on an optically accessible, turbocharged four-cylinder DISI en-
gine in an engine testbed.

437
Series were systematically initiated by advancing the spark timing of one
single combustion cycle to an early crank angle (subsequent cycles re-
maining at regular spark timing), well before the knock limit. This yielded
an artificial single PI event in one of the cylinders, which was sometimes
followed by an alternating series of PI and regular cycles. The processes
leading to PI within a series were investigated by an endoscopic high-
speed camera visualization system through an optical access to the com-
bustion chamber of cylinder no. 1. The images were recorded in the visible
light spectrum, without additional illumination. Therefore, luminescence
phenomena could be easily distinguished from the background.

The experiments showed that pre-ignition series can be induced by the

artificial initiation of one single PI-cycle, and established a causality of
the later PI-cycles within a series and the initial PI. The high-speed image
sequences revealed occasional luminescent small spots which are mo-
ving through the combustion chamber. A detailed analysis of all optically
recorded PI-series showed that all PIs within a series were accompanied
by luminescent spots in the compression stroke, and, vice versa, that all
occurrences of luminescence also were followed to PI events.

A large number of spots was detected in the power stroke of regular com-
bustion cycles immediately following a PI. However, in the compression
stroke of the same regular cycles, no detectable luminescence was pre-
sent. The spot density in subsequent regular cycles of a PI series was
decreasing with increasing time from the initial PI event. Only about 10%
of all artificial PIs also caused PI-series; temporally close attempts to in-
itiate PI series were less successful than attempts with larger temporal
separation.

If we interpret the luminescent spots as hot particles in the combustion

chamber emitting thermal radiation, the different observations could be
explained by the following picture:

Particles can be released from deposits at the piston (land or crevice) by

a PI (possibly by support of strong pressure waves), but do not leave the
piston before or shortly after gas exchange TDC. They get exposed to
the bulk gas phase in the combustion chamber, possibly by inertial forces
at the strongly accelerated piston. During the intake stroke, particles are
still too cold to emit detectable radiation or to initiate combustion. Howe-
ver, in the following regular combustion cycle, the particles get heated by
the exhaust gas, and some of them manage to remain in the combustion

438
chamber till the next cycle, where they are hot enough to ignite the gas
phase and cause a PI. Each time this sequence of PI and regular com-
bustion is repeated, part of the deposit “reservoir” is consumed, until it is
empty some cycles later. Refilling this “reservoir” would then be a neces-
sary condition for another PI-series to occur, explaining the comparatively
small number of PI-sequences that could be artificially created.

Further investigations using high-speed-imaging and an illumination of the

combustion chamber can bring additional information about the processes
in the exhaust- and intake stroke, collecting more evidence to corroborate
or refute this picture.

439
References
[1] Amann, M.; Alger, T.; Mehta, D.: “The effect of EGR on low-speed
pre-ignition in boosted SI engines”, SAE Int. J. Engines 4(1):235-
245, 2011
[2] Amann, M.; Alger, T.; Westmoreland, B.; Rothmaier, A.: “The effects
of piston crevices and injection strategy on low-speed pre-ignition in
boosted engines“, SAE Int. J. Engines 5(3):1216-1228, 2012
[3] Bradley, D.; Dixon-Lewis, G.; El-Din Habik, S.; Mushi, E.M.J.: “The
oxidation of graphite powder in flame reaction zones”, Proceedings
of the Combustion Institute 20(1984), 931-940
[4] Bradley, D.; Kalghatgi, G.; Andrae, J.; Harrison, A.: “The nature of
super-knock and its origins in SI engines”, International Combustion
Engines: Performance, Fuel Economy and Emissions, Institution of
Mechanical engineers, 2009
[5] Dahnz, C.; Han, K.-M.; Spicher, U.; Magar, M.: “Investigations on
pre-ignition in highly supercharged SI engines”, SAE Int. J. Engines,
2010, 3(1), 214-224
[6] Dahnz, C.; Magar, M.; Kubach, H.; Schiessl, R.; Spicher, U.; Maas,
U.: Abschlussbericht FVV-Vorhaben „Vorentflammung in Ottomoto-
ren“, Universität Karlsruhe (TH), Institut für Kolbenmaschinen, Insti-
tut für Technische Thermodynamik, 2009
[7] Dahnz, C. and Spicher, U.: “Irregular combustion in supercharged
spark ignition engines – pre-ignition and other phenomena”, Interna-
tional Journal of Engine Research 2010 11: 485-498
[8] Dahnz, C.; Palaveev, S.; Schwenger, C.; Spicher, U.: “Faseroptische
Untersuchungsmethoden zur Analyse von Verbrennungsanomalien
in Ottomotoren“, published in „Knocking in Gasoline Engines – Irre-
gular Combustion“; ISBN: 978-3-8169-3047-1; Expert Verlag; 2010
[9] Golloch, R.: “Downsizing bei Verbrennungsmotoren“, Springer Ver-
lag, 2005
[10] Grigo, M.; Wurms, R.; Budack, R.; Helbig, J.; Langa, Z.; Trost, W.:
“Der neue 2,0-l-TFSI-Motor mit Audi valvelift system“,ATZextra,
Springer Automotive Media, 2008, 13, 30-35
[11] Günther, M.; Tröger, R.; Zwahr, S.: “Enthalpie basierter Ansatz zur
Quantifizierung und Vermeidung von Vorentflammungen“, MTZ -
Motortechnische Zeitschrift, Springer Automotive Media, 2011, 72,
296-301

440
[12] Haenel, P.; Seyfried, P.; Kleeberg, H.; Tomazic, D.: “Systematic
approach to analyze and characterize pre-ignition events in turboch-
arged direct injected gasoline engines”, SAE Technical Paper 2011-
01-0343
[13] Han, K.-M.: “Lichtleiterbasierte Methoden zur optischen Analyse von
räumlichen Verbrennungsprozessen und Verbrennungsanomalien
in Ottomotoren“, Logos Verlag, 2010
[14] Han, K.-M.; Velji, A.; Spicher, U.: “A New Approach for Three-Dimen-
sional High-Speed Combustion Diagnostics in Internal Combustion
Engines”, SAE Paper 2006-01-3315
[15] Maas, U. and J. Warnatz: “Ignition Processes in Hydrogen-Oxygen
Mixtures”, Combustion and Flame, 74:53-69, 1988
[16] Magar, M.; Schiessl, R.; Maas, U.: “Premature flame initiation in SI
engines: Modeling studies on the role of residual gas” presented at
ICDERS 23rd, Irvine CA, USA, July 2011
[17] Palaveev, S.; Magar, M.; Kubach, H.; Schiessl, R.; Spicher, U.;
Maas, U.: “Premature flame initiation in a turbocharged DISI engine
– Numerical and experimental investigations, SAE Int. J. Engines
6(1):54-66, 2013
[18] Peters, N.; Kerschgens, B.; Paczko, G.: „Super-Knock Predic-
tion Using a Refined Theory of Turbulence,“ SAE Int. J. Engines
6(2):953-967, 2013
[19] Pinto da Costa, J.M.C; Cracknell, R.F.; Seaton, N.A.; Sarkisov, L.:
“Towards predictive molecular simulation of normal and branched
alkane adsorption in carbonaceous engine deposits. Carbon 49
(2011) 445-456.
[20] Stauch, R. and U. Maas: “Transient detailed numerical simulation
of the combustion of carbon particles”, International Journal of Heat
and Mass Transfer 01/2009
[21] Stojkovic, B.D.; Fansler, T.D.; Drake, M.C.; Sick, V.: ” „High-Speed
Imaging of OH* and Soot Temperature and Concentration in a Stra-
tified-Charge Direct-Injection Gasoline Engine.“ Proceedings of the
Combustion Institute 30(2): 2657 -2665
[22] Takeuchi, K.; Fujimoto, K.; Hirano, S.; Yamashita, M.: “Investigation
of engine oil effect on abnormal combustion in turbocharged direct
injection-spark ignition engines”, SAE Technical Paper 2012-01-
1615, 2012

441
[23] Willand, J.; Daniel, M.; Montefrancesco, E.; Geringer, B.; Hofmann,
P.; Kieberger, M.: “Grenzen des Downsizing bei Ottomotoren durch
Vorentflammung“, MTZ - Motortechnische Zeitschrift, Springer Auto-
motive Media, 2009, 70, 422-428
[24] Zahdeh, A.; Rothenberger, P.; Nguyen, A.; Anbarasu, M.;
Schmuck-Soldan, S.; Schaefer, J.; Goebel, T.: “Fundamental Appro-
ach to Investigate Pre-Ignition in Boosted SI Engines”, SAE Int. J.
Engines 4(1):246-273, 2011

Acknowledgments
The authors wish to express their most sincere gratitude to Audi AG for
providing the experimental engine and control unit and Uwe Walther for
the developing and im-plementing the ignition system. Many thanks to Phil
for proofreading.

442
The Authors
Dipl. Ing. Matthias Kratzsch
IAV GmbH
Berlin / Germany
Dipl. Ing. Michael Günther
IAV GmbH
Chemnitz / Germany
Dr.-Ing. Philipp Adomeit
FEV GmbH
Aachen / Germany
Dipl. Ing. Matthias Biehl
Robert Bosch GmbH
Stuttgart / Germany
Dipl-Ing. (FH) Dominik Biehl
Continental Automotive GmbH
Regensburg / Germany
Dipl.-Ing. Nikola Bobicic
TU Wien
Vienna / Austria
Dr.-Ing. André Brunn
FEV GmbH
Aachen / Germany
Filippo Cavanna
Magneti Marelli S.p.A.
Bologna / Italy
Dr.-Ing. Franz Chmela
Large Engines Competence
Center
TU Graz
Graz / Austria
Dr. Peter Christiner
Large Engines Competence
Center
TU Graz
Graz / Austria
Dipl.-Ing. Carsten Dieterich
FEV GmbH
Aachen / Germany
Dipl.-Ing. Christian Disch
Institut für Kolbenmaschinen
Karlsruhe Institute of Technology
(KIT)
Karlsruhe / Germany
DI Thomas Dobes
AVL List GmbH
Graz / Austria
Dipl.-Ing. Arndt Döhler
International Development Centre
Powertrain Advanced Engineering
Adam Opel AG
Rüsselsheim / Germany
Dr.-Ing. Jürgen Fischer
Mercedes-AMG GmbH
Affalterbach / Germany
B. Eng (Hons) Paul Freeland Dipl.-Ing. Werner Holly
MAHLE Powertrain Limited TU Wien
Northampton / UK Vienna / Austria

Dipl.-Ing. Alfred Frommelt Prof. Dr.-Ing. Karl Huber

Daimler AG Technische Hochschule Ingolstadt
Stuttgart / Germany Ingolstadt / Germany

Nicola Garagnani Dipl.-Ing. Bernhard Jochim

Magneti Marelli S.p.A. Institute of Combustion Technology
Bologna / Italy RWTH-Aachen University
Aachen / Germany
Dipl.-Ing. Christopher Gessenhardt
Volkswagen AG Dipl.-Ing. Arnold Kaden
Wolfsburg / Germany Daimler AG
Stuttgart / Germany
Dipl.-Ing. Helmut Gildein
Daimler AG Dr. Paul E. Kapus
Stuttgart / Germany AVL List GmbH
Graz / Austria
Dipl.-Ing. Marco Günther
RWTH Aachen University Dipl.-Ing. Bruno Kerschgens
Aachen / Germany Institute of Combustion Technology
RWTH-Aachen University
Prof. Dr.-Ing. Christian Hasse Aachen / Germany
Institut für Energieverfahrenstechnik
und Chemieingenieurwesen Dipl.-Ing. Dr. techn.
Lehrstuhl für Numerische Markus Kieberger
Thermofluiddynamik Continental Automotive GmbH
TU Bergakademie Regensburg / Germany
Freiberg / Germany
Dipl.-Ing. Martin Kirsten
Dipl.-Ing. (FH) Johann Hauber Large Engines Competence
Technische Hochschule Ingolstadt Center
Ingolstadt / Germany TU Graz
Graz / Austria
Dipl.-Ing. Michael Heiss
TU Wien Dr.-Ing. Andreas Kleemann
Vienna / Austria BMW Group
Munich / Germany
Dr. Alois Hirsch
AVL List GmbH
Graz / Austria
B.Sc. Silja Klier Assistant Prof. Dipl.-Ing. Dr.techn.
RWTH Aachen University Thomas Lauer
Aachen / Germany TU Wien
Vienna / Austria
Prof. Dr. sc. techn. Thomas Koch
Institut für Kolbenmaschinen Dipl.-Ing. Dirk Linse
Karlsruhe Institute of Technology BMW Group
(KIT) Munich / Germany
Karlsruhe / Germany
Prof. Dr. rer. nat. habil.
Dr. Taku Kondo Ulrich Maas
Honda R&D Institute of Technical
Saitama / Japan Thermodynamics
Karlsruhe Institute of Technology
Dr. Eberhard Kraus (KIT)
Daimler AG Karlsruhe / Germany
Stuttgart / Germany
Dipl.-Ing. Max Magar
Dipl.-Ing. Benjamin Kraus Institute of Technical
Mercedes-AMG GmbH Thermodynamics
Affalterbach / Germany Karlsruhe Institute of Technology
(KIT)
Dr.-Ing. Heiko Kubach
Karlsruhe / Germany
Institut für Kolbenmaschinen
Karlsruhe Institute of Technology Dr. Bernd Mahr
(KIT) MAHLE Powertrain Limited
Karlsruhe / Germany Northampton / UK
Dr. Assoc. Prof. Dr. Karsten Michels
Tatsuya Kuboyama Volkswagen AG
Chiba University Wolfsburg / Germany
Chiba / Japan
Prof. Dr. Yasuo Moriyoshi
Dipl. Ing. Philipp Langfritz Chiba University
IAV GmbH Chiba / Japan
Chemnitz / Germany
Ing. Mario Ninaus
Riccardo Lanzoni AVL List GmbH
Magneti Marelli S.p.A. Graz / Austria
Bologna / Italy
DI Martin Ogris
AVL List GmbH
Graz / Austria
Dipl.-Ing. Sebastian Ohler
Institute of Automotive Engineering
and Power Train Engineering
Helmut Schmidt University
University of the Federal Armed
Forces Hamburg
Hamburg / Germany
Günter Paczko
Institute of Combustion Technology
RWTH-Aachen University
Aachen / Germany
Dipl.-Ing. Stefan Palaveev
Institut für Kolbenmaschinen
Karlsruhe Institute of Technology
(KIT)
Karlsruhe / Germany
M. Eng. Elvira Perleß
Robert Bosch GmbH
Stuttgart / Germany
Dr.-Ing. Dr. h.c. mult.
Norbert Peters
Institute of Combustion Technology
RWTH-Aachen University
Aachen / Germany
Dr. Harald Philipp
AVL List GmbH
Graz / Austria
Dr. Gerhard Pirker
Large Engines Competence
Center
TU Graz
Graz / Austria
Dipl.-Ing. Stefan Pritze
International Development Centre
Powertrain Advanced Engineering
Adam Opel AG
Rüsselsheim / Germany
Prof. Dr.-Ing Hermann
Rottengruber
Institut für mobile Systeme IMS
Otto von Guericke University
Magdeburg
Magdeburg / Germany
Dipl.-Ing. Hubert Schaumberger
GE Jenbacher GmbH & Co OG
Jenbach / Austria
Dipl.-Ing. Uwe Schaupp
Daimler AG
Stuttgart / Germany
Dr.-Ing. Robert Schießl
Institute of Technical
Thermodynamics
Karlsruhe Institute of Technology
(KIT)
Karlsruhe / Germany
Dr.-Ing. Erik Schünemann
BMW Group
Munich / Germany
Prof. Dr.-Ing. Christian Schwarz
BMW Group
Munich / Germany
Dr.-Ing. Markus Selder Dipl.-Wirtsch. Ing Eike Christian
BMW Group Todsen
Munich / Germany Institut für mobile Systeme IMS
Otto von Guericke University
David Serrano Magdeburg
IFP Energies nouvelles Magdeburg / Germany
Rueil-Malmaison Cedex / France
Dr.-Ing. Christian Trapp
Stefano Sgatti GE Jenbacher GmbH & Co OG
Magneti Marelli S.p.A. Jenbach / Austria
Bologna / Italy
Dr. Marco Warth
Dipl.-Ing. Robert Sloboda MAHLE Powertrain Limited
Robert Bosch GmbH Northampton / UK
Stuttgart / Germany
Prof. Dr. Andreas Wimmer
Hiroshi Sono Large Engines Competence
Honda R&D Center
Saitama / Japan TU Graz
Graz / Austria
Prof. i.R. Dr.-Ing. Ulrich Spicher
Institut für Kolbenmaschinen Dr. Ernst Winklhofer
Karlsruhe Institute of Technology AVL List GmbH
(KIT) Graz / Austria
Karlsruhe / Germany
Prof. Dr. techn. Andreas Witt
Dipl. Ing. Alexander Stahr BMW Group
IAV GmbH Munich / Germany
Chemnitz / Germany
Dipl.-Ing. Sebastian Wolf
Dr.-Ing Karl Georg Stapf Mercedes-AMG GmbH
Ingenieurbüro TWB Affalterbach / Germany
Fulda / Germany
Jean-Marc Zaccardi
Dr. Jörg Theobald IFP Energies nouvelles
Volkswagen AG Rueil-Malmaison Cedex / France
Wolfsburg / Germany
Massimo Zanotti
Magneti Marelli S.p.A.
Bologna / Italy