Impact of the cycle-to-cycle variation of an internal combustion ...

167
ARISTOTLE UNIVERSITY OF THESSALONIKI FACULTY OF ENGINEERING MECHANICAL ENGINEERING DEPARTMENT LABORATORY OF APPLIED THERMODYNAMICS Impact of the cycle-to-cycle variation of an internal combustion engine to gaseous pollutants emissions Apostolos Karvountzis-Kontakiotis Dipl. Mechanical Engineer Thessaloniki, August 2015

Transcript of Impact of the cycle-to-cycle variation of an internal combustion ...

ARISTOTLE UNIVERSITY OF THESSALONIKI

FACULTY OF ENGINEERING

MECHANICAL ENGINEERING DEPARTMENT

LABORATORY OF APPLIED THERMODYNAMICS

Impact of the cycle-to-cycle variation of an

internal combustion engine to gaseous

pollutants emissions

Apostolos Karvountzis-KontakiotisDipl. Mechanical Engineer

Thessaloniki, August 2015

Apostolis Karvountzis-Kontakiotis

Impact of the cycle-to-cycle variation of an

internal combustion engine to gaseous

pollutants emissions

2015

Supervisor: Assistant Professor L. Ntziachristos

Members of Examination Committee:

Full Professor Z. Samaras

Associate Professor G. Koltsakis

Assistant Professor L. Ntziachristos

This thesis is submitted in partial fullfillment of the requirements for the degree of

Doctor of Engineering

c©Aristotle University, 2015. All rights reserved. No part of this publication my be

reproduced without written permission of the copyright holder.

This work is dedicated to

all the members of my family,

and to my only one Paschalena.

Acknowledgements

First of all, I would like to thank my supervisor Prof. Leonidas Ntziachristos

for his time, ideas and contribution on this PhD thesis. Our long discussions

instigated many of the themes presented here.

I would like to express my gratitude to the director of the laboratory (LAT)

Prof. Zissis Samaras for his trust and his support, particularly on the experi-

mental part of this project. I also wish to thank Prof. Grigoris Koltsakis for

teaching me internal combustion engines and aftertreatment modeling and for

his confidence in me during our cooperation with automotive industry. I also

wish to extend my gratitude to all the teachers that inspired and shaped me

as an engineer and person throughout my studies.

Special thanks to the 7th European Framework ”LESSCCV” for the financial

support. The discussions and ideas of the partners of this project were source

of inspiration to me. I am especially thankful to Prof. Ananias Tomboulides

for his advice and support during the LESSCCV project. I also want to

thank Prof. Anestis Kalfas who inspires me to work in the research area and

motivates me to publish my research work.

Special thanks to all the staff of the Laboratory of Applied Thermodynamics

(LAT) for their valuable services and support. Thanks are due to my colleague

and friend Dr. Athanasios Dimaratos for our interesting discussions on the

topics of internal combustion engines and their measurement techniques as

well as for his great contribution on the experimental part of this thesis. I owe

a lot to Dr. Mark Peckham, Director at Cambustion Ltd, for his technical

support and his grant of the NDIR500 CO/CO2 analyzer. Special thanks to

Dr. Costas Michos for the stimulating discussions and his advice on emission

modeling. Thanks are due to Dr. Elias Vouitsis for his fruitful and interesting

discussions on engine design and emissions trends. The assistance of the Aris-

totle Racing Team on the engine measurements is gratefully acknowledged.

Many thanks to my colleagues Dimitrios, Thomas, Elias, Christos and my

tolerant officemate Dora, for sharing our thoughts and worries, and for making

this personal research journey a bit less lonely. I also want to thank my younger

colleagues and wish them good luck to their new start. Thanks are also due

to all staff of Exothermia S.A for their support and assistance during my

collaboration with Exothermia S.A.

I had the chance to work with many MSc students and gain through them

much experience and knowledge. Many thanks to George Giannakopoulos,

Orestis Mpouras, Thomas Souliotis, Lazaros Kefalidis, Nikos Zigkopis and

Pavlos Evagelidis.

I am deeply grateful to my old friends and classmates Teo, Panos, Christos,

Vasilis, Thanos and Orestis for giving me so many nice memories all these

years.

Last but not least, I would like to thank my parents, my sister, my uncle

and his family and my patient Paschalena, for all their help, support and

encouragement all these years.

Abstract

This work focuses on the computational and experimental investigation of cyclic emis-

sion variability, particularly on NOx emissions. In this context, this thesis contributes

towards clarifying the following gray aspects of future engine development:

• the development and validation of a detailed chemistry emission model for the ki-

netically prediction of the gaseous pollutants emissions

• the experimental investigation of cyclic emission variability, using fast response an-

alyzers and the correlation with combustion characteristics

• the utilization of the detailed chemistry model for the improved prediction of emis-

sion variability under engine homogeneous conditions

The first and third objectives requires the development of an advance emission model

that presents improved accuracy compared to conventional emission approaches. This is

performed with the utilization of a detailed chemical reaction scheme, assuming that each

burned zone is a well-stirred reactor. Compared to existing emission models, the model

is characterized by the absence of both the use of chemical equilibrium and any tuning

parameter or calibration factor. The model presents computational time in the same order

of magnitude of conventional emission models. As this model presents improved accuracy

of NO emissions and reduced computational time, it can contribute to the design of more

efficient engines.

Regarding the second objective, in this study is experimentally explained how com-

bustion variability affects emissions variability. Although the impact of cyclic combustion

variability has been extensively investigated in the past, only very little concern has been

paid on cyclic emission variability. This experimental study presents the impact of various

engine parameters, such as engine load, equivalence ratio, and ignition timing on cyclic

emission variability.

i

ii

Contents

Abstract i

Table of Contents ii

List of Figures vi

List of Tables x

Nomenclature xi

1 Introduction 1

1.1 Research scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Fundamentals of spark ignition engines . . . . . . . . . . . . . . . . . . . . 2

1.2.1 Engine operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.2.2 Mixture formation . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.2.3 The ignition process . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.2.4 The combustion process . . . . . . . . . . . . . . . . . . . . . . . . 4

1.2.5 Pollutant formation . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.3 Thesis aim and objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.4 Thesis structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Literature Review 9

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.2 Pollutants Formation Modeling . . . . . . . . . . . . . . . . . . . . . . . . 10

2.2.1 NOx Emissions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2.2 CO Emissions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.2.3 HC Emissions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.2.4 Pollutants Formation Models . . . . . . . . . . . . . . . . . . . . . 15

2.2.4.1 Simplified Kinetic Models . . . . . . . . . . . . . . . . . . 16

2.2.4.2 Detailed Kinetic Models . . . . . . . . . . . . . . . . . . . 17

2.3 Cycle-to-Cycle Variability . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

iii

2.3.1 The Nature of Cyclic Variation . . . . . . . . . . . . . . . . . . . . 19

2.3.2 Origins of Cyclic Variation . . . . . . . . . . . . . . . . . . . . . . . 20

2.3.3 Measures of Cyclic Variation . . . . . . . . . . . . . . . . . . . . . . 21

2.3.4 Effects of Engine Variables on Cyclic Variability . . . . . . . . . . . 24

2.3.4.1 Equivalence Ratio . . . . . . . . . . . . . . . . . . . . . . 24

2.3.4.2 Fuel Type . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.3.4.3 Combustion Chamber Geometry . . . . . . . . . . . . . . 26

2.3.4.4 Compression Ratio . . . . . . . . . . . . . . . . . . . . . . 27

2.4 Cyclic Emissions Variability . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2.5 Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

3 Proposed Emissions Model 31

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.2 Emission Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3.2.1 Mass Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.2.1.1 New Burned Mass . . . . . . . . . . . . . . . . . . . . . . 33

3.2.1.2 Inlet/Outlet Concentrations . . . . . . . . . . . . . . . . . 35

3.2.2 Energy Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.3 Validation Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.4 Modeling Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.4.1 Engine Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.4.2 Modeling Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.4.3 Coupling Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.5 Emission Model Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.5.1 Mean Cycle Validation . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.5.2 Cycle-to-Cycle NO Variation . . . . . . . . . . . . . . . . . . . . . . 42

3.5.2.1 Contribution of Prompt Mechanism on the CCV of NO

Emissions . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.5.3 Deviation Between Mean Cycle Values and Mean CCV Values . . . 47

3.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4 Experimental Investigation of Cyclic Emissions Variability 51

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4.2 Experimental Set up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4.2.1 Engine Test Bed . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4.2.2 Experimental Configuration . . . . . . . . . . . . . . . . . . . . . . 53

4.2.3 Measurement Protocol . . . . . . . . . . . . . . . . . . . . . . . . . 54

4.3 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4.3.1 In cylinder pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4.3.2 Engine and Cylinder Emissions . . . . . . . . . . . . . . . . . . . . 56

iv

4.3.3 Exhaust Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . 59

4.3.4 Data Acquisition Card . . . . . . . . . . . . . . . . . . . . . . . . . 60

4.3.5 Engine Control Unit . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4.4 Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

4.4.1 Combustion Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 61

4.4.2 Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 63

4.4.3 Emissions Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

4.4.4 Cyclic Variability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

4.5.1 Impact of Engine Load on Variability . . . . . . . . . . . . . . . . . 67

4.5.2 Impact of Stoichiometry on Variability . . . . . . . . . . . . . . . . 72

4.5.3 Impact of Ignition Timing on Variability . . . . . . . . . . . . . . . 79

4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

5 Emission Model Validation and Application for CCV Prediction 89

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

5.2 Modeling Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

5.2.1 Two Zone Combustion Analysis . . . . . . . . . . . . . . . . . . . . 90

5.2.1.1 Cylinder Model . . . . . . . . . . . . . . . . . . . . . . . . 90

5.2.1.2 Integration Flow . . . . . . . . . . . . . . . . . . . . . . . 91

5.2.1.3 Error Formula . . . . . . . . . . . . . . . . . . . . . . . . 93

5.2.2 Emissions Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

5.2.2.1 Proposed Modeling . . . . . . . . . . . . . . . . . . . . . . 94

5.2.2.2 Simplified Modeling . . . . . . . . . . . . . . . . . . . . . 94

5.3 Cycle Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

5.4 Mean Cycle Emissions Prediction . . . . . . . . . . . . . . . . . . . . . . . 96

5.4.1 Engine Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

5.4.2 Mixture Stoichiometry . . . . . . . . . . . . . . . . . . . . . . . . . 98

5.4.3 Ignition Timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

5.5 Sensitivity Analysis of the Emission Model . . . . . . . . . . . . . . . . . . 102

5.6 Application of the model to predict CCV . . . . . . . . . . . . . . . . . . . 104

5.7 Cycle to Cycle Emission Prediction . . . . . . . . . . . . . . . . . . . . . . 106

5.7.1 Engine Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

5.7.2 Mixture Stoichiometry . . . . . . . . . . . . . . . . . . . . . . . . . 110

5.7.3 Ignition Timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

5.8 Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

v

6 Conclusions & Future Work 119

6.1 Experimental Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

6.2 Emissions Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

6.3 Novelty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

6.4 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

A A FORTRAN Program for Predicting Homogeneous Gas Phase Chem-

ical Kinetics 127

A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

A.2 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

References 132

vi

List of Figures

2.1 Spark ignition engine emissions for different air/fuel ratios [16]. . . . . . . . 10

2.2 Cross-sections of combustion chambers investigated in Young [55] with var-

ious spark locations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.1 Schematic of the two zone detailed chemistry emission model. . . . . . . . 32

3.2 Schematic of the Rover K4 model developed in BOOST. . . . . . . . . . . 38

3.3 Comparison of measured and simulated NO molar fractions for stoichio-

metric combustion (λ=1.0). Results without prompt mechanism are also

presented. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.4 Comparison of measured and simulated NO molar fractions for lean (λ=1.5)

operating engine conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . 42

3.5 Impact of slight stoichiometry variation on NO formation. . . . . . . . . . 43

3.6 CCV of in-cylinder pressure and temperature evolution, which results on

the distribution of imep and NO emissions (P1015 point). . . . . . . . . . . 44

3.7 CCV of in-cylinder pressure and temperature evolution, which results on

the distribution of imep and NO emissions (W1030 point). . . . . . . . . . 46

3.8 CCV NO distribution w/o the prompt mechanism. . . . . . . . . . . . . . 48

4.1 Schematic of the experimental configuration . . . . . . . . . . . . . . . . . 53

4.2 Sample of processing pressure raw data at 4000 rpm, 50% throttle position

and rich operating conditions (λ=0.93). . . . . . . . . . . . . . . . . . . . . 56

4.3 Photo from the experimental configuration regarding the positioning of the

probes for the cylinder and engine exhaust emissions sampling. . . . . . . . 57

4.4 The Cambustion fast response analyzers . . . . . . . . . . . . . . . . . . . 59

4.5 The LabVIEW interface and the data acquisition card. . . . . . . . . . . . 61

4.6 Typical NO recording in the exhaust with reference to the cylinder pressure. 64

4.7 IMEP and emissions distributions at 4000 rpm, 80% open throttle position,

stoichiometric mixture and MBT ignition timing. . . . . . . . . . . . . . . 65

vii

4.8 Pressure variability at 4000 rpm, 80% open throttle position, stoichiometric

mixture and MBT ignition timing. Mean pressure curve, maximum and

minimum gross heat release rates for the 150 consecutive cycles are also

illustrated. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

4.9 Mean pressure and gross heat release traces for stoichiometric conditions,

4000 rpm, identical spark timing and various throttle positions. . . . . . . 67

4.10 Relationship between IMEP and ϑ5% at 4000 rpm, stoichiometric mixture,

identical ignition timing and various throttle positions (TP: 20% and TP:

80%). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4.11 Relationship between IMEP and NO at 4000 rpm, stoichiometric mixture,

identical ignition timing and various throttle positions. . . . . . . . . . . . 70

4.12 Relationship between Pmax and NO concentration at 4000 rpm, stoichio-

metric mixture, identical ignition timing and various throttle positions. . . 71

4.13 Mean pressure and gross heat release traces for rich, stoichiometric and

lean mixture, 6000 rpm, 80% TP, and identical spark timing. . . . . . . . . 72

4.14 Relationship between ϑ5% and mixture stoichiometry at 6000 rpm, 80% TP

and identical ignition timing. . . . . . . . . . . . . . . . . . . . . . . . . . . 73

4.15 Relationship between IMEP and mixture stoichiometry at 4000 rpm, 80

per cent throttle position and constant ignition timing. . . . . . . . . . . . 75

4.16 Relationship between λ and NO concentration at 6000 rpm, 80% throttle

position and identical spark ignition timing. . . . . . . . . . . . . . . . . . 76

4.17 Relationship between NO and early flame kernel development (ϑ5%) at

4000 rpm, 80% TP and identical ignition timing. . . . . . . . . . . . . . . . 77

4.18 Relationships between λ and carbon monoxide and carbon dioxide concen-

trations at 6000 rpm, 80% TP and identical spark ignition timing. . . . . . 78

4.19 Mean pressure and gross heat release traces for rich conditions (λ=0.93),

4000 rpm, 80% throttle position and various spark timings. . . . . . . . . . 80

4.20 Relationship between IMEP and maximum cylinder pressure at 4000 rpm,

80% TP and λ=0.94, for different ignition timings. . . . . . . . . . . . . . 81

4.21 Relationship between ignition delay and maximum cylinder pressure at

4000 rpm, at 4000 rpm, 80% TP and λ=0.94, for different ignition timings. 82

4.22 Relationship between maximum cylinder pressure and crank angle at which

maximum cylinder pressure occurs for different ignition timings at 4000

rpm, 80% TP and λ=0.94. . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

4.23 Relationship between NO formed and crank angle at which maximum cylin-

der pressure occurs for different ignition timings at 4000 rpm, 80% TP and

λ=0.94. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

viii

4.24 Relationship between NO formed and crank angle at which maximum cylin-

der pressure occurs for different ignition timings at 4000 rpm, 80% TP and

λ=0.94. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

5.1 Flowchart of the two zone heat release model. . . . . . . . . . . . . . . . . 92

5.2 Evolution of cylinder pressure, zone temperatures, NO and CO concentra-

tions at 4000 rpm, 80% throttle position, stoichiometric mixture and MBT.

(a) measured cylinder pressure of the closed thermodynamic cycle, (b) the

calculated gas temperature (Tg), the burned zone temperature (Tb) and

the unburned temperature (Tun), (c) calculated NO concentrations under

conditions of equilibrium/kinetics control, (d) CO concentrations under

conditions of equilibrium/kinetics control. . . . . . . . . . . . . . . . . . . 95

5.3 Effect of engine load on NO and CO formation for stoichiometric mixture

and various engine speed conditions: (left column) NO simulation, (right

column) CO simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

5.4 Effect of mixture lambda on NO and CO formation for 80% throttle po-

sition and various engine speed conditions: (left column) NO simulation,

(right column) CO simulation. . . . . . . . . . . . . . . . . . . . . . . . . . 99

5.5 Effect of ignition timing on NO and CO formation at 4000 rpm, 80% throt-

tle position and stoichiometric conditions: (left column) NO simulation,

(right column) CO simulation. . . . . . . . . . . . . . . . . . . . . . . . . . 101

5.6 Effect of ignition timing on NO and CO formation at 6000 rpm, 20% throt-

tle position and stoichiometric conditions: (left column) NO simulation,

(right column) CO simulation. . . . . . . . . . . . . . . . . . . . . . . . . . 102

5.7 Impact of wall temperature on pollutants formation using the detailed

chemistry model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

5.8 Impact of residual gas fraction on pollutants formation using the proposed

detailed chemistry model. . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

5.9 Evolution of maximum and minimum IMEP cycle in relationship to crank

angle: (a) measured cylinder pressure, (b) calculated heat release and

burned zone temperature, (c) NO formation and (d) CO formation. . . . . 105

5.10 Prediction of cyclic NO variability for 20% (left column) and 80% (right

column) throttle position. (a) Relationship between NO and measured

cylinder pressure, (b) NO return map, (c) Relationship between measured

and calculated NO. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

5.11 Prediction of cyclic CO variability for 20% (left column) and 80% (right

column) throttle position. (a) CO return map, (b) Relationship between

measured and calculated CO. . . . . . . . . . . . . . . . . . . . . . . . . . 109

ix

5.12 Prediction of cyclic NO variability for rich (left column) and lean (right

column) mixture composition. (a) Relationship between NO and measured

cylinder pressure, (b) NO return map, (c) Relationship between measured

and calculated NO. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

5.13 Prediction of cyclic CO variability for rich (left column) and lean (right col-

umn) mixture composition. (a) CO return map, (b) Relationship between

measured and calculated CO. . . . . . . . . . . . . . . . . . . . . . . . . . 112

5.14 Prediction of cyclic NO variability at 4000 rpm, stoichiometric mixture,

80% TP, 52o[CA] BTDC ignition timing (first row), 48o[CA] BTDC ignition

timing (second row) and 42o[CA] BTDC ignition timing (third row). Left

column: NO return map, Right column: Relationship between measured

and calculated NO. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

5.15 Prediction of cyclic CO variability at 4000 rpm, stoichiometric mixture,

80% TP, 52o[CA] BTDC ignition timing (first row), 48o[CA] BTDC ignition

timing (second row) and 42o[CA] BTDC ignition timing (third row). Left

column: CO return map, Right column: Relationship between measured

and calculated CO. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

x

List of Tables

2.1 Speed coefficients for the forward reaction of the Zeldovich mechanism [16] 12

3.1 New burned mass composition based on λ . . . . . . . . . . . . . . . . . . 34

3.2 Rover K4 engine characteristics . . . . . . . . . . . . . . . . . . . . . . . . 37

3.3 Validation of the thermodynamic model using BOOST. . . . . . . . . . . . 40

3.4 Comparison of COV values for imep and NO with and without the prompt

mechanism. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

3.5 Comparison of mean cycle values (MC) and mean CCV values for imep

and NO with and without the prompt mechanism. . . . . . . . . . . . . . . 47

4.1 Honda CBR600 engine characteristics . . . . . . . . . . . . . . . . . . . . . 52

4.2 Experimental engine test conditions. . . . . . . . . . . . . . . . . . . . . . 54

4.3 Specifications of the cylinder pressure experimental equipment . . . . . . . 55

4.4 Specifications of the emission gas analyzers and sensors. . . . . . . . . . . . 57

4.5 Mean and COV experimental data for two different throttle positions at

4000 rpm, identical ignition timing (48o [CA] BTDC) and stoichiometric

operating conditions (λ=0.99). . . . . . . . . . . . . . . . . . . . . . . . . . 68

4.6 Mean and COV experimental data for two different throttle positions at

6000 rpm, identical ignition timing (41o [CA] BTDC) and stoichiometric

operating conditions (λ=1.0). . . . . . . . . . . . . . . . . . . . . . . . . . 68

4.7 Mean and COV experimental data for various mixture lambda operating

conditions at 4000 rpm, 80 % throttle position and identical ignition timing

(48o [CA] BTDC). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

4.8 Mean and COV experimental data for various mixture lambda operating

conditions at 6000 rpm, 80 % throttle position and identical ignition timing

(41o [CA] BTDC). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

4.9 Mean and COV experimental data for various spark ignition timings at

4000 rpm, 80% throttle position and rich operating conditions (λ=0.94). . 80

4.10 Mean and COV experimental data for various spark ignition timings at

6000 rpm, 20% throttle position and rich operating conditions (λ=0.93). . 86

xi

xii

Nomenclature

Roman Symbols

Acyl cylinder wet area [m2]

aht heat transfer coefficient [-]

bht heat transfer coefficient [-]

εht radiation coefficient [-]

xb burn rate [-]

xi species mole fraction [-]

ycyl cylinder position [m]

yi species mass fraction [-]

Greek Symbols

α the carbon moles of the fuel molecular formula

β the hydrogen moles of the fuel molecular formula

δ the nitrogen moles of the fuel molecular formula

ε the stoichiometric fuel/air molar ratio

γ the oxygen moles of the fuel molecular formula

λ mixture lambda

φ equivalence ratio

σ Stefan-Boltzmann constant [W/m2K]

ϑ crank angle [degrees]

xiii

Subscripts

b burned

bb blowby

cyl cylinder

f flame

ht heat transfer

j number of elements

k number of species

L Loss

mix mixture

u unburned

w wall

Acronyms

A wet area [m2]

CCV Cyclic Combustion Variability

CD Combustion Duration

CEV Cyclic Emissions Variability

CO Carbon Monoxide

CO2 Carbon Dioxide

COV Coefficient Of Variation

D Piston Bore [m]

EOC End of Combustion

EVO Exhaust Valve Open

H2O Steam

xiv

HC Unburned hydrocarbons

ICEs Internal Combustion Engines

IMEP Indicated Mean Effective Pressure

IVC Intake Valve Close

LDVs Light Duty Vehicles

MBT Maximum Breaking Torque

MFB Mass Fraction Burned

MW Molecular mass

N2 Nitrogen

NO Nitric Oxide

NO2 Nitrogen Dioxide

NOx Nitrogen Oxides

O2 Oxygen

PFI Port Fuel Injection

PM Particulate Matter

PMEP Pumping Mean Effective Pressure

P Pressure [Pa]

RON Research Octane Number

SI Spark Ignition

SOI Start Of Ignition

SUVs Sport Utility Vehicles

TDC Top Dead Center

T Temperature [K]

TWC Three Way Catalyst

WGS Water Gas Shift reaction

WOT Wide Open Throttle

xv

Chapter 1

Introduction

1.1 Research scope

Transportation is responsible for approximately one third of global CO2 emissions due

to human activity [1]. Road transportation accounts for the great majority of energy

consumption (∼ 85%) [2] and therefore the great majority of greenhouse emissions. At

the same time, light duty vehicles (LDVs) - vehicles that people use in their daily lives

such as cars, pickup trucks and sport utility vehicles (SUVs) - represent more than 20%

of transportation, as the world population grows and economic activity is developed.

During the twentieth century, the internal combustion engines (ICEs) became the

predominant source of power for light duty vehicles. This is expected to be continued

during the twenty first century, as the forecast of electric and plug-in vehicles (LDVs) for

2040 is that they reach about 70 million cars, or less than 5% of the total fleet [3]. The

rest 95% of engine propulsion vehicles use at about 90% the well-known spark ignition

(SI) engine in their powertrain [3], mostly utilizing gasoline.

Gasoline engine technology is currently entering a new phase of intense development

and technology deployment due to three basic determinants: legislation limits on exhaust

gas emissions, relatively low manufacturing costs and advanced engine performance char-

acteristics. European legislation exerts an ever increasing pressure on automotive industry

for low exhaust gas pollutants [4] and greenhouse emissions [5]. The latter is described

by European CO2 regulation to reach 95 g/km by 2020; this can be achieved by adopting

new technologies on SI engines [6]. Lowering CO2 emissions has also a benefit on fuel

consumption. On the other side, fuel economy and low exhaust gas emissions should be

combined with drivability refinement and torque characteristics in a new technology SI

engine. It is meanwhile widely accepted [7] that engine downsizing in combination with

pressure charging, such as turbocharging, is the most appropriate technology bundle to

combine fuel economy and engine performance. It also has been discussed in depth that

1

1. Introduction

direct injection is a technology with high synergy potential to further enhance the merits

of turbocharging [8].

The complete stoichiometric homogeneous combustion of the fuel hydrocarbon (CxHy)

results to exhaust gas with components such as nitrogen (N2), carbon dioxide (CO2) and

steam (H2O). In a real combustion chamber of an internal combustion engine, incomplete

combustion occurs. Nitrogen oxides (NOx), carbon monoxide (CO), unburned hydrocar-

bons (HC) and particulates are components of incomplete combustion that are emitted

together with the complete combustion components. These substances are detrimental

to human health and should be limited by the engine operation during the combustion

process or by using the engine aftertreatment system. Conventional SI engines operate

in the narrow stoichiometric window and therefore are characterized by generally low

exhaust gas emissions. The Three-Way Catalyst (TWC) which is located downstream

the SI engine is a simple and low cost device which ensures the complete combustion of

gaseous phase exhaust gas emissions.

Cycle combustion variability is an undesirable characteristic of SI engines due to fluc-

tuations in both early flame kernel development and turbulent flame propagation. Com-

bustion in engines evolves differently in each operation cycle at nominally steady state

operating conditions. Experimentally, cycle-to-cycle variability is best observed by the

scatter of the measured cylinder pressure around the mean pressure curve. Such fluctua-

tions of the cylinder pressure have an impact on engine performance [9], fuel consumption

[10] and pollutant emission [11–14], while in some extreme cases such as highly diluted

lean mixtures could result in misfiring or knocking [10].

The scope of this thesis is to contribute towards understanding the impact of cyclic

combustion variability on emissions in homogeneous combustion; from a literature point

of view, this belongs to a gray research area. For this purpose, an experimental and

computational investigation of cycle to cycle emissions variability on SI engine was the

objective of this thesis.

1.2 Fundamentals of spark ignition engines

As this thesis is focused on homogeneous combustion and SI engines, it would be essential

to preliminary present a literature review on the fundamentals of SI engines. Although

further details of the literature review on cycle combustion variability and pollutant for-

mation will be given within each chapter, this section aims to briefly introduce the tech-

nology of SI engines, regarding the injection, ignition and combustion procedures, as it is

presented in a large number of textbooks [15–17].

2

1.2. Fundamentals of spark ignition engines

1.2.1 Engine operation

In SI engines, the fuel vapor and air are mixed together either in the intake system or by

direct fuel injection in the cylinder. In the first case, so-called as single or multi point

injection [16], the homogeneous mixture is mixed with the residual gas in the cylinder

during the cylinder charging phase; then it is compressed. Under normal operating con-

ditions, combustion is initiated towards the end of the compression stroke at the spark

plug by an electric discharge. During the combustion process, the initial flame kernel is

transformed to a turbulent flame which propagates through the homogeneous mixture.

The end of the combustion comes when the flame quenches on the cylinder or piston walls.

Although the combustion process is qualitatively repeated from cycle to cycle, quanti-

tatively it is observed that cylinder pressure evolves differently in each operation cycle at

nominally steady state operating conditions. The root of this observation is combustion

variability; the flame development and flame propagation vary from cycle to cycle giving

a different mass fraction burned and heat release at each working cycle. This is because

flame growth depends on local mixture motion and composition. These quantities vary

not only from cycle to cycle but also from cylinder to cylinder. Cycle to cycle and cylinder

to cylinder fluctuations are of highly importance as the extreme cycles limit the operating

regime of the engine.

1.2.2 Mixture formation

The main target of engine fuel systems is to prepare an air-fuel mixture that fulfills the

requirements of the engine over various operating conditions. Although the carburetor

has been the most common device for the control of the fuel flow into the intake manifold,

today electronic fuel injection is the most common fuel flow control system used in SI

engines. The latter is further divided into multi point port injection when external mixture

formation is used and direct injection when internal mixture formation is applied.

In conventional SI engines with external mixture formation, the multi point port

injection is applied in common. It is recognized that the fuel injection system is a source

of cyclic combustion variability, the origins of which are in detail presented in chapter 2.

Furthermore, the fuel and air distribution between engine cylinders is not uniform and

also varies from cycle to cycle. Finally, these fluctuations of the equivalence ratio have an

impact on fuel consumption and at extreme cases could lead to misfiring and drivability

issues.

1.2.3 The ignition process

In external mixture formation SI engine, ignition of the air-fuel mixture takes place at

the minimum cylinder volume; that occurs slightly before the top dead center via spark

3

1. Introduction

discharge between the spark plug electrodes. The ignition procedure through which energy

is transmitted to the mixture through spark discharge is well known as thermal explosion

and differs significantly from the chemical or chain explosion which is met in diesel engines.

The ignition process is summarized into the three phases, as it is presented by Stone [17].

1. Pre-breakdown. Before the discharge occurs, the mixture in the cylinder is a per-

fect insulator. As the spark pulse occurs, the potential difference across the plug

gap increases rapidly (typically 10-100kV/ms). This causes electrons in the gap to

accelerate towards the anode. With a sufficiently high electric field, the acceler-

ated electrons may ionise the molecules they collide with, which leads to the second

phase.

2. Breakdown. Once enough electrons are produced by the pre-breakdown phase, an

overexponential increase in the discharge current occurs. This can produce currents

of the order of 100 A within a few nanoseconds. This is concurrent with a rapid

decrease in the potential difference and electric field across the plug gap (typically to

100 V and 1kV/cm respectively. The breakdown causes a very rapid temperature

and pressure increase. Temperatures of 60000K give rise to pressures of several

hundred bars locally. These high pressures cause an intense shock wave as the

spark channel expands at supersonic speed. Expansion of the spark channel allows

the conversion of potential energy to thermal energy, and facilitates cooling of the

plasma. Prolonged high currents lead to thermionic emission from hot spots on the

electrodes and the breakdown phase ends as the arc phase begins.

3. Arc discharge. The characteristics of the arc discharge phase arc controlled by

the external impedances of the ignition circuit. Typically, the burning voltage is

about 100 V and the current is greater than 100 mA, and is dependent on external

impedances. The arc discharge is sustained by electrons emitted from the cathode

hot spots. Depending on the conditions, the efficiency of the energy-transfer process

from the arc discharge to the thermal energy of the mixture is typically between 10

and 50 per cent.

1.2.4 The combustion process

In SI engines that operate homogeneously, the combustion process is initiated by the

thermal explosion and it is ended by wall flame quenching. For the characterization of

the combustion process completeness, it is well spread in literature to use the mass fraction

burned profiles as a function of crank angle. In some other cases, the gross heat release,

which represents the total chemical energy of the burned fuel, is also used. Both profiles

of mass fraction burned and gross heat release present the characteristic S-shape; that is

divided into three main areas:

4

1.2. Fundamentals of spark ignition engines

• Early flame kernel development is defined as the time period between spark dis-

charge start and combustion start. Generally start of combustion corresponds to

a significant small percent of mass fraction burned. In this thesis, it was adopted

that the 0-5% of mass fraction burn (MFB) duration corresponds to the early flame

kernel development. The index of 5% is adopted by the experimental observation

of the measured data: it is hard or even impossible to measure lower values of MFB

from the indicator diagram because of uncertainties in the measured in-cylinder

pressure [18, 19].

• Turbulence flame propagation represents the main combustion process, which is

defined as the 5-95% of the mass fraction burn duration. During this phase the

transition of initial laminar combustion to fully turbulence combustion occurs. This

second stage of combustion ends after the peak pressure during the operating cycle;

however this is ill-defined in most of the cycles.

• Wall flame quenching is the final stage of combustion. At this stage, the flame front

area is contacting more of the combustion chamber wall, therefore the flame front

area is reduced and the remaining unburned mixture is being burnt more slowly.

The final stage of combustion is too slow and may not be completed by the time

of exhaust valve open. The result of the incomplete combustion is the unburned

hydrocarbons fraction measured downstream of the engine exhaust valve. A typical

value of 1000 to 3000 ppm C1 unburned hydrocarbons corresponds to 1 to 2.5% of

unbunred fuel [15].

Those steps correspond to normal combustion conditions; however at few cycles ab-

normal combustion occurs. Abnormal combustion takes place due to pre-ignition or self-

ignition. In the case of pre-ignition, fuel is ignited by a hot spot; these could be located

at the exhaust valves or could even be hot carbon combustion deposits. On the other

hand, self-ignition is when pressure and temperature of a particular air/fuel composition

mixture are such that the unburned gas upstream of the flame front ignites spontaneously.

Both cases could lead to knocking.

During operation of a SI engine, not all cycles can be ideal, leading to cyclic combustion

variability. Cyclic dispersion occurs because the turbulence within the cylinder varies from

cycle to cycle and mixture composition and uniformity randomly fluctuate. The impacts

of combustion variability are on the variability of engine-out emissions, an increase on

fuel consumption and on limitation of engine performance.

1.2.5 Pollutant formation

In homogeneous SI engines, the fuel vapor, air and any residual exhaust gas from the

previous combustion cycle are essentially uniformly mixed. In such a case, the pollu-

5

1. Introduction

tants formation is a function of cylinder overall equivalence ratio, combustion rate and

temperature as well as residence time of the gas in the post-flame burned zone.

Nitric Oxide (NO) formation comes from both flame front and post-flame gases [15].

In the case of turbulence premixed flames, the flame reaction zone is extremely thin (∼0.1 mm) and residence time of the gas is too short. What is more, cylinder pressure

and temperature rise during most of the combustion process; this explains the higher

temperature at the post-flame zone. Therefore post-flame gases contribute to the majority

of the final nitric oxide concentration.

Carbon monoxide (CO) is also formed by the combustion process. Its formation is

mainly affected by stoichiometry. On the one hand, in rich fuel-air mixtures, fuel carbon

cannot be completely converted to CO2 due to lack of oxygen. On the other hand, in stoi-

chiometric and slightly lean fuel-air mixtures, CO2 dissociation also results to substantial

CO level. Due to lambda control of the engine operation close to stoichiometric, engine

carbon monoxide emissions are not constant but follow the engine lambda target.

Hydrocarbons (HC) emissions in the exhaust of a SI engine are usually related with the

engine combustion efficiency. Rich mixtures increase HC emissions, as oxygen availability

is decreased. Flame quenching at the combustion chamber walls and the filling of crevice

volumes with air/fuel mixture which escapes the primary combustion process are the

main reasons of incomplete combustion, even at lean conditions. Additionally, at very

lean mixtures (λ > 1.2) combustion temperature drops and HC emissions increase.

In real operation of a SI engine operating homogeneously, engine-out emissions are

not constant from cycle to cycle. Mixture composition is affected by fuel and residual gas

fluctuations [20] that results to a significant perturbation of stoichiometry from cycle to

cycle. In addition perturbations of the combustion rate, maximum burned zone temper-

ature and residence time are also observed at each operating cycle. All the above reasons

result to engine emissions variability. The experimental observation and computational

prediction of homogeneous SI engines emissions variability are the main scope of this

thesis.

1.3 Thesis aim and objectives

The aim of this thesis is demonstrate the development and validation of a novel detailed

engine-out emissions model, in order to explore the cyclic emission variability of SI engines.

The work is split into the following individual objectives:

1. The development of a novel engine-out emissions model, which compared to existing

emissions models utilizes a detailed reaction scheme that includes all the potentially

possible chemical pathways for pollutant formation, particularly for NO. The model

6

1.4. Thesis structure

assumes that the burned zone of the cylinder is a homogeneous reactor and concen-

trations are kinetically controlled in the post flame region.

2. The validation of the proposed emissions model in mean IMEP cycle simulations

under various engine conditions and engine types. The model is validated against

experimental NO data found in literature from a methane fueled engine and against

experimental NO and CO data from a high speed SI engine. The model is vali-

dated under various engine operating conditions, such as engine load, speed, ignition

timing and mixture lambda.

3. The comparison of the proposed emissions model with widely used simplified emis-

sions models, which highlights the advantages of the new modeling methodology.

The comparison of the different modeling methodologies is focused on the accuracy

and computational cost of the algorithms for both NO and CO prediction.

4. The experimental investigation of the cyclic emission variability of a high speed SI

engine. The impact of various engine parameters, such as engine load, equivalence

ratio, and ignition timing on cyclic emission variability is studied. The overall objec-

tive is primarily to explain emissions variability based on observations of combustion

variability.

5. The predictability of cyclic emission variability, using the proposed emissions model

and comparison of this prediction with simplified models. As cyclic variability is

the result of fluctuations in both mixture and combustion characteristics, detailed

reaction schemes seem to be more sensitive on these perturbations and more accurate

at final prediction.

1.4 Thesis structure

The thesis starts with chapter 2, where an extensive literature review on emission modeling

and engine cyclic variability is performed. First, the main chemical pathways of pollutant

formation are reviewed and the current emission models are presented. Then, a detailed

literature investigation on cyclic variability is performed. This includes the nature, the

origins and measures of cyclic variation as well as the effects of engine variables on cyclic

variability. Finally this summarizes the limited literature on the effects of combustion

variability on emissions variation.

Chapter 3 presents the mathematical background of the proposed emissions model,

which corresponds to the objective 1 of this study. The emission model is coupled with a

two zone combustion model to predict literature experimental data from a methane fueled

research engine. The efficiency of the emission model is tested under various engine load,

7

1. Introduction

ignition timing, and equivalence ratio conditions. The validated model is utilized on a

cycle to cycle emissions analysis which is focused on determining the impact of prompt

NO formation mechanism on cyclic emissions variability.

Chapter 4 deals with the experimental investigation of cyclic dispersion in objective

4. This includes the investigation of combustion, performance, and emissions variability

under various engine operating conditions. The investigation is performed on a high speed,

port fueled injected, spark ignition engine. In this chapter it is attempted to correlate the

combustion variability with the observed emissions fluctuations.

Chapter 5 presents the validation of the proposed emissions model against simplified

emissions models in order to accomplish objectives 2, 3 and 5. First mean IMEP cycle

simulations are utilized to validate the mean NO and CO emissions. Then, the validated

model is used in a cycle to cycle simulation basis, to predict cyclic NO and CO variability.

Widely used simplified emissions models are compared to the new proposed emissions

model to highlight the benefits. The computational cost of each modeling approach is

also estimated.

Final chapter 6 summarizes the most important findings of the experimental and

modeling work, identifies the novelty and the contribution to knowledge of this thesis and

ends with some potential future work directions.

8

Chapter 2

Literature Review

2.1 Introduction

In order to better understand the process of cyclic emission variability, some of the funda-

mental theories of pollutants formation for conventional SI engines and the causes of cycle

to cycle variability are reviewed in this chapter. The current literature review gives good

insights into the chemical pathways of emissions formation as well as the mechanisms

that affect the combustion process and its variability. The main purpose of this chapter

is to review the current status regarding emissions modeling and the feasibility to predict

cyclic emissions variability, as this is described in published literature.

The chapter starts by laying the pollutants formation theory, regarding NOx, CO and

HC emissions at homogeneous charge conditions. Different types of emissions models are

reviewed to answer the question whether the accurate estimation of engine-out emissions

is primarily a thermodynamic or a kinetic issue. The sensitivity of these emission models

to residual gas fraction or burning rate is also reviewed. The second part of this chapter

includes a detailed literature investigation on cyclic variability. Firstly, the nature of cyclic

variability is defined, which is a combination of stochastic and deterministic aspects. Then

the origins and measures of cyclic variability are presented. Finally, the effects of each

parameter on cyclic variability are reviewed, such as mixture composition, small and

large scale turbulence, fuel type, engine geometry etc. The final section of this chapter

reviews the effect of combustion variability on emissions variability, an area which is of

high interest, particularly to this work. This part is the most essential of this chapter, as

cyclic emissions variability is a rather grey research area with only a handful number of

studies available in published literature.

9

2. Literature Review

2.2 Pollutants Formation Modeling

In the complete combustion of the fuel hydrocarbons, the exhaust gas consists of oxygen

(O2), nitrogen (N2), carbon dioxide (CO2) and steam (H2O), which are not detrimental

to human health. In fact, CO2 is responsible for the greenhouse effect but it doesn’t

pose a direct health hazard. However, in incomplete combustion that is observed in

actual operation of the internal combustion engines, carbon monoxide (CO), unburned

hydrocarbons (HC), nitrogen oxides (NOx) and particulate matter (PM) also appear as

exhaust gas components. These components are controlled by the emissions legislation,

as the exposure of humans to high concentrations of these species pose a direct health

hazard.

In conventional PFI SI engines, the emissions of interest are NOx, CO and HC. These

emissions vary between different engines and are dependent on such variables as ignition

timing, engine load, engine rotational speed and, in particular, air/fuel ratio. Figure 2.1

shows typical variations of the emissions with air/fuel ratio for a spark ignition engine

[16].

Figure 2.1: Spark ignition engine emissions for different air/fuel ratios [16].

The mechanisms that describe the NOx, CO and HC pollutants formation is the field

of interest of this section.

10

2.2. Pollutants Formation Modeling

2.2.1 NOx Emissions

Nitric oxide (NO) and nitrogen dioxide (NO2) are usually grouped together as nitrogen

oxides (NOx) emissions; however over 95% of a conventional SI engine NOx comes out

as NO. The emitted NO will subsequently oxidize to NO2 in the environment, and it

is the NO2 that can react with unburnt non-methane hydrocarbons in the presence of

ultra-violet light to form ozone. [17].

NO formation mechanisms are well investigated in the open literature. The most

well-known NO formation mechanism is the thermal one, usually called as the extended

Zeldovich mechanism [21, 22], which is responsible for the majority of the formed NO in

the post flame gas zone. The prompt mechanism was proposed by Fenimore [23], who

found that NO was also being formed in the flame region of hydrocarbon-air mixtures

and this mechanism dominates for close to stoichiometric and richer mixtures [24, 25], as

well as for ultra lean conditions, where the combustion temperature is extremely low [25].

More recent studies [26–28] propose a new chemical pathway for NO formation where the

N2H atoms, which are formed at slightly rich conditions, interact with O atoms. This

route has been shown to be particularly important in the combustion of hydrogen [29] and

for hydrocarbon fuels with large carbon-to-hydrogen ratios [30]. Miller et al. [31] observed

that this reaction results to a minor decrease, about 2.4% of the total NOx prediction.

The N2O related mechanism is also proposed in some studies [22, 25, 31–33] as a possible

chemical pathway for the NO formation near stoichiometric and leaner conditions, while

at richer conditions, its effect is negligible. The last way to form NO is the oxidation of

the fuel contained nitrogen that increases the final formed NO via the thermal mechanism

[21], when the fuel contains atomic nitrogen in its molecule.

NO forms in both the flame front and the post flame region. In reciprocating engines,

combustion occurs at high pressure, so that flame reaction zone is extremely thin (∼ 0.1

mm) and residence time within this zone is extremely short. Additionally, the cylinder

pressure increases during the combustion process, so that early formed burned gases are

compressed to higher burned temperature. Thus, total formedNO in the post flames gases

(thermal mechanism) dominates the flame-front NO concentration (prompt mechanism).

The thermal NO mechanism consists of three elementary reactions:

O +N2 � NO +N (2.1)

N +O2 � NO +O (2.2)

N +OH � NO +H (2.3)

The rate constant for these reactions are reported in numerous literature sources, a

few of which are summarized in table 2.1. Heywood [15] proposes that the combustion

11

2. Literature Review

Table 2.1: Speed coefficients for the forward reaction of the Zeldovich mechanism [16]

Reaction i ki,r[cm3/mol · s] Author

1

1.8 · 104 · exp[− 38, 400

T

]Baulch at al. [34]

0.544 · 1014 · T 0.1 · exp[− 38, 020

T

]GRI-MECH 3.0 [35]

0.76 · 1014 · exp[− 38, 000

T

]Heywood [15]

2

6.4 · 109 · T · exp[− 3, 150

T

]Baulch at al. [34]

9.0 · 109 · T · exp[− 3, 280

T

]GRI-MECH 3.0 [35]

1.48 · 108 · T 1.5 · exp[− 2, 860

T

]Pattas [33]

3

3.0 · 1013 Baulch at al. [34]

3.36 · 1013 · exp[− 195

T

]GRI-MECH 3.0 [35]

4.1 · 1013 Heywood [15]

and thermal NO formation are decoupled and that equation (2.4) describes the kinetically

controlled NO rate, based on the equilibrium NO value. Typical values of R1, R2 and

R3 are given by Heywood [15] and are strongly depended on mixture equivalence ratio.

d[NO

]dt

=

2R1

[1−

([NO

]/[NO

]e

)2]1 +

[NO

]/[NO

]eR1

R2 +R3

(2.4)

The formation of the prompt NO in the flame front itself is much more complicated

that thermal NO formation, because this process is closely related to the formation of

the CH radical, which can react in many ways. The general scheme of the Fenimore

mechanism is that hydrocarbon radicals react with molecular nitrogen to form amines or

cyano compounds. The amines and cyano compounds are then converted to intermediate

compounds that ultimately form NO.

12

2.2. Pollutants Formation Modeling

CH +N2 � HCN +N (2.5)

C +N2 � CN +N (2.6)

Ignoring the processes that form CH radicals to initiate the mechanism, the prompt

mechanism is described by equation 2.7. For equivalence ratios higher that 1.2, other

routes open up and the chemistry becomes more complex [16, 24].

d[NO

]dt

= k10,r[HCN ][O] + k11,r[CN ][O2] (2.7)

k10,r = 2.3 · 104 · T 1.71exp

(− 3, 521

T

)m3

kmol · s(2.8)

k11,r = 8.7 · 109 · T 1.71exp

(216

T

)m3

kmol · s(2.9)

The N2O intermediate mechanism is important in fuel lean mixtures and low temper-

ature conditions. The three steps of this mechanism are:

O +N2 +M � N2O +M (2.10)

H +N2O � NO +NH (2.11)

O +N2O � NO +NO (2.12)

The N2H mechanism is a two key step mechanism, particularly important in the

combustion of fuel rich hydrogen mixtures [36]:

N2 +H → NNH (2.13)

NNH +O → NO +NH (2.14)

Finally, in the combustion of fuels with bound nitrogen, the nitrogen in the parent

fuel is rapidly converted to hydrogen cyanide (HCN) or ammonia (NH3), following the

above described reaction pathways.

13

2. Literature Review

2.2.2 CO Emissions

Carbon monoxide (CO) is abundant in rich-combustion products [15]. In SI engines, rich

mixtures are employed during engine startup to prevent stalling, as well as at wide open

throttle (WOT) to provide maximum power. At stoichiometric and slightly lean mixtures,

CO is produced at typical combustion temperatures as a result of the dissociation of

CO2. Turns [37] describes the relationship between CO and mixture temperature, where

carbon monoxide concentration rapidly falls as equilibrium temperature is decreased.

In SI engines, temperature falls rapidly during the expansion and exhaust process and

equilibrium is likely not to prevail. Conversely, CO exhaust concentration is “frozen”

between the procedures of combustion process and expansion/exhaust blowdown processes

[38, 39].

The observation that, in the combustion of premixed hydrocarbon-air mixtures, CO

concentration differs from the equilibrium value, implies that CO level is kinetically con-

trolled [15, 40]. The most important reaction that controls the CO exhaust concentration

is the Water Gas Shift (WGS) reaction (2.15), and appears to be significant in the post-

flame:

CO +OH � CO2 +H (2.15)

Calculations indicate that the WGS reaction is equilibrated during the expansion

and exhaust blowdown processes; however the three-body recombination processes (e.g

H + OH + M � H2O + M) are not fast enough to maintain equilibrium among all the

radicals [37]. This causes the partial-equilibrium CO concentrations to be well above

those of full equilibrium [39, 41].

A widely used model for CO emissions is given by D’Errico et al. [40, 42]. This model

uses the WGS reaction, given at Eq. (2.15), and expresses the resulting CO reaction rate

as follows:

d[CO]

dt=(R1 +R2

)(1−

[CO][

CO]e

)(2.16)

where [CO]e is the predicted equilibrium concentration of CO, and values for the rates

R1 and R2 are given by Heywood [15] and D’Errico et al. [40, 42]. This model appears

more accurate compared to the equilibrium assumption [43], however its kinetics needs

calibration for different type of engines.

Other CO production mechanisms include quenching by cold surface, where CO is

formed though the WGS reaction as well as partial oxidation of unburned fuel, in crevices.

Both these two mechanisms produce mainly hydrocarbons and are described in detail in

the next section.

14

2.2. Pollutants Formation Modeling

2.2.3 HC Emissions

Hydrocarbons (HC) are the consequence of incomplete combustion of the hydrocarbon

fuel. In the exhaust of the vehicles, unburned hydrocarbons are measured as total hydro-

carbon concentration in C1 base, usually in parts per million (ppm). Unburned hydrocar-

bon levels in the exhaust of a SI engine under normal operating conditions are typically in

the range of 1000 to 3000 ppm C1, which correspond to between about 1 and 2.5% of the

fuel flow into the engine [15]. The hydrocarbon profile in the exhaust of engines consists

of a combination of paraffins, olefins, acetylene and aromatics compounds, as a result of

unburned or partially burned fuel hydrocarbons and engine oil. Polyaromatic species are

also present in less abundance.

Heywood’s textbook [15] provides four possible HC emissions formation mechanisms

for conventional SI engines: a) flame quenching at the cold surfaces of the combustion

chamber; b) the contribution of crevice volume with unburned mixture that due to flame

quenching can not follow the main combustion rate; c) absorption of fuel vapor into oil

layers during compression stroke; a fraction of which is desorbed during the expansion

and the exhaust blowflow; and finally d) incomplete flame propagation in the bulk of the

charge.

2.2.4 Pollutants Formation Models

The previous section presented the most important formation mechanisms for main pollu-

tants emitted by conventional SI engines, such as NOx, CO, HC. These mechanisms can

be classified into mainly physical or chemical processes. Reaction mechanisms are suitable

to describe in detail the overall chemical reaction pathway; that is the formation of NOx

and CO emissions, as HC emissions are more a complex physical issue, as it was described

above. Such reaction mechanisms can consist of a single or limited steps of elementary

reactions or a detailed reaction scheme. In this study, it is adopted that the first case

belongs to the category of simplified kinetic models and the second to the detailed kinetic

models. Both models can be coupled with combustion models, as their input requires the

temperature and pressure cylinder profile and some other cylinder mixture characteris-

tics, such as A/F ratio and residual gas fraction. Simplified kinetic models exhibit short

execution time and reduced requirements for input information, but they suffer from low

sensitivity to various engine conditions and they require calibration. Conversely, detailed

kinetic models are computationally more expensive and require more input data; however

they can give a better picture of the step by step sequence of the elementary reactions,

they can more accurately estimate the exhaust concentration of many chemical species

and they do not need any calibration. In several studies simplified or detailed kinetic

models are utilized, the results of which are reviewed in this section.

15

2. Literature Review

2.2.4.1 Simplified Kinetic Models

Many researchers in the past have tried to answer the question whether the engine out

pollutants formation is primarily a thermodynamic or a chemistry problem. Adopting

temperature stratification during combustion can improve the predicted in-cylinder NO

concentration [32, 44]. In these studies, it was observed that the extended Zeldovich

mechanism is able to predict NO emissions within an error of 15-60%. This result implies

that the NO simulation is not just a temperature issue, but also a significant chemistry

problem. The first NO kinetic control mechanisms [21, 22, 33] were utilized at older

studies, instead of the inaccurate assumption of equilibrium [43].

In 1979, Heywood et al. [45] proposed a thermodynamic engine cycle simulation

and a chemical model to predict NOx emissions from a conventional SI engine. The

required inputs were engine geometry, engine speed, overall equivalence ratio, residual

gas fraction and intake manifold pressure and temperature. For the heat release, Vibe

model was applied [46]. NOx emissions are calculated by using the extended Zeldovich

kinetic scheme, with the steady state assumption for the N concentration and equilibrium

values used for H, O, O2 and OH concentrations in the adiabatic core. A three zone

combustion model was used to describe the combustion process: an adiabatic burned

zone, an unburned zone and a burned gas boundary layer where combustion occurs.

The simulation was compared against experimental data from a research single cylinder

SI engine. It was found that NOx emissions are very sensitive to equivalence ratio,

residual gas fraction and combustion duration. Although further comparison between

experimental and simulation NOx data is not given, this study is important as it is

the first cycle simulation investigation in published literature, including performance and

NOx emissions calculations.

In a more recent study, D’Errico et al. [40] presented a pollutant emissions model

for SI engines, which included an eddy burning model for the accurate prediction of heat

release, the extended Zeldovich mechanism for the prediction of NOx emissions, a two

step reactions CO model and a single Arrhenius equation for HC post oxidation, given

by Lavoie and Blumberg [47]. The temperature distribution is achieved by dividing the

mixture into an adiabatic core and a thermal boundary layer, where this core was further

divided into an equal mass of adiabatic zones. The emission model was calibrated, based

on one engine operating point. The simulation values were validated against experimental

data from a 2.0L Fiat-Alfa Romeo four cylinder engine. On the results of this study, it

was found that the model was able to predict engine-out emissions qualitatively and

quantitatively at full load conditions over the entire engine speed range. However, the

study does not present the validation of the model on partial load conditions.

The multizoning technique is performed for the more accurate prediction of the tem-

perature distribution, which significantly affects the chemical kinetics and finally the es-

timated exhaust emissions. Michos et al. [44] utilized a multizone turbulent combustion

16

2.2. Pollutants Formation Modeling

model in coupling with the extended Zeldovich mechanism to predict NOx emissions.

In this model, the number of zones are user defined and each zone exchanges heat with

the chamber wall but zone mass remains constant. They found that by increasing the

number of zones, NO error estimation initially decreases and then slightly increases and

converges to a constant value after the fifth zone assumption. At partial engine load, NO

error is higher while this deviation decreases as engine load increases. The five burned

zone emission model was able to qualitatively predict the measured NOx concentrations

at various engine load conditions but not quantitatively.

Finally, Richard et al. [48] used a physical 0D combustion model for simulating heat

release and simple kinetics to predict NO and CO emissions. The experimental data

were from a turbocharged SI engine. They noticed that the reduced physical combustion

model was able to accurately predict pressure rise and heat release, although engine

emissions were not well predicted. Under the entire engine speed range, NOx emissions

were underestimated at partial load and slightly overestimated at full load conditions,

while CO emissions were consistently underestimated and the error was increased at

higher engine speed conditions. This last study is informative, as the same experimental

data and the same combustion model were used in coupling with a detailed kinetic model

and much more accurate emissions prediction [49]. These results are discussed in the next

section.

2.2.4.2 Detailed Kinetic Models

Detailed chemistry models contain more reaction pathways compared to simplified kinetics

models; therefore they are able to predict and explain NO formation in a more analytical

way. A lot of experimental and theoretical studies present the significant impact of residual

gas fraction on charge dilution, which finally steeply decreases NO formation [50–52].

However, it was found that simplified kinetic models overestimate NO from 12 to 25%

when engine load increases and residual fraction drops [31, 32, 53]. Burn rate is another

critical engine variable where NO prediction is very sensitive on [32]. Cyclic combustion

variability studies can be used to examine fast and slow burning rates on NO formation

[54]; however it is known that by only changing the burning rate is not possible to predict

the scatter of NO variability with a simplified NO kinetic model [13]. Besides, the

most important feature of detailed chemistry models for realistic NO predictions is the

capability to predict NO and CO emissions with a good accuracy under various engine

operating conditions and without any calibration methodology.

Miller et al. [31] proposed a super extended Zeldovich mechanism (SEZM) for the pre-

diction of NOx, which consists of 67 reactions and 13 chemical species, while it assumes

equilibrium concentrations for all other species. Steady-state engine cycle modeling was

performed by GESIM, a quasi-dimensional cycle simulation which takes into consideration

turbulent entrainment and eddy burnup, swirl, tumble, chamber geometry on flame de-

17

2. Literature Review

velopment, burn rate, and fuel consumption. In conjunction with the combustion model,

a quasi one-dimensional dynamics model (MANDY) for the wave dynamics in the pipes

was also applied in this study. The target of this study was to improve the NOx predic-

tion, as the extended Zeldovich mechanism over predicted NOx by more than 25%. It

was found that both termolecular/unimolecular reactions and N2H chemistry give a 2.4%

reduction on observed NOx; however, predicted NOx levels are in excess of the experi-

mental data by more than 20%. This observation could be explained by pressure effects

on the radical pool, which was assumed to be in equilibrium in this study. In addition,

the proposed modified pressure dependent reaction one rate of the Zeldovich mechanism

shows improved accuracy on NOx prediction as a function of load. Last but not least,

the proposed super extended Zeldovich model presented good accuracy within 10% for

lean, rich and EGR diluted mixtures.

Rublewski and Heywood [32] tried to answer to the question if accurately prediction

of engine out NO concentration is primarily a thermodynamic problem, or a chemistry

problem. Therefore, temperature stratification during combustion in combination with

a crevice/combustion inefficiency routine was applied in their modeling. This multizone

approach with an adiabatic core improves the NO simulation; however the NO formation

was found to be mainly affected by the residual gas fraction and the burn rate. In addition,

it was shown that model accuracy would be modestly improved if the equilibrium burned

gas radical pool assumption was removed and a kinetically controlled radical pool was

adopted. Last but not least, in this study it was found that theN2O mechanism little affect

the NO formation, although other researchers [22, 31] suggested that this mechanism can

be significant under highly dilute conditions.

Last but not least, Bougrine et al. [49] introduced a tabulated chemistry approach in

coupling with a detailed chemistry model (SENKIN) to predict NO and CO emissions.

The initial values are given by equilibrium, while detailed chemistry calculates the differ-

ence of the equilibrated value from the kinetic value. Compared with the results of their

previous investigation [48], the new emission approach shows smaller errors, especially

for the estimation of CO emissions. However, NOx emissions demonstrate that errors in

calculations may occur due to temperature stratification and the equilibrium assumption

in the early stage.

2.3 Cycle-to-Cycle Variability

Cyclic combustion variability is experimentally identified by the variation of cylinder

pressure evolution from one cycle to another. Since the pressure development is closely

related to the combustion process, substantial variations in the combustion process on

a cycle to cycle basis are occurring. CCV has a negative effect on engine performance,

fuel consumption as well as engine-out emissions [10, 12, 55, 56]. The nature, the origins

18

2.3. Cycle-to-Cycle Variability

and the indicators of cyclic combustion variability are discussed in this section. Finally,

a literature review is performed on the engine parameters that affect cyclic variation.

2.3.1 The Nature of Cyclic Variation

Cyclic combustion variability is experimentally identified by the variation of cylinder

pressure evolution from one cycle to the next [57], as a result of both stochastic and

deterministic combustion related effects [58, 59]. Although much of the research on cyclic

variability has focused on stochastic aspects, such as two well known review studies on

cyclic variability [10, 55], it is widely accepted that observed cyclic variations can present

a high degree of low-dimensional deterministic structure [58]. This deterministic aspect

implies some degree of predictability and potential for real-time control [60].

In order to better understand the nature of cyclic variability, it is of interest to de-

fine the terms ”stochastic” and ”deterministic”. A stochastic size takes a random value

between the minimum and the maximum limits, which can be predicted in a probability

sense by using a probability density function. In the case of combustion variability, small-

scale and large-scale turbulence of the gas mixture is known for its stochastic nature. On

the other hand, the term ”deterministic” is used for a system that follows explicit laws

of cause and effect, such that if we know precisely what the initial conditions are, we can

predict the state of the system at any time in the future [60]. Regarding CCV, residual

gas fraction is a deterministic parameter, as the residual gas of the previous cycle affects

the combustion and emissions of the next, which implies a strong ”memory” effect from

cycle to cycle.

The stochastic and deterministic dimensions of cyclic variability have been adopted

by many researchers in the past, in order to model the cyclic variations in spark ignition

engines, although not by using exactly the same terminology. Dai et al. [61] classified the

causes of cyclic variations into two major groups: (a) prior-cycle effects and (b) same-

cycle effects. The prior-cycle effects stands for residual gas variation in the combustion

chamber, whereas the same-cycle effects refer to variations in air/fuel mixtures in the

cylinder.

The main difference between the stochastic and deterministic aspects of cyclic variabil-

ity is that the deterministic structure can be potentially limited through engine control.

Therefore, exploring the deterministic nature of cyclic variability gives the potential op-

portunity to engine manufactures of further improving the engines emissions and fuel

economy by the limitation of the deterministic nature of CCV. In fact, the deterministic

nature of cyclic variability can be highlighted close to the unstable combustion limits

[61–63]; that is at lean, non-homogeneous charged and high diluted engine conditions. At

this combustion area, limitation of the deterministic combustion variations can extend the

engine limits to leaner and more diluted mixtures, while this can benefit the engine ther-

19

2. Literature Review

mal efficiency. As cyclic variability is a nonlinear complex phenomenon with very large

degrees of freedom, the control of such systems can be performed by the chaos theory

[64, 65].

2.3.2 Origins of Cyclic Variation

Cycle-to-cycle variability (CCV) is an undesirable characteristic of internal combustion

engines. This is caused by the non identical in-cylinder conditions from one combustion

cycle to the other, at nominally identical operating conditions. As air pollutant and green-

house gas emissions standards are getting more stringent around the world, mitigation of

CCV can improve fuel efficiency and minimize engine emissions. In addition, limitation

of CCV allows better engine calibration and enables performance close to the optimum

conditions [10].

The origins of CCV in a spark-ignition (SI) engine can be summarized as follows

[10, 15, 17, 63]:

i the variations in the gas motion and the mixture strength near the spark plug location

at time of ignition [66],

ii the variation in air-fuel ratio supplied to each cycle [20, 67–69]

iii non-uniform mixture composition in the cylinder due to incomplete mixing between

air, fuel, and residual gas [68, 70]

iv ability of different ignition systems to ignite the local mixture [68, 71–74]

Early flame kernel development corresponds to the combustion process from SI initi-

ation to the transition to a fully turbulent combustion. This early burn period, which is

often approximated by the 5% mass fraction burned (MFB) limit, is recognized as the

most crucial stage of the combustion evolution [75, 76]. Early flame kernel development

greatly affects the development of combustion and build-up of cylinder pressure [77, 78].

Slow early burning rate cycles continue to burn slowly during flame propagation and sub-

sequently lead to lower peak pressures [79]. Actually, the initial laminar flame initiated

by the spark has been found to be very repeatable. Curry [79] found that the burning

rate was very repeatable within some 5 millimeters from the spark plug, and significant

deviations occurred only thereafter. Variations in the early flame kernel development

should not so much depend on spark intensity, as long as the mixture can be ignited [20],

but on follow up in-cylinder conditions. However, in case that the mixture is not easily

ignited, such as very lean and highly diluted mixtures, the spark ignition energy has a

remarkable impact on the combustion process [80].

20

2.3. Cycle-to-Cycle Variability

Cylinder mixture composition also affects peak cylinder pressure. Slightly rich mix-

tures exhibit higher burning rates, higher peak pressure and the minimum heat release

variability [18]. Cyclic variability in the fuel flow and of water content in the inlet charge

were strongly linked to combustion variability [81]. Grunefeld et al. [20] also showed that

high combustion pressure of the preceding cycle results to a higher quantity of residual

gas and a lower combustion pressure in the following cycle. Local mixture composition

also affects burning rate. Experimental studies on a lean-burn SI engine showed that

locally rich mixtures around the spark plug give fast burn cycles, while too lean mixtures

may not even ignite [11, 82].

Turbulence is also a major cause of combustion variability. It is divided into large-

scale (mean) and small-scale motions where the first affects flame propagation [83] and the

second one affects mostly the initial flame kernel [14]. Many engine parameters have an

effect on mixture turbulence but this primarily depends on engine speed. By increasing

engine speed, mixture turbulence is also increased which results to higher flame speed

variations and finally higher combustion variability [12, 18].

Turbulence or fuel charging (mixture lambda) are stochastic aspects of cyclic combus-

tion variability. However, cyclic variability also originates from deterministic aspects, such

as the residual gas of the previous cycle, due to which the combustion process presents

a strong ”memory” effect from one cycle to the next. Daw et al. [58] found that the

primary mechanism for deterministic oscillations in dilute combustion was the highly

nonlinear impact of residual gases left behind from a previous cycle on the combustion

rate in a succeeding cycle. In order for this effect to be significant, they noticed that the

nominal in-cylinder charge dilution level had to be close to the dilute flammability limit.

On this evidence it can be concluded that the major cause of cyclic combustion varia-

tion originates from the early flame kernel period, while the nature of this effect on initial

flame is both stochastic and deterministic. Moreover, a great number of engine param-

eters are affected by cyclic combustion variability. Therefore, the next chapter attempts

to distinguish the most useful indicators of cyclic variability, which are also utilized later

on in the current work.

2.3.3 Measures of Cyclic Variation

Cyclic combustion variation can be experimentally identified by using different measuring

techniques, such as the following:

i ionization probes [84] or laser beams [85] to measure flame front area, flame radius

and flame arrival at given location,

ii high speed cameras or Schlieren photography [67, 71] in order to capture different

combustion frames,

21

2. Literature Review

iii piezoelectric crystal for the measurement of the cylinder pressure over crankangle.

In most studies two techniques are applied simultaneously, in order to identify cyclic

cylinder pressure and flame variation. Based on Matekunas [71] and Heywood [15], the

cyclic variations can be grouped into four main categories, depending on the variation of

different type of parameters. The following quantities have been used [10]:

1. Pressure-related parameters. The maximum cylinder pressure Pmax; the crank

angle at which this maximum pressure occurs ϑPmax; the maximum rate of pressure

rise (dP

dϑ)max; the crank angle at which the maximum rate of pressure rise occurs

ϑ(dP

dϑ)max

; the IMEP of the individual cycles.

2. Combustion-related parameters. The maximum rate of heat release (net or gross);

the maximum rate of mass burning or maximum rateof change of burnt mass frac-

tion in the cylinder (dxbdϑ

)max; the early flame kernel development angle ϑ5%; the

combustion duration ϑ5−90%.

3. Flame front-related parameters. Flame radius, flame front area enflamed or burned

volume at given times; flame arrival time at given locations.

4. Exhaust gas-related parameters. Concentration of different components in the ex-

haust gases.

One important measure of cyclic variability, derived from pressure data, is coefficient

of variation of indicated mean effective pressure. It is defined as the standard deviation

in imep divided by the mean imep, and is usually expressed in percent:

COVimep =σimepimep

× 100 (2.17)

It defines the cyclic variability in indicated work per cycle and it has been found that

vehicle drivability problems usually result when COVimep exceeds about 10 percent. The

same definition of COV is also used for the other parameters defined above.

Brown et al. [86] conducted tests on a research Ricardo E6 engine at various engine

loads, ignition timings and equivalence ratio conditions. They have noticed that the cycle

to cycle variations in combustion should be characterized by the COV of the imep. Often

the COV of the maximum cylinder pressure (Pmax) is used since it is easier to quantify.

However, it has been proved in this study that this can lead to misleading results when

the ignition timing is varied in the region of the MBT ignition timing. It has also been

demonstrated that when the ignition timing is varied there is no correlation between the

COV of the imep and COV of the maximum cylinder pressure. The experimental data

22

2.3. Cycle-to-Cycle Variability

also show that the COV of the imep is at minimum in the region of the MBT ignition

timing, and this also corresponds to a minimum in the 0-10 % burn time.

The most characteristic relationships between maximum cylinder pressure (Pmax),

crank angle that maximum cylinder pressure occurs (ϑPmax) and imep are described by

Matekunas [71]. In that study, a single cylinder flame photography engine was employed

under various ignition timing and equivalence ratio conditions at constant inlet manifold

pressure (55 kPa). It was found that as the spark timing approaches the MBT conditions,

the relative variations in imep are very small as compared with the variations in ϑPmax,

and become more significant with advanced ignition timing.

Early flame kernel development also presents a strong relationship with imep. Aleiferis

et al. [80] experimentally investigated the flame behavior in a lean-burn optical stratified-

charge single cylinder SI engine. A strong relationship between imep and ϑ5% was ob-

served, especially at lean operating conditions. However, ϑ5% is not a very informative

parameter, as it does not provide any information about the reasons for early combustion

from cycle to cycle. Analysis of the projected flame-front shape and its wrinkling (small

scale) showed that the fastest lean flames were associated with a large standard devia-

tion (σd) of the local displacement between the instantaneous and filtered flame contours

and with a small number of crossing points (M) between the instantaneous and filtered

contours. Additionally, the fastest lean flames possessed values of d and M similar to

those measured from the images of stoichiometric flames. This suggested that the fastest

flames on a cycle-by-cycle basis might have been burning with a richer than average A/F

ratio around the spark plug. The measured shape factors (large scale) did not show any

strong correlation between the crank angle of 5% mass fraction burned and the ratio of

the length of the instantaneous flame contour to the length of the filtered flame contour,

or to the perimeter of a circle with the same radius to the equivalent radius of the flame.

As it follows from the relevant discussion, a limited number of parameters are preferred

as CCV indicators. The most widely used quantities are the pressure related parameters

( Pmax, ϑPmax and imep), while flame radius and displacement of the flame kernel center

from the spark plug center can also be found in a number of studies. The first three are

the easiest to acquire, as the experimental equipment which is needed is only the pressure

transducer. When ignition timing is set to MBT, using Pmax as the CCV indicator will be

the most prominent among others. However, if the interest is limited to CCV in the stage

of the early flame kernel development, then the most suitable parameter would be ϑPmax.

The imep is most useful for indicating the engine response to CCV in the combustion

process. The two flame front-related parameters seem to provide valuable information for

a better understanding of the in-cylinder combustion and the fundamental reasons for the

cyclic variations.

23

2. Literature Review

2.3.4 Effects of Engine Variables on Cyclic Variability

Many studies investigate the impact of various engine variables on combustion variation.

Young [87] categorized engine operating and design variables into two main groups, i.e.

chemical and physical factors.

Chemical factors are those variables that determine the composition of the trapped

cylinder charge. Such factors are equivalence ratio, the residual gas that dilutes the

charged mixture and the fuel type. These factors were thought to influence cycle-to-cycle

variations by affecting laminar flame speeds and expansion velocity. In fact, equivalence

ratio and residual gas fraction are factors that mainly relate with the deterministic na-

ture of cyclic variability; therefore their control is of primary importance as it gives the

potential opportunity to improve fuel economy and engine-out emissions.

On the other hand, physical factors are related with the physical conditions under

which combustion proceeds, and can also affect the cycle-to-cycle combustion variations.

These factors are combustion chamber geometry, ignition system, compression ratio, en-

gine speed, mixture preparation and motion.

The factors that influence cyclic variability are reviewed in this section.

2.3.4.1 Equivalence Ratio

Equivalence ratio is highlighted as the mixture parameter that has a strong influence on

cyclic combustion variability. A number of CCV studies have investigated the impact of

equivalence ratio on combustion variability. Slightly rich mixtures exhibit higher burning

rates, higher peak pressure and the minimum heat release variability [80]. Generally,

the minimum cyclic variations occur at slightly rich equivalence ratios corresponding

approximately to the mixture strength for best power [87].

Using ionization gaps located 16.5 mm (probe 1) and 65 mm (probe 2) from the spark

plug, Harrow and Orraan [22] observed minimum variations in flame arrival times at

1.15 to 1.25 equivalence ratio (13 to 12 air-fuel ratio (A/F)), as shown in Fig. 3a. The

variations in arrival times were characterized by the time interval between the arrival of

10% and 90% of the cycles monitored. Minimum variations in arrival times were 0.49

and 0.35 at probes 1 and 2, respectively. For a normally distributed random variable,

this spread would correspond to 2.45 times the standard deviation. Flame speeds, calcu-

lated from arrival times at the two probes, were showed highest values at approximately

1.25 equivalence ratio (12 A/F) for engine speeds from 750 to 2500 r/min. Warren and

Hinkamp [23] noted similar variations of flame arrival times at a distance of 73 mm from

the spark gap as a function of air-fuel ratio. Test results showed that minimum variations

in flame arrival times occurred at about the mixture strength for maximum power, 13:1

A/F (1.15 equivalence ratio). As a fraction of the mean arrival time, the minimum spread

was approximately 0.38. A non-normal distribution in arrival times was observed. In

24

2.3. Cycle-to-Cycle Variability

both studies, minimum variation in arrival times occurred where the mean arrival times

were at minimum, i.e., flame speeds were highest.

Matthew et al. [88] investigated the impact of equivalence ratio on cyclic variability

using both an SI engine model and experimental data. The SI engine model was a quasi-

dimensional code that utilized the concepts of fractal geometry to simulate the effects of

turbulence on the flame propagation process. A single cylinder research engine, fueled

with propane, operated at 600 rpm and at constant intake manifold conditions (170kPa,

65degC) was utilized for the model validation. It was found that the equivalence ratio

affects flame stretch and the distribution of the flame wrinkling scales. In fact, variations

in equivalence ratio also result in variations of the residual gas fraction and therefore

independent effects of variations in the equivalence ratio and in the residual fraction have

not been examined. Variations in the equivalence ratio have a dominant effect on cyclic

variability in the peak cylinder pressure, but have little effect on the crank angle that

peak pressure occurs and an insignificant effect on the early flame kernel development.

Although the early flame kernel is controlled by laminar flame chemistry, it was found that

if the engine conditions are sufficiently turbulent, flame wrinkling can begin to dominate

very early in the combustion process.

2.3.4.2 Fuel Type

Different types of fuel can be injected in SI engines, either in liquid or in gaseous phase.

Gasoline, iso-octane, methane, propane, hydrogen, liquid petroleum gas (LPG) and com-

pressed natural gas (CNG) are the most widely used fuels in the open literature. Each

of these fuels present different chemical and physical properties, while the fuel type also

affects the combustion process, its variability as well as the pollutants formation. How-

ever, it is recognized that fuels which are rich in hydrogen present faster burning rates

and lower combustion variability due to the high flame velocity of H2.

The first study that presents the impact of fuel type on cyclic variability was performed

by Soltau [89]. In these test, a variety of liquid fuels (gasoline, benzene, and methanol)

and gaseous fuels (town gas, i.e., coal gas, butane, and methane) were experimentally

investigated. The single cylinder research engine had a compression ratio of 9:1, while

engine speed was varied from 500 to 3500 rpm with throttle setting. On the results, it was

found that town gas which was hydrogen-rich presented the lower combustion variability.

With the exception of town gas, no significant difference in combustion variation was

observed among the other fuels.

Sen et al. [90] investigated the effect of hydrogen addition on cyclic variability in a

natural gas SI engine. The engine was a three cylinder one, operating at 3000 rpm and

an excess air ratio of 1.4. Three cases were examined: natural gas with no hydrogen

added, and natural gas with the addition of 23% and 40% hydrogen by volume, while

192 consecutive cycles were analyzed, using the wavelet multiresolution analysis. It was

25

2. Literature Review

found that by increasing the hydrogen volume in the fuel, this had a pronounced effect on

reducing the cyclic variability of the imep. Finally, it was observed that the 23% hydrogen

by volume mixture contributed more to the limitation of CCV, while the limitation of

CCV from the 23% to the 40% hydrogen by volume mixture was lower.

Wang et al. [91] examined the cyclic variation in a hydrogen-enriched gasoline SI

engine under various hydrogen volume fractions. The test was carried out on a modified

four-cylinder gasoline engine equipped with an electronically controlled hydrogen injection

system. The engine was operated at 1400 rpm, on different hydrogen volume fractions in

the total intake, excess air ratios, spark timings and manifold absolute pressures. The test

results demonstrated that the coefficient of variation in indicated mean effective pressure

was distinctly decreased with the increase of hydrogen blending ratio. At manifold abso-

lute pressure of 61.5 kPa, the relevant excess air ratio for the engine lean burn limit was

extended from 1.45 to 2.55 when the hydrogen volume fraction in the intake was raised

from 0% to 4.5%. By performing the same investigation at 790 rpm, it was found that

hydrogen addition was more effective on reducing engine cyclic variation at low loads.

2.3.4.3 Combustion Chamber Geometry

The combustion chamber geometry is mainly refers to the chamber shape in conjunction

with the spark plug location and affects primarily the length of the main combustion

period and secondarily its variation from cycle to cycle [87]. The most effective combustion

chamber geometry that improves engine stability is the design that reduces the combustion

process duration (on a crankangle basis) through mainly chamber openness and turbulence

- generating features.

Young [55] used four variables to describe the combustion chamber geometry: (a)

intake valve area, (b) squish area, (c) maximum flame travel distance and (d) a chamber

openness parameter. Only squish area and a chamber openness parameter, were found to

be important. Squish represents the normalized maximum flame-travel distance at TDC

and was computed for each chamber from measurements of piston - crown and cylinder-

head contours. The openness parameter is related to the accessibility of the combustion

chamber volume to a spherical flame propagating from the spark plug location, and is

influenced by both the shape of the combustion chamber and the location of the spark plug.

Figure 2.2 illustrates the cross-sections of the three combustion chambers investigated in

this study. The more desirable location of the spark plug is given for the case of the

opened chamber, which presented a faster burning rate, due to the more spherical flame

propagating from the spark plug.

The impact of combustion chamber geometry on cyclic variability and engine stability

was also validated by Lucas and Brunt [92]. In this experimental study, the effects of

various combustion chamber designs on combustion rate in an SI engine were explored,

by measuring the flame propagation rate and cylinder pressure. It was noticed that

26

2.3. Cycle-to-Cycle Variability

Figure 2.2: Cross-sections of combustion chambers investigated in Young [55] with various

spark locations.

combustion chamber shape has a trivial effect on flame speed; however the combustion

rate and cyclic variation (measured as COVPmax) were greatly affected.

Chanchaona [93] studied the effects of a disc and a bowl piston in a natural gas SI

engine. The engine was a single cylinder Ricardo E6, running over a range of compression

ratios and operating conditions. It was detected that the bowl piston presented a slightly

higher resistance to knock than the ”flat” piston. Additionally, imep variations for bowl

piston combustion chambers were low and slightly affected by increasing compression

ratio, while the disc piston presented much higher variability at higher compression ratios.

2.3.4.4 Compression Ratio

Compression ratio in SI engines is directly related with the thermal efficiency of the

thermodynamic operating cycle. By increasing compression ratio and by keeping the

same ignition timing, the initial temperature and pressure conditions are higher, which

finally leads to higher average laminar flame speed. Ozdor et al. [10] mentions that a

higher compression ratio would improve engine thermodynamic efficiency, but is limited

by the fast burn cycles. for which the knock tendency is higher.

Chanchaona [93] tested a variety of combustion chambers at various compression ra-

tios. He found that COV imep is not essentially affected by compression ratio but by

27

2. Literature Review

combustion chamber shape, since increasing compression ratio for various combustion

chambers gave inconsistent results for imep variation. Altering compression ratio results

in distorting the combustion chamber geometry and flame propagation patterns.

2.4 Cyclic Emissions Variability

Cyclic dispersion of mixture composition and turbulence, even at nominally steady state

operating conditions, are linked to fluctuations on burning rate and finally on cycle to cycle

variation of engine-out emissions. As mixture characteristics and combustion evolution

seems to affect each other, a deeper understanding of the crucial parameters that affect

pollutants formation can be given by engine simulation [12, 31, 32, 44, 45, 53].

Despite combustion CCV has been studied rather extensively, its effect on engine

emissions is not well understood. Milkins et al. [94] found that the mean combustion

cycle corresponds to relatively low CO and HC emissions while faster and slower burning

cycles always lead to higher emissions. It is known from mean cycle studies that at partial

load engine conditions, CO emissions are correlated with heat release characteristics of

premixed burn and at higher load conditions, CO emissions are mainly controlled by

mixing [95]. However the impact of these parameters on the scatter of CO variability

has not been investigated in literature. On the other hand, NO formation is linked to

oxygen availability, burning rate and maximum cylinder pressure [54, 96, 97]. Cyclic

variations of combustion and pressure development [98], cyclic fluctuations of mixture

stoichiometry [99], and heterogeneities of cylinder charge [70] are the main sources of NO

variation. One study [12] showed that NO formation presents a non-linear correlation

with indicated mean effective pressure (IMEP). However, a detailed experimental study

that investigates the impact of various engine parameters such as stoichiometry, load or

ignition timing on NO cycle-to-cycle variation has yet to be conducted.

Ball et al. [13] presented a methodology for the estimation of NO formation on

a cycle to cycle basis. The experimental data consisted of cyclic resolved NO values

obtained with a fast CLD, while the modeled data were calculated using the extended

Zeldovich mechanism. Wiebe model was used for the prediction of the heat release on the

thermodynamic model, while combustion duration and completeness of combustion were

obtained by the measured cylinder pressure, on a cycle to cycle basis. The results of the

NO modeling showed good agreement with experimental data, in terms of the mean NO

level. However, the modeled data showed less scatter in the plots than the experimental

data, as residual gas fraction, temperature at IVC and mixture lambda were not cyclic

varied.

28

2.5. Summary and Conclusions

2.5 Summary and Conclusions

Cyclic combustion variation in SI engines occurs in all operating conditions and becomes

worse when engines operate under lean and highly diluted conditions. The main impact of

cyclic variability is on early flame kernel development, that affects the combustion process.

The control of the deterministic aspect of cyclic variability, which is mainly defined as the

elimination of the residual gas fraction and its variability, could lead to improvements on

fuel consumption and to lower engine-out emissions.

Pollutants formation can be described by chemical or physical submodels. Reaction

mechanisms are suitable to describe in detail the overall chemical reaction pathway. Con-

versely to simplified kinetic models, detailed kinetic models describe better the step by

step sequence of the elementary reactions, they calculate accurately the exhaust concen-

tration of the total chemical species and they do not need a separate calibration. Such

models present all the necessary physical and chemical properties for accurate estimation

of cyclic emission variability.

Although the engine parameters that affect the pollutants formation are widely known,

cyclic emission variability is a gray aspect of internal combustion engine modeling. Emis-

sions prediction is not always accurate due to modeling assumptions and the nature of

cyclic variability. The limited literature on emissions variability proves that this topic has

to be better investigated, as automotive industry is under stricter emission legislation.

29

2. Literature Review

30

Chapter 3

Proposed Emissions Model

3.1 Introduction

In this chapter, an engine-out emissions model for the prediction of NO formation in

homogeneous engine conditions is presented. The model can be coupled with any multi-

zone thermodynamic model, where the burned zones are assumed as perfectly-stirred

reactors. In this study, a two zone approach is adopted, which consists of an unburned

and a burned region. The computational time of the proposed detailed chemistry model

is slightly increased but at the same order of magnitude to the Zeldovich mechanism. The

basic benefits of this detailed chemistry model are its simplicity and the absence of any

calibration parameter that makes it a generalized application, possible to be applied in

any type of homogeneous combustion engine.

This chapter aims to investigate the impact of a detailed chemical mechanism on NO

prediction. The presented detailed chemistry model is coupled with an existing two zone

combustion model. The Vibe model [46] is used for the prediction of heat release; then the

calculated thermodynamic data are fed into the detailed chemistry model to predict NO

evolution at each degree of crank angle. Different chemical reaction schemes are examined

to qualitatively evaluate the impact of prompt mechanism in a detailed chemical kinetic

model. Experimental data obtained from literature were used to validate the mean NO

levels calculated. Then the validated model is utilized to predict the impact of cyclic

variability on mean NO and the amplitude of its variation. The cyclic variability was

simulated by introducing random perturbations, which followed a normal distribution,

to the Vibe function parameters. The results of this approach show that the proposed

emission model better predicts effectively mean NO formation. Also, it shows that due

to the non linear formation rate of NO with temperature, cycle-to-cycle variation leads

to higher mean NO emission levels than what one would predict without taking cyclic

variation into account.

31

3. Proposed Emissions Model

3.2 Emission Model

The proposed emission model uses the fundamentals of the widely known two zone pol-

lutant formation models [33, 45]; however in this model the burned zone is assumed as a

well-stirred reactor. A well-stirred, or perfectly-stirred, reactor is an ideal reactor in which

perfect mixing is achieved inside the control volume. The equations describing the reactor

are a set of coupled nonlinear algebraic equations, which are solved by using SENKIN,

a FORTRAN based code developed in Sandia Laboratories [100]. SENKIN solves the

chemical kinetics differential equations and predicts the formation rate of products. The

calculations of kinetics are based on the GRI 3.0 mechanism, a reaction scheme which was

initially proposed for methane-air combustion and consists of 53 species and 325 reactions

[35]. The reaction scheme involved of a number of carbon-nitrogen species and radicals

which are relevant in the NO formation chemistry, including HCN , H2CN , CN , HCNO

and HOCN . Further information about SENKIN solver can be found in Appendix A.

Figure 3.1: Schematic of the two zone detailed chemistry emission model.

Figure 3.1 illustrates the detailed chemistry model developed in the current study. The

solver requires the thermodynamic data of the burned zone such as the calculated burned

32

3.2. Emission Model

temperature and cylinder pressure at each crank angle. The calculated mass fraction

burned (MFB) defines the newly burned moles which enter from the flame front to the

burned zone. The newly burned moles are calculated from the oxidation rate of the fuel,

according to the stoichiometry of the combustion.

The newly burned moles and the composition from the previous step are assumed

to be homogeneously mixed and are imported as the initial input composition of the

burned zone in homogeneous reactor (SENKIN). The homogeneous reactor calculates as

an output the new composition of the burned zone which will be again imported in the

next crank angle. At the end of the combustion process, no new moles are assumed in the

SENKIN input speciation. The loop ends at the EVO and kinetics are thereafter tend to

be frozen, regarding the burned zone temperature.

In the magority of the existing two zone emissions models [14, 40, 45, 48, 101], only

the thermal mechanism is considered. Additionally, the necessary species for the ther-

mal mechanism (H2, H, O2, O, OH, H2O) are calculated assuming equilibrium. The

proposed emission model uses a detailed chemical mechanism which includes all the pos-

sible chemical pathways of NO formation, while the other necessary species are calculated

not from equilibrium but from detailed kinetics. This improves the precision in NOx

prediction, with a cost though in computational time [32]. However, a reduced detailed

chemical mechanism but with explicit kinetics for intermediate species could serve as a

compromise between accuracy and computational time. Last but not least, the proposed

emission model calculates not only NO emissions, but also CO emissions as it will be

presented on model validation (Chapter 5).

3.2.1 Mass Balance

The mass balance equation is utilized for the calculation of the new burned mass at each

crank angle as well as for the calculation of the inlet concentrations to the homogeneous

reactor. The first step considers the fuel oxidation, which products are imported to the

burned zone. The second step considers a mixing process between the six (6) new burned

species and the fifty three (53) species calculated from the previous time step by the

well-stirred reactor.

3.2.1.1 New Burned Mass

By representing the chemical formula of a fuel as CαHβOγNδ, the overall combustion

reaction as a function of mixture lambda can be written as:

φεCαHβOγHδ + (O2 + 3.76N2)→ ν1CO2 + ν2H2O + ν3N2 + ν4O2 + ν5CO + ν6H2 (3.1)

where ε is the stoichiometric fuel/air molar ratio and it is given by equation 3.2.

33

3. Proposed Emissions Model

ε =

(nfna

)st

=

[(1 + 3.76

)(α +

β

4− γ

2

)]−1(3.2)

Considering that the atoms are conserved, the mass balance equation [Eq 3.3] for

each element (C,H,N,O) can be applied. This gives four equations with six unknowns.

In equation 3.3, j represents the number of elements and k is the number of species of

equation 3.1.K∑k=1

ν∗kjνk = 0 (3.3)

Convenient approximations for lean and rich combustion are:

λ > 1 ν5 = ν6 = 0

λ ≤ 1 ν4 = 0

For the lean or stoichiometric cases, atom-balance equations are sufficient to deter-

mine the product composition (four equations and four unknowns). For the rich case we

introduce an equilibrium constant for the Water Gas Shift (WGS) reaction:

CO2 +H2 � CO +H2O

K =ν2ν5ν1ν6

(3.4)

where K is a constant. Values of 3.8 to 3.5 are commonly used for K, while in this model

the 3.5 value was adopted. Solutions for both rich and lean cases are given in table 3.1.

Table 3.1: New burned mass composition based on λ

i Species λ ≤ 1 λ > 1

1 CO2 αφε αφε− ν52 H2O βφ ε

20.42− φε(2α− γ) + ν5

3 N2 0.79 + δφ ε2

0.79 + δφ ε2

4 O2 0.21(1− φ) 0

5 CO 0 ν56 H2 0 0.42(φ− 1)− ν5

In the rich case the parameter ν5 is given by the solution of a quadratic equation:

ν5 =−β +

√b2 − 4ac

2a(3.5)

34

3.2. Emission Model

where

a = 1−Kb = 0.42− φε

(2α− γ

)+K

[0.42big(φ− 1

)+ αφε

]c = −0.42αφε

(φ− 1

)K

In reciprocating engines, residual gas is well mixed with the fresh fuel/air mixture. At

each crank angle step, the new burned mass consists of the combustion products and the

residual gas products. Ferguson [102] has presented a subroutine for the calculation of

the composition of fuel-air residual gas mixtures. By rewriting the combustion equation

3.1 it gives us:

φεCαHβOγHδ + ν ′4O2 + ν ′3N2 → ν ′′1CO2 + ν ′′2H2O + ν ′′3N2 + ν ′′4O2 + ν ′′5CO + ν ′′6H2 (3.6)

where

ν ′: reactant coefficient

ν ′′: product coefficient

Adopting similar notation for other symbols, it should be clear that for a mixture of

residual gas and premixed fuel-air, the mass fraction which finally enters the burned zone

is given by equation 3.7.

yi = (1− f)y′i + fy′′i (3.7)

Similarly, the species molar fractions are given by equation 3.8:

xi = (1− xr)x′i + xrx′′i (3.8)

where the residual mole fraction is:

xr =

[1 +

M ′′

M ′

(1

f− 1

)](3.9)

3.2.1.2 Inlet/Outlet Concentrations

By the time the combustion starts, it is assumed that air, fuel and residual gas are perfectly

mixed. During the combustion process, the cylinder mass is divided into the burned and

unburned zones, while the mass balance is described by equation 3.10. Although during

the combustion process the cylinder pressure increases remarkably, which benefits the

blow-by of the cylinder mass through crevices, in this model it is assumed that blowby

does not occur. Thus, at each crank angle step, the new burned mass that enters the

burned zone is equal to the unburned mass that ”leaves” the fresh zone (eq. 3.11).

dmcyl

dϑ=dmb

dϑ+dmu

dϑ+dmbb

dϑ(3.10)

35

3. Proposed Emissions Model

dmb

dϑ= −dmu

dϑ(3.11)

In the previous section it was presented that the new burned mass that enters the

burned zone consists of six (6) species (CO2,CO,H2O,H2,O2,N2) which originate from the

fuel-air reaction and the residual gas of the fresh zone. These new species are assumed

to be perfectly mixed with the fifty three (53) species that burned zone consists of, so as

the sum of all the constituent mass fractions must be unity (eq 3.12).∑i

Yi = 1 (3.12)

The homogeneous reactor solver requires the mole fractions of the 53 species. The

mass fraction is converted to mole fraction by using the molecular mass of the species of

interest and of the mixture:

χi =YiMWmix

MWi

(3.13)

where the mixture molecular mass is given by equation 3.14:

MWmix =1∑

i

(Yi/MWi

) (3.14)

3.2.2 Energy Balance

There are many possibilities for the chemical kinetics problems that one may need to

solve for various applications (Appendix A). In this case, where SENKIN is used for the

prediction of pollutant formation, it is considered that the pressure and temperature of

the system are constant. In fact, heat release, pressure and burned zone temperature are

calculated by the two zone combustion model. In such cases, where the pressure is held

constant, the energy equation is replaced by the condition that the temperature is known.

Therefore, the calculated burned zone temperature is considered.

3.3 Validation Data

Experimental data are necessary for the validation of the model presented in this chapter.

Ball et al. [12] used experimental data from a Rover K4 optical engine in order to perform

combustion analysis in a SI engine and to correlate the combustion process with the NO

emissions. The fuel used in those experiments was methane. That engine from the Ball

et al. work is simulated in the present study, as many engine specifications necessary for

the modeling are contained in that publication and are summarized in table 3.2. The

36

3.4. Modeling Approach

Table 3.2: Rover K4 engine characteristics

Main SpecificationsBore 80 [mm]

Stroke 89 [mm]

Con. Rod Length 160 [mm]

Compression Ratio 10 [mm]

Cam timingIVO 12 BTDC

IVC 52 ABDC

Peak Lift Inlet 8.8 (mm) at 70 BBDC

EVO 52 BBDC

EVC 12 ATDC

Peak Lift Exhaust 8.8 (mm) at 70 ABDC

Cylinder HeadType Rover K16 1.4 MPI

Pent Angle 45

Inlet Valve Seat Diameter 24 [mm]

Exhaust Valve Diameter 19.6 [mm]

Number Of Valves 4 [-]

model was applied to this engine and the results of the simulations were compared with

the experimental data for validation.

This optical engine was measured under partial load and Wide Open Throttle condi-

tions (WOT), for different crank angle ignition duration and lambda values. Information

about the engine performance and the engine emissions (NOx, HC) was also available for

each measured engine point.

3.4 Modeling Approach

In order to run the detailed chemistry model, engine cylinder data are required. The

simulation of the Rover K4 optical engine was performed by using a commercial engine

simulation tool (AVL BOOST). The combustion process is predicted by using the two

zone Vibe function.

37

3. Proposed Emissions Model

3.4.1 Engine Model

As already mentioned, the commercial engine simulation package AVL BOOST is applied

for the simulation of the single cylinder SI engine. The model configuration is illustrated

on figure 3.2.

Figure 3.2: Schematic of the Rover K4 model developed in BOOST.

The simulation calculates the heat release rate and the in-cylinder thermodynamic

properties. The combustion submodel used for the prediction of heat release was a two-

zone Vibe model. The Vibe function describes the MFB at a given crank angle [46]:

Qf

(ϑ)

Qf,total

= 1− e−a

(ϑ−ϑSOI∆ϑCD

)m+1

(3.15)

In equation 3.15, ϑSOI is the degree of crank angle where ignition starts, ϑCD is the

duration of combustion in crank angle degrees, m is the shape parameter for Vibe function,

and a is the Vibe combustion efficiency parameter.

The two-zone approach consists of a burned zone with a calculated burned temperature

for the combustion products and an unburned zone with a unburned temperature for

the fresh compressed mixture and any residuals from the previous combustion cycle. A

uniform pressure for both zones is assumed. As Vibe function is an empirical model

and not a ”physical” model, it was used in this study for the estimation of the pressure

and temperature history at each operating point and not for further detailed combustion

investigation.

3.4.2 Modeling Assumptions

The following assumptions are considered in this modeling approach:

38

3.5. Emission Model Validation

1. uniform pressure in the cylinder (burned and unburned zone at the same pressure);

2. a complete combustion of hydrocarbon fuel with air;

3. uniform composition in the burned zone;

4. NOx emissions solely consisting of NO.

The validity and impact of these assumptions in the final results is investigated in the

results section.

3.4.3 Coupling Simulation

The coupling between the BOOST model and the detailed chemistry model is performed

offline. Firstly the thermodynamic characteristics of the cylinder are calculated by the

engine simulation tool, then the crank angle series data of the cylinder pressure, the

burned zone temperature and the MFB during the combustion phase are imported in the

emission model as a post processing procedure. The final NO emissions are calculated at

the end of the closed thermodynamic cycle, when the exhaust valves open.

3.5 Emission Model Validation

3.5.1 Mean Cycle Validation

The Rover K4 was simulated with the AVL BOOST model for mean cycle Vibe parameters

and the results were compared with the experimental cycle-averaged engine data found in

Ball et al. [12]. The comparison between experimental and simulated data refers to the

IMEP, the maximum pressure during the combustion phase, the crank-angle degree where

maximum pressure occurs, and the crank angle degree where 10% of MFB is observed. All

these data are presented in table 3.3. The designation of each point in table 3.3 is done

with the P and W initials corresponding to partial load or wide open throttle operation,

respectively, followed by two digits corresponding to the lambda value (10 corresponding

to λ = 1 and 15 corresponding to λ = 1.5), followed by two digits of crank angle degree

where ignition starts before top dead center.

The predicted thermodynamic data of the ten simulated operating points were used

as an input for the NO prediction. The simulated NO emissions are compared with the

experimental NO emissions in figure 3.3 for stoichiometric combustion and in figure 3.4

for lean combustion. In stoichiometric combustion, NO predictions are shown with and

without the effect of the prompt mechanism. The prompt mechanism has been switched

off by zeroing the HCN radicals in the chemical mechanism.

39

3. Proposed Emissions Model

Table 3.3: Validation of the thermodynamic model using BOOST.

Cases λ ϑign [oCA] IMEP [bar] Pmax [bar] ϑPmax [oCA] ϑ10% [oCA]

Partial Load

P1015exp

1.00 -153.84 18.05 20 5

sim 3.84 16.45 20.11 4.94

P1030exp

1.00 -303.27 25.92 7 -10

sim 3.27 20.47 10.92 -11.42

P1045exp

1.00 -453.15 29.34 2 -18

sim 3.14 32.37 3.07 -18.32

P1515exp

1.50 -151.46 10.23 0 19

sim 1.44 12.20 1.34 19.37

P1530exp

1.50 -302.33 15.13 12 -2

sim 2.33 14.96 11.84 -2.24

P1545exp

1.50 -452.25 21.44 6 -16

sim 2.23 25.48 6.26 -15.91

WOT

W1015exp

1.00 -156.14 33.02 18 3

sim 6.15 35.76 19.99 2.27

W1030exp

1.00 -305.55 45.68 5 -11

sim 5.55 55.34 7.45 -11.18

W1515exp

1.50 -153.83 17.33 10 11

sim 3.82 21.07 3.51 10

W1530exp

1.50 -304.38 29.20 12 -3

sim 4.38 29.67 13.83 -3.23

The model appears to have a rather good accuracy over a wide NO range, that is from

NO concentrations of less than 10 ppm (P1515) to more than 2000 ppm (W1030). For

these cases where large differences can be seen (e.g. W1015), one should also observe re-

lated differences in the thermodynamic data and not only in the reaction modeling. Cases

with lower thermodynamic error show better prediction in NO results (example P1015).

By using a more sophisticated combustion model [48, 103], the burn rate prediction could

be improved with significant improvement in NO prediction as well.

The availability of oxygen is key variable affecting NOx prediction. This may be

an additional reason of difference between measured and experimental data. Within a

typical stoichiometric window of (0.95 ≤ λ ≤ 1.05) that appears in actual engines during

stoichiometric operation, slight differences in lambda could affect the total amount of

40

3.5. Emission Model Validation

Figure 3.3: Comparison of measured and simulated NO molar fractions for stoichiometric

combustion (λ=1.0). Results without prompt mechanism are also presented.

NOx formed during combustion. The stoichiometric cases of the experimental data were

also simulated with a slightly rich (λ = 0.95) and slightly lean (λ = 1.05) mixture. The

results are presented in figure 3.5. It is observed that the measured NOx concentration

is almost always between these slightly lean and rich simulated values. Hence, slight

departures from the set lambda in the experimental data may be a significant reason for

the difference between experiment and simulation.

Figure 3.3 also shows that the prompt mechanism increases the total amount of NOx

concentrations by 10%-15% in case of stoichiometric combustion. Including the prompt

formation one can increase the accuracy of the chemical mechanism. Bachmaier et al. [24]

used an experimental configuration to define the equivalence ratio in which the prompt

mechanism becomes significant in terms of total NOx formation for various hydrocarbon

mixtures. They found that the prompt NO formation starts to become significant as the

mixture moves towards stoichiometry from λ = 1.33, in the case of methane. The prompt

mechanism was negligible for leaner (λ ≥ 1.5) conditions. Our results confirm the signif-

icance of the prompt mechanism in addition to the thermal one, even for stoichiometric

combustion.

The thermodynamic input scenario is also important in lean conditions; however the

oxygen availability does not affect the final results as much as in the stoichiometric case.

41

3. Proposed Emissions Model

Figure 3.4: Comparison of measured and simulated NO molar fractions for lean (λ=1.5)

operating engine conditions.

In lean combustion, it seems that non-homogeneities in the burned zone can become

important for accurately predicting final NO emissions. Multi-zoning is mostly used in

0D-engine models to take into account mixture stratification. In multi-zone modeling,

different lambda and temperatures are assumed in each zone and NO prediction can be

improved.

Another reason for differences between the simulated and experimental results could

be the uncertainty in the high concentration of hydrocarbons (HC) that this engine emits

(up to 9000 ppm). By assuming the measured concentration of HC in the model, the

prompt mechanism appears very significant, even in the lean case. As this engine is an

optical and not a production one, these HC were assumed to be generated from crevices

in the piston/cylinder interface and oil oxidation, rather than from fuel combustion itself.

Although these HC do not participate in combustion, they could have an effect in a cold

outer zone of a multi-zone model.

3.5.2 Cycle-to-Cycle NO Variation

The detailed chemical mechanism was then used for the investigation of NO CCV. From

the various engine points in figure 3.3, four engine points were chosen for the CCV analysis;

two in partial load (P1015, P1030) and two at wide open throttle operation (W1015,

42

3.5. Emission Model Validation

Figure 3.5: Impact of slight stoichiometry variation on NO formation.

W1030). All engine points were selected in stoichiometric conditions, to also include the

effect of the prompt mechanism in NO formation.

NO variability was investigated using a statistical analysis. Vibe combustion param-

eters such as the ignition timing (SOI), the CD and the Vibe shape coefficient (m) were

randomly varied within limits, assuming that these parameters follow a normal distribu-

tion. The mean values for these distributions were equal to the values used in the case

of mean cycle modeling. The range of the variation considered was taken from a relevant

analysis in the framework of the FP6 LESSCCV research project [104] and differed for

partial load and WOT operation. Full load points correspond to higher CCV than low

load engine points [10]. One hundred engine cycles were simulated in each engine point

and the results of imep and NOx concentrations are presented in distributions. Differences

between mean cycle indices and CCV values are discussed.

Cycle-to-cycle variation of pressure and temperature are illustrated in figures 3.6a and

3.6b, respectively, for the engine point of partial load and ignition timing of 15o BTDC

(P1015). The mean value of maximum pressure is 16.6 bar and the standard deviation is

0.98 bar, while the peak temperature has a mean value of 2172 K and a standard deviation

of 20 K.

Figures 3.6c and 3.6d illustrate the distributions of imep and NO concentration, re-

spectively, for the same engine point (P1015), due to the variation of the combustion

43

3. Proposed Emissions Model

Figure 3.6: CCV of in-cylinder pressure and temperature evolution, which results on the

distribution of imep and NO emissions (P1015 point).

44

3.5. Emission Model Validation

parameters. Both figures include the statistical characteristics of the distributions such

as the mean value and the Standard Deviation (SD). Mean Cycle imep (MC imep) and

the mean imep value of the CCV analysis coincide perfectly, while MC NOx value and

CCV mean NOx value seems to have a slight deviation.

The same approach was also followed for the operation point at partial load (P1030)

with ignition timing 30o BTDC. The peak pressure distribution has a mean value of 20.6

bar and the standard deviation is 1.05 bar. The peak temperature has a mean of 2173 K

and 30 K respectively. Distributions of this engine point for imep present no difference

for the MC imep and the CCV imep value. In the case of NO, a small difference between

MC NOx and CCV NOx values is again shown.

In the case of WOT, the same approach with a higher range of Vibe parameters

was used for the CCV analysis. Pressure and temperature plots of the engine point

of 30o BTDC (W1030) are presented in figures 3.7a and 3.7b, respectively. Pressure and

temperature peak values have a higher range as a result of higher range in the combustion

parameters. In the case of W1015, the mean value of the maximum pressure is 35.7 bar

and SD is equal with 3.1 bar, while the peak temperature varies from 2121 K to 2310 K

with a mean value and standard deviation equal with 2200 K and 39 K, respectively. Same

order of magnitude differences are noticed for the case of W1030, where peak pressure

varies from 45.5 bar to 61.7 bar (mean 55.5 bar, standard deviation 2.9 bar) and peak

temperature varies from 2281 K to 2503 K (mean 2410 K, standard deviation 42 K).

The distributions of imep and NO are illustrated in figures 3.7c and 3.7d. Due to

higher CCV, MC imep and CCV imep are slightly different in both cases. Thus, MC

NOx value and CCV NOx values present a higher deviation compared with the partial

load. This also indicates that deviation between MC and CCV NO values is affected by

the range of change of the combustion parameters.

3.5.2.1 Contribution of Prompt Mechanism on the CCV of NO Emissions

The impact of the prompt mechanism on NOx CCV has been also investigated. In the

case of mean cycle modeling, it was observed that the prompt mechanism accounts for an

additional 10% to 15% in the final NOx concentration. Therefore, it is expected that the

prompt mechanism should have a corresponding effect on NO CCV.

Figure 3.8 demonstrate the distributions of NO for a partial load and a full load

engine point of CCV analysis, considering only the thermal mechanism. Mean values of

NO distribution show a decrease compared to the mean CCV NO values using the full

mechanism. In addition, for the case of using the detailed chemical mechanism without

the prompt one, a slight difference between MC NO values and CCV NO values is also

observed. However, the prompt mechanism has an additional impact in the statistic

characteristics of the NO distribution, which is described in the next section.

45

3. Proposed Emissions Model

Figure 3.7: CCV of in-cylinder pressure and temperature evolution, which results on the

distribution of imep and NO emissions (W1030 point).

46

3.5. Emission Model Validation

Table 3.4: Comparison of COV values for imep and NO with and without the prompt

mechanism.

Cases COVimep [%]COVNO [%] COVNO [%]

(Full model) (w/o prompt)

P1015 1.81 14.30 14.35

P1030 0.84 14.50 13.77

W1015 0.81 26.03 26.89

W1030 3.81 11.71 16.94

3.5.3 Deviation Between Mean Cycle Values and Mean CCV

Values

CCV does not only result in a range of values for NOx emissions but, due to the non-

linearity of NOx formation with combustion parameters and primarily with temperature,

it may also have an impact on the average NOx emitted. Hence, comparison between

cycle- averaged values and the mean CCV value is important.

Table 3.5: Comparison of mean cycle values (MC) and mean CCV values for imep and

NO with and without the prompt mechanism.

Points CasesIMEP

Diff [%]NO [ppm]

Diff [%]NO [ppm]

Diff [%][bar] (Full model) (w/o prompt)

P1015MC 3.84

0.00735

0.99623

1.71CCV 3.84 728 613

P1030MC 3.27

0.001320

0.651230

4.14CCV 3.27 1311 1181

W1015MC 6.15

0.49646

2.02516

1.78CCV 6.12 633 507

W1030MC 5.55

0.901990

8.051720

0.97CCV 5.5 1842 1703

Partial load and full load are two cases that exhibit different variability for the com-

bustion parameters. In partial load, the mean value of imep CCV distribution is almost

the same to the mean cycle imep value. On the other hand, the CCV imep values are

always lower than the mean cycle imep values in full load operation (Table 3.5). This

47

3. Proposed Emissions Model

(a) P1015 point. (b) W1030 point.

Figure 3.8: CCV NO distribution w/o the prompt mechanism.

means that the impact of CCV on average is a degradation of the engine performance.

NO formation is also affected by the variability in the combustion parameters. Both

in full and partial load CCV NOx values are always less than MC NOx values, which

reflects the CCV impact in NO formation (Table 3.5). In WOT operation, this impact is

higher than in partial load. This result is related with the nonlinearity of NO formation

and for this reason it can not quantitatively correlated with imep variation. As shown in

Table 4, the higher the difference between CCV imep and MC imep is, the higher is this

difference between CCV NOx and MC NOx, too.

The coefficient of variation is used as a metric of the intention of the NO CCV in table

3.4. The impact of the prompt mechanism is also separated in this table. NO in general

presents higher variability due to CCV than imep does. Also, the results show that it

is not possible to establish a direct link between imep CCV and NO CCV. The latter

is dependant on both the operation point and the CCV of imep. Finally, the impact of

the prompt mechanism on CCV is also specific to the engine point considered. In one

of the WOT conditions examined, the prompt mechanism led to a significant increase in

NO CCV, that is not obvious in the other cases. This means that the combination of

heat release rate with reaction kinetics is unique for each engine point that results to a

behaviour which cannot be generalized at this stage. Simulations with other engines and

further refinements in the model may lead to a more consistent behaviour of CCV NO

with CCV in other combustion parameters.

48

3.6. Conclusions

3.6 Conclusions

This chapter presents the mathematical backround of the proposed detailed chemistry

emission model. The model simulates the burned zone as a well-stirred reactor and

utilizes a chemical mechanism for the prediction of homogeneous engine-out emissions.

The emission model can predict NO, CO and other engine-out emissions, however in this

chapter the NOx prediction was preliminary presented.

Literature experimental data were used for the validation of the detailed chemistry

model, assuming steady state conditions. The model satisfactorily predicts NO emissions,

ranging from a few ppm to a couple of thousand of ppm of NO molar fraction, in both sto-

ichiometric and lean conditions and under various ignition timing conditions. Cases with

over 10% error were considered to be either inaccurately thermodynamically simulated or

highly affected by minor mixture lambda fluctuations.

Then, the validated model was used for the simulation of NO variation due to com-

bustion CCV. It was found that CCV NO distributions exhibit a higher COV compared

to the IMEP distributions. In addition, mean CCV NO values are always lower than

the average cycle NO values up to 8%. This clearly declares that elimination of engine

combustion variability could reduce NOx emissions.

Finally, the impact of prompt mechanism in NO cyclic emission variability was also

investigated. The detailed chemistry model was employed without considering the prompt

chemical pathway, which gave a up to 15% deviation on mean cycle NOx prediction

from the ”full model” at stoichiometric conditions. In the case of CCV investigation,

it was found that the prompt mechanism has an impact in the COV and mean value

of NO distributions, although the impact was dependant on the engine operation point

considered.

49

3. Proposed Emissions Model

50

Chapter 4

Experimental Investigation of Cyclic

Emissions Variability

4.1 Introduction

Cycle-to-cycle combustion variation (CCV) is an undesirable characteristic of internal

combustion engines, the origin of which is in detailed presented on chapter 2.3. In this

chapter the impact of various engine operating parameters is explored on performance,

combustion process, early flame kernel development, and especially on pollutants for-

mation of a SI engine on a cycle-to-cycle basis. Although in most of the experimental

studies the CCV trends of various engine calibrations on performance and early flame

kernel are well-known, only a handful number of experimental studies can be found in

published-literature regarding cyclic emissions variability (CEV).

The present work aims at experimentally exploring the impact of various engine operat-

ing parameters on performance, early flame kernel development, and pollutants formation

of a SI engine on a cycle-to-cycle basis. A piezoelectric sensor, integrated in the spark-

plug, and fast response analyzers are employed for the cycle-resolved measurements. The

investigation includes a wide range of engine operating conditions with varying engine

load, equivalence ratio, and ignition timing. The analysis of the results aims at correlat-

ing various interactions among combustion parameters (such as IMEP, maximum pressure

and crank angle of 5% mass fraction burned), and pollutant emissions (NO, CO, CO2).

The overall objective of this study is primarily to explain how combustion variability

affects emissions variability, under various engine operating conditions.

51

4. Experimental Investigation of Cyclic Emissions Variability

4.2 Experimental Set up

4.2.1 Engine Test Bed

The experiments were conducted on an in-line four cylinder PFI SI engine, modified from

an original Honda CBR 600RR motorcycle. Table 4.1 shows the basic features of the

engine. The engine was controlled by an open-access engine control unit (ECU), which

enabled alteration and recording of the entire set of engine parameters (e.g., spark timing,

equivalence ratio, fuel injection timing and duration, etc.) and was connected to a Schenk

steady-state hydraulic brake. The fuel used was a market gasoline with a nominal octane

number of 98 RON.

Table 4.1: Honda CBR600 engine characteristics

Main SpecificationsBore 67 [mm]

Stroke 42.5 [mm]

Con. Rod Length 91.8 [mm]

Compression Ratio 12.2 [mm]

Cam timingIVO 67 BTDC

IVC 42.5 ABDC

EVO 91.8 BBDC

EVC 12.2 ATDC

Valvetrain Chain driven, DOHC

Cylinder HeadType Honda CBR 600RR

Inlet Valve Diameter 26 [mm]

Exhaust Valve Diameter 22.5 [mm]

Number Of Valves 4

Fuel SystemFuel Delivery Port Fuel Injection

52

4.2. Experimental Set up

4.2.2 Experimental Configuration

The experimental configuration is schematically illustrated in figure 4.1. The measure-

ments involved recording of two sets of parameters, one at high and the other at low

sampling frequency. High frequency sampling was conducted for cylinder pressure, ex-

haust gas temperature, crankshaft and camshaft position and cycle resolved emissions

such as CO, CO2, NO and NOx. Pollutants measurement was conducted in the exhaust

manifold directly downstream of the exhaust valve. A National Instruments data acquisi-

tion card was used for high frequency sampling, capable of sampling rates up to 500kHz.

In this experimental study, a set of 150 consecutive cycles was sampled at each engine

operating point, while the sampling rate was adjusted to correspond to 23

degree of crank

angle.

Figure 4.1: Schematic of the experimental configuration

The low sampling frequency set of parameters comprised ECU derived signals, includ-

ing lambda value, ignition timing, throttle position, inlet air pressure and temperature

and coolant temperature, all recorded at a frequency of 10Hz. Average CO, CO2, and

HC emissions were also recorded at low frequency in the exhaust tailpipe, upstream of

the muffler.

53

4. Experimental Investigation of Cyclic Emissions Variability

A spark plug with an integrated piezoelectric transducer was used to measure cylinder

pressure. This is manufactured to fit the particular engine cylinder head and can be used

as a direct replacement of the original spark plug. The signal from the transducer was

fed to the charge amplifier. The exhaust gas temperature was measured by a K-type

thermocouple connected to the relevant transducer.

4.2.3 Measurement Protocol

The test protocol consisted of a three dimensional matrix of measurements at different

throttle positions, equivalence ratios, and ignition timings. The selected engine speed

values were 4000, 5000 and 6000 rpm. Throttle positions were varied from 20% and 50%

to 80% of wide open throttle. Three different lambda settings from the whole range of

fuel/air ratios were tested. The target settings were at λ = 0.93 (rich mixture), λ = 1

(stoichiometric mixture) and λ = 1.07 (lean mixture). Finally, different ignition timings

were examined, including the default setting corresponding to maximum breaking torque

(MBT), a 5o [CA] advanced from MBT ignition case and three retarded ignition from

MBT such as 5o, 10o and 15o [CA]. Table 4.2 presents in detail the test conditions at each

engine speed and throttle position operating point.

Table 4.2: Experimental engine test conditions.

CasesEngine Speed [rpm]

4000 5000 6000

TP: 20[%]L/S/R L/S/R L/S/R

+5/MBT/-5/-10/-15 MBT +5/MBT/-5/-10/-15

TP: 50[%]L/S/R L/S/R L/S/R

MBT MBT MBT

TP: 80[%]L/S/R L/S/R L/S/R

+5/MBT/-5/-10/-15 MBT MBT

4.3 Measurements

4.3.1 In cylinder pressure

In-cylinder pressure was measured by a Kistler 6113B spark plug piezoelectric pressure

transducer that was flush mounted on cylinder number 2. The transducer signal was

amplified using a Kistler 5011B charge amplifier and acquired using a data acquisition

54

4.3. Measurements

card, National Instruments 6341 USB module [NI manual], before finally recorded on a

personal computer for processing. Table 4.3 presents the specifications of the cylinder

pressure experimental equipment. As a crankshaft encoder was not available at the time

period the measurements performed, the engine crankshaft and camshaft signals were

used to synchronize the cylinder pressure signal with the crank angle.

Table 4.3: Specifications of the cylinder pressure experimental equipment

Pressure SensorKistler 6113B Pressure Range: 0-200 [bar]

Natural frequency: ∼ 65% [kHz]

Linearity: ≤ ±0.5% FSO

Sensitivity drift: ≤ ±1%

Signal AmplifierKistler 5011B Linearity: ≤ ±0.05% FSO

Error: ≤ ±0.5% FSO

Aquisition CardNI 6341 USB Channels Number: 8 differential

16 single ended

ADC resolution: 16 bit

Sample rate: 500kS/s

Three sets of 150 consecutive in-cylinder pressure cycles were recorded at each operat-

ing point. As the engine dyno was hydraulic, the set with the minimum speed variability

was selected for further processing. From the selected raw data set, in-cylinder measured

pressure is firstly synchronized with the crank angle signal (Fig. 4.2a) and then pres-

sure cycles are rearranged to the same four stroke crank angle reference area (Fig. 4.2b).

Figure 4.2a illustrates a sample pressure cycles for the case of 4000 rpm, 50% throttle

position and rich conditions (λ=0.93). The pressure cycles are then corrected for the drift

in the vertical direction observed at Fig. 4.2a by pegging the total pressure signal to the

MAP index given by ECU. The result of this process is shown in figure 4.2b, where all 150

cycles are superimposed. Finally, the magnitude of the cycle-to-cycle variation at these

operating engine conditions can be clearly seen.

55

4. Experimental Investigation of Cyclic Emissions Variability

(a) Raw data from one set of 150 consecutive

pressure cycles.

(b) Pegging the pressure signal to the MAP

index at IVC.

Figure 4.2: Sample of processing pressure raw data at 4000 rpm, 50% throttle position

and rich operating conditions (λ=0.93).

4.3.2 Engine and Cylinder Emissions

Owing to the rapid fluctuations of emissions, fast response analyzers are required for

the measurement of the cycle resolved emission levels. Fast response analyzers present

short response time step, at the order of few milliseconds, which makes them suitable

for transient emissions studies [105, 106], as well as for the investigation of cycle-to-cycle

pollutants formation [14, 98, 107]. For the cylinder NO/NOx emissions measurements,

a Cambustion CLD400 analyzer was employed, while CO and CO2 measurements were

conducted using the Cambustion NDIR500 analyzer. Figure 4.3a illustrates the position-

ing of the three probes, regarding NO/NOx and CO/CO2 emissions respectively. At

the engine tailpipe, a Signal 3000HM HC analyzer was employed to detect the engine

average HC exhaust concentration level. The second channel of the CO/CO2 detector

also measured engine averaged concentrations just upstream of the exhaust muffler. The

positioning of the HC heated line as well as the CO/CO2 probe and the lambda sensor are

illustrated on Fig. 4.3b. The main technical characteristics of the total exhaust emission

instrumentation are provided by table 4.4.

A fast response NO detector combines the standard CLD measurement technique with

a rapid sampling method, which slightly varies on different instrumentation [99, 108]. This

Cambustion fNOx400 analyzer utilizes the same sampling system of the fast FID, which

is well-proven and details have appeared in many publications [109, 110]. The sampling

system deliver the sample to the chemiluminescence chamber where it is mixed with a

flow of ozone. The resulting chemiluminescent reaction produces photons in the 700-1500

56

4.3. Measurements

Table 4.4: Specifications of the emission gas analyzers and sensors.

NO/NOX analyzerCambustion fNOX400 Response time: NO:4ms / NOX :8ms

Linearity: ≤ ±1% FSO

CO/CO2 analyzerCambustion NDIR-500 Response time: ≤ 8ms

Linearity: ≤ ±2% FSO

HC analyzerSignal-3000HM Response time: l.5 ms

Linearity: ≤ ±0.5% FSO

Lambda SensorBosch LSU 4.2 Lambda Range: 0.65 to ∞

Exhaust Gas Pressure: ≤ 2.5 [bar]

Exhaust Gas Temperature Range: 930oC

(a) Cylinder exhaust emissions sampling. (b) Engine exhaust emissions sampling.

Figure 4.3: Photo from the experimental configuration regarding the positioning of the

probes for the cylinder and engine exhaust emissions sampling.

57

4. Experimental Investigation of Cyclic Emissions Variability

nm wavelength range which are detected with a photomultiplier tube. The linearity of

this device from 0 to 4000 ppm is within 1% (Table 4.4). Additionally, the fNOx400s

gas reaction chamber is housed in a remote sampling head so that sampling from close to

the exhaust system is facilitated. The light is transferred from the reaction chamber to a

photomultiplier in a main control unit via a fibre-optic bundle. The main control unit also

contains the power supply, electronics, ozone generator, and vacuum regulators to operate

the sampling head. This procedure is performed only to measure NO concentration. The

NO2 can be also be measured by using a pre-reaction to convert NO2 in the sample to

NO, where thereafter both NO species can be detected with the same technique, as NOx

emissions. This process, though easily achieved, would cause unacceptable deterioration

in the time response. However, NO typically accounts for more than 95% of the total

NOx emissions produced from SI engines. In this study, one channel for NO measurement

and one for NOx emissions were used. Figure 4.3a illustrates the locations of the two

NOx and NO probes. NO measurement presented response time in the order of 4ms,

while the measurement of the total NOx presented response time in the order of 8ms,

due to the upstream converter. Finally, after the data processing, it was found that the

NOx analyzer response time was not enough to measure cycle to cycle NOx emissions

and therefore it was not used in the results of this experimental study.

CO and CO2 measurements were conducted using the Cambustion NDIR500 analyzer.

This analyzer also comprised two channels, each able of detecting both CO and CO2

signals. CO and CO2 concentrations are determined using the non-dispersive infrared

(NDIR) technique. The sample is drawn through a narrow heated sample probe capillary

in to the low pressure sample optical chamber. Radiation is produced by an Infra-Red

emitter, then directed through the heated gas sample. Optical filters for monitoring the

absorption of IR at suitable wavelengths for CO and CO2 but avoiding interference from

other exhaust gas constituents (e.g. H2O, H2) are mounted in a chopping wheel which

rotates at 15000rpm and is located beneath the optical chamber. A sensitive IR detector

is located beneath the chopping wheel and is therefore exposed to IR radiation referencing

CO, CO2, a reference filter (for feedback control of emitter intensity, monitoring a wave-

length not affected by sample gas constituents) and a blank sector (for feedback control

of detector temperature). The detectors rapidly changing output from the 4 sectors is

digitally transmitted to the main control unit where it is linearised and output as a 0-10V

analogue DC signal. The system has two sample heads, each measuring CO and CO2

simultaneously and therefore four analogue outputs are available. The response time of

the system is less than 8ms (10-90%), while it is achieved by miniaturizing the sampling

system [98, 111, 112]. The first channel measured cyclic resolved cylinder exhaust port

concentrations while the second one measured averaged concentrations just upstream of

the exhaust muffler.

Furthermore, a conventional analyzer was used for the detection of HC average emis-

58

4.3. Measurements

(a) Cambustion fNOx400. (b) Cambustion NDIR500.

Figure 4.4: The Cambustion fast response analyzers

sions at the engine exhaust tailpipe. The Signal 3000HM HC analyzer presents a response

time at the order of 1-2sec. The analyser uses the well established principle of flame ion-

ization to detect volatile organic compounds in a gas stream. The analyzer uses either H2

or H2/He mixture for calibration and detects methane equivalence (C1 base). A double

insulated heated sampling line at a constant temperature of 220o was also used, in order

to avoid condensation of water vapor. Last but not least, at the exhaust pipe, the engine

lambda sensor was also located (Figure 4.3b). The lambda sensor was a wide-band Bosch

LSU 4.2, where its main characteristics are given in table 4.4.

4.3.3 Exhaust Temperature

Cyclic combustion variability results on exhaust gas temperature variability. For the

measurement of this temperature variation, a thin 0.5mm K-type thermocouple was used

together with the relevant transducer Phoenix Contact for exhaust gas temperature mea-

surement. The position of the thermocouple is presented on figure 4.3a, exactly after the

NOx probe. The thermocouple enters the exhaust pipe manifold through a small hole

59

4. Experimental Investigation of Cyclic Emissions Variability

which was locally blocked using cement, in order to avoid exhaust flow blow-by. Finally,

it was found by the recordings that the measured temperature profile was not that much

fast, possibly affected by the exhaust pipe thermal inertia or the thermocouple mass.

Therefore, the measured temperature variability is used as a reference point to validate

the observations of the combustion analysis and no further analysis was performed.

4.3.4 Data Acquisition Card

A National Instruments X-Series 6341 data acquisition (DAQ) card was used for the ac-

quisition of the high frequency engine data. The DAQ device employs eight (8) differential

channels for the recording of the cylinder pressure, the crank angle signal, the camshaft

signal, the temperature as well as the cylinder exhaust emissions such as the NO, NOx,

CO and CO2. As the maximum device sampling rate was 500kS/s, each channel maxi-

mum sampling rate was 62.5kS/s. Regarding the engine speed, the sampling frequency

equation 4.1 determines the maximum crank angle recording step. Maximum crank angle

recording step is increased as rotational engine speed is increased for constant frequency.

The maximum engine speed, based on the test protocol, was 6000 rpm; therefore the

crank angle step that was constantly adopted at all operating conditions was two third of

the crank angle (23

[oCA]).

fs =360 ·N60 ·∆ϑ

=6 ·N∆ϑ

(4.1)

National Instruments LabVIEW is a software development environment for creating

custom applications that interact with real-world data or signals. In this experimental

study, a LabVIEW interface was developed, in order to import, record and plot the eight

differential signals into a computer. Figure 4.5 illustrates the LabVIEW interface which

utilizes the NI DAQ card. Each recording was saved as a separate file, while at each

operating conditions at least two recordings were performed.

4.3.5 Engine Control Unit

The tested Honda engine was controlled by an open access ECU. The MOTEC M800

allowed the control and the recording of the total engine parameters, such as ignition tim-

ing, lambda, throttle position, MAP index, water and oil temperature etc. Additionally,

MOTEC was utilized as a low frequency data acquisition card, as the CO, CO2 and HC

exhaust emissions were recorded by the ECU.

60

4.4. Data Processing

Figure 4.5: The LabVIEW interface and the data acquisition card.

4.4 Data Processing

The collected data were post-processed by using an in-house MATLAB R© code. The

code, initially transforms the recorded signals to physical magnitudes and then matches

in-cylinder pressure with pollutants (NO, NOx, CO, CO2) traces for each individual

engine cycle (chapter 4.4.3). The code performed at each individual cycle combustion

analysis, such as the calculation of the heat release and the ϑ5%, performance analysis,

such as the calculation of the IMEP and emission analysis, such as the calculation of the

cycle-resolved emissions.

4.4.1 Combustion Analysis

The measured in-cylinder pressure and the geometrical characteristics of the combustion

chamber are used for the combustion analysis at each individual cycle. The cylinder

volume and the differential of the cylinder volume can be calculated as follows:

V (ϑ) = Vc +D2π

4s(ϑ) (4.2)

61

4. Experimental Investigation of Cyclic Emissions Variability

dV

dϑ= D2π

4

ds

dϑ(4.3)

where s(ϑ) is the piston position.

Then, the single zone heat release analysis is performed according to Heywood [15]:

dQnet

dϑ=

γ

γ − 1PdV

dϑ+

1

γ − 1VdP

dϑ(4.4)

where P is the measured cylinder pressure, γ is the specific heat ratio and V is the

instantaneous cylinder volume.

Both cylinder temperature and mixture equivalence ratio influence the specific heat

ratio [113, 114], with the effect of the cylinder temperature being more prominent. In this

study, the specific heat ratio was calculated using the empirical formula of Engel [115], as

described by equation 4.5:

γ = γo − k1 exp(−k2T

) (4.5)

where γo is a reference value (γo=1.38), k1 and k2 are constants (0.2 and 900 respec-

tively) and T is the cylinder gas temperature calculated assuming an ideal gas.

The empirical formula of Han et al. [116] was adopted for the calculation of the heat

transfer coefficient. This formula is described in equations 4.6 and 4.7, while the heat

losses to the cylinder wall are calculated by equation 4.8.

hg(ϑ) = 687P (ϑ)0.75U(ϑ)0.75D−0.25T (ϑ)−0.465 (4.6)

U(ϑ) = 0.494Up + 0.73 · 10−6(

1.35PdV

dϑ+V dP

)(4.7)

dQwall

dϑ=∑

hg,i · Ai · (Tw,i − Tg) (4.8)

where P is the cylinder pressure, U is the instantaneous mean gas velocity in the

combustion chamber, given from Eq. 4.7, D is the cylinder bore, T is the cylinder gas

temperature, Up is the mean piston speed, A is the combustion chamber surface in contact

with the gas and Tw is the mean cylinder wall temperature, assumed uniform in the whole

combustion chamber.

The gross heat release rate (Eq. 4.9) is calculated as the sum of net heat release rate

(Eq. 4.4) and heat losses (Eq. 4.8). Integrating the gross heat release rate (at each crank

angle step) and dividing by the fuels lower heating value (LHV) provides the fuel mass

burned at each crank angle step (Eq. 4.10).

dQgross

dϑ=dQnet

dϑ+dQwall

dϑ(4.9)

62

4.4. Data Processing

mf =

∑Qgross

LHV(4.10)

By using equation 4.10, the crank angle at which 5% of the fuel mass has been burned

determines the ϑ5%, which characterizes the crank angle that early flame kernel occurs.

4.4.2 Performance Analysis

The measured in-cylinder pressure over the operating cycle of the engine can be used

to calculate the work transfer from the gas to the piston. For the work calculation the

measured cylinder pressure and the calculated cylinder volume (Eq. 4.2) should be used.

The indicated work per cycle Wc,i (per cylinder) is obtained by integrating the term PV

over the entire four-stroke cycle (Eq. 4.11).

Wc,i =

∮PdV (4.11)

imep =

∮PdV

Vd(4.12)

The gross IMEP is defined as the indicated work per unit displacement volume done

by the gas during the compression and expansion stroke (Eq. 4.12). The work per unit

displacement volume required to pump the working fluid into and out of the engine during

the intake and exhaust strokes is termed the pumping mean effective pressure (PMEP).

Finally, the work per unit displacement volume of the whole cycle corresponds to the

net indicated mean effective pressure, and is described by equation 4.13. In this study,

the term of used IMEP corresponds from the IVC to the EVO, which approximately

corresponds the two stroke (compression-expansion) definition.

IMEPnet = IMEP − PMEP (4.13)

4.4.3 Emissions Analysis

The cycle-resolved emissions are synchronized with the measured cylinder pressure from

the same combustion cycle. Figure 4.6 demonstrates an example of processing the traces

of cylinder pressure and NO emissions from the same cylinder. While the exhaust valve

remains closed, the analyzer measures NO levels in the gas remaining in the exhaust

manifold from the preceding cycle. When the exhaust valve opens (EVO), a portion of

in-cylinder gas exits rapidly (blow-down) and washes the old gas out. A delay between

EVO and analyzer response (point A in figure 4.6) is observed, attributed to the finite

time it takes to the exhaust gas to travel between the valve outlet and the sampling

point, and to the instrument response time [14]. During the exhaust stroke and while the

63

4. Experimental Investigation of Cyclic Emissions Variability

exhaust valve closes (EVC), the analyzer signal shows a slight instability (section B-C in

figure 4.6), indicating a variation of NO concentration in the cylinder gas or interference

from flow phenomena in the exhaust port. The shape of this section may vary in each

cycle due to the bulk gas motion and turbulence [108]. After the exhaust valve closes

again for the next cycle, the analyzer signal remains rather constant (part C-D in Figure

4.6). The mean of this rather stable level is taken as the NO emissions of the latest cycle.

Figure 4.6: Typical NO recording in the exhaust with reference to the cylinder pressure.

Similar procedure is followed for the CO and CO2 recorded signals. Regarding the

NOx signal, the corresponding C-D line was found to be either very short or unsteady at

same operating points, due to the higher response time step of the converter and possibly

due to its conflict with the exhaust pressure, as this device was located very close to the

exhaust valve. Therefore, in this study it is assumed that NOx and NO emissions are

equal.

4.4.4 Cyclic Variability

The intensity of the variability is derived from pressure data and it is expressed by the

coefficient of variation (COV) of indicated mean effective pressure in percent, as it is

64

4.5. Results

described by equation 2.17. In this work, COV index is utilized not only for the variability

of performance (IMEP), but also for maximum cylinder pressure (Pmax) and the cycle

resolved NO, CO, and CO2 emissions. The mathematical definitions of mean value (x),

standard deviation (σ) and coefficient of variation (COV) are also presented in 2.

4.5 Results

Variability in combustion evolution manifests itself as a scatter of the measured pressure

traces around the mean pressure curve. In turn, the variation in pressure build-up leads

to variations of engine performance and emissions from cycle to cycle.

Figure 4.7: IMEP and emissions distributions at 4000 rpm, 80% open throttle position,

stoichiometric mixture and MBT ignition timing.

A typical example of combustion, performance, and emission variability is illustrated

on figure 4.7. This refers to an engine speed of 4000 rpm, 80% throttle position, stoi-

chiometric mixture, and MBT ignition timing. The variation of combustion parameters

is significant under these conditions. The mean IMEP value of 8.8 bar varies with a stan-

dard deviation (SD) of 0.3 bar (COVIMEP=3%). NO emissions show an average content

of 1600 ppm and a standard deviation of 300 ppm (COVNO ∼ 19%), CO levels are at an

65

4. Experimental Investigation of Cyclic Emissions Variability

average of 0.87% and a SD equal to 0.20% (COVCO=22.5%), while CO2 is hardly variable

with an average of 12.9% and a SD equal to 0.17% (COVCO2=1.35%). This example

shows that the CCV severity is different for the various parameters of concern.

Figure 4.8: Pressure variability at 4000 rpm, 80% open throttle position, stoichiometric

mixture and MBT ignition timing. Mean pressure curve, maximum and minimum gross

heat release rates for the 150 consecutive cycles are also illustrated.

Figure 4.8 illustrates the variation of cylinder pressure from cycle to cycle for the

previous example, in order to better understand CCV. The maximum cylinder pressure

ranges between 30 and 58 bar with an average at 43 bar. Based on the heat release

analysis model developed, figure 4.8 also shows the burn rates for the highest and lowest

pressure traces in this diagram. High maximum cylinder pressure cycles are related with

fast burn rates. This is because earlier initial flame kernel development leads to timely

combustion evolution before TDC, which results to higher cylinder pressure. On the

other hand, lower maximum cylinder pressure occurs due to slower combustion process;

late early flame kernel development shifts main flame propagation phase significantly after

TDC where cylinder volume increases and suppresses pressure build up. The origins of

these variations are examined in the next sections.

66

4.5. Results

4.5.1 Impact of Engine Load on Variability

In this presented conventional spark ignition (SI) engine, load is controlled by the throttle

valve. At various throttle angles, air and fuel flow are throttled together, so as mixture

composition is essentially unchanged. In fact, throttle valve controls the volumetric ef-

ficiency of the engine, while it also has an impact on residual gas fraction and an effect

on the pressure waves in the intake manifold. The sensitivity of residual gas to intake

pressure has been shown to affect combustion and emission variability [13, 40, 41]. In

this section, the impact of engine load on combustion, performance and emissions cyclic

variability is in detail experimentally investigated.

Figure 4.9: Mean pressure and gross heat release traces for stoichiometric conditions, 4000

rpm, identical spark timing and various throttle positions.

Volumetric efficiency is maximized at wide open throttle (WOT). Figure 4.9 illustrates

the mean imep cycles for 20% and 80% open throttle, 4000 rpm, stoichiometric mixture

and identical ignition timing. It is observed that pressure traces vary even during the

compression stroke, as the trapped mass at 80% TP is higher. However, during the

combustion phase, the difference between the two pressure traces is further increased, not

only due to the higher fuel mass at 80% TP which is burned, but also due to the faster

combustion process at higher engine load. Heat release rate shows that combustion at

67

4. Experimental Investigation of Cyclic Emissions Variability

higher engine load starts earlier owning to higher air and fuel masses (higher cylinder

pressure at that crank angle), while residual gas fraction that dilutes mixture is lower

which benefits faster flame propagation.

The faster early flame kernel development is also described at table 4.5, which cor-

responds to the averaged and COV values of the 150 consecutive engine cycles. It is

presented that the ϑ5% is 8o [CA] delayed at 20% TP compared to the higher load engine

point for identical ignition timing. Additionally, table 4.5 shows cyclic variability values

for performance, combustion and emissions. Higher load decreases COVimep, which is also

confirmed by literature [12, 117].

Table 4.5: Mean and COV experimental data for two different throttle positions at 4000

rpm, identical ignition timing (48o [CA] BTDC) and stoichiometric operating conditions

(λ=0.99).

CasesIMEP Pmax ϑ5% NO CO CO2 HC Torque

[bar] [bar] [oCA] [ppm] [%] [%] [ppm] [Nm]

TP 20[%]mean 6.55 29.94 357.9 516 1.91 12.69

2913 28COV [%] 8.60 13.78 1.32 34.04 24.03 2.10

TP 80[%]mean 8.77 43.35 350.7 1618 0.87 12.90

3846 39COV [%] 3.49 13.94 1.15 18.54 22.53 1.35

Table 4.6: Mean and COV experimental data for two different throttle positions at 6000

rpm, identical ignition timing (41o [CA] BTDC) and stoichiometric operating conditions

(λ=1.0).

CasesIMEP Pmax ϑ5% NO CO CO2 HC Torque

[bar] [bar] [oCA] [ppm] [%] [%] [ppm] [Nm]

TP 20[%]mean 7.22 36.72 352.6 1291 0.66 13.07

3070 31COV [%] 4.57 10.75 1.04 21.69 33.94 1.09

TP 80[%]mean 10.45 50.44 351.6 2229 0.73 12.99

6497 49COV [%] 2.59 11.52 0.99 13.48 45.36 1.42

An explanation of cycle to cycle variation can be developed from the initial flame

kernel of combustion development and its dependence on mixture residual gas fraction.

At partial load conditions, the initial flame kernel develops at more diluted conditions

68

4.5. Results

due to the relatively high fraction of residual gas. Figure 4.10 illustrates the relationship

between IMEP for low (TP=20%) and high (TP=80%) engine loads. At 20% throttle

position, ϑ5% is delayed compared to the 80% setting, despite identical ignition timing.

Additionally, at 20% throttle position, ϑ5% presents a higher range of crank angle that it

occurs, compared with the 80% of throttle position. This higher scatter of ϑ5% at low load

conditions is accompanied by a steep decrease of IMEP at very late early flame kernel

cycles.

Figure 4.10: Relationship between IMEP and ϑ5% at 4000 rpm, stoichiometric mixture,

identical ignition timing and various throttle positions (TP: 20% and TP: 80%).

At partial load operating conditions, not only IMEP presents a higher variability

but also emissions present the same trend (Table 4.5). The most significant impact of

engine load on emissions variability is on NO, while CO and CO2 COV slightly vary.

However, it was found a clear index that higher engine load leads to a more stable engine

operation and to lower combustion, performance and emissions CCV (Table 4.5). Figure

4.11 illustrates the relationship between IMEP and NO exhaust content at three different

engine load conditions. The minimum increase of NO at high IMEPs, shows that IMEP

is not a good predictor of NO levels; NO may vary from 1500 to almost 2500 ppm at

an almost constant IMEP close to 9 bar. This plateau could be related with the non

69

4. Experimental Investigation of Cyclic Emissions Variability

Figure 4.11: Relationship between IMEP and NO at 4000 rpm, stoichiometric mixture,

identical ignition timing and various throttle positions.

linear relationship between combustion temperature and NO formation. At high load

conditions, small deviations of the mixture temperature lead to higher scatter of NO

emissions but finally to lower emission variability (Table 4.5).

Nitric oxide is mainly formed at high temperatures, through the well-known Zeldovich

thermal mechanism. Therefore, peak combustion temperature has a remarkable impact

on nitrogen oxidation. Increasing throttle position leads to higher engine load and max-

imum cylinder pressure and also to higher peak combustion temperature. As cylinder

temperature increases, NO emission level also increases for a constant cyclic mixture

composition (figure 4.12). The relationship between maximum cylinder pressure and NO

emissions presents a close to linear trend. In fact, deviations from the linear correlation

occur due to fuel and residual gas variations from cycle to cycle.

Carbon monoxide and carbon dioxide are also affected by engine load. Table 4.5

shows that the average CO emissions are decreased when engine load is increased, which

is accompanied by a minor increase on CO2 concentrations. The latter declares that

completeness of combustion is improved at higher loads, mainly due to the lower residual

gas fraction and the higher combustion temperature. In the same table it is observed that

70

4.5. Results

Figure 4.12: Relationship between Pmax and NO concentration at 4000 rpm, stoichiomet-

ric mixture, identical ignition timing and various throttle positions.

CO variability (COVCO) was found to be slightly decreased at 80% throttle position, while

a clear similar trend is also observed for CO2 variability. Therefore, it is clearly resulted

that residual gas fraction increases combustion, performance and emissions variability and

at higher loads lower COV values can be achieved.

Both table 4.5 and 4.6 present that unburned hydrocarbons are increased at higher

engine load conditions. Higher engine load leads to higher combustion temperature which

benefits the oxidation rate of the fuel hydrocarbons. Therefore, it is clearly believed

that the mechanism which increases the unburned hydrocarbons concentration at higher

engine load condition is through the mixture trapped mass in the crevices. As engine load

is increased, cylinder pressure is also increased which leads more of the air-fuel mixture

into crevices (most significantly the space between the piston crown and cylinder walls),

where the flame is quenched and mixture is left unburned.

Although the trends of the impact of engine load on combustion, performance and

emissions variability are more than clear, it is highly interesting to validate the same

trends at higher engine speed conditions. Table 4.6 presents the impact of engine load on

average and CCV values at 6000 rpm engine operating conditions. Similar to the 4000

71

4. Experimental Investigation of Cyclic Emissions Variability

rpm, variability of IMEP, ϑ5% and NO emissions is decreased at wider open throttle po-

sition. However, it is observed that the variability of CO and CO2 emissions is increased,

conversely to the 4000 rpm observation. This result could be an impact of turbulence in-

tensity on mixing fuel and air, as it is known from other researchers [118] that turbulence

intensity is slightly decreased at higher BMEP (or IMEP), which could affect mixture

homogeneity, completeness of combustion and finally CO and CO2 formation.

4.5.2 Impact of Stoichiometry on Variability

Mixture stoichiometry has a major effect on combustion efficiency and pollutants forma-

tion. On one hand, lean-burn engine operation favors fuel efficiency but diluted mixtures

generally lead to combustion variations, which limit the engine operation range [18]. On

the other hand, rich mixtures ignite easier and lead to more repeatable combustion, with

a cost on fuel consumption.

Figure 4.13: Mean pressure and gross heat release traces for rich, stoichiometric and lean

mixture, 6000 rpm, 80% TP, and identical spark timing.

Mixture composition also has an impact on combustion evolution and appears as an

important source of CCV [119]. Figure 4.13 shows the mean cycle pressure and gross heat

72

4.5. Results

release rate traces for three different mixtures: slightly rich, slightly lean and (almost)

stoichiometric. In all cases, the engine operates at 6000 rpm, 80% throttle position and

constant spark timing. It should be noted here that, owing to the constant throttle

position and the different lambda values, engine load slightly varies. Therefore the results

presented here may also be affected by the slightly differentiated load. The results show

that a slightly rich mixture leads to the maximum cylinder pressure and that leaner

mixtures lead to lower combustion pressure. In fact, both cases of λ=0.92 and λ=0.99

exhibit identical initial early flame kernel development; however in the rich mixture,

flame propagates faster (after the initial flame kernel development) over a smaller cylinder

volume (as the piston is closer to TDC), which leads to higher maximum pressure. On

the other hand, lean mixture not only ignites later but leads to a flame that propagates

at a slower pace.

Figure 4.14: Relationship between ϑ5% and mixture stoichiometry at 6000 rpm, 80% TP

and identical ignition timing.

The observation of the mean cycle heat release rates that mixture lambda has an

impact on combustion process is also validated by the CCV investigation. Initiation of

combustion (ϑ5%) delays when the mixture becomes leaner and this drives the complete

combustion to evolve later (after the TDC). This results to a reduced work by the piston

73

4. Experimental Investigation of Cyclic Emissions Variability

during compression and so is the work done by the piston during the expansion stroke,

which finally leads to a lower IMEP. On the other side, rich mixtures present a more

repeatable (lower scatter of ϑ5%) and initial flame kernel development, which leads to

faster burning rate and finally to high maximum cylinder pressure and a lower scatter of

IMEP. Table 4.7 shows that early flame kernel development presents similar variability

at lean, rich and stoichiometric mixtures, which declares that the observed variability on

combustion, performance and emissions is mainly determined by the mean early flame

kernel value and the impact of mixture lambda on flame propagation.

Table 4.7: Mean and COV experimental data for various mixture lambda operating

conditions at 4000 rpm, 80 % throttle position and identical ignition timing (48o [CA]

BTDC).

CasesIMEP Pmax ϑ5% NO CO CO2 HC Torque

[bar] [bar] [oCA] [ppm] [%] [%] [ppm] [Nm]

λ=0.94mean 8.90 46.95 348.1 1021 2.94 11.90

3913 41COV [%] 2.24 12.64 1.13 17.49 10.45 1.80

λ=1.00mean 8.77 43.35 350.7 1618 0.87 12.90

3846 39COV [%] 3.49 13.94 1.15 18.54 22.53 1.35

λ=1.11mean 8.42 39.90 353.0 1716 0.22 12.63

2911 37COV [%] 5.08 13.11 1.17 26.45 42.57 1.16

Table 4.8: Mean and COV experimental data for various mixture lambda operating

conditions at 6000 rpm, 80 % throttle position and identical ignition timing (41o [CA]

BTDC).

CasesIMEP Pmax ϑ5% NO CO CO2 HC Torque

[bar] [bar] [oCA] [ppm] [%] [%] [ppm] [Nm]

λ=0.92mean 10.70 58.15 348.8 1387 2.91 11.89

6798 49COV [%] 1.71 9.44 0.91 22.44 25.09 3.86

λ=0.99mean 10.50 51.13 351.3 2069 1.06 12.84

6497 49COV [%] 2.62 10.98 0.92 20.10 65.88 2.80

λ=1.06mean 9.87 43.23 354.8 1961 0.26 12.92

5608 46COV [%] 5.22 13.08 1.09 21.91 56.57 1.44

74

4.5. Results

Table 4.8 gives the mean combustion values for these three different composition

mixtures. It was found that slightly lean mixture present higher combustion efficiency

(ηc ∼ 0.95) and at the same time higher COVimep; this shows that engine operation at

leaner and more fuel efficient conditions can be limited by the level of cyclic variability.

At the same time, slightly rich mixture shows lower combustion efficiency (ηc ∼ 0.86)

but also lower COVimep, which means that combustion at such engine conditions is more

stable. It is clearly understood that the final thermal efficiency is a trade-off between

combustion efficiency and coefficient of variation of IMEP.

Figure 4.15: Relationship between IMEP and mixture stoichiometry at 4000 rpm, 80 per

cent throttle position and constant ignition timing.

Figure 4.15 illustrates the relationship between IMEP and mixture lambda that were

measured at nominally rich, lean and stoichiometric mixture engine conditions. Due to

the delayed early flame kernel development at leaner mixtures, mean indicated work is

reduced, while the maximum IMEP value is observed for slightly rich mixtures. Addition-

ally, the scatter of IMEP is increased from rich to leaner mixtures, as flame propagation

at leaner mixtures evolves during the expansion stroke at lower cylinder pressures. In the

same figure it is also observed that same cycles which present ϑ5% close or after TDC may

lead to misfiring or to partial burning.

75

4. Experimental Investigation of Cyclic Emissions Variability

Figure 4.16: Relationship between λ and NO concentration at 6000 rpm, 80% throttle

position and identical spark ignition timing.

Nitric oxide formation is a function of peak combustion temperature, mixture residence

time at high temperature and oxygen availability. Slightly lean mixtures combine high

combustion temperature and oxygen availability; therefore these mixtures present the

highest mean NO value compared to stoichiometric and rich mixtures (table 4.7). In

addition, slightly lean mixtures present higher variations of heat release and variations of

excess oxygen which both contribute to the highest COVNO index (table4.8).

Mixture composition is one of the most important causes of emissions variability. Cycle

to cycle fuel charging and residual gas variations have a significant impact on combustion

rate and pollutant formation. In the case of NO formation, both burning rate and mix-

ture lambda have an impact on the final NO cycle concentration amplitude. Figure 4.16

illustrates the relationship between cyclic mixture composition (λ) and NO concentra-

tion. Nitric oxide formation extends between the fast and slow burning rates. The slow

burning rate curve shows that the maximum NO concentration is produced within the

stoichiometric window. On the other hand, fast burn curve shows that the maximum NO

concentration results from a slightly lean mixture. Oxygen availability and combustion

temperature are the main reasons of this observation. At low burning rates, the maxi-

76

4.5. Results

Figure 4.17: Relationship between NO and early flame kernel development (ϑ5%) at 4000

rpm, 80% TP and identical ignition timing.

mum NO formation is controlled by combustion temperature, while at fast combustion

conditions, this is controlled by oxygen availability. Additionally, as all cycles shown

have the same ignition timing, it has to be concluded that combustion in rich cycles is

more advanced than lean cycles, as equivalence ratio has an impact on laminar flame

speed during early flame kernel development [87]. This behavior is more prominent at the

slow burn limit, also due to the fact that rich cycles present lower combustion variability.

More advanced cycles (which are the rich ones) exhibit higher combustion temperature,

an observation that justifies that NO formation is mainly controlled by combustion tem-

perature at that limit. Nitric oxide variability is also governed by oxygen availability and

combustion temperature. At rich mixture conditions, NO formation presents lower vari-

ation as the faster and slower burning rates do not present significant difference (figure

4.16). On the other hand, stoichiometric and lean mixtures present higher NO emission

variability. As it was revealed in table 4.8 above, stoichiometric and lean mixtures lead

to higher COVimep, translated into more intense combustion variability, which leads to

higher NO emission variability. Finally, figure 4.16 shows that the maximum NO con-

centration is produced by slightly lean mixtures, which is also the area where the higher

77

4. Experimental Investigation of Cyclic Emissions Variability

NO emission variation is observed.

However the question still remains: burn rate or oxygen availability is the most impor-

tant parameter on NO formation. Figure 4.17 illustrates the relationship between nitric

oxide formation and early flame kernel development for rich and lean mixture composi-

tion. In lean mixtures, early flame kernel has a more intensive impact on NO formation

compared to rich operating conditions, mainly due to the excess of oxygen. Cycles that

initial flame kernel occurs early in combination with excess of oxygen benefits the NO

emissions level. Delayed early flame kernel seems to be independent from the overall

mixture equivalence ratio value as the same low NO level is presented at rich and lean

mixtures. In conclusion, oxygen availability significant benefits NO formation at cycles

which present early initial flame kernel and declares fast burning rate. Conversely, excess

of oxygen has no impact on NO formation at cycles which present delayed initial flame

kernel and slow burning rate.

Figure 4.18: Relationships between λ and carbon monoxide and carbon dioxide concen-

trations at 6000 rpm, 80% TP and identical spark ignition timing.

Carbon monoxide is a product of incomplete combustion, due to oxygen depletion in

the bulk flow (rich combustion) or flame quenching (lean combustion), or CO2 dissociation

at high post combustion temperatures. Highest CO levels appear in rich operating condi-

78

4.5. Results

tions where COVCO is minimized (Table 4.8). This is because stoichiometry is basically

the sole parameter affecting CO at rich conditions, while variation of other parameters

may be responsible at stoichiometric and lean conditions. Figure 4.18 illustrates the re-

lationship between measured CO and CO2 at various mixture compositions. CCV of CO

is very low for a given lambda value while it increases exponentially as lambda drops

below unity. Fluctuations of the burn rate and the initial phases of combustion little

affect CO formation, which is known from other studies as well [98, 107]. Similar to CO,

unburned hydrocarbons were maximized at rich conditions due to mixture stoichiometry

(Table 4.8). Therefore, averaged CO2 concentration at that conditions is minimized.

The observed trends for the impact of mixture lambda on combustion and performance

variability are also validated at 4000 rpm and 80% throttle position (Table 4.7). The

minimum COVimep is presented at slightly rich conditions, which is accompanied by the

highest averaged IMEP value. Nitric oxide and carbon monoxide emissions also present

the lowest CCV value at rich conditions, while mixture at rich conditions ignites earlier

(ϑ5%).

4.5.3 Impact of Ignition Timing on Variability

Engine ignition timing setting is usually at the Maximum Braking Torque (MBT) point.

Earlier or later ignition timings lead to lower piston net work. Slightly retarded or ad-

vanced ignition may occur in actual application as a result of random perturbations in

the in-cylinder conditions, such as fuel availability in the spark plug area. Dependence of

combustion and, in particular, emission variability on the ignition timing setting is little

studied in published literature [15, 71].

Table 4.9 shows performance values and emission levels at five different ignition points.

Slightly rich operating conditions were selected in order to limit the effect of lambda fluc-

tuations on the investigation of ignition timing on combustion variability. The maximum

mean IMEP is accompanied by the minimum COVimep at the MBT case. MBT operating

point also presents the minimum COV value for the nitric oxide and carbon monoxide

exhaust gas concentrations. From ”late” to ”early” ignition timings, averaged NO and

CO concentrations are increased, and so does the mean unburned hydrocarbons emissions

measured at the exhaust pipe.

Variations in spark timing lead to heat release rate fluctuations, such as different early

flame kernel phasing and burning rate, which result in different pressure evolutions. Figure

4.19 illustrates the mean pressure trace and gross heat release rate for the same operating

conditions of table 4.9. The MBT ignition timing is the case that presents the higher

average value and the lower COV of IMEP. Advanced ignition timing (MBT +5) presents

the higher maximum cylinder pressure; this occurs because the maximum heat release

rate is achieved before the TDC. In this case, gross heat release traces present that flame

79

4. Experimental Investigation of Cyclic Emissions Variability

Table 4.9: Mean and COV experimental data for various spark ignition timings at 4000

rpm, 80% throttle position and rich operating conditions (λ=0.94).

CasesIMEP Pmax ϑ5% NO CO CO2 HC Torque

[bar] [bar] [oCA] [ppm] [%] [%] [ppm] [Nm]

MBT -15mean 8.81 39.48 353.9 858 2.62 12.14

3518 39COV [%] 3.27 12.72 1.04 19.99 12.12 1.83

MBT -10mean 8.86 42.88 351.2 970 2.69 12.08

3681 40COV [%] 2.90 12.13 1.04 18.24 9.89 1.61

MBT -5mean 8.90 46.85 348.2 1016 2.94 11.90

3913 41COV [%] 2.30 12.99 1.11 16.70 9.41 1.71

MBTmean 8.91 51.60 345.3 1298 2.72 12.03

4030 41COV [%] 1.76 11.61 1.30 11.33 9.38 1.77

MBT +5mean 8.91 54.86 343.2 1326 2.92 11.90

4158 40COV [%] 1.90 11.86 1.26 13.55 10.35 1.70

Figure 4.19: Mean pressure and gross heat release traces for rich conditions (λ=0.93),

4000 rpm, 80% throttle position and various spark timings.

80

4.5. Results

development and propagation do not have any significant difference with the MBT case,

except from the ignition timing which leads to lower net work during the combustion

cycle. On the other hand, as ignition timing becomes retarded (MBT -5, MBT -15),

maximum cylinder pressure is reduced and occurs later in the cycle. At ”late” ignition,

gross heat release traces present that early flame kernel development takes place close to

TDC, where mixture pressure and temperature are high and this lead to a faster ”ignition”

phase. This could explain the reason why ”MBT -15” case presents a small difference at

the start of heat release rate compared to the other ignition timings. However, the phase

of flame propagation occurs during the expansion of the piston, where pressure increase

during combustion is limited by the volume expansion. This results in a slower burning

rate and a lower peak pressure. In such retarded ignition timing (”MBT -15”), it was also

observed that a few cycles result in incomplete fuel combustion.

Figure 4.20: Relationship between IMEP and maximum cylinder pressure at 4000 rpm,

80% TP and λ=0.94, for different ignition timings.

In most engine operating conditions, ignition timing is usually set at MBT. Figure

4.20 illustrates the relationship between the IMEP and the maximum cylinder pressure

for five different ignition timings. Maximum IMEP is given at the MBT ignition tim-

ing, while advanced ignition timings can lead to higher maximum cylinder pressure but

81

4. Experimental Investigation of Cyclic Emissions Variability

to lower IMEP. In fact, advanced ignition timing over-increases compression work and

although expansion work is also increased, the net work is lower than the MBT case.

Also, with early ignition the peak pressure and temperature may be sufficient to cause

knock, therefore tested ignition cases over ”MBT +5” were avoided in this investigation.

Conversely, retarded ignition leads to lower pressure rise, as combustion evolution occurs

during piston expansion where combustion volume is increased and final maximum cylin-

der pressure and expansion work are also decreased. Finally, it is observed on figure 4.20

that few cycles presented sufficient incomplete combustion that leads to less engine power

and increases IMEP variability.

Figure 4.21: Relationship between ignition delay and maximum cylinder pressure at 4000

rpm, at 4000 rpm, 80% TP and λ=0.94, for different ignition timings.

The effect of ignition timing on burning rate has to be investigated. Figure 4.21

presents the relationship between the required time for the development of the early flame

kernel (ϑ5% - ϑign) and the maximum cylinder pressure (Pmax). Retarded ignition timings

lead to faster development of the ”initial” flame kernel, although they result to a lower

maximum pressure. The spark discharge in relation to piston position can explain this.

Retarded ignition occurs closer to the TDC, where cylinder pressure and temperature are

relatively higher when spark is discharged. The mixture ignites faster but this does not

82

4.5. Results

Figure 4.22: Relationship between maximum cylinder pressure and crank angle at which

maximum cylinder pressure occurs for different ignition timings at 4000 rpm, 80% TP

and λ=0.94.

lead to high mixture pressure because the flame propagation meets the piston phasing

during expansion. Conversely, in the case of advanced ignition timings, higher maximum

cylinder pressure is achieved but the mixture takes longer to ignite.

Combustion variability is limited between the faster and the slower burning cycles.

Figure 4.22 illustrates the maximum pressure versus the crank angle at which the max-

imum pressure occurs for five different ignition timings. The characteristic shape of the

formed limits is well-understood in published literature [71] and correspond to the fast

and the slow burn combustion boundaries. Retarded ignition results in much lower max-

imum cylinder pressure (Pmax) compared to the MBT case. These cases present either

high ϑPmax when they have a fast burning behavior or ϑPmax closer to the TDC when they

present a slow burning behavior. Advanced ignition timing points correspond at higher

maximum cylinder pressure (Pmax) and ϑPmax slightly after TDC. In this area, combustion

variability, which is defined as the difference between the fast and the slow burn curve,

is significantly low. It is finally shown that ignition timing has an impact on combustion

variability and this variability is limited between the fast and the slow burn rates.

83

4. Experimental Investigation of Cyclic Emissions Variability

With regard to ignition timing impact on NO variability, table 4.9 presents that MBT

ignition timing leads to the lowest COVNO, similarly to COVimep. Figure 4.23 illustrates

the relationship between NO and crank angle at which the maximum pressure occurs. The

fast burning rate curve corresponds to higher NO levels, compared to the slow burning

rate. In advanced ignition timings, flame propagation evolves close to or even before TDC,

which results to higher maximum pressure, higher peak combustion temperature and

finally higher level of NO formation. On the other hand, the slow burn limit represents

the incompleteness of combustion; this concludes to lower fuel mass burned per cycle,

lower heat release and finally lower level of nitric oxide formation. As combustion rate

is switched from slow to fast burning rate curve during real operation CCV, nitric oxide

concentration is increased, which declares the relationship between NO formation and

combustion rate.

Figure 4.23: Relationship between NO formed and crank angle at which maximum cylin-

der pressure occurs for different ignition timings at 4000 rpm, 80% TP and λ=0.94.

Ignition timing determines the combustion efficiency and therefore it has an impact

on unburned hydrocarbons and exhaust gas temperature. Table 4.9 shows that retarded

ignition timing leads to lower unburned hydrocarbons concentration, while it was also

found that the mean exhaust temperature was increased by 20K. The same trends are also

84

4.5. Results

validated by another experimental study [120]. Furthermore, mean value of CO emissions

was found to be decreased as ignition timing occurs ”late” and opposite to this, the

minimum COVCO was found at MBT. Figure 4.24 illustrates the CO emissions from each

individual cycle versus the crank angle at which the maximum pressure occurs for various

spark timings. The trend line reveals that advanced ignition timing leads to a higher CO

emissions level. In fact, ”early” ignition timing increases cylinder pressure that much,

so that flame front ”pushes” more air-fuel mixture mass into crevices gap where flame is

quenching on the ”cold” chamber wall and this increase combustion incompleteness [120].

On the contrary, retarded ignition timing leads to higher combustion duration and lower

cylinder pressure which benefits combustion completeness. Finally, the higher combustion

efficiency (more burned fuel) explains the increase of exhaust gas temperature that was

observed at ”late” ignition timings.

Figure 4.24: Relationship between NO formed and crank angle at which maximum cylin-

der pressure occurs for different ignition timings at 4000 rpm, 80% TP and λ=0.94.

The variability of carbon monoxide has to be further investigated. Figure 4.24 also

shows that the scatter of CO emissions increased at retarded ignition timings for the

same crank angle that maximum cylinder pressure occurs; that is the difference between

the fast and the slow burn curve. In fact, such an observation has its roots on the

85

4. Experimental Investigation of Cyclic Emissions Variability

Table 4.10: Mean and COV experimental data for various spark ignition timings at 6000

rpm, 20% throttle position and rich operating conditions (λ=0.93).

CasesIMEP Pmax ϑ5% NO CO CO2 HC Torque

[bar] [bar] [oCA] [ppm] [%] [%] [ppm] [Nm]

MBT -15mean 7.45 35.05 356.5 772 1.96 12.61

2973 31COV [%] 3.04 10.51 0.87 22.81 14.15 1.30

MBT -10mean 7.50 39.09 353.3 907 2.77 12.08

3141 32COV [%] 2.96 10.29 0.94 32.08 17.99 2.80

MBT -5mean 7.51 43.14 346.9 1017 2.66 12.11

3261 33COV [%] 2.75 11.35 1.11 23.75 24.74 3.45

MBTmean 7.54 48.38 345.5 1206 2.84 12.00

3497 33COV [%] 2.41 9.05 1.04 22.85 21.47 2.96

MBT +5mean 7.49 52.66 342.9 1410 2.84 12.00

3500 33COV [%] 2.50 9.48 1.02 21.54 22.37 3.03

combustion process, as the achieved cylinder pressure controls the trapped mass into

crevices. Cylinder pressure varies from cycle to cycle and finally the trapped mass and

the completeness of combustion also cyclic varies. Cycles that present higher cylinder

pressure lead to more air-fuel mixture into the piston wall gap that increases CO and HC

emissions while cycles that present lower cylinder pressure have an opposite effect. It is

finally understood that the nature of retarded ignition, which is higher cylinder pressure

variability (COVPmax at table 4.9), is the parameter that leads to higher CO variability

(COVCO).

The impact of ignition timing at combustion and emissions variability has been also in-

vestigated at 6000 rpm and 20% throttle position and rich operating conditions (λ=0.93).

The results of this investigation are briefly presented on table 4.10 and the observed trends

are in agreement with table 4.9. It was found that the MBT case concludes the minimum

COV regarding imep, ϑ5% and emissions. In fact, the lower COVNO is observed at ”MBT

+5” case, however it has to be considered that the observed torque difference between

the two cases is negligible. Unburned hydrocarbons are decreased as ignition timing is re-

tarded, while exhaust temperature was also found to raise. Compared to the table 4.9, the

most important difference is that the COV of carbon monoxide concentrations is higher at

6000 rpm and 20% TP. Although at higher engine speed conditions turbulence intensity

is increased which improves homogeneity and CO consumption, the higher residual gas

fraction at partial load conditions leads to higher COVCO.

86

4.6. Conclusions

4.6 Conclusions

The operation of spark-ignition engines is limited by combustion variations from cycle to

cycle, which results a higher variation on pollutant emissions. In this study, the cycle-

to-cycle variability of NO and CO, as well as of CO2 and HC to a limited extend,

was investigated and the combustion factors which are related with the performance and

emission variability were presented. To that aim, a four cylinder PFI spark ignition engine

was employed, operating in a wide range of load, lambda and ignition timing.

Engine load has a significant impact on the variability of engine performance and

emissions. At low load operating conditions, engine performance variability (COVimep) is

higher. A throttled engine is characterized by higher residual gas fraction which dilutes

mixture and increases combustion and emissions variability. Lower NO mean value was

found at these conditions, due to lower cylinder pressure and mixture temperature, how-

ever the COVNO was higher due to the higher per cent of residual gas and therefore to

the more diluted mixture.

The impact of combustion stoichiometry on the variability of engine performance and

emissions was also investigated. Slightly rich mixtures showed the higher IMEP and

the lower COVimep. Higher variation of combustion in lean mixtures has its roots on

early flame kernel development, as slightly lean mixtures ignite slower and later than

rich ones. This leads to slower flame propagation and lower cylinder pressure. In lean

mixtures, higher variation of the crank angle that early flame kernel development occurs

was detected as the main reason of NO variability, as NO formation depends on burning

rate and mixture temperature.

Last but not least, the impact of ignition timing on the variability of engine per-

formance and emissions has been investigated. MBT ignition timing was found to be

accompanied by the highest IMEP value and the lowest COVIMEP . Advanced ignition

timings are characterized by faster combustion and higher cylinder pressure. On the other

hand,retarded ignition results to a shorter delay in the development of early flame kernel,

as this occurs closer to TDC. At retarded cycles, flame propagation is shifted to the pis-

ton expansion phase, where cylinder volume is increased and finally maximum cylinder

pressure is decreased. Because of the expansion, both cylinder pressure and crank an-

gle at which maximum cylinder pressure occur vary more in each cycle. The significant

variation of cylinder pressure at retarded ignition conditions was found to be the main

source of NO and CO variation in this investigation. On the one hand, NO formation is

strongly linked with the temperature vs time profile, and the higher perturbation of the

latter at retarded cycles is the main root of the high NO variability. On the other hand,

CO variability is linked with the variation of cylinder pressure which affects the trapped

mixture mass into crevices at each cycle and finally the combustion efficiency.

87

4. Experimental Investigation of Cyclic Emissions Variability

88

Chapter 5

Emission Model Validation and

Application for CCV Prediction

5.1 Introduction

This chapter aims to investigate the cyclic emission variability by using the novel emissions

model developed in this study. The new emissions model, which was explicitly presented

on chapter 3 is validated against the experimental data of chapter 4. The results of the

proposed model are compared against the extended Zeldovich mechanism for NO and a

simplified two step reaction kinetic mechanism for CO, which both consist a simplified

modeling approach. The improved predictability of the new model against the simplified

emission model is explored under various engine conditions.

The chapter begins with the detailed presentation of a two zone heat release model.

The model uses the measured cylinder pressure to calculate the net and gross heat release

and the thermodynamic characteristics of the two zones. The two zone thermodynamic

data are utilized not only by the proposed detailed emission model but also by the simpli-

fied model approach. In a first step, the two zone emission model is validated under mean

cycle simulations, and is compared with the simplified emission model. Then, the detailed

emission model is applied on a cyclic emission variability simulation study. Under various

engine operating conditions, the new model is validated against cycle-resolved measured

emission data and is compared with the cyclic prediction of the simplified model. The

comparison between the simplified emissions model and the proposed emissions model

allows to understand whether pollutants formation is primarily a thermodynamic or a

chemistry problem.

89

5. Emission Model Validation and Application for CCV Prediction

5.2 Modeling Approach

The proposed detailed chemistry emission model requires the thermo-physical properties

of the in-cylinder mixture as an input. The cylinder thermodynamic properties are cal-

culated by applying a two zone heat release analysis model on the measured cylinder

pressure. This model requires the measured crank-angle resolved cylinder pressure and

other macroscopic combustion parameters (lambda, residual gas fraction, combustion ef-

ficiency), in order to calculate the heat release, the burn rate, the temperature profiles

and the thermodynamic properties of the two zone mixture. The calculation of NO and

CO concentrations is then performed for each step of the calculation on the burned zone.

Using the calculated thermodynamic properties, the examined detailed chemistry model

[54] is compared against simplified kinetic models for NO and CO prediction [15, 40, 45].

5.2.1 Two Zone Combustion Analysis

5.2.1.1 Cylinder Model

The two-zone heat release analysis is a widely known methodology that offers the possi-

bility of avoiding combustion modeling (when cylinder pressure is measured), to rapidly

calculate the thermodynamic properties of the burned and unburned mixture, while the

number of iterations required for convergence is significantly low [121–123]. In the pro-

posed model the well-known subroutines of Ferguson [102] are used. Ferguson uses a

ten (10) species scheme for the combustion products, which consists of CO2, H2O, N2,

O2, CO, H2, H, O, OH, and NO. The thermodynamic properties of these species are

given by the well-known polynomial fitted curves of JANAF data, as these described by

Buttsworth [124]. Regarding fuel thermodynamic properties, a similar approach with

polynomial curves is given by Heywood [15].

Ferguson [102] calculates mixture temperature separately for burned and unburned

zone. In the unburned zone, which corresponds to mixture temperature lower than 1000K,

thermodynamic properties are calculated by the mixing of fuel, air and residual gas.

Unburned zone temperature is calculated as follows:

dTudϑ

= −

dQL,u

dϑm(1− xb

)cp,u

+υucp,u

∂ ln υu∂ lnTu

dP

dϑ(5.1)

At elevated temperatures, typical of the combustion process (burned zone), speciation

is calculated by the equilibrium state of Olikara and Borman [43] for the combustion

products H, O, N , H2, OH, CO, NO, O2, H2O, CO2 and N2.

At the combustion start crank angle, the initial temperature value of the burned zone

is calculated by assuming adiabatic combustion. Then, at each time interval, the burned

90

5.2. Modeling Approach

zone temperature differential depends on the burn rate

(xb,

dxbdϑ

), the measured cylinder

pressure and the heat losses of this zone, according to Eq. 5.2).

dTbdϑ

= −

dQL,b

dϑmxbcp,u

+υbcp,b

∂ ln υb∂ lnTb

dP

dϑ+hu − hbxbcp,b

dxbdϑ

(5.2)

Heat losses at each zone are calculated by applying the Annand model [125] at each

zone, which takes into account not only heat convection but also radiation [44].

dQL,j

dϑ= Aw,j

[aht

kjDRebht

(Tj − Tw

)+ εhtσ

(T 4j − T 4

w

)](5.3)

The wet area of each zone is calculated by equations 5.4 and 5.5 :

Aw,u = Acyl(1−√xb)

(5.4)

Aw,b = Acyl

(1√xb

)(5.5)

and the total wet area is geometrical characteristic of the cylinder (Eq. 5.6).

Acyl = πD2

2+ πDycyl (5.6)

5.2.1.2 Integration Flow

The developed model uses an iterative method for the calculation of the two zone mixture

properties during the closed thermodynamic cycle. A similar methodology is used by

Guezennec and Hamama [122]; however here, instead of calibrating the heat transfer

coefficient, the amount of trapped fuel is adjusted. Finally, the model is terminated when

the calculated and measured combustion efficiency converge. The measured combustion

efficiency is calculated from the exhaust emissions using the following equation:

ηc =[CO2] +

(0.5× [CO]

)[CO2] + [CO] +

(n× [HC]

) (5.7)

where n is the number of carbon in the fuel type.

91

5. Emission Model Validation and Application for CCV Prediction

Figure 5.1: Flowchart of the two zone heat release model.

92

5.2. Modeling Approach

The iterative process of the model is schematically outlined in figure 5.1. Firstly, the

model reads all the required engine data, such as the cylinder geometrical characteristics,

the species thermodynamic coefficients, the fuel type data and the user defined data (λ,

f , ηcomb). Before the two zone heat release analysis, the initial condition of the cylinder

model is given by a single zone heat release analysis (Eq. 4.4).

Through the single zone heat release analysis it is possible to estimate the initial values

for the start of combustion (SOC) and the end of combustion (EOC). Furthermore, the

initial value for air mass flow is estimated by assuming a volumetric efficiency, regarding

the throttle position, while the fuel mass flow is calculated using the lambda index. The

cylinder overall lambda was indirectly estimated by the cycle-resolved emissions using the

Brettschneider formula [126].

The main program can be divided into three sub-models; the compression, the com-

bustion and the expansion routines. Compression duration is from intake valve close

(IVC) up to start of combustion (SOC). During this phase, it is calculated the unburned

zone temperature and mixture properties (h,s,u,γ). Combustion phase, which runs from

SOC to EOC, calculates the temperature and thermodynamic properties of the two zones

as well as the net and gross heat release (instantaneous and cumulative), the burned fuel

and the mass fraction burned (xb). The last expansion phase that follows combustion up

to the EVO calculates the temperature and thermodynamic properties of the two zones

and the averaged mixture indexes.

After the main program, the convergence criteria are checked. This includes that the

SOC and the EOC are identical for two iterations (the heat release change due to the

new specific heat ratio at each iteration), while the combustion efficiency is also compared

with Eq. 5.7.

5.2.1.3 Error Formula

The evaluation of modeling is performed by using the percentage error formula. In order

to calculate the percentage error between the measured value and the calculated value,

the following formula is used:

Error[%] =|Simulation− Experiment|

|Experiment|(5.8)

It has to be noted that the error of a group of measured points and a group of simulated

points is performed by utilizing the mean values of each group.

93

5. Emission Model Validation and Application for CCV Prediction

5.2.2 Emissions Modeling

5.2.2.1 Proposed Modeling

The mathematical background of the detailed chemistry model is in detailed presented

in chapter 3. This model provides improved prediction of NOx emissions, while in this

chapter the CO emissions predictability of the model is also presented. The model can be

coupled with any multi-zone thermodynamic model, where the burned zones are assumed

as well-stirred reactors. In this application, a single burned region is adopted. The

computational time of the proposed detailed chemistry model is slightly increased but at

the same order of magnitude to existing simplified emission models. The basic benefits

of this detailed chemistry model are its simplicity and the absence of any calibration

parameter that makes it a generalized application, possible to be applied in any type of

homogeneous combustion engine.

5.2.2.2 Simplified Modeling

The simplified emission model consists of a simplified NO and CO scheme. Regarding

NO emissions, the widely known extended Zeldovich mechanism is utilized. The chemical

pathways of this mechanism are described in Eq. 2.1-2.3. The final NO production rate

is given by Heywood [15], and is described in Eq. 2.4.

As for CO model, a kinetic model for the WGS reaction (Eq. 2.15) is utilized [15, 40].

The model assumes that CO concentration increases rapidly in the flame front zone to

be successively oxidized to CO2 via a kinetically controlled mechanism. Simplified CO

formation rate is expressed through equation 2.16.

5.3 Cycle Simulation

The two-zone heat release analysis estimates the thermodynamic properties of the closed

thermodynamic cycle, from IVC to EVO, by using the measured cylinder pressure. These

thermodynamic properties are utilized from different pollutant formation approaches, such

as equilibrium [43] , the simplified emission model and the detailed chemistry model. As

simplified model uncertainty factor is in the order of 20% to 80% [15], the kinetics of both

NO and CO submodels were adjusted to measured emissions data at 4000 rpm and the

lowest IMEP (20% TP).

An example of the measured pressure trace which corresponds to 4000 rpm, 80%

of wide open throttle and stoichiometric mixture is illustrated in figure 5.2a. The IVC

corresponds to 23o [CA] ABDC, EVO at 40o [CA] BBDC, while the charge is ignited at 48o

[CA] BTDC. Figure 5.2b illustrates the temperature profiles of the two zones, which are

94

5.3. Cycle Simulation

(a) Cylinder pressure. (b) Zones temperature.

(c) NO emissions. (d) CO emissions.

Figure 5.2: Evolution of cylinder pressure, zone temperatures, NO and CO concentra-

tions at 4000 rpm, 80% throttle position, stoichiometric mixture and MBT. (a) measured

cylinder pressure of the closed thermodynamic cycle, (b) the calculated gas tempera-

ture (Tg), the burned zone temperature (Tb) and the unburned temperature (Tun), (c)

calculated NO concentrations under conditions of equilibrium/kinetics control, (d) CO

concentrations under conditions of equilibrium/kinetics control.

95

5. Emission Model Validation and Application for CCV Prediction

calculated by the cylinder model. The mean gas temperature never reaches the maximum

calculated burned zone temperature because of the non-perfect combustion efficiency.

Figure 5.2c illustrates the evolution of NO concentrations assuming different pollutant

formation assumptions, while the calculated values are compared against the measured

NO value experimentally determined at the EVO for this specific combustion cycle. In

the equilibrium model, the NO profile follows the shape of the burned zone temperature,

which seems to be incorrect as the NO formation rate is independent from the burn

rate. The equilibrium model error is about 95%. The simplified model presents faster

NO formation compared to the detailed chemistry model, although both models estimate

finally the same NO level with a minor error of 3%.

CO concentration is also controlled by kinetics. However the air-fuel ratio is more im-

portant for CO formation than temperature [16]. Figure 5.2d depicts that the equilibrium

model corresponds to faster CO formation and destruction, where the final CO predicted

value presents an error of 98%. The simplified model predicts faster CO formation and

almost kinetics equilibrium during the piston expansion, with an error in the order of

74%. On the other hand, the detailed chemistry model predicts slower CO formation and

CO destruction even at EVO. In fact, CO destruction may also occur at the blow down

exhaust process, which can justify the observed 12% error.

5.4 Mean Cycle Emissions Prediction

After this initial validation, the models are tested at various engine conditions to explore

their prediction capabilities. The error magnitude on NO and CO prediction of both the

detailed chemistry model and the simplified model is examined under various engine load,

mixture stoichiometry, ignition timing and engine speed conditions. The effect of each

separate engine variable on emissions modeling error is the field of concern of this section.

5.4.1 Engine Load

In a conventional SI engine, fuel and air are mixed together in the intake system and

further mixed with the residual gas inside the cylinder during intake process. The throttle

valve position controls engine load as it controls the inlet mixture and residual gas fraction.

Lower residual gas fraction improves flame speed and increases combustion temperature

[127]. In this modeling approach, a typical profile of residual gas fraction was adopted

based on previous studies [32, 53], where the residual gas fraction is a function of inlet

pressure and is reduced as IMEP is increased.

The left column of figure 5.3 represents the NO predicted values for both the simplified

and detailed kinetic models which are compared to the cycle resolved experimental values

for various engine speed and IMEP conditions. As IMEP increases, the burned zone

96

5.4. Mean Cycle Emissions Prediction

Figure 5.3: Effect of engine load on NO and CO formation for stoichiometric mixture

and various engine speed conditions: (left column) NO simulation, (right column) CO

simulation.

97

5. Emission Model Validation and Application for CCV Prediction

temperature increases accordingly while the residual gas fraction decreases. The latter

benefits the NO formation.

The simplified NO model was initially calibrated at 4000 rpm and low IMEP con-

ditions, similar to other studies [31]. This low load operating point was selected, as the

thermal NO mechanism underestimates the NO emissions at low temperature conditions.

Without considering this calibrated operating point, the simplified model presents error

within 2% and 81.5% over the eight (8) presented points. Additionally, it is observed that

the error increases, as engine speed also increases.

On the other hand, the detailed chemistry model shows an improved prediction of

formed NO. Its maximum error is 2.6% over the nine (9) presented points, which is found

at 5000 rpm and 80% TP. However the detailed chemistry model presents a typical error

of less than 1.3%. The improved prediction can be explained on the kinetic calculation of

the other important species (CO, O2, H2) and on the detailed chemical pathways, which

are included for the NO formation and destruction.

CO predicted values for the two emission models compared to the experimental values

are also shown in fig 5.3. Similarly to the NO investigation, the simplified CO model is

calibrated to the measured CO value at 4000 rpm and low IMEP (20% TP), as residual gas

fraction and CO2 are increased at this engine point. Considering the remaining operating

points investigated the simplified model presents a mean error of 45.6%. It is also depicted

overall the engine speed conditions that error increases as engine load increases.

The detailed chemistry model can better estimate the CO values at various load

conditions. Its maximum error is 65%, observed at 6000 rpm and high load conditions,

which is an expected behavior of this model. In fact, as IMEP (and speed) are increased,

maximum cylinder pressure is also increased, which ”pushes” more mixture in the crevices

volume, a phenomenon which is non considered by the model, as only one burned region

is adopted. However that, the detailed chemistry model presents an average error of less

than 30%, which is 35% lower than the average observed error of the simplified CO model.

5.4.2 Mixture Stoichiometry

The precision of predicting engine out emissions over different lambda values around

stoichiometry is very important for contemporary gasoline engines, due to its application

on the optimization of the emission control systems. The predictability of the simplified

emission model and the detailed chemistry model for CO and NO engine out emissions

is given in Fig. 5.4.

Regarding NO emissions, the detailed chemistry model presents better accuracy com-

pared to the simplified emission model. In the entire investigated range of engine speeds

and equivalence ratios, the average observed error of the detailed chemistry model is 3.8%

over the nine investigated operating points, while for the simplified model this is 74%.

98

5.4. Mean Cycle Emissions Prediction

Figure 5.4: Effect of mixture lambda on NO and CO formation for 80% throttle position

and various engine speed conditions: (left column) NO simulation, (right column) CO

simulation.

99

5. Emission Model Validation and Application for CCV Prediction

This 74% average observed error mainly originates from stoichiometric and slightly rich

conditions. At these conditions, engine torque and mixture temperature are maximized

but oxygen is expected to be faster consumed by CO and H2 and not by the thermal

NO mechanism. The latter can occur by considering other chemical pathways, which

the detailed chemistry model does. Finally, at both stoichiometric and lean conditions,

detailed chemistry model presents a 2.7% average error while the error of the simplified

model is about 47%, which is comparable to other literature studies [31, 32, 44].

CO formation, at homogeneous combustion conditions, is primarily controlled by

fuel/air equivalence ratio. However, CO is also kinetically controlled, especially at rich

conditions through the WGS reaction [15, 16]. The detailed chemistry model was applied

for the prediction of CO emissions under various lambda conditions. Under lean condi-

tions and especially at 4000 rpm and 6000 rpm, the detailed model underestimates the CO

emissions by 92% and 67% respectively. In fact, under lean conditions, CO is primarily

controlled by other mechanisms such as crevices and mixture non-homogeneneities, which

are not considered in this model. By comparing the two CO models at overall lambda

and engine speed conditions, the detailed chemistry model presents average error of 40%

for the nine investigated operating points, which is better than the 63% average observed

error of the simplified model. Although CO prediction of the detailed chemistry model

was found less efficient at this investigation, it has to be noted that this mixture lambda

investigation occurred at 80% throttle position, where the formed CO in crevices (due

to the high cylinder pressure) is not considered by the single burned zone (see Chapter

5.4.1).

In this investigation, where the effect of lambda mixture is explored, it is worth noting

the benefits of using more complicated chemistry at a cost of computational time. The

most simplified approach is the utilization of the equilibrium, as it was presented by

Olikara and Borman [43]. The model was able to run the total combustion cycle, from

the SOC to the EVO in less than 15 seconds at each operating point and its average error

was 78% for NO emissions and 67% for CO emissions for the nine investigated points.

The simplified kinetic model, which utilizes the Zeldovich mechanism for NO and the

WGS reaction for CO, requires less than two minutes to be executed and its average error

was found to be about 74% for NO and 63% for CO for the same nin points. Additionally,

the kinetics of this simplified model usually require calibration, which time cost is not

considered here [31, 32]. The execution time of the detailed chemistry model is highly

depending on the number of combustion steps and is shorter for rich combustion which

presents lower combustion duration compared to lean conditions. However, its average

execution time is at the order of five to ten minutes and its accuracy was found to be in

the order of 3.8% for NO and 40% for CO emissions. Compared to multi-dimensional

combustion codes, the proposed detailed chemistry model is proposed as a satisfactory

trade-off between computational cost and accuracy. Finally, by increasing the number

100

5.4. Mean Cycle Emissions Prediction

of burned zones, the accuracy of CO can be improved, while the computational cost is

expected to slightly increase, due to the tools of parallel processing.

5.4.3 Ignition Timing

Spark timing controls the initiation of the combustion process and through that the en-

gine IMEP and engine-out emissions. The optimum timing gives the MBT, the maximum

IMEP and the lower cyclic variability of IMEP [17]. On the other hand, retarded ignition

timing is selected at engine warm-up conditions, in order to increase exhaust gas temper-

ature and rapidly warm up the catalyst, with a negative impact on fuel consumption.

Figure 5.5: Effect of ignition timing on NO and CO formation at 4000 rpm, 80% throttle

position and stoichiometric conditions: (left column) NO simulation, (right column) CO

simulation.

Figure 5.5 shows the impact of ignition timing on formed NO and CO for stoichiomet-

ric mixtures, at 4000 rpm and 80% TP. The detailed chemistry model predicts accurately

the decrease of formed NO concentration at retarded ignition timings. At 4000 rpm and

80% TP, the error of the detailed model is between 0.6% and 3.7%, while the error of

the simplified model is between 2.4% and 59%. In the case of CO emissions, it is ex-

perimentally observed that retarded ignition increases CO up to a rapid drop at very

”late” ignition timings. This CO trend is successfully predicted by the detailed chemistry

model, within an average error of 26%, while the simplified model presents a ”flat” area

and an average error of 89%. In fact, simplified model at both NO and CO cases predicts

small differences at various ignition timings and fails not only quantitatively but also

qualitatively.

A similar investigation of ignition timing at NO and CO emissions is presented at 6000

101

5. Emission Model Validation and Application for CCV Prediction

Figure 5.6: Effect of ignition timing on NO and CO formation at 6000 rpm, 20% throttle

position and stoichiometric conditions: (left column) NO simulation, (right column) CO

simulation.

rpm and partial load conditions (20% of WOT). Regarding NO emissions, the detailed

chemistry model accurately predicts the formed NO both qualitatively and quantitatively,

while the simplified model seems to follow only qualitatively. The average error of the

detailed chemistry model is about 1.7%, while the simplified model presents an average

error of 44.5%. In the case of CO emissions, the simplified model can not follow the

experimental CO trend. The mean error of the simplified CO model is about 157%, while

the average observed error for the detailed chemistry model is 36%. However, it has to

be reminded that the simplified model was calibrated at 4000 rpm and 20% and it is

expected to present higher errors at 6000 rpm.

In conclusion, it was found that the detailed chemistry model is able to qualitatively

and quantitatively predict theNO and CO emissions at various ignition timing conditions,

without any calibration and with a limited cost on computational time. On the other

hand, the simplified model was found inefficient for the investigation of the impact of

ignition timing on NO and CO emissions. The usefulness of the proposed model can be

found on engine design and optimization from the automotive industry, considering the

engine ignition timing optimization during warm-up.

5.5 Sensitivity Analysis of the Emission Model

Combustion process is influenced by several flow and chemical properties, such as large

and small scale turbulence, mixture stoichiometry, residual gas, crevices and heat transfer.

All these parameters also affect pollutants formation. In fact, several in-cylinder mixture

102

5.5. Sensitivity Analysis of the Emission Model

properties present cyclic dispersion which in the case of pollutant formation may give

significant deviation from the average cycle value. As residual gas fraction and wall

temperature were not cyclically measured values, their impact is investigated by using

the detailed chemistry model under a parametric analysis.

(a) NO emissions. (b) CO emissions.

Figure 5.7: Impact of wall temperature on pollutants formation using the detailed chem-

istry model.

Heat transfer is a significantly important term of the in-cylinder energy balance and

can affect the final calculated temperature. Figure 5.7 illustrates the impact of wall

temperature on NO and CO emissions at 4000 rpm, 80% throttle position, close to sto-

ichiometric mixture and 3% residual gas fraction. The impact of wall temperature from

300 to 600 [K] increases peak burned zone from 2630 to 2670[K]. This small temperature

deviation increases NO emissions by 470[ppm] and CO by 500[ppm], which is 34% and

0.6% deviation respectively. This means that wall temperature mainly affects NO for-

mation rather than CO formation. Additionally, the maximum emission limits assuming

adiabatic combustion are also illustrated.

The residual gas fraction also affects the prediction of pollutants formation. In this

model approach, by increasing the residual gas fraction the cylinder model calculates

less burned fuel and heat release, as the measured cylinder pressure remains constant.

Figure 5.8 presents the impact of the residual gas fraction at 4000 rpm, 80% TP and

stoichiometric mixture on NO and CO formation. Residual gas increase from 2% to 10%

reduces formed NO by 160% and CO by 13%. Values over 10% of residual gas fraction are

non realistic for this operating point (4000rpm, 80% throttle position). However figure 5.8

shows that small differences of residual gas fraction are more important at small amount

of residual gas fraction and less important at residual gas fractions over 10%.

103

5. Emission Model Validation and Application for CCV Prediction

(a) NO emissions. (b) CO emissions.

Figure 5.8: Impact of residual gas fraction on pollutants formation using the proposed

detailed chemistry model.

5.6 Application of the model to predict CCV

SI engine operation is limited by cyclic combustion variability. Combustion variability

not only affects engine performance but is linked with cyclic emissions variability (CEV).

While CCV is extensively studied in published literature, emission variability and in

particular NO variability has not received significant attention. In few computational

studies, simplified kinetic models are utilized which cannot predict the scatter of the

emission data [14]. Simplified models have been used so far because they are fast enough

to be used in many replications of the same nominal operation conditions. The method

developed in this study is of sufficient speed and of improved detail to be applied on CCV

pollutants prediction.

Figure 5.9 illustrates the maximum and minimum measured cylinder pressures for 150

consecutive engine cycles at 4000 rpm, 80% throttle position and stoichiometric mixture

of the engine used in this study. Variations in gas motion and composition result to

a variability not only to the cylinder pressure, but also to the burn rate, the cylinder

burned temperature and finally the pollutants formation. The 22 bar difference between

the maximum and minimum peak pressures results in more than 1000ppm deviation in

NO emissions and 5000 ppm for CO emissions. The model predictions also show the

different formation rates.

The detailed chemistry model presents an error of 6% on the prediction of NO emis-

sions at both fast and slow burn rates. On the other hand, the simplified emission model

overestimates the NO difference between the fast and slow burn rate cycles by 36%.

However the main error of the simplified NO model is displayed at the early phase of

104

5.6. Application of the model to predict CCV

Figure 5.9: Evolution of maximum and minimum IMEP cycle in relationship to crank

angle: (a) measured cylinder pressure, (b) calculated heat release and burned zone tem-

perature, (c) NO formation and (d) CO formation.

NO formation, where the fast and slow NO rates slightly differ. This declares that Zel-

dovich mechanism is more sensitive to the zone temperature profile and less to the burn

rate. Therefore, the predicted NO difference with the Zeldovich mechanism is more a

temperature effect than a kinetic issue.

CO variability is well described by the detailed chemistry model, while the error of

3.6% can be justified by mixture inhomogeneities, which are not considered here. On

the other hand, the simplified CO kinetic model systematically overestimates the CO

emissions and underestimates the CO deviation between fast and slow burn rate cycles by

67%. This model shows at the early stage of CO formation no significant variation between

the fast and slow burn rates, although the heat release rates differ greatly. Additionally,

the late stage of CO formation appears frozen (constant) with simplified modeling and

105

5. Emission Model Validation and Application for CCV Prediction

not constant with detailed kinetics. The reason that CO is overestimated at the slow

burn cycle from the simplified kinetic model is the mixture temperature, which is higher

during expansion, and the simplification of the WGS reaction. Finally the CO simplified

kinetic model appears to be a temperature dependent model, while detailed chemistry

appears to be more suitable for the estimation of CO variability.

In conclusion, the proposed detailed chemistry model was applied on the maximum

and minimum IMEP cycle at nominally steady state engine operating conditions. It was

found that this model predicts different production rates at various engine IMEP cycles

and finally different emission altitudes. On the other side, the simplified emission model

is less sensitive to the burn rate and may underestimate the cyclic emission variability.

The predictability of cyclic emission variability is investigated in the next section.

5.7 Cycle to Cycle Emission Prediction

In this section, the emission models are applied on a cycle to cycle emission simulation,

in order to explore the error magnitude under various engine operating conditions. In

order to characterize the efficiency of the different emission methodologies, three different

types of plots are utilized, such as the relationship between NO emissions and cylinder

pressure, the return map of each emission variable and the relationship of each variable

between measured and calculated value.

The relationship between NO emissions and maximum cylinder pressure is widely used

on a number of cycle to cycle emission variability studies [13, 14]. Such a figure depicts

the impact of combustion variability (pressure trace) on NO variability and highlights,

whether this relationship is linear or not, the impact of other mixture parameters on NO

formation. The maximum cylinder pressure represents the maximum cylinder temperature

that occurs in the cylinder, which is related with the thermal NO mechanism.

Another widely used method for characterizing cycle-resolved engine variables is the

return map, where a set of data (at nominally constant engine operating point) separated

by a fixed number of cycles (delay pairs) are plotted against each other [60, 128–130].

The resulting aggregate plot of all time-series delay pairs forms the return map, which

can reveal any low-dimensional relationship between two engine variables. In most cases,

the preferred delay interval is one engine cycle, so that next-cycle correlation is most

apparent.

Last but not least, the efficiency of the different modeling approaches is performed by

plotting the measured values versus the calculated values. Values close to the diagonal

represent successful modeling prediction, while the modeling error is higher for values far

from the diagonal. This is a standard methodology for the valuation of modeling approach

and it is adopted in this study.

106

5.7. Cycle to Cycle Emission Prediction

All types of plots have the same target; to depict the impact of detailed chemistry and

the absence of equilibrium on modeling of cyclic emission variability.

5.7.1 Engine Load

It was experimentally observed on chapter 4.5.1 that cyclic emission variability is in-

creased as low engine load conditions. The validation of emission dispersion modeling is

performed at two different engine load conditions (20% and 80%), 6000 rpm, stoichio-

metric mixture and identical ignition timing (41o [CA] BTDC). The predictability of the

simplified emission model and the detailed chemistry model for NO and CO emissions

variability is illustrated in figure 5.10 and in figure 5.11 respectively.

The prediction of cyclic NO variability for 20% throttle position is illustrated on

the left column of figure 5.12. The detailed chemistry model presents a higher scatter

as well as a better accuracy of the experimental data. The detailed chemistry model

underestimates the average NO measured data by 1%, while the simplified model by 43%.

In the third row of plots it is observed that the detailed chemistry model underestimates

the NO concentrations under 1500 ppm and overestimates the higher NO concentrations,

therefore the mean error is balanced to 1% error. A possible root of this observation may

be the mixture composition. In this investigation, HC were not cyclic measured and

mixture lambda may be richer at low IMEP cycles and leaner at high IMEP cycles.

The results of cyclic NO variability at 80% throttle position are illustrated on the

right column of figure 5.12. By comparing the average values of the measured and cal-

culated NO data, detailed chemistry model presents 26.5% error while simplified model

returns a 35.6% error. Compared to the 20% TP, the detailed chemistry model presents

higher error as the minor inaccuracies of the mixture lambda estimation contribute more

at higher cylinder temperature and pressure conditions. At the same time, simplified

model estimation is improved as thermal mechanism is more important at high load con-

ditions. Although that, detailed chemistry model appears more proper to predict NO

cyclic dispersion.

The modeling of cyclic emission variability was performed with a constant residual

gas fraction, and without adopting a residual gas estimation model [51, 131, 132]. Such

kind of models are able to estimate residual gas fraction based on cylinder pressure during

blowdown process. By introducing a residual gas model, NO scatter variability increases

and the standard deviation of the NO data is improved. In the case of 80% TP, the

calculated standard deviation error is about 50% for the detailed chemistry model and

more than 75% for the simplified NO model.

The efficiency of the detailed chemistry model and the simplified model on the pre-

diction of CO variability is illustrated on figure 5.13. Regarding the 20% TP, the error of

the averages value between the detailed chemistry model and the measured CO concen-

107

5. Emission Model Validation and Application for CCV Prediction

Figure 5.10: Prediction of cyclic NO variability for 20% (left column) and 80% (right

column) throttle position. (a) Relationship between NO and measured cylinder pressure,

(b) NO return map, (c) Relationship between measured and calculated NO.

108

5.7. Cycle to Cycle Emission Prediction

Figure 5.11: Prediction of cyclic CO variability for 20% (left column) and 80% (right

column) throttle position. (a) CO return map, (b) Relationship between measured and

calculated CO.

109

5. Emission Model Validation and Application for CCV Prediction

tration is 5.7% while the average error between the simplified model and the measured

values is 69%. As for the case of 80% throttle position, the two models present almost

equal predictability. The detailed chemistry model presents an average observed error of

82%, while the same error of the simplified model is 91%. The observation of both NO

and CO error show that lambda index slightly shifts to richer values. Shifting lambda to

leaner conditions could benefit both the NO and CO prediction. The lambda estimation

utilizes the mean HC value [126], which concludes the oil consumption and not only the

unburned hydrocarbons. Therefore, the qualitative prediction of CO variability from the

detailed chemistry model is satisfactory.

Finally, an important observation from the CO investigation is the scatter of the

predicted data. Detailed chemistry model presents a higher data scatter, which proves

to be primarily a chemistry issue and not just a ”initial” mixture composition problem.

Detailed chemistry model is more sensitive on the burn rate, while the burn rate is related

with the CO and H2 kinetic rates and finally with the CO engine-out emissions. Simplified

model, which assumes that only WGS reaction is kinetically controlled, underestimates

the scatter of CO emissions, conversely to the detailed chemistry model that calculates

all species kinetically.

5.7.2 Mixture Stoichiometry

The impact of mixture lambda on cyclic emission variability, which was experimentally

observed on chapter 4.5.2 is now approached by emission modeling. The rich and lean

operating conditions at 6000 rpm, 80% throttle position and identical ignition timing (41o

[CA] BTDC), which have been primarily presented in table 4.8, were selected to validate

the modeling efficiency. The final results are depicted in figures 5.12 and 5.13 for NO and

CO emissions.

In the case of rich operating conditions (λ ∼ 0.92), the detailed chemistry and the

simplified models seem to predict NO emissions equally. The simplified emission model

overestimates NO emissions over a few high pressure cycles, therefore its average error is

about 15%, while the detailed chemistry model presents on average error of about 41%.

However the detailed chemistry model has lower error on the standard deviation, which

is 1.3% compared to the 124% of the simplified model and finally the COVNO value is

overestimated by 74% compared to the 164% of the simplified model. Additionally, the

constant offset between the detailed chemistry model and the diagonal may be the result

of a constant lambda offset error.

On the right part of figure 5.12 and at slightly lean conditions (λ ∼ 1.06), the detailed

chemistry model is shown to predict cyclic NO variability more accurately. The error of

the detailed chemistry model is less than 10% compared to slightly more than 37% for the

simplified model. The COVNO value of the experimental data is also better estimated by

110

5.7. Cycle to Cycle Emission Prediction

Figure 5.12: Prediction of cyclic NO variability for rich (left column) and lean (right

column) mixture composition. (a) Relationship between NO and measured cylinder pres-

sure, (b) NO return map, (c) Relationship between measured and calculated NO.

111

5. Emission Model Validation and Application for CCV Prediction

Figure 5.13: Prediction of cyclic CO variability for rich (left column) and lean (right

column) mixture composition. (a) CO return map, (b) Relationship between measured

and calculated CO.

112

5.7. Cycle to Cycle Emission Prediction

the detailed chemistry model. As the residual gas fraction was adopted constant, and the

lean and diluted conditions present high residual gas fluctuations, the error in the COV

value was more than 50% for the detailed chemistry model and more than 100% for the

simplified one.

The constant lambda shift error at rich conditions is also confirmed in the CO investi-

gation. Despite that, the detailed chemistry model qualitatively predicts the experimental

trends and the data scatter. On the other side, the simplified CO model seems to be less

sensitive to various burn rates, and presents a ”flat” behavior close to 4% CO concen-

tration. Regarding lean conditions, the detailed chemistry model shows better accuracy

compared to the simplified approach. Their average errors on the mean value are 13%

and 280% for the detailed chemistry and the simplified models respectively. This large

difference was expected, as at lean conditions the WGS reaction is less important as it

is the CO oxidation and the kinetic control of oxygen (O2). Finally at more lean and

diluted conditions, the mixture becomes less homogeneous and the detailed chemistry

error is expected to increase.

5.7.3 Ignition Timing

The potential of the detailed chemistry model to predict cyclic emission variability at

various ignition timing conditions is examined in this sub-section. The test cases include

engine operation at 4000 rpm, 80% TP, stoichiometric mixture and five different ignition

timings (42o, 45o, 48o, 50o, 52o [CA] BTDC), where the 48o [CA] corresponds to the MBT

case. The results of cyclic NO and CO variability, regarding the retarded (42o [CA]

BTDC), advanced (52o [CA] BTDC) and MBT ignition timings are illustrated in figures

5.14 and 5.15.

The cyclic NO variability is investigated on figure 5.14 which includes the return map

and the error plot of NO. Detailed chemistry model has average error of 18%, 42% and

23% for advanced, MBT and retarded ignition timing while the simplified model presents

24%, 70% and 11% average error respectively. Considering all the ignition cases, the

detailed chemistry is better by 20%. At retarded ignition timing, burn rate is slower and

sometimes incomplete. At this limit, the return map of detailed chemistry model shows

a more deterministic rather than stochastic nature of NO cyclic variability, which is not

clearly observed by the measured NO emissions.

The right column of Fig. 5.14 describes the relationship between measured and cal-

culated NO values. It is observed that under all ignition timing conditions (advanced,

MBT and retarded), the detailed chemistry model can not predict the diagonal line, which

corresponds to zero error. Instead, it appears that at low IMEP cycles the model under-

estimates the NO emissions while at high IMEP cycles overestimates formed NO with

greater intensity at MBT and retarded ignition timings. This appears as at 4000 rpm

113

5. Emission Model Validation and Application for CCV Prediction

Figure 5.14: Prediction of cyclic NO variability at 4000 rpm, stoichiometric mixture, 80%

TP, 52o[CA] BTDC ignition timing (first row), 48o[CA] BTDC ignition timing (second

row) and 42o[CA] BTDC ignition timing (third row). Left column: NO return map, Right

column: Relationship between measured and calculated NO.

114

5.7. Cycle to Cycle Emission Prediction

Figure 5.15: Prediction of cyclic CO variability at 4000 rpm, stoichiometric mixture, 80%

TP, 52o[CA] BTDC ignition timing (first row), 48o[CA] BTDC ignition timing (second

row) and 42o[CA] BTDC ignition timing (third row). Left column: CO return map, Right

column: Relationship between measured and calculated CO.

115

5. Emission Model Validation and Application for CCV Prediction

engine lambda fluctuates more compared to the 6000 rpm. High IMEP cycles present a

faster burning rate due to richer mixture conditions and conversely a slower burning rate

leads to lower IMEP cycles. Therefore, it is considered that this wrong trend is the result

of inaccurate lambda estimation.

Figure 5.14 shows the prediction of CO variability for both the detailed and the sim-

plified models. Regarding the model errors, in the five tested ignition timing cases, the

detailed chemistry model presents an average error from 3.5% to 16% while the simplified

CO model has an average error from 79% up to 116%, which implies that the detailed

chemistry model is significantly better on predicting CO. More specifically, the detailed

chemistry model presents 8% error at 52o [CA] BTDC ignition timing, 16% at 48o [CA]

BTDC ignition timing and 11% at 42o [CA] BTDC ignition timing. On the other hand, in

the three presented (left) plots, the simplified CO model predicts average CO concentra-

tion close to 1.8% and seems to be less sensitive on burn rate variability. Its error is 89%,

113% and 86% at advanced, optimum MBT and retarded ignition timings respectively.

The right column of the figure illustrates that the detailed chemistry model predicts qual-

itatively and quantitatively the diagonal of the three plots. Last but not least, the return

maps of CO emissions show that under all examined ignition timing engine conditions,

CO variability nature is more a stochastic rather than a deterministic.

5.8 Summary and Conclusions

In this chapter, the proposed detailed chemistry model was validated against the experi-

mental data of chapter 4, which included various engine operating conditions with different

engine speed, load, stoichiometry and ignition timing. Under these engine conditions, it

was tested whether the detailed chemistry model can better predict NO and CO emis-

sions at mean IMEP cycle simulations and cycle to cycle simulations. Simplified kinetic

mechanisms were compared to the detailed emission model to depict the advantages of

the detailed chemical kinetics.

The engine load investigation on mean cycle NO formation showed that the error of the

simplified emission model is between 2% and 81.5%, while the detailed chemistry model

exhibits a mean prediction error of less than 1.3% over the nine investigated operating

points. Additionally, the mean cycle CO for the detailed chemistry model is less than

30% and for the simplified model about 45.6%. The improved prediction is explained by

the detailed chemical kinetics calculations of the other important species (O2, H2, H, OH

etc.) and on the detailed chemical pathways, which are considered for the NO formation

and destruction in the detailed chemistry model.

The effect of mean cycle mixture lambda on NO and CO formation was also explored.

NO predictions using the detailed chemistry and the simplified models show average errors

of 3.8% and 74% respectively. Furthermore, it was found that the simplified model returns

116

5.8. Summary and Conclusions

mainly fails at rich conditions for NO and at lean conditions for CO. Under all lambda

cases, the detailed chemistry model predicts CO emission better by 36% over the nine

investigated operation points on average, compared to the simplified model.

The prediction of mean cycle formed NO was also investigated under various ignition

timing conditions at two different engine speed and load conditions (4000rpm and 80%TP,

6000rpm and 20%TP). The detailed chemistry model presents better sensitivity on burn

rate, as the total species in the burned zone are kinetically controlled. The average NO

error for the two investigated operating points is 1.8% and 37.5% for the detailed chemistry

and the simplified models, respectively. Regarding the mean cycle CO prediction, the

detailed chemistry model can qualitatively and quantitatively predict the CO trend. As

for the average CO error for the two investigated operating points, this is 31% and 123%

for the detailed chemistry and the simplified models, respectively.

Then, the impact of wall temperature and residual gas fraction on detailed chemistry

model prediction was investigated. The two parameters are difficult to be determined

experimentally, therefore a sensitivity analysis on NO and CO pollutants was presented.

The latter showed that NO formation is mainly affected by these parameters, while CO

formation seems to be less sensitive to these parameters and more on mixture homogeneity

and mixture lambda.

In the next section of this chapter, the detailed chemistry model was applied to sim-

ulate the maximum and minimum engine cycles on a cyclic variability study. It was

observed that the emission variation is highly related to the variation on burn rate for

each combustion cycle. The simplified emission model significantly overestimates the early

pollutant formation at both NO and CO emissions by 36% and 67% respectively, which

declares that these mechanisms are more sensitive on the zone temperature profile and

less on the burn rate. The detailed emission model is more suitable to predict emission

variability and its errors on NO and CO predictions are 6% and 3.6% respectively. The

detailed emission model also illustrates that at slow burn rate cycles, CO emissions may

not achieve the frozen kinetic area until the EVO.

The validated detailed chemistry model was then utilized on the prediction of cyclic

emission variability. The CCV engine load investigation was performed at 6000 rpm,

stoichiometric conditions and two different engine load conditions (20% TP and 80% TP).

At 20% TP, the average NO emission error was 1% for the detailed chemistry model and

43% for the simplified model, while the average CO error was 5.7% and 69% respectively.

At 80% throttle position, the prediction of the average NO error for the two models is

converged, as the thermal mechanism becomes more significant. The average NO error

is for the detailed chemistry model 26.5% and for the simplified model 35.6%.

The detailed chemistry model was able to predict the average concentration in the

CCV mixture lambda investigation. At rich conditions, detailed chemistry model and

simplified model predict average NO emissions similarly with an average error of about

117

5. Emission Model Validation and Application for CCV Prediction

15%, while the average CO error is 13% and 280% respectively for the two models. At

lean conditions, detailed chemistry models predicts 10% average NO error which is lower

than the 37% average NO error of the simplified model.

The prediction of cyclic NO variability at various ignition timing conditions is per-

formed with the detailed chemistry model within an 20% better accuracy compared to

the simplified model. In addition, the average CO error is from 3.5% to 16% for the

detailed chemistry model, while the simplified CO model has an average error from 79%

up to 116%. The return maps of NO and CO emissions showed that NO is becoming

more deterministic as ignition timing is retarded, while CO nature is stochastic under all

ignition timing conditions.

The above results conclude to the fact that the proposed detailed chemistry model

can accurately predict NO and CO emissions without requiring any calibration of kinetic

parameters, is not based on arbitrary equilibrium assumptions, and can be executed in

a comparable computational time with simplified methods. This means that the method

proposed can be used to predict pollutants formation with high accuracy even including

the number of iterations required when studying cyclic combustion variability. Since the

method can be coupled to practically any 0D or 1D combustion model, it can be used for

more precise engine optimization, with expected benefits in both pollutants emission and

engine efficiency.

118

Chapter 6

Conclusions & Future Work

6.1 Experimental Work

Chapter 2 presented an extensive literature survey on the experimental investigation of

cycle to cycle variability. Although the impact of CCV on engine combustion and per-

formance has been investigated in a great number of studies, little effort has been spent

on cyclic emission variability. In fact, the literature review showed that only a handful of

studies have experimentally explored the cyclic emissions variability.

In this thesis, an experimental investigation of NO and CO variation under premixed

combustion was presented. The experiments were conducted on an in-line four cylinder

port fuel injection spark-ignition engine, modified from an original Honda CBR600RR

motorcycle. A Kistler measuring spark plug with an integrated piezoelectric transducer

was used to measure cylinder pressure. The exhaust gas temperature was measured by

a K-type thermocouple connected to the relevant transducer. High frequency pollutants

measurement of cycle-resolved emissions was conducted using fast response analyzers,

which were located close to the exhaust valve. The investigation was performed over a wide

range of engine operating conditions, including different engine load, equivalence ratio,

and ignition timing. The analysis of the results aims at correlating various interactions

among combustion parameters (such as IMEP, maximum pressure and crank angle of 5%

mass fraction burned), and pollutants emissions (NO, CO). The overall objective of this

study was primarily to explain how combustion variability affects emissions variability.

Engine load has a significant impact on the variability of engine performance and

emissions. At low load operating conditions, engine operates with higher residual gas

fraction which dilutes mixture and increases combustion and emissions variability. At

higher engine load conditions and 4000 rpm, it was found that COVimep is decreased by

60%, COVNO drops by 45% and COVCO by 6%, while at 6000 rpm COVimep is decreased

by 43% and COVNO by 38%, compared to low load conditions. At low load conditions,

119

6. Conclusions & Future Work

the early flame kernel development (ϑ5%) exhibits more variance which is accompanied

by a steep decrease at very late early flame kernel cycles.

The impact of combustion stoichiometry on the variability of engine performance and

emissions was also investigated. Slightly rich mixtures showed the higher IMEP and

the lower COVimep. Higher variation of combustion in lean mixtures has its roots on

early flame kernel development, as slightly lean mixtures ignite slower and later than

rich ones. This leads to slower flame propagation and lower cylinder pressure. In lean

mixtures, higher variation of the crank angle that early flame kernel development occurs

was detected as the main reason of NO variability, as NO formation depends on burn

rate and mixture temperature.

Furthermore, the impact of ignition timing on the variability of engine performance and

emissions has been investigated. MBT ignition timing was found to be accompanied by the

highest IMEP value and the lowest COVimep. Advanced ignition timings are characterized

by faster combustion and higher cylinder pressure. On the other hand, retarded ignition

results to a shorter delay in the development of early flame kernel, as this occurs closer

to TDC. At retarded cycles, flame propagation is shifted to the piston expansion phase,

where cylinder volume is increased and finally maximum cylinder pressure is decreased.

Because of the expansion, both cylinder pressure and crank angle at which maximum

cylinder pressure occur vary more in each cycle. The significant variation of cylinder

pressure at retarded ignition conditions was found to be the main source of NO and CO

variation in this investigation. On the one hand, NO formation is strongly linked with

the temperature vs time profile, and the higher perturbation of the latter at retarded

cycles is the main root of the high NO variability. On the other hand, CO variability is

linked with the variation of cylinder pressure which affects the trapped mixture mass into

crevices at each cycle and finally the combustion efficiency.

In conclusion, NO variability was found to be primarily affected by cylinder mixture

characteristics such as mixture equivalence ratio and residual gas fraction. At engine con-

ditions that the residual gas fraction decreases, COVNO and COVCO are simultaneously

decreased. Lean mixture conditions increase COVNO due to the unstable combustion

rate and the oxygen availability. At these conditions, COVNO depends on burn rate and

maximum cylinder pressure and temperature. The impact of burning rate on COVNO

has been also investigated by altering the ignition timing. Faster burning rates are more

repeatable, average NO value is increased and COVNO is decreased, while slower burning

rates present higher variability, average NO value is decreased and COVNO is increased.

6.2 Emissions Modeling

The main contribution of this thesis is the implementation and validation of a novel

emissions model, that predicts NO and CO emissions within an error of less than 10%

120

6.2. Emissions Modeling

and without significant increase of the computational time. The basic benefits of this

proposed emissions model are its simplicity and the absence of any calibration parameter

that makes it a generalized application, possible to be applied in any type of homogeneous

combustion engine. Although this model utilizes detailed kinetics to predict engine-out

emissions, it does not require a sophisticated or advanced combustion model, and can be

combined with any 0D or 1D combustion engine model.

Chapter 3 presents the mathematical background of the proposed detailed chemistry

model. The detailed chemistry model is based on SENKIN, a FORTRAN based code

developed by Sandia Laboratories. The calculations of kinetics are based on the GRI 3.0

mechanism, a reaction scheme which was initially proposed for methane-air combustion

and consists of 53 species and 325 reactions.

The novel emissions model assumes that each burn zone is a homogeneous reactor and

can be applied in coupling with any multi-zone combustion model. In this thesis, a two

zone approach is adopted, which consists of an unburned and a burned region. In the

burned region, at each crank angle step the inlet mass is calculated based on the principle

of mass balance. The inlet species are calculated from the combustion of the air, fuel

and residual gas. The emission composition takes place in the post flame area, where the

homogeneous reactor calculates the combustion speciation with good accuracy.

First, the model was demonstrated in coupling to a engine simulation tool to predict

the NO emissions of a single cylinder research engine. The two zone Vibe model was used

to calculate the burned zone temperature, pressure and the burn rate which were then

imported to the detailed chemistry emission model. The model satisfactorily predicted NO

emissions, ranging from a few ppm to a couple of thousand of ppm of NO molar fraction,

in both stoichiometric and lean conditions and under various ignition timing conditions.

Cases with over 10% error were considered to be either inaccurately thermodynamically

simulated or highly affected by minor mixture lambda fluctuations.

However the literature experimental data were insufficient for a detailed model valida-

tion. The validation of the detailed model is presented in chapter 5, using the experimental

data of chapter 4. The model was first validated in mean IMEP cycle simulations and

then it was used in a cycle to cycle simulation basis. Simplified kinetic mechanisms were

compared to the detailed emission model to depict the advantages of the detailed chem-

ical kinetics intensively. The validation was conducted under various engine operating

conditions such as engine speed, load, stoichiometry and ignition timing.

The engine load investigation on mean cycle NO formation showed that error of the

detailed chemistry model is less than 1.3%, compared to the 42% of the simplified model.

The improved prediction is explained by the detailed chemical kinetics calculations of

the other important species (CO, O2, H2) and on the detailed chemical pathways, which

are considered for the NO formation and destruction in the detailed chemistry model.

The effect of mixture lambda on mean cycle NO and CO formation was explored. NO

121

6. Conclusions & Future Work

predictions using the detailed chemistry and the simplified models show average errors of

3.8% and 74% respectively. It was also found that the simplified model returns mainly

fails at rich conditions for NO and at lean conditions for CO. Under all lambda cases, the

detailed chemistry model predicts CO emission better by 36%, compared to the simplified

model. Under various ignition timing conditions, at two different engine speed and load

conditions, detailed chemistry has better mean cycle prediction. The detailed chemistry

model presents better sensitivity on burn rate, as the total species in the burned zone are

kinetically controlled. The average error for the two investigated operating points is 1.8%

and 37.5% for the detailed chemistry and the simplified models, respectively.

Then, the detailed chemistry model was applied on a cyclic variability study. It was

observed that the emission variation is highly related to the variation on burn rate for each

combustion cycle. The simplified emission models significantly overestimates the early

pollutant formation rate at both NO and CO emissions by 36% and 67% respectively,

which declares that these mechanisms are more sensitive on the zone temperature profile

and less on the burn rate. The detailed emission model is more suitable to predict emission

variability and its errors on NO and CO predictions are 6% and 3.6% respectively. The

model also illustrates that at slow burn rate cycles, CO emissions may not achieve the

frozen kinetic area until the EVO.

The validated detailed chemistry model was then utilized on the prediction of cyclic

emission variability. The CCV engine load investigation performed at 6000 rpm, stoi-

chiometric conditions and 20% TP presented average NO emission error of about 1% for

the detailed chemistry model and 43% for the simplified model, while the average CO

error was 5.7% and 69% respectively. At 80% throttle position, the average NO error

for the detailed chemistry model is better by 34% compared to the simplified model. At

rich conditions, detailed chemistry model and simplified model predict similar average

NO emission error, while the average CO error is 13% and 280% respectively for the

two models. At lean conditions, detailed chemistry models predicts 10% average NO

error which is 270% better to the simplified model. The cyclic NO variability at various

ignition timing conditions is predicted with the detailed chemistry model within an 20%

better accuracy compared to the simplified model. Similar trends were observed for the

average CO error. This is from 3.5% up to 16% for the detailed chemistry model, while

the simplified CO model has an average error from 79% up to 116%.

These results point to the fact that the proposed detailed chemistry model can ac-

curately predict NO and CO emissions without requiring any calibration of kinetic pa-

rameters, is not based on arbitrary equilibrium assumptions, and can be executed in a

comparable computational time with simplified methods. This means that the method

proposed can be used to predict pollutants formation with high accuracy even including

the number of iterations required when studying cyclic combustion variability. Since the

method can be coupled to practically any 0D or 1D combustion model, it can be used for

122

6.3. Novelty

more precise engine optimization, with expected benefits in both pollutants emission and

engine efficiency.

The results of the simulation study indicate that the proposed detailed chemistry

model can be used to predict pollutants formation with high accuracy even including

the number of iterations required when studying cyclic combustion variability. Since the

method can be coupled to practically any 0D or 1D combustion model, it can be used for

more precise engine optimization, with expected benefits in both pollutants emission and

engine efficiency.

6.3 Novelty

In the extensive literature review of this thesis (see Chapter 2), the effect of combus-

tion variability on emissions variability, particularly on NO variability, and the potential

possibility of emission modeling to predict cyclic emission variability have been studied.

Taking into account these studies, this work contributes in the following novelties:

1. A novel emissions model for the improved prediction of NO and CO emissions

has been proposed on chapter 3. The model assumes that each burned zone is a

well-stirred reactor and calculates kinetically the concentrations of all combustion

species, using a detailed reaction scheme. Compared to existing emission models,

the model is characterized by the absence of both the need of chemical equilibrium

assumption and the absence of any tuning parameter or other calibration factor.

The model presents computational time in the same order of magnitude of widely

used simplified emission models. The mean cycle simulation results showed that

the proposed model predicts NO and CO emissions within a reasonable average

error of less than 10%, over a variety of engine operation points. As this model

presents improved accuracy of NO emissions and reduced computational time, it

can contribute to the design of more efficient engines.

2. The impact of CCV on engine combustion and performance had been investigated

in a great number of studies, but only very little concern has been paid on cyclic

emission variability. In fact, the literature review showed that only a handful of

studies experimentally explore the cyclic emission variability. The research gap be-

tween cyclic variability and pollutant formation is bridged into chapter 4 of this

thesis with extensive tests on a high-speed engine. The experimental investigation

studied the impact of various engine parameters, including the engine load, equiv-

alence ratio, and ignition timing on cyclic emission variability. The analysis of the

results aimed at correlating various interactions among combustion parameters and

pollutants emissions (NO, CO). The overall objective of this study was primarily

to explain how combustion variability affects emissions variability.

123

6. Conclusions & Future Work

3. This thesis also contributed on importing the predictability of cyclic emission vari-

ability. A small number of studies earlier presented in literature that used simplified

NO models to calculate cyclic NO variability failed to predict the NO scatter. Also

errors on qualitative NO trends prediction have been often observed, while model-

ing CO variability has not been performed in the earlier published literature. The

new emissions model was developed to predict NO and CO variation, due to CCV,

under various engine conditions. The results were satisfactory and proved that de-

tailed chemistry is more sensitive on burn rate and mixture composition impacts,

compared with simplified emission models.

The results of this work have been published (or submitted) in the following peer-

reviewed journals:

1. Karvountzis-Kontakiotis A., Ntziachristos L., (2015) “Investigation of cycle-

to-cycle variability of NO in homogeneous combustion”, Oil and Gas Science and

Technology, vol: 70 (1), pg: 111-123, doi: 10.2516/ogst/2013199.

2. Karvountzis-Kontakiotis A., Ntziachristos L., (2015) “Improvement of NO and

CO predictions for a homogeneous combustion SI engine using a detailed chemistry

model” (Submitted at Journal of Applied Energy, under review)

3. Karvountzis-Kontakiotis A., Dimaratos A., Ntziachristos L., Samaras Z., (2015)“Ex-

ploring the stochastic and deterministic aspects of cyclic emission variability in a

high speed SI engine” (Submitted at Journal of Energy, under review)

The results of this work have been also presented in the following international con-

ferences:

1. Karvountzis-Kontakiotis A., Ntziachristos L., “A detailed chemical mechanism

to predict NO cycle-to-cycle variation in homogeneous engine combustion”, IFAC

Workshop on Engine and Powertrain Control: Simulation and Modeling, IFP En-

ergies nouvelles, France, 23-25 Oct 2012.

2. Karvountzis-Kontakiotis A., Ntziachristos L., Samaras Z., Dimaratos A., Peck-

ham M., “Experimental investigation of cyclic variability on combustion and emis-

sions of a high-speed SI engine”, 2015, SAE Technical Paper 2015-01-0742.

This work has been initially funded by FP7 (7th Framework Programme). A part of

the results of this work have been also presented in three meetings in Milano, Thessaloniki

and Paris during the LESSCCV project of the 7th European framework.

124

6.4. Future Work

6.4 Future Work

The following aspects could be further pursued in the future in order to improve and

continue the work presented in this thesis:

1. Cylinder temperature is recognized as a very important variable on emissions predic-

tion. Increasing the number of burned zones leads to a more detailed temperature

stratification in the cylinder chamber that could improve the pollutant formation

predictability. Increasing the number of zones close to the chamber wall and in the

crevices volume could improve CO prediction, as CO is formed in this area due to

flame quenching.

2. Although this study was performed assuming homogeneous engine conditions, a

variable degree of stratification occurs in reality, in particular for diesel and direct

injection SI engines. The proposed detailed chemistry model could also be applied

on non-homogeneous mixture conditions by increasing the number of zones, where

each zone is characterized by different oxygen availability and is assumed as a ho-

mogeneous reactor. Mass diffusion between the zones could further improve the

prediction of pollutant formation, however this would complicate the analysis.

3. The novel emissions model proposed reaction schemes are based on the GRI 3.0

mechanism, a reaction scheme which was initially proposed for methane-air com-

bustion and consists of 53 species and 325 reactions. A reduced reaction scheme

with less reactions and species will reduce the computational time without a sig-

nificant impact on the model accuracy. A sensitivity analysis is required in such a

directions.

4. The modeling of cyclic emission variability was performed assuming constant resid-

ual gas fraction, and without adopting a residual gas estimation model [51, 131, 132].

Such kind of models are. By introducing a residual gas model, which would be able

to estimate residual gas fraction based on cylinder pressure during blowdown pro-

cess, NO scatter variability would be improved as well as the estimation of the

standard deviation of the NO data.

5. The cyclic combustion variability is analyzed in chapter 2 to be of both stochastic

and deterministic nature. In most of the CCV literature surveys, no significant

deterministic effects are noted. The lack of an observable determinism is, in part,

due to the nature of the experimental conditions investigated, which tended to

be dominated by operating conditions far from unstable combustion limits, such as

at stoichiometric, homogeneous-charge, low-dilution conditions. More specifically in

this experimental study, no significant deterministic effects on the cyclic combustion

125

6. Conclusions & Future Work

variability were noted, as this study was also performed at engine conditions far from

unstable combustion limits. The definition of the deterministic aspects of cyclic

emission variability would be a great contribution to the published literature.

126

Appendix A

A FORTRAN Program for

Predicting Homogeneous Gas Phase

Chemical Kinetics

A.1 Introduction

SENKIN is a FORTRAN computer program that predicts the time-dependent chemical

kinetics behavior of a homogeneous gas mixture in a closed system. In addition to predict-

ing the species and temperature histories, the program can also compute the first-order

sensitivity coefficients with respect to the elementary reaction rate parameters.

There are many possibilities for the chemical kinetics problems that one may need to

solve for various applications. In this program, six different problem types can be solved.

The distinction between the problems comes from the externally imposed thermodynamic

conditions. The six problems are: A. an adiabatic system with constant pressure, B.

an adiabatic system with constant volume, C. an adiabatic system with the volume a

specified function of time, D. a system where the pressure and temperature are constant,

E. a system where the volume and temperature are constant, and F. a system where the

pressure and temperature is specified functions of time.

For our combustion applications, conditions A through C generally apply to sponta-

neous ignition problems. These options may find use for modeling combustion bombs,

rapid compression machines, and perhaps reflected-wave shock tubes. The assumptions

of constant pressure (case A) and constant volume (case B) are limiting conditions for

a fixed mass of mixture that is reacting in an adiabatic system. For a closed, adiabatic

system at constant pressure, the mixture is free to expand and the enthalpy of the system

is constant. For a closed, adiabatic system with constant volume, no expansion work can

be done on the surroundings and the internal energy of the mixture is constant. Case C

127

A. A FORTRAN Program for Predicting Homogeneous Gas Phase ChemicalKinetics

considers a time-varying volume, and is intended for use in modeling configurations such

as rapid compression machines.

Another general type of homogeneous kinetics problem is considered by problem types

D and E. In these problems, heat release is not important, because either the reaction

mechanism is not strongly exothermic, or the mixture contains such a large fraction of

a diluent that the heat released per mass of mixture is relatively small. In these cases,

the pressure or volume are held constant and the the energy equation is replaced by

the condition that the temperature is known. Case F allows computation of mixture

composition histories for specified temperature and pressure histories. This capability is

useful for modeling situations such as a plug flow reactor, where the temperature history

can be measured accurately.

The computational solution is accomplished by a code called DASAC, which was

written by Caracotsios and Stewart [133]. The software is a modification and exten-

sion of Petzold’s differential/algebraic equation solver called DASSL. DASAC handles the

solution of the governing differential equations together with an efficient simultaneous

computation of the first-order sensitivity coefficients. The numerical method is based on

the backwards differentiation formulas and is especially well suited for solving the stiff

equations that are common in chemical kinetics applications.

The following sections describe the governing equations for each of the cases handled

by the program and provide instructions on how to operate the code. In the last section,

an example problem illustrates the capabilities of the program.

A.2 Governing Equations

In this section, the equations for mass and energy conservation are described for the six

problem types considered by the program. The reacting mixture is treated as a closed

system with no mass crossing the boundary, so the total mass of the mixture m =K∑k=1

mk

is constant, and dm/dt = 0. Here mk is the mass of the kth species and K is the total

number of species in the mixture. The individual species are produced or destroyed

according to

dmk

dt= V ωkWk (A.1)

where t is time, ωk is the molar production rate of the kth species by elementary reaction,

Wk is the molecular weight of the kth species, and V is the volume of the system, which

may vary in time. Since the total mass is constant, this can be written in terms of the

mass fractions as

128

A.2. Governing Equations

dYkdt

= υωkWk (A.2)

where Yk = mk/m is the mass fraction of the kth species and υ = V/m is the specific

volume. The species equations (A.2) are the same in all cases, A through F. For cases

D through F, the temperature is known, so the energy equation is unnecessary and the

problem is completely defined by equations (A.2). For cases A through C, the energy

equation must be derived in light of the specific constraints used in each case.

The first law of thermodynamics for a pure substance in an adiabatic, closed system

states that

de+ pdυ = 0 (A.3)

where e is the internal energy per mass and p is the pressure. This relation holds for an

ideal mixture of gases, with the internal energy of the mixture given by

e =K∑k=1

ekYk (A.4)

where ek is the internal energy of the kth species. Differentiating the internal energy of

the mixture leads to the expression

de =K∑k=1

Ykdek +K∑k=1

ekdYk (A.5)

Assuming calorically perfect gases, we write dek = cv,kdT , where T is the temperature of

the mixture, and cv,k is the specific heat of the kth species evaluated at constant volume.

Defining the mean specific heat of the mixture, cv =K∑k=1

Ykcv,k and differentiating with

respect to time, the energy equation becomes

cvdT

dt+

K∑k=1

ekdYkdt

+ pdυ

dt= 0 (A.6)

Substitution of equation (A.1) for the species production rate gives

cvdT

dt+ p

dt+ υ

K∑k=1

ekωkWk = 0 (A.7)

where cv =K∑k=1

Ykcv,k. The ideal gas equation of state is used to compute the pressure,

p =%RT

W(A.8)

129

A. A FORTRAN Program for Predicting Homogeneous Gas Phase ChemicalKinetics

where R is the universal gas constant, W is the mean molecular weight of the mixture,

and % is the mass density. In case C, we presume that the volume is provided as a function

of time, so the specific volume and its rate of change are

υ(t)

=V(t)

m(A.9)

and

dt=

1

m

dV

dt(A.10)

The system of equations for case C consists of equation (A.7) for the energy, and the K

equations (A.2) for the species mass fractions. In case B, the volume is held constant, so

equation (A.7) reduces to

cvdT

dt+ υ

K∑k=1

ekωkWk = 0 (A.11)

In case A, the first law of thermodynamics reduces to the condition that the enthalpy

of the mixture is constant. The definition of enthalpy is, h = e+ pυ, which differentiated

becomes

dh = de+ υdp+ pdυ (A.12)

The pressure is constant, so the term involving dp drops out and the first law (equation

A.3) simplifies to the condition

dh = 0 (A.13)

The mixture enthalpy is

h =K∑k=1

Ykhk (A.14)

where hk is the specific enthalpy of the kth species. Proceeding as before, the energy

equation for the constant pressure case becomes

cpdT

dt+ υ

K∑k=1

hkωkWk = 0 (A.15)

where the mean specific heat of the mixture is cp =K∑k=1

Ykcp,k. The system of equations

for case A consists of equation (A.15) for the energy, and the K equations (A.2) for the

species mass fractions.

130

A.2. Governing Equations

The net chemical production rate ωk of each species results from a competition between

all the chemical reactions involving that species. Each reaction proceeds according to the

law of mass action and the forward rate coefficients are in the modified Arrhenius form

kf = AT βexp

(−ERT

)(A.16)

where the activation energy E, the temperature exponent β, and the pre-exponential

constants A are parameters in the model formulation. The details of the chemical reaction

equations and the thermochemical properties are found in [35].

The initial value problem for each of the different cases formulated above requires

initial conditions for the temperature, pressure, and composition of the mixture. The

initial density is computed from the equation of state. These are intensive variables, so

the problem is independent of the absolute quantity of mixture in question. However,

case C requires input of the system volume V (t), which is an extensive variable. This

forces the computation of another extensive variable, namely the mass of mixture, which

is a constant during the solution. So in case C, the mass is computed from the initial

density and volume, m = %(0)V (0).

The system of ordinary differential equations described in the previous section is gen-

erally stiff, and thus is most efficiently solved by implicit techniques. A software package

called DASAC [133] (Differential Algebraic Sensitivity Analysis Code) performs the time

integration and first-order sensitivity analysis. The DASAC package is based on the dif-

ferential/algebraic system solver DASSL, which performs the time integration using a

backward differentiation formula (BDF). These BDF methods are in regular use for solv-

ing a wide range of stiff problems, including chemical kinetics problems. The notions of

stiffness and implicit numerical methods are described elsewhere (see Kee et al. [134]).

The details of the numerical methods in DASSL are described by Petzold [135], and the

DASAC enhancements by Caracotsios and Stewart [133]. Therefore, it is only briefly

outline some of the central features of the sensitivity methods.

131

A. A FORTRAN Program for Predicting Homogeneous Gas Phase ChemicalKinetics

132

References

[1] Environmental Protection Agency (EPA). U.s. greenhouse gas inventory report:

1990-2013. 2013.

[2] Z. Samaras and I. Vouitsis. 3.13 - Transportation and Energy, pages 183–205.

Academic Press, Oxford, 2013.

[3] ExxonMobil. The outlook for energy: A view to 2040, 2014.

[4] EC (2007a). Regulation (ec) no. 715/2007 of the european parliament and of the

council of 20 june 2007on type approval of motor vehicles with respect to emissions

from light passenger and commercial vehicles (euro 5 and euro 6) and on access to

vehicle repair and maintenance information. Technical report, Official Journal of

the European Union L 171, pg 1-16.

[5] S. Larsson. Commercial vehicles, fuel efficiency and co2: Challenges and possible

solutions. IEA Freight Truck Fuel Economy Workshop. Challenge Bibendum, 20-21

May 2011.

[6] T. V. Johnson. Vehicular emissions in review. SAE Technical Paper, 2012-01-0368,

2012.

[7] M. Wirth and H. Schulte. Downsizing and stratified operation an attractive com-

bination based on a spray-guided combustion system, 2006.

[8] C. Reulein, E. Schnemann, C. Schwarz, and M. Wetzel. Thermodynamics of the

bmw three-cylinder engine. MTZ, 74(5):4–9, 2013.

[9] D.D Brehob and C.E Newman. Monte carlo simulation of cycle by cycle variability.

SAE Technical Paper, 922165, 1992.

[10] N. Ozdor, M. Dulger, and E. Sher. Cyclic variability in spark ignition engines a

literature survey. SAE Technical Paper, 940987, 1994.

[11] E. Milkins, H. Watson, L. Goldsworthy, and R. Hallworth. Cycle by cycle variability

in emissions of a spark ignition engine. SAE Technical Paper, 741034, 1974.

133

REFERENCES

[12] J.K. Ball, R.R. Raine, and C.R. Stone. Combustion analysis and cycle-by-cycle vari-

ations in spark ignition engine combustion part 2: a new parameter for completeness

of combustion and its use in modelling cycle-by-cycle variations in combustion. Pro-

ceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile

Engineering, 212:507–523, 1998.

[13] J. Ball, C. Stone, and N. Collings. Cycle-by-cycle modelling of no formation and

comparison with experimental data. Proceedings of the Institution of Mechanical

Engineers, Part D: Journal of Automobile Engineering, 213(2):175–189, 1999.

[14] J.K. Ball, J.B. Martin, C.R. Stone, and N. Collings. Validation of a cyclic no

formation model with fast no measurements. SAE Technical Paper, 2001-01-1010,

2001.

[15] J.B. Heywood. Internal Combustion Engines Fundamentals. McGraw-Hill, New

York, 1988.

[16] G.P. Merker, C. Schwarz, G. Stiesch, and F. Otto. Simulation of combustion and

pollutant formation for engine-development. Springer, 2004.

[17] R. Stone. Introduction to Internal Combustion Engines. Macmillan, 1992.

[18] P.G. Aleiferis, Y. Hardalupas, A.M.K.P. Taylor, K. Ishii, and Y. Urata. Flame

chemiluminescence studies of cyclic combustion variations and air-to-fuel ratio of the

reacting mixture in a lean-burn stratified-charge spark-ignition engine. Combustion

and Flame, 136:72–90, 2004.

[19] M.F. Brunt and A.L. Emtage. Evaluation of burn rate routines and analysis errors.

SAE Technical Paper, 970037, 1997.

[20] G. Grnefeld, V. Beushausen, P. Andresen, and W. Hentschel. A major origin of

cyclic energy conversion variations in si engines: Cycle-by-cycle variations of the

equivalence ratio and residual gas of the initial charge. SAE Technical Paper, 941880,

1994.

[21] C.T. Bowman. Kinetics of pollutant formation and destruction in combustion.

Progress in Energy and Combustion Science, 1(1):33–45, 1975.

[22] G. Lavoie, J. Heywood, and J. Keck. Experimental and theoretical study of nitric

oxide formation in internal combustion engines. Combustion Science and Technol-

ogy, 1(4):313–326, 1970.

[23] C.P. Fenimore. Formation of nitric oxide in premixed hydrocarbon flames. Sympo-

sium (International) on Combustion, 13(1):373–380, 1971.

134

REFERENCES

[24] F. Bachmaier, K. H. Eberius, and TH. Just. The formation of nitric oxide and the

detection of hcn in premixed hydrocarbon-air flames at 1 atmosphere. Combustion

Science and Technology, 7(2):77–84, 1973.

[25] J.A. Miller and C.T. Bowman. Mechanism and modeling of nitrogen chemistry in

combustion. Progress in Energy and Combustion Science, 15(4):287–338, 1989.

[26] J.W. Bozzelli and A.M. Dean. O+nnh–a possible new route for nox formation in

flames. International Journal of Chemical Kinetic, 27(11):1097–1109, 1995.

[27] A.M. Dean and J.W. Bozzelli. Combustion Chemistry of Nitrogen, chapter 2, pages

125–341. Springer New York, 2000.

[28] A. N. Hayhurst and E. M. Hutchinson. Evidence for a new way of producing

no via nnh in fuel-rich flames at atmospheric pressure. Combustion and Flame,

114(12):274–279, 1998.

[29] A. A. Konnov, G. Colson, and J. De Ruyck. The new route forming no via nnh.

Combustion and Flame, 121(3):548–550, 2000.

[30] T Rutar, J. C Y. Lee, P Dagaut, P. C. Malte, and A. A. Byrne. Nox formation path-

ways in lean-premixed-prevapourized combustion of fuels with carbon-to-hydrogen

ratio between 0.25 and 0.88. Proceedings of the Institution of Mechanical Engineers,

Part A: Journal of Power and Energy, 221(3):387–398, 2007.

[31] R. Miller, G. Davis, G. Lavoie, C. Newman, and T. Gardner. A super-extended

zel’dovich mechanism for nox modeling and engine calibration. SAE Technical Pa-

per, 980781, 1998.

[32] M. Rublewski and J. Heywood. Modeling no formation in spark ignition engines

with a layered adiabatic core and combustion inefficiency routine. SAE Technical

Paper, 2001-01-1011, 2001.

[33] K. Pattas and G. Hafner. Stickoxydbildung bei der ottomotorischen verbrennung.

MTZ, 12:397–404, 1973.

[34] D. L. Baulch, C. J. Cobos, R. A. Cox, C. Esser, P. Frank, Th. Just, J. A. Kerr,

M. J. Pilling, J. Troe, R. W. Walker, and J. Warnatz. Evaluated Kinetic Data for

Combustion Modelling, volume 21. 1992.

[35] G.P. Smith, D.M. Golden, M. Frenklach, N.W. Moriarty, B. Eiteneer, M. Golden-

berg, C.T. Bowman, R.K. Hanson, S. Song, W.C. Gardiner Jr., V.V. Lissianski,

and Z. Qin. http : //www.me.berkeley.edu/gri−mech/.

135

REFERENCES

[36] D.D Thomsen, F.F Kuligowski, and N.M Laurendeau. Modeling of no formation in

premixed, high-pressure methane flames. Combustion and Flame, 119(3):307–318,

1999.

[37] S.R. Turns. An Introduction to Combustion: Concepts and Applications. McGraw-

Hill, New York, 2012.

[38] H.K. Newhall. Kinetics of engine-generated nitrogen oxides and carbon monoxide.

Symposium (International) on Combustion, 12(1):603–613, 1969.

[39] J.B Heywood. Pollutant formation and control in spark-ignition engines. Progress

in Energy and Combustion Science, 1(4):135–164, 1976.

[40] G. DErrico, G. Ferrari, A. Onorati, and T. Cerri. Modeling the pollutant emissions

from a s.i. engine. SAE Technical Paper, 2002-01-0006, 2002.

[41] J.C. Keck and D. Gillespie. Rate-controlled partial-equilibrium method for treating

reacting gas mixtures. Combustion and Flame, 17(2):237–241, 1971.

[42] A. Onorati and G. Ferrari. Modeling of 1-d unsteady flows in i.c. engine pipe

systems: Numerical methods and transport of chemical species. SAE Technical

Paper, 980782, 1998.

[43] C. Olikara and G. Borman. A computer program for calculating properties of equi-

librium combustion products with some applications to i.c. engines. SAE Technical

Paper, 750468, 1975.

[44] C.D. Rakopoulos and C.N. Michos. Development and validation of a multi-zone

combustion model for performance and nitric oxide formation in syngas fueled spark

ignition engine. Energy Conversion and Management, 49(10):2924–2938, 2008.

[45] J. Heywood, J. Higgins, P. Watts, and R. Tabaczynski. Development and use of a

cycle simulation to predict si engine efficiency and nox emissions. SAE Technical

Paper, 790291, 1979.

[46] A. Jante. The wiebe combustion law (das wiebe-brenngesetz; ein fortschritt in der

thermodynamik der kreisprozesse von verbrennungsmotoren). Kraftfahrzeugtechnik,

9:340–346, 1960.

[47] G. A. Lavoie, A. A. Adamczyk, E. W. Kaiser, J. W. Cooper, and W. G. Roth-

schild. Engine hc emissions modeling: Partial burn effects. Combustion Science

and Technology, 49(1-2):99–105, 1986.

136

REFERENCES

[48] S. Richard, S. Bougrine, G. Font, F.-A. Lafossas, and F. Le Berr. On the reduction

of a 3d cfd combustion model to build a physical 0d model for simulating heat

release, knock and pollutants in si engines. Oil Gas Sci. Technol. Rev. IFP Energies

nouvelles, 64(3):223–242, 2009.

[49] S. Bougrine, S. Richard, J. B. Michel, and D. Veynante. Simulation of co and no

emissions in a si engine using a 0d coherent flame model coupled with a tabulated

chemistry approach. Applied Energy, 113(0):1199–1215, 2014.

[50] W. Li, Z. Liu, Z. Wang, and Y. Xu. Experimental investigation of the thermal and

diluent effects of egr components on combustion and nox emissions of a turbocharged

natural gas si engine. Energy Conversion and Management, 88:1041–1050, 2014.

[51] P. Giansetti, C. Perrier, P. Higelin, Y. Chamaillard, A. Charlet, and S. Couet. A

model for residual gas fraction prediction in spark ignition engines. SAE Technical

Paper, 2002-01-1735, 2002.

[52] Y. Kaneko, H. Kobayashi, and R. Komagome. The effects of exhaust gas recircula-

tion and residual gas on engine emissions and fuel economy. SAE Technical Paper,

750414, 1975.

[53] R. Miller, S. Russ, C. Weaver, E. Kaiser, C. Newman, G. Davis, and G. Lavoie.

Comparison of analytically and experimentally obtained residual fractions and nox

emissions in spark-ignited engines. SAE Technical Paper, 982562, 1998.

[54] A. Karvountzis-Kontakiotis and L. Ntziachristos. Investigation of cycle-to-cycle

variability of no in homogeneous combustion. Oil Gas Sci. Technol. Rev. IFP

Energies nouvelles, 70(1):111–123, 2015.

[55] M. Young. Cyclic dispersion some quantitative cause-and-effect relationships. SAE

Technical Paper, 800459, 1980.

[56] N. Ozdor, M. Dulger, and E. Sher. An experimental study of the cyclic variability

in spark ignition engines. SAE Technical Paper, 960611, 1996.

[57] D. Patterson. Cylinder pressure variations, a fundamental combustion problem.

SAE Technical Paper, 660129, 1966.

[58] C.S. Daw, C.E.A. Finney, J.B. Green, M.B. Kennel, J. F. Thomas, and F. T.

Connolly. A simple model for cyclic variations in a spark-ignition engine. SAE

Technical Paper, 962086, 1996.

[59] J. Roberts, J. Peyton Jones, and K. Landsborough. Cylinder pressure variations as

a stochastic process. SAE Technical Paper, 970059, 1997.

137

REFERENCES

[60] C.E.A. Finney, B.C. Kaul, C.S. Daw, R.M. Wagner, D.K. Edwards, and J.B. Green.

A review of deterministic effects in cyclic variability of internal combustion engines.

International Journal of Engine Research, 2015. 10.1177/1468087415572033.

[61] W. Dai, N. Trigui, and Y. Lu. Modeling of cyclic variations in spark-ignition engines.

SAE Technical Paper, 2000-01-2036, 2000.

[62] M. Belmont, M. Hancock, and D. Buckingham. Statistical aspects of cyclic vari-

ability. SAE Technical Paper, 860324, 1986.

[63] A. Medina, P.L. Curto-Risso, A.C. Hernndez, L. Guzmn-Vargas, F. Angulo-Brown,

and A.K. Sen. Quasi-Dimensional Simulation of Spark Ignition Engines. Springer-

Verlag London, 1 edition, 2014.

[64] E. Schll and H.G. Schuster. Handbook of Chaos Control, volume 2. Wiley, New

York, 2008.

[65] L. Chew, R. Hoekstra, J. Nayfeh, and J. Navedo. Chaos analysis on in-cylinder

pressure measurements. SAE Technical Paper, 942486, 1994.

[66] K. Hamai, H. Kawajiri, T. Ishizuka, and M. Nakai. Combustion fluctuation mecha-

nism involving cycle-to-cycle spark ignition variation due to gas flow motion in s.i.

engines. Symposium (International) on Combustion, 21(1):505–512, 1988.

[67] M. Namazian, S. Hansen, E. Lyford-Pike, J. Sanchez-Barsse, J. Heywood, and

J. Rife. Schlieren visualization of the flow and density fields in the cylinder of

a spark-ignition engine. SAE Technical Paper, 800044, 1980.

[68] C. F. Li, Q. C. Duan, F. Maroteaux, and M. Murat. Time-resolved measurement

of fuel transient behaviour and cycle to cycle variation of local fuel/air equivalence

ratio in gasoline engines. SAE Technical Paper, 940989, 1994.

[69] J. Bedford, G. Brereton, H. Schock, and R. Herrin. Measurements of cycle to cycle

variability of the inlet flow of fuel injectors using lda. SAE Technical Paper, 2006-

01-3314, 2006.

[70] B. Pundi, V. Zvonow, and C. Gupta. Effect of charge non-homogeneity on cycle-by-

cycle variations in combustion in si engines. SAE Technical Paper, 810774, 1981.

[71] F.A. Matekunas. Modes and measures of cyclic combustion variability. SAE Tech-

nical Paper, 830337, 1983.

[72] M. Nakai, Y. Nakagawa, K. Hamai, and M. Sone. Stabilized combustion in a spark

ignited engine through a long spark duration. SAE Technical Paper, 850075, 1985.

138

REFERENCES

[73] T. Alger, J. Gingrich, B. Mangold, and C. Roberts. A continuous discharge ignition

system for egr limit extension in si engines. SAE Int. J. Engines, 4:677–692, 2011.

[74] W. Chen, D. Madison, P. Dice, and J. Naber. Impact of ignition energy phasing

and spark gap on combustion in a homogenous direct injection gasoline si engine

near the egr limit. SAE Technical Paper, 2013-01-1630, 2013.

[75] P.G Hill. Cyclic variations and turbulence structure in spark-ignition engines. Com-

bustion and Flame, 72(1):73–89, 1988.

[76] M.J. Rauckis and W.J. McLean. The effect of hydrogen addition on ignition delays

and flame propagation in spark ignition engines. Combustion Science and Technol-

ogy, 19(5-6):207–216, 1979.

[77] V. Arrigoni, F. Calvi, G. Cornetti, and U. Pozzi. Turbulent flame structure as

determined by pressure development and ionization intensity. SAE Technical Paper,

730088, 1973.

[78] B. Peters and G. Borman. Cyclic variations and average burning rates in a s. i.

engine. SAE Technical Paper, 700064, 1970.

[79] S. Curry. A three-dimensional study of flame propagation in a spark ignition engine.

SAE Technical Paper, 630487, 1963.

[80] P.G. Aleiferis, A.M.K.P. Taylor, K. Ishii, and Y. Urata. The nature of early flame

development in a lean-burn stratified-charge spark-ignition engine. Combustion and

Flame, 136(3):283–302, 2004.

[81] A.A. Burluka, A.M.T.A. El-Dein Hussin, Z.Y Ling, and C.G.W. Sheppard. Effects

of large-scale turbulence on cyclic variability in spark-ignition engine. Experimental

Thermal and Fluid Science, 43(0):13–22, 2012.

[82] S.H. Joo, K.K. Srinivasan, K.C. Lee, and S.R. Bell. The behaviourt of small- and

large-scale variations of in-cylinder flow during intake and compression strokes in a

motored four-valve spark ignition engine. International Journal of Engine Research,

5(4):317–328, 2004.

[83] D. Hana, A.M. Ickes, S.V. Bohac, Z. Huanga, and D.N. Assanis. HC and CO

emissions of premixed low-temperature combustion fueled by blends of diesel and

gasoline. Fuel, 99:13–19, 2012.

[84] S. Petrovic. Cycle by Cycle Variations of Flame Propagation in a Spark Ignition

Engine. SAE Technical Paper, 820091, 1982.

139

REFERENCES

[85] M. Swords, G. Kalghatgi, and A. Watts. An Experimental Study of ignition and

Flame Development in a Spark Ignited Engine. SAE Technical Paper, 821220, 1982.

[86] A. Brown, C. Stone, and P. Beckwith. Cycle-by-cycle variations in spark ignition

engine combustion - part i: Flame speed and combustion measurements and a

simplified turbulent combustion model. SAE Technical Paper, 960612, 1996.

[87] M. Young. Cyclic dispersion in the homogeneous-charge spark-ignition engine - a

literature survey. SAE Technical Paper, 810020, 1981.

[88] R. Matthews, M. Sarwar, M. Hal, D. Filipe, D. Miller, and N. Cernansky. Predic-

tions of cyclic variability in an si engine and comparisons with experimental data.

SAE Technical Paper, 912345, 1991.

[89] J. P. Soltau. Cylinder pressure variations in petrol engines. Proceedings of the

Institution of Mechanical Engineers: Automobile Division, 14(1):99–117, 1960.

[90] A.K. Sen, J. Wang, and Z. Huang. Investigating the effect of hydrogen addition

on cyclic variability in a natural gas spark ignition engine: Wavelet multiresolution

analysis. Applied Energy, 88(12):4860–4866, 2011.

[91] S. Wang and C. Ji. Cyclic variation in a hydrogen-enriched spark-ignition gaso-

line engine under various operating conditions. International Journal of Hydrogen

Energy, 37(1):1112–1119, 2012.

[92] G. Lucas and M. Brunt. The effect of combustion chamber shape on the rate of

combustion in a spark ignition engine. SAE Technical Paper, 820165, 1982.

[93] S. Chanchaona. Cyclic Variability in a Natural Gas Fuelled Spark Ignition Engine.

PhD thesis, 1990.

[94] E. Milkins, H. Watson, L. Goldsworthy, and R. Hallworth. Cycle by cycle variability

in emissions of a spark ignition engine. SAE Technical Paper, 741034, 1974.

[95] Jonathan Etheridge, Sebastian Mosbach, Markus Kraft, Hao Wu, and Nick Collings.

Modelling cycle to cycle variations in an si engine with detailed chemical kinetics.

Combustion and Flame, 158(1):179–188, 2011.

[96] L. Ben, N. Raud-Ducros, R. Truquet, and G. Charnay. Influence of air/fuel ratio

on cyclic variation and exhaust emission in natural gas si engine. SAE Technical

Paper, 1999-01-2901, 1999.

140

REFERENCES

[97] B. P. Pundir, V. A. Zvonow, and C. P. Gupta. Nitric oxide formation in spark-

ignition engines with in-cylinder charge non-homogeneity of a random nature. Pro-

ceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile

Engineering, 199(3):227–236, 1985.

[98] P. Davis and M. Peckham. Measurement of cycle-by-cycle afr using a fast response

ndir analyzer for cold start fuelling calibration applications. SAE Technical Paper,

2006-01-1515, 2006.

[99] S. Baltisberger and K. Ruhm. Fast no measuring device for internal combustion

engines. SAE Technical Paper, 940962, 1994.

[100] A.E Lutz, R.J Kee, and J.A. Miller. Senkin: A fortran program for predicting

homogeneous gas phase chemical kinetics with sensitivity analysis. SAND87-8248

Revised, 1988.

[101] U. Kesgin. Study on prediction of the effects of design and operating parameters

on nox emissions from a leanburn natural gas engine. Energy Conversion and Man-

agement, 44(6):907–921, 2003.

[102] C. R. Ferguson. Internal Combustion Engines, Applied Thermosciences. John Wiley

and Sons, New York, 1986.

[103] J. P. Duclos, M. Zolver, and T. Baritaud. 3d modeling of combustion for di-si

engines. Oil and Gas Science and Technology - Rev. IFP, 54(2):259–264, 1999.

[104] A. Shuemie, R. Fairbrother, Ch Ptsch, and R. Tatschl. Lessccv project meeting,

milano, wp4. 2011.

[105] M. Peckham, A. Finch, and B. Campbell. Analysis of transient hc, co, nox and co2

emissions from a gdi engine using fast response gas analyzers. SAE Int. J. Engines,

4(1):1513–1522, 2011.

[106] C. Rakopoulos, A. Dimaratos, E. Giakoumis, and M. Peckham. Experimental as-

sessment of turbocharged diesel engine transient emissions during acceleration, load

change and starting. SAE Technical Paper, 2010-01-1287, 2010.

[107] P. Davis and M. Peckham. Cycle-by-cycle gasoline engine cold start measurement

of residual gas and afr using a fast response co/co2 analyzer. SAE Technical Paper,

2008-01-1649, 2008.

[108] K. St. J. Reavell, N. Collings, M. Peckham, and T. Hands. Simultaneous fast

response no and hc measurements from a spark ignition engine. SAE Technical

Paper, 971610, 1997.

141

REFERENCES

[109] I. Finlay, D. Boam, J. Bingham, and T. Clark. Fast response fid measurement

of unburned hydrocarbons in the exhaust port of a firing gasoline engine. SAE

Technical Paper, 902165, 1990.

[110] C. Cheng, W. Cheng, J. Heywood, D. Maroteaux, and N. Collings. Intake port

phenomena in a spark-ignition engine at part load. SAE Technical Paper, 912401,

1991.

[111] T. Hands, M. Peckham, B. Campbell, and C. Sutela. Transient si engine emissions

measurements on the ftp75 drive cycle with a fast response co instrument. SAE

Technical Paper, 2001-01-3540, 2001.

[112] C. Sutela, N. Collings, and T. Hands. Fast response co2 sensor for automotive

exhaust gas analysis. SAE Technical Paper, 1999-01-3477, 1999.

[113] M. Brunt, H. Rai, and A. Emtage. The calculation of heat release energy from

engine cylinder pressure data. SAE Technical Paper, 981052, 1998.

[114] M. A. Ceviz and . Kaymaz. Temperature and airfuel ratio dependent specific heat

ratio functions for lean burned and unburned mixture. Energy Conversion and

Management, 46(1516):2387–2404, 2005.

[115] R. Egnell. Combustion diagnostics by means of multizone heat release analysis and

no calculation. SAE Technical Paper, 981424, 1998.

[116] S. Han, Y. Chung, Y. Kwon, and S. Lee. Empirical formula for instantaneous heat

transfer coefficient in spark ignition engine. SAE Technical Paper, 972995, 1997.

[117] C. Lacour and C. Pera. An experimental database dedicated to the study and

modelling of cyclic variability in spark-ignition engines with les. SAE Technical

Paper, 2011-01-1282, 2011.

[118] R. Prucka, T. Lee, Z. Filipi, and D. Assanis. Turbulence intensity calculation from

cylinder pressure data in a high degree of freedom spark-ignition engine. SAE

Technical Paper, 2010-01-0175, 2010.

[119] G.T. Kalghatgi. Early flame development in a spark-ignition engine. Combustion

and Flame, 60(3):299–308, 1985.

[120] J. Zareei and A.H. Kakaee. Study and the effects of ignition timing on gasoline

engine performance and emissions. European Transport Research Review, 5(2):109–

116, 2013.

142

REFERENCES

[121] K.M. Chun and J.B. Heywood. Estimating heat-release and mass-of-mixture burned

from spark-ignition engine pressure data. Combustion Science and Technology, 54(1-

6):133–143, 2007.

[122] Y. Guezennec and W. Hamama. Two-zone heat release analysis of combustion data

and calibration of heat transfer correlation in an i. c. engine. SAE Technical Paper,

1999-01-0218, 1999.

[123] E. Ortiz-Soto, G. Lavoie, J. Martz, M. Wooldridge, and D. Assanis. Enhanced heat

release analysis for advanced multi-mode combustion engine experiments. Applied

Energy, 136(0):465–479, 2014.

[124] D. Buttsworth. Spark ignition internal combustion engine modelling using matlab.

Technical report, University of Southern Queensland, 2002.

[125] W.J.D. Annand. Heat transfer in the cylinders of reciprocating internal combustion

engines. Proceedings of the Institution of Mechanical Engineers, 177(1):973–996,

1963.

[126] J. Brettschneider. Extension of the equation for calculation of the air-fuel equiva-

lence ratio. SAE Technical Paper, 972989, 1997.

[127] C. Pera, S. Chevillard, and J. Reveillon. Effects of residual burnt gas heterogene-

ity on early flame propagation and on cyclic variability in spark-ignited engines.

Combustion and Flame, 160(6):1020–1032, 2013.

[128] B. Kaul, R. Wagner, and J. Green. Analysis of cyclic variability of heat release for

high-egr gdi engine operation with observations on implications for effective control.

SAE Int. J. Engines, 6(1):132–141, 2013.

[129] D. Scholl and S. Russ. Air-fuel ratio dependence of random and deterministic cyclic

variability in a spark-ignited engine. SAE Technical Paper, 1999-01-3513, 1999.

[130] Guo-xiu Li and Bao-feng Yao. Nonlinear dynamics of cycle-to-cycle combustion vari-

ations in a lean-burn natural gas engine. Applied Thermal Engineering, 28(56):611–

620, 2008.

[131] J. Larimore, E. Hellstrm, S. Jade, A. G. Stefanopoulou, and Li Jiang. Real-time in-

ternal residual mass estimation for combustion with high cyclic variability. Interna-

tional Journal of Engine Research, 16(3):474–484, 2015. 10.1177/1468087414552616.

[132] M. Mladek and C. Onder. A model for the estimation of inducted air mass and the

residual gas fraction using cylinder pressure measurements. SAE Technical Paper,

2000-01-0958, 2000.

143

REFERENCES

[133] M. Caracotsios and W. E. Stewart. Sensitivity analysis of initial-boundary-value

problems with mixed pdes and algebraic equations: Applications to chemical and

biochemical systems. Computers and Chemical Engineering, 19(9):1019–1030, 1995.

[134] R.J. Kee, J.A. Miller, and T.H Jefferson. Chemkin: A general-purpose, problem-

independent, transportable, fortran chemical kinetics code package. Technical re-

port, Sandia National Laboratories, 1980.

[135] L.R Petzold. A description of dassl: a differential/algebraic system solver. Technical

report, Sandia National Laboratories, 1982.

144