by
Matthew J. Rublewski
B.S.M.E., Pennsylvania State University
(1998)
Submitted to the Department of Mechanical Engineering in Partial Fulfillment of the Requirements for the Degree of
Master of Science in Mechanical Engineering at the
Massachusetts Institute of Technology
February 2000
@ 2000 Massachusetts Institute of Technology
All rights reserved
MASSACHUSETTS INSTITUTE
OF TECHNOLOGY
SEP 2 0 2000
Signature of A uthor ....... .. .
. . . . . .
.......................
Department of Mechanical Engineering
February 3, 2000
Certified by ............................................................. ........
John B.Heywood
Sun Jae Professor of Mechanical Engineering
Thesis Supervisor
Accepted by ............................ .......
Ame A. Sonin
Chairman, Department Committee on Graduate Students

Nitric Oxide Formation and Thermodynamic Modeling in
Spark Ignition Engines by
Matthew J. Rublewski
Submitted to the Department of Mechanical Engineering
February 4, 2000 in Partial Fulfillment of the Requirements for the Degree of Master of Science in Mechanical Engineering
ABSTRACT
An assessment of a thermodynamic based cycle simulation's ability to predict steady state engine out NO concentration over a wide range of operating conditions has been conducted. An experimental data base, which included measurements of NO concentration, cylinder pressure, and residual gas fraction, was obtained from a 2.0 liter Nissan engine while firing on propane.
Using experimentally derived bum rate information, ensuring that the correct total mass and mixture composition is considered, modeling combustion inefficiency effects, and increasing the amount of heat transfer during combustion were all concluded to be necessary for making accurate predictions of in-cylinder pressure. Based on experimental data, accounting for temperature stratification during combustion with a layered adiabatic core routine improved the slope of model predictions in comparison to a fully mixed adiabatic core. With a layered model, the three reaction extended Zeldovich mechanism, alone, was capable of predicting NO concentration as a function of equivalence ratio, spark timing, and intake manifold pressure to within 15% accuracy. This was achieved by using a forward reaction rate of 1.66E+12T'.
3 for the first reaction of the Zeldovich mechanism. A sensitivity analysis was performed which showed NO predictions to be fairly insensitive to the heat transfer amount and size of the crevice used to model combustion inefficiency. This analysis also confirmed that residual fraction and bum rate are the most critical engine variables for making NO predictions, and indicated that performing a kinetic calculation instead of assuming an equilibrium radical pool may further improve model accuracy.
A fast response NO meter was then placed in the exhaust port of the Nissan engine to investigate the amount of cyclic variation during steady state operation. Cycle resolved values of NO were determined by mass weighting the fast NO signal over the shifted exhaust valve open period.
Under light loads, lean conditions, and EGR operation, NO concentration was shown to correlate almost linearly with peak pressure. However, NO variation increased with engine load. The large amount of observed high load scatter was attributed to mixture non-uniformity effects and error introduced by the single point measurement. By using bum rate, alone, as an input variable, the previously calibrated cycle simulation was capable of following general trends in the cycle by cycle NO data.
Thesis Advisor: Professor John B. Heywood
Title: Sun Jae Professor of Mechanical Engineering
3

Acknowledgments
Its over, and I want to sincerely thank many people.
It has been a pleasure to have Professor John Heywood as a supervisor. I would like to thank him for picking up a student and project in progress, and for providing expert advice and guidance during all of our meetings. I would also like to thank Professor Simone
Hochgreb for giving me the opportunity to come to MIT and for providing good direction in the early stages of this thesis. Professor Wai Cheng also provided many helpful suggestions.
I thank God for my office mate Jim Cowart, without him I would not have had a prayer of finishing this work. He is a gifted person who truly cares about others; I hope to become more like him. Thanks to Brian Corkum for being an outstanding technician who taught me a great deal about engineering and provided many a good lunch; the lab has not been the same since he left. Special thanks to Mark Dawson, Gary Landsberg, and Chris
O'Brien for their direct help with my project, and to John Baron for his support of my cycling career. I would also like to say thank you to all the members of the Sloan Lab who served with me for their friendship and help in various ways.
I would like to thank the members of the MIT Engine and Fuels Consortium for their financial support of my two years of work at MIT.
Finally, I would like to thank the following very important people: My parents for their continued love and support. Chet and Mona for their steady advice and for always welcoming me at their home away from MIT. Kate for being there for me in the beginning.
4

TABLE OF CONTENTS
L IST O F T A B L E S ............................................................................................ 7
L IST O F FIG U R E S .........................................................................................
CHAPTER 1 INTRODUCTION........................................................................... 11
1.1 Role of Oxides of Nitrogen, NOx, in Air Pollution ...................................... 11
1.2 Role of Automotive Industry in Overall NO Emission and Control.................... 12
1.3 Background on NO Formation and Modeling ............................................. 12
1.4 P revious W ork ................................................................................. 14
1.5 Structure of T hesis ............................................................................ 16
1.6 Steady State Modeling Objectives ......................................................... 16
1.7 Cycle by Cycle NO Variation Objectives ................................................ 17
8
CHAPTER 2 EXPERIMENTAL METHOD ............................................................. 21
2.1 Experimental Test Matrix ................................................................... 21
2 .2 T est E ngine .................................................................................. . . 22
2.3 M ixture Preparation .......................................................................... 23
2.4 Cylinder Pressure Measurement ............................................................ 24
2.5 C ylinder Pressure A nalysis .................................................................... 26
2.6 Residual Fraction Predictions ................................................................. 27
2.6.1 Residual Measurement with Atmospheric CP Chamber ..................... 28
2.6.2 Residual Measurement with CP Chamber Under Vacuum .................. 29
2.6.3 Verification and Extension of Limited Experimental Data .................. 30
2.7 N O M easurem ent ............................................................................... 32
CHAPTER 3 MODELING METHOD ................................................................... 45
3.1 Modeling Approach .......................................................................... 45
3.2 Cycle Simulation General Description ..................................................... 46
3.3 Valve Flow Sub-Model ........................................................................ 47
3.4 Combustion Sub-Model ..................................................................... 48
3.5 Heat Transfer Sub-Model ................................................................... 48
3.6 Combustion Inefficiency Sub-Model ....................................................... 49
3.7 NO Formation Sub-Model ................................................................... 51
3.8 Temperature Profiles within NO Sub-Model ............................................. 52
3.8.1 Fully Mixed Temperature Profiles .............................................. 53
3.8.2 Unmixed / Layered Temperature Profiles ...................................... 55
CHAPTER 4 STEADY STATE MODELING RESULTS & DISCUSSION ...................... 65
4.1 Model Calibration Approach................................................................ 65
4.2 V alve Flow C alibration ........................................................................ 66
4.3 Heat Transfer and Crevice Calibration ..................................................... 68
5

4.4 Calibration of Kinetic Routine .............................................................
4.5 Load Sweep Modeling Results .............................................................
70
72
4.6 Equivalence Ratio Sweep Modeling Results .............................................. 73
4.7 Exhaust Gas Recirculation Sweep Modeling Results ..................................... 75
4.8 Spark Timing Sweep Modeling Results ................................................... 76
CHAPTER 5 SENSITIVITY ANALYSIS ...............................................................
5.1 Effect of Adding N20 Mechanism ........................................................
87
87
5.2 Effect of Considering the Residual NO Concentration .................................. 89
5.3 Upgrade to Full Equilibrium Calculation ................................................... 90
5.4 Kinetically Controlled Radical Pool Investigation ....................................... 92
5.5 General Sensitivity Analysis ................................................................. 95
5.5.1 Variables of Uncertainty Residual, Heat Transfer, Crevice Size ............ 95
5.5.2 Input Parameter Perturbation ......................................................
5.5.3 NO Sub-Model Variables ........................................................
98
100
5.5.4 Fully M ixed Sensitivity .......................................................... 100
5.6 Summary and Conclusions Steady State Modeling .................................... 101
CHAPTER 6 CYCLE BY CYCLE NO VARIATION STEADY STATE OPERATION..... 113
6.1 F ast N O M eter N otes ........................................................................ 113
6.2 Signal Characteristics During Lean Operation PHI=0.914 ........................... 114
6.3 Processing the Fast NO Exhaust Data ..................................................... 115
6.3.1 Signal Characteristics as a Function of Load .................................. 115
6.3.2 Plug Flow Modeling of the Exhaust Event .................................... 116
6.4 Three Different Methods of Determining a Cycle Resolved NO Value ............... 117
6.4.1 Analysis of Three Methods for MAP = 0.5 bar , PHI=0.91 .................. 118
6.4.2 Analysis of Three Methods for MAP = 0.8 bar, Stoichiometric ............ 119
6.4.3 Analysis of Three Methods for MAP = 0.5 bar, Stoichiometric ............ 120
6.5 Load Sweep Cycle by Cycle Variations ................................................... 120
6.6 Cycle by Cycle NO Variation EGR, Equivalence Ratio, and Spark Sweeps ....... 122
6.6.1 E G R Sw eep ........................................................................ 122
6.6.2 L ean O peration ..................................................................... 124
6.6.3 Rich Operation and Spark Sweep ............................................... 124
6.7 Cycle by Cycle Bum Rate Modeling ......................................................
6.8 Observations and Recommendations ......................................................
125
6.7.1 Load Sweep CBC Modeling ..................................................... 125
6.7.2 Lean and EGR Points CBC Modeling ........................................ 127
127
B IB L IO G R A PH Y ..........................................................................................
A P PE N D IC E S .............................................................................................
139
143
6

LIST OF TABLES
Table 2.1: Experimental Operating Conditions .......................................................
Table 2.2: Nissan SR20DE Specifications ............................................................
21
22
Table 2.3: Experimentally Measured Residual Fraction with Atmospheric Sampling ........... 29
Table 2.4: Experimentally Measured Residual Fraction with CP Chamber held at 0.46 bar .... 30
Table 2.5: Assumed Residual Mass Fraction Values Load Sweep ............................... 32
Table 2.6: Assumed Residual Mass Fraction Values PHI, EGR, and Spark Sweep ............ 32
Table 3.1: Cycle Simulation Input Variables ........................................................... 58
Table 4.1: Model and Experimental Comparison of IMEP -
1500 rpm , $ = 1.0, Load Sweep ............................................................
Table 4.2: Model and Experimental Comparison of Air and Residual Fraction
1500 rpm, $ = 1.0, Load Sweep...........................................................
Table 4.3: Comparison of Recommended Rate Constants for Reaction 1,
N + N O = N
2
+ O ...........................................................................
65
66
71
Table 5.1: Percent Difference between SENKIN and Simulation Predicted-
L ayer N O P rofiles .............................................................................. 94
Table 5.2: Input Parameter Sensitivity Analysis Load Points .................................... 96
Table 5.3: Input Parameter Sensitivity Analysis Lean and EGR Points ........................ 97
Table 6.1: Comparison of Different Methods for Calculating a Cycle Resolved NO Value -
0 .5 b ar - L ean .................................................................................
Table 6.2: Comparison of Different Methods for Calculating a Cycle Resolved NO Value -
0.8bar Stoichiom etric .....................................................................
Table 6.3: Linear Correlation Analysis Results for Cyclic NO Concentration -
P eak P ressu re .................................................................................
1 18
120
123
7

LIST OF FIGURES
Figure
Figure
Figure
1.1
1.2
1.3
Photochemical Smog Formation Time Scales and NOx Ozone Cycle .............. 18
Major Sources of Smog Formation Pollutants ....................................... 18
In-cylinder Strategies for Reducing Engine Out NO Emissions .................... 19
Figure
Figure
Figure
2.1
2.2
2.3
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11
2.12
Figure
Figure
Figure
Figure
2.13
2.14
2.15
2.16
Figure 2.17
Pressure Transducer Experimental Location Side View ............................ 35
Pressure Transducer Experimental Location Top View............................ 35
Motoring Log P Log V Correct Phasing with Respect to Volume ............... 36
Motoring Log P Log V 1' Off in Phasing with Respect to Volume .............. 36
Investigation of Cylinder Pressure Accuracy ......................................... 37
Experimentally Derived Burn Rate with Wiebe Function ........................... 37
Fast Flame Ionization Detector Experimental Set-up ................................. 38
Residual Fraction Measurements with Atmospheric CP Chamber .................. 38
-
Residual Fraction Measurements with CP Chamber Held Under Vacuum ........ 39
Cyclic V ariability of Residual Signals .................................................. 39
Experimentally Measured Valve Lift for Nissan Engine ........................... 40
N issan V alve O verlap Period .......................................................... 40
Comparison of Experimental Residual Data Sets and Correlation ................ 41
SAE 982046 Ford Residual Data Sets for Equivalence Ratio and EGR ........... 41
Schematic of Cambustion Fast NO Meter Sampling System ...................... 42
Steady State Experimental Engine Out NO -
Load and Equivalence Ratio Sweeps ................................................. 43
Steady State Experimental Engine Out NO -
EGR and Spark Tim ing Sweeps ......................................................... 43
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
3.5
3.6
3.7
3.8
3.9
3.1
3.2
3.3
3.4
Modeling Approach Flow Chart .........................................................
Thermodynamic Representation of Cylinder Contents ...............................
Cycle Simulation Valve Flow Discharge Coefficient Map ...........................
NO Sub-model Temperature Profile Options ..........................................
Fully Mixed Temperature and Boundary Layer Profiles ..............................
Fully Mixed NO Concentration Profiles ................................................
Unmixed / Layered Temperature Profiles ..............................................
Unmixed / Layered NO Concentration Profiles .......................................
Comparison of Layered and Fully Mixed Temperature Profiles ....................
Figure 4.1 Uncalibrated Model NO Predictions Load Sweep .................................. 77
Figure 4.2 Uncalibrated Model IMEP Predictions Load Sweep ................................ 77
Figure 4.3 Baseline Operating Condition Pressure Trace Analysis -
Uncalibrated Model Comparison ...................................................... 78
Figure 4.4 Uncalibrated Model Air and Residual Fraction Predictions Load Sweep ........ 78
Figure 4.5 Baseline Operating Condition Pressure Trace Analysis-
Correct Charge Mass and Composition ...............................................
Figure 4.6 Baseline Operating Condition Pressure Trace Analysis-
Heat Transfer and Crevice Effects .....................................................
79
79
57
59
60
60
61
61
62
62
63
8

Figure 4.7 Baseline Operating Condition Pressure Trace Analysis-
Fully Calibrated Model Comparison ................................................... 80
Figure
Figure
Figure
Figure
Figure
4.8 Calibrated Model Heat Transfer Predictions Load Sweep ........................ 80
4.9 Calibrated Model IMEP and Peak Pressure Predictions -
L o ad S w eep ............................................................................. .. 8 1
4.10 Calibrated Model IMEP and Peak Pressure Predictions -
4.11 -
Equivalence R atio Sw eep ..............................................................
Load Sweep Modeling Comparison with k
1
=3.3E+12T
3
-
81
82 1500rpm - P H I = 1.0 .....................................................................
4.12 Equivalence Ratio Sweep Modeling Comparison with k
1
=3.3E+12T'.3 _
1500rpm M A P = 0.5bar ................................................................ 82
Figure
Figure
Figure
Figure
4.13 Final Load Sweep Modeling Comparison with k
1
=1.66E+12T03
-
1500rpm - P FH = 1.0 .....................................................................
4.14 Amount of NO Reduction After Factor of 2 Adjustment to ki -
Layered A .C . Load Sweep ............................................................
83
83
4.15 Final Equivalence Ratio Sweep Modeling Comparison with k
1
=1.66E+12T03
-
1500rpm M A P = 0.5bar ................................................................ 84
4.16 Amount of NO Reduction After Factor of 2 Adjustment to kI -
Layered A.C. Equivalence Ratio Sweep ............................................... 84
Figure
Figure
Figure
4.17 Comparison of Layered Model Predictions with Previous Work .................. 85
4.18 - Final EGR Sweep Modeling Comparison with k
1
=1.66E+12T
0 3
-
1500rpm - PH I = 1.0 .....................................................................
4.19 Final Spark Sweep Modeling Comparison with ki=1.66E+12T0.3 _
85
1500rpm - PH I = 1.0 ..................................................................... 86
Figure 5.1
-
Effect of Adding N20 Mechanism Equivalence Ratio Sweep with ki=1.66E+12TO 3 -1500 rpm - MAP = 0.5bar................................... 104
Figure 5.2 Effect of Adding N20 Mechanism Equivalence Ratio Sweep with k
1
=1.5E+12T
0 3
1500 rpm MAP = 0.5bar.....................................104
Figure 5.3 Overall Effect of Adding N20 Mechanism All Sweeps ........................... 105
Figure 5.4 Amount of Residual NO Concentration in Unburned Mixture Load Sweep.....105
Figure 5.5 Effect of Modeling Residual NO Concentration Load Sweep .................... 106
Figure 5.6 Overall Effect of Modeling Residual NO Concentration ........................... 106
Figure 5.7 Baseline Operating Condition Thermodynamic Property Analysis -
Burned Zone Temperature Comparison ..............................................
Figure 5.8 Baseline Operating Condition Thermodynamic Property Analysis -
107
Burned Zone Specific Heat and Gamma Comparison .............................. 107
Figure 5.9 Baseline Operating Condition Thermodynamic Property Analysis -
Burned Zone Enthalpy and Density Comparison..................................... 108
Figure 5.10 Layered Model A.C.Temperature Profiles Baseline Operating Condition ...... 108
Figure 5.11 Constant Enthalpy Combustion Process from Tunburned = 850K -
109 1500rpm Stoichiometric M AP =0.5bar..............................................
Figure 5.12 Cycle Simulation and SENKIN NO Profiles for Selected Layers 0.5 bar ......
Figure 5.13 Cycle Simulation and SENKIN NO Profiles for Selected Layers 0.3 bar ......
Figure 5.14 Cycle Simulation and SENKIN NO Profiles for Selected Layers 0.8 bar ......
109
110
110
9

Figure
Figure
5.15 Cycle Simulation and SENKIN NO Profiles for Selected Layers Lean ......... 111
5.16 Cycle Simulation and SENKIN NO Profiles for Selected Layers EGR ......... 111
Figure
Figure
Figure
Figure
Figure
Figure
6.1 Fast NO Meter Sampling System Specifications .....................................
6.2 Fast NO Detector Output for Five Consecutive Cycles 0.5 bar Lean ..........
6.3 Cycle by Cycle Modeling Comparison Layered A.C. 0.5 bar Lean ..........
6.4 Exhaust Port NO Signal Variation with MAP ...........................
6.5 Cycle Simulation Calculated Exhaust Mass Flowrate Profiles ....................
6.6 Exhaust Port NO Profiles for Five Consecutive Cycles 0.5 bar Lean .........
Figure
Figure
6.7 Exhaust Port NO Profiles for Five Consecutive Cycles 0.8 bar ..................
6.8 Exhaust Port NO Profiles for Five Consecutive Cycles 0.3 bar ..................
Figure 6.9 Cycle by Cycle Variation of Exhaust Port NO with Peak Pressure -
Figure
Figure
L o ad S w eep ..............................................................................
6.10 Different Techniques for Determining Cycle Resolved NO Value ...............
6.11 Cycle by Cycle Variation of Exhaust Port NO with Peak Pressure -
E G R S w eep ..............................................................................
Figure 6.12 Cycle by Cycle Variation of Exhaust Port NO with Peak Pressure -
Figure
Figure
Figure
L ean O peration ...........................................................................
6.13 Cycle by Cycle Variation of Exhaust Port NO with Peak Pressure -
L ean O peration cont. ....................................................................
6.14 Cycle by Cycle Variation of Exhaust Port NO with Peak Pressure -
R ich O p eration ...........................................................................
6.15 Cycle by Cycle Variation of Exhaust Port NO with Peak Pressure -
S p ark S w eep ..............................................................................
Figure
Figure
Figure
6.16 Individual Cycles Selected for Modeling Comparison Load Sweep .............
6.17 Cycle by Cycle Modeling Comparison Layered A.C. Load Sweep ............
6.18 Cycle by Cycle Modeling Comparison Layered A.C. Lean EGR ............
133
133
134
134
135
135
136
136
137
137
129
129
130
130
131
131
132
132
10

INTRODUCTION
1.1 Role of Oxides of Nitrogen, NOx, in Air Pollution
Nitric oxide and nitrogen dioxide, has been studied and regulated extensively since the 1950s and for good reason. Two harmful components of spark ignition engine exhaust, unburned hydrocarbons (HG) and nitric oxide (NO), are responsible for the formation of low level ozone which is also known as photochemical smog. Figure 1.1 gives a good general description of the time scales involved with smog formation. During summer months, the morning rush hour traffic can lead to high concentrations of NO and HC in the lower atmosphere. When the midday sun provides the necessary ultraviolet light, NO is quickly oxidized to NO
2 with the aid of available HC. Finally, NO
2 leads to the formation of low level ozone according to the reaction mechanism next to figure 1.1. One last requirement for smog formation is an extended period of stagnant air conditions which allows the necessary reactions to take place before the pollutants are dispersed. Weather patterns can also dictate where the low-level ozone will end up, since smog formed in
highly populated areas can be carried hundreds of miles down wind.
Low-level ozone, PAN, and other oxidizing species can be harmful to architecture, are responsible for millions of dollars of crop damage yearly, and have devastated large forest areas in the past. For people, smog can irritate the eyes and extended exposure can lead respiratory problems. Therefore the risk groups include anyone who spends a significant amount of time outdoors in problem areas during summer months, such as young children, outdoor workers, joggers, and cyclists. The environmental protection agency has defined unhealthy levels of ozone as greater than 0. lppm and, in recent years, has developed online monitoring services to check air quality during summer months everywhere in the country. [1]
11

1.2 Role of Automotive Industry in Overall NO Emission and Control
Figure 1.2 details the major sources for the two smog formation pollutants. The automotive industry is responsible for approximately half of the total NO emitted into the environment. In most major cities, the NO contribution from motor vehicles easily exceeds 50%, and the vast majority of those vehicles are powered by spark ignition engines. This does not mean that automobiles today are not extremely clean; it is more a volume issue now because of the shear number of vehicles in operation, over 150 million in the United States alone. Consider that a vehicle is producing over 4g/mile of engine out
NOx prior to catalyst after treatment. In 1973, the first regulations of NOx were set at
3g/mile. In California today an Ultra Low Emission Vehicle is required to emit less than
0.2g/mile over its lifetime. Tier II regulations set to take affect for all vehicle manufactured after 2004, would require vehicles in California to emit less than 0.05g/mile of NO over the first 50,000 miles of operation. Two orders of magnitude reduction is quite an engineering task.
In reaction to the first wave of regulation, the automotive community was doing extensive research in the early 1970s to understand and control NO formation at the source, in the cylinder. The amount of research on NO related topics, in recent years, has been on a slight decline, mainly due to the effectiveness of three way catalyst after treatment advances. On a modern vehicle, a cold catalyst can convert over 95% of the engine out
NOx. Once the catalyst has warmed up after a few minutes of operation, over 99% of the
NOx is successfully converted along with converting HC and CO at the same time.
However, the catalytic converter does require engine operation that cycles around the stoichiometric amount of fuel, and this limits overall engine fuel economy.
1.3 Background on NO formation and Modeling
Although the catalyst is very effective at meeting the regulations of today, controlling NO production in the cylinder will continue to be a concern as design strategies aim at tier II levels. Whenever, a fuel is burned and flame temperatures exceed 1800K a significant amount of the nitrogen present in atmospheric air will start to become oxidized.
Zeldovich [2] was the first to recognize and model the NO formation process in 1946, and
12

research done at MIT in the early 1970s by Lavoie, Keck, and Heywood [3] completed the reaction scheme known as the extended Zeldovich mechanism shown below.
N+NO
N+02 >NO+O
N+OH
3 >NO+H
By making a few assumptions, which will be discussed in detail later, about the state of the cylinder gases after the flame has passed, NO concentration levels can be calculated from a single in-cylinder temperature profile. The level of NO concentration will be a function of the cylinder temperatures and the amount of available oxygen. The only way to limit the amount of oxygen in the cylinder would be to run under fuel rich conditions. Though, this
is not an option since HC concentration would grow substantially.
Therefore, controlling NO formation in the cylinder, becomes as simple as limiting the temperature of the burned gases during combustion and expansion, without running rich. The most dramatic effect can be seen when the engine is forced to operate under lean conditions. As the engine goes slightly lean, NO concentration reaches a maximum since temperatures are still high and more oxygen is now available. However, if the mixture can be reduced to fuel air equivalence ratios of approximately 0.7, flame temperature drops substantially and NO concentration can be reduced by an order of magnitude. Lean operation today is limited by an available catalyst that can operate while being fed steady lean exhaust products. Based on the current conversion efficiency, it would not be helpful to reduce NO concentration by one order of magnitude without a catalyst, if three way catalysts can reduce stoichiometric NO concentration by two orders of magnitude.
Another method of limiting NO concentration is by increasing the amount of burned gases present in the intake mixture by recirculating a fraction of the exhaust back into the intake manifold; this control strategy is known as exhaust gas recirculation, EGR.
Burned gas present during combustion increases the thermal capacity of the burnable mixture, thus reducing overall flame temperatures. If the level of EGR is increased to approximately 20% of the incoming fresh mixture, NO reduction can again approach an order of magnitude. In addition, this strategy allows the three way catalyst to be used, since it does not significantly change the stoichiometry of the exhaust gases. The effect of
13

lean and EGR operation on NO concentration was illustrated nicely by Blumberg and
Kummer [4] in an early fundamental work. Even without EGR, the unburned mixture always has some burned gas present due to residual gases from the previous cycle. This natural EGR is mainly a function of engine load and will be discussed extensively in this study. One final strategy commonly used to reduce NO concentration while not affecting the catalyst operation is to retard the spark timing from its maximum brake torque (MBT) point. This will effectively move the fuel heat release process out of phase with the piston compression process, reducing peak pressures and corresponding temperatures. However, all of the above mentioned strategies come at the sacrifice of combustion stability and engine performance, so the challenge of controlling NO is not at all simple. Figure 1.3 presents a preview of experimental data that illustrate the changes in NO concentration seen with the above mentioned parameters.
Since emission regulations will force NO concentration to always be one of the most critical design parameters to meet, effective computer models must be available to the automotive industry. By using a model to predict NO concentration and performance accurately as a function of operating conditions instead of doing extensive dynamometer testing, new engine design and control methods can be explored much faster and at a fraction of the cost. The availability of accurate computer simulations is also a valuable tool to the large number of researchers working in the area of exhaust after treatment.
1.4 Previous Work
NO formation and modeling in spark ignition engines has being studied in detail for the past four decades, and the amount of published work is immense. Therefore only a brief discussion of the previous work that has shaped this thesis directly will be discussed now. Two investigations at MIT by Poulos and Heywood [5] and McGrath [6] laid the foundation for the modeling predictions shown later in this thesis. Poulos originally developed the quasi-dimensional MIT cycle simulation for the purpose of investigating the effects of combustion chamber geometry on bum rate, but the main structure of the model still remains although several minor changes have occurred over time. McGraths main focus was to investigate the effect of engine crevices on NO predictions. This work also
14

developed a representation of the burned gases that combined both a thermal boundary layer and a model for temperature stratification. However, the McGrath study was limited
by not having a detailed enough experimental data set to validate the model with. In addition, several minor errors in handling the mass balance of the crevice routine and calculating the temperature profiles of the stratification model clouded the conclusions of the study. Regardless, the work of McGrath marked an excellent starting point for this investigation.
Ford Motor Company has published two recent papers [7,8] that deal with accuracy of NO predictions from thermodynamic based cycle simulations such as the one used at
MIT. Both of these papers were published during the course of this thesis and three main conclusions were drawn. First, Fords cycle simulation was predicting NO concentration nearly 25% lower than experimental data under lean operating conditions. Therefore, the
NO chemistry considered was extended from the three reaction Zeldovich mechanism to over seventy reactions in order to improve lean operation accuracy. Secondly, the reaction rate used for the first reaction of the Zeldovich mechanism had to be adjusted by a factor of five over the load range in order to obtain good agreement with experimental data. Finally, an investigation of modeling predictions of residual gas fraction was performed and correcting cylinder charging errors improved the NO concentration predictions as well. It was unclear if a model for temperature stratification was used in these studies, and the effect of engine crevices was neglected.
Two papers, both with collaboration by Stone, were also consulted extensively during the course of this thesis. First, in 1995, Raine and Stone [9] demonstrated that using a layered representation of the cylinder gases during combustion improved the NO predictions from a relatively simple engine simulation. Recently in 1999, Ball, Stone, and
Collings [10] used a similar model to make predictions of cycle by cycle NO concentration measured with a fast response NO detector. This worked attempted to model combustion inefficiency based on the estimated mass fraction burned calculated from a pressure trace heat release analysis code. However, both of these studies gave few details concerning model calibration techniques and considered a limited range of engine operating conditions.
15

1.5 Structure of Thesis
Discussion will begin with a presentation of the specific objectives of this project.
From there, the work will be presented in five additional chapters. Chapter 2 gives an overview of the experimental test matrix. Chapter 3 presents the important features of and assumptions made by the cycle simulation used to generate the modeling results in this thesis. Chapter 4 details the steady state model calibration methodology used to accurately represent of the thermodynamic state of the gases. This chapter will then explore the reaction rate constants used with the extended Zeldovich mechanism, and a comparison of modeling results and experimental data will be presented and discussed. Chapter 5 will conclude the steady state modeling section after a detailed sensitivity analysis is performed.
Finally, Chapter 6 will present an initial investigation of NO cyclic variability while operating under steady state conditions. This chapter will attempt to explain the cycle by cycle trends observed in fast NO meter data by applying the earlier modeling results. Then, the thesis will end with some brief conclusions and recommendations for future work in the area of cyclic variability.
The main focus of this thesis throughout will be to investigate the ability of thermodynamic cycle simulations to predict, both average and cycle by cycle, engine out
NO concentration over a wide range of operating conditions. The specific objectives of this thesis can be summarized as follows:
1.6 Steady State Modeling Objectives
" develop an experimental data base that effectively covers the entire operating range of the spark ignition engine and includes accurate measurements of cylinder pressure and mixture composition
* develop a model calibration methodology to determine whether the challenge of predicting NO concentration is primarily a thermodynamic or chemistry problem. Then an assessment of the completeness of the extended Zeldovich mechanism alone will be made after confidence in model predictions of cylinder pressure and temperature is gained
16

* complete a sensitivity analysis that will explore the limitations of the current MIT cycle simulation, identify which model components are potential sources of error, and identify which engine variables are most critical for predicting NO concentration
1.7 Cycle by Cycle NO Variation Objectives
* demonstrate the amount of NO cycle by cycle variation over the complete range of operating conditions considered in the steady state modeling discussion
* develop an appropriate data analysis method to be used for determining a cycle resolved value of NO concentration based on the output signal of the fast NO meter placed in the exhaust port
* apply what was learned from the modeling results and sensitivity analysis to explain the observed trends in cyclic NO data
17

-
N02+hv ->NO+O
-03+M
03+
NO--N02+02
NO
HC
-
hv
N02 -> 03,
PANoxidizing species
HC
5%
18

E
CL 1500
0 z
0
0
500
-
-
Lean w
M
M a
M
.
M w
U
0
0.6 0.8 1.0
(b) EGR Operation
E
0
C
()
0)
.C
%%0, 1500 "
0 z
1000_-
500
0
0
M
U
N
N
5 10 15
19

20

EXPERIMENTAL METHOD
2.1 Experimental Test Matrix
Because NO formation is highly dependent on in-cylinder temperatures and pressures, an experimental test matrix had to be developed that would cover most of the parameters affecting burned gas temperature. The review of previous work done in the area indicated that equivalence ratio, spark timing, and burned gas fraction are the key parameters affecting in-cylinder temperatures. The last of these can be considered in two different ways, either residual fraction alone or residual plus exhaust gas recirculation
(EGR). The main parameter affecting residual fraction is engine load or intake manifold absolute pressure (MAP) because of backflow effects, which will be discussed later.
The baseline operating condition was chosen to be 1500 rpm with MAP equal to
0.5 bar, because this is close to what the auto-industry considers to be a typical part load operating point. Speed was not considered in this study because it should have only minor effects on temperature through time available for heat transfer, and the single cylinder setup of the test engine gave a limited speed range. This leaves four independent sweeps of equivalence ratio, MAP, EGR, and spark timing which all revolve around the baseline operating point: 1500 rpm, stoichiometric, MAP = 0.5 bar. Table 2.1 shown below characterizes all of the experimental test points in more detail. It should be noted that the engine was fully warmed up and fired on propane for all test points. Aside from the spark sweep, the timing was optimized for maximum brake torque (MBT). MBT was defined as the timing which yielded maximum IMEP. This corresponds well with peak pressure location between 13' and 15' ATDC, and maximum heat release rate location near
70 ATDC.
Table 2.1 Experimental Operating Conditions
Engine Speed
Spark Timing
FuelType
Coolant Temperature
1500 rpm
Optimized for MBT
Propane
75 - 800 C
21

Load
($=
1.0)
Equivalence Ratio, $
(MAP = 0.5 bar)
Spark Timing
(MAP=0.5 bar, $ = 1.0)
EGR
(MA P=0.5 bar, $ = 1.0)
0.33, 0.4, 0.5, 0.6, 0.9 bar
Rich 1.25, 1.12, 1.06, 1.00
Lean 0.96, 0.91, 0.84, 0.77, 0.71
Retarded 50, 100
Advanced 50, 100, 150, 200
4%, 8%,12%,16%
2.2 Test Engine
All experiments were performed on a Nissan SR20DE production four cylinder spark ignition engine. The engine was designed with a pentroof combustion chamber, four valves per cylinder, and a centrally located spark plug. Complete technical specifications are listed below in Table 2.2.
Table 2.2 Nissan SR20DE Specifications
Engine Type 4 valve/cylinder DOHC
Aluminum Head/Block
Displacement / Cylinder (cm
3
)
Clearance Volume (cm
3
)
500
58.77
Bore x Stroke 8.6 x 8.6
Compression Ratio 9.5
Intake Valves
(34 mm Diameter/ 10.2 mm Max Lift)
Exhaust Valves
(30 mm Diameter / 9.4 mm Max Lift)
Valve Overlap Period
Open 130 BTDC
Close 235' ATDC
Open 4830 ATDC
Close 7230 ATDC
160
The engine was modified to fire on a single cylinder, prior to this study, to make mixture preparation and exhaust gas analysis easier to interpret. To accomplish this, the intake manifold has been modified so that three of the intake runners are sealed off from the plenum and vented to the atmosphere. This allows fuel and air from the throttle to only enter the first cylinder for combustion. The exhaust manifold runner for the firing cylinder has also been isolated from the other three and attached to a ten gallon damping tank before reaching a trench. The fuel/air ratio was monitored at all times with a Horiba, Model
22

MEXA-I10 ,universal exhaust gas oxygen sensor, which was mounted approximately 10 cm downstream of the exhaust port. Engine coolant flow and temperature are controlled by an external pump, water heater, and heat exchanger unit. This gave precise control of the engine block temperature and allowed preheating to take place prior to experimental runs.
An external oil cooling circuit was also used to maintain the oil sump temperature at approximately 750 C for all experiments.
The Nissan engine is coupled to a 100HP Dynamatic dynamometer which is capable of motoring the engine or absorbing the output while firing. Because of the intake manifold design, three of the four cylinders are motoring at all times. Therefore, even while firing on one cylinder, the engine load must increase to approximately 70% of maximum to overcome the friction of the other three and switch to absorption on the dyno.
Regardless of load however, the engine speed was maintained with a Digalog controller.
The distributor has also been modified, and a separate custom ignition system was used for the active cylinder.
2.3 Mixture Preparation
The engine air supply was monitored with a Kurz, model 505-9A, air mass flow meter and is displayed in units of grams per second. Since the instrument had not been calibrated since its installation, verification experiments were performed using a Meriam laminar flow element connected in series with the Kurz meter. The two methods showed agreement to within 3% over the range of engine operation studied. The engine test cell environment was not controlled, therefore intake air temperature and humidity level varied day to day in the laboratory. For all of the experimental data sets shown later, the intake air temperature, measured just after the throttle body, ranged from 23' to 28' C and relative humidity levels, measured in the test cell, ranged from 25% to 40%. A discussion of humidity and intake air temperature effects on NO emissions will be held until Chapter 4.
To avoid liquid fuel effects and limit mixture non-uniformity, the engine was fired with propane as the fuel for all experiments. The port fuel injectors were removed from the manifold and properly sealed. A continuous flow of propane was introduced to the back of the intake runner, approximately 30 cm upstream of the intake valves. The amount
23

of fuel supplied was determined with the aid of a critical flow orifice. As long as the ratio of the manifold pressure to supply pressure was larger than 0.522, the flow of propane through the orifice has reached sonic velocities, and mass flow is only a function of the upstream supply pressure. The calculation of propane mass flow in grams per second is shown in the equation 2.1 below:
r n
= Psupply VyR TA eff (Eq. 2.1)
This flow rate was also verified off the engine using a soap film bubble test setup. The final test of whether the fuel and air flow rates were accurate was based on a comparison with the average air fuel ratio readout from the oxygen sensor.
Finally, for several test points, EGR was simulated with a mixture of 83% N2 and
17% C02, by volume, and introduced continuously into the intake manifold after the throttle body, at the entrance to the plenum. This mixture composition was chosen based on a method detailed by Hinze [11], because it has approximately the same molar heat capacity as natural exhaust gas, only without water vapor. EGR mass flow rates are typically described as percent of ingested mass according to equation 2.2 below:
M
EGR
% E GR =
MEGR +
Mpropane
+ Mair (Eq. 2.2)
Again the flow was regulated with critical flow orifices and ranged from 4% to 16% of the total ingested mass. The artificial EGR was left at room temperature for ease while performing the experiments. Since the upcoming modeling analysis takes EGR temperature as an input, this will not lead to any modeling discrepancies.
2.4 Cylinder Pressure Measurement
Cylinder pressure was measured with a side-mounted Kistler 605 1B piezoelectric pressure transducer located in the engine head approximately 1 cm above the travel of the piston. The transducer sensor sits in a cylindrical crevice with an approximate depth of
6mm and a diameter of 3mm. Figures 2.1 and 2.2, illustrate the location of the transducer from a top and side view. An output signal from the transducer was sent to a Kistler model
5010A dual mode charge amplifier and then sampled by a pc based data acquisition
24

system. Pressure data was recorded at 1 degree intervals using the output from a 360' degree per revolution optical shaft encoder as an external trigger. An additional reference flag at BDCC was also recorded from a combination of the shaft encoder pulse and cam shaft position sensor.
In order to get an absolute pressure measurement from a piezoelectric transducer, the pressure signal must be pegged to the manifold pressure at some point in the cycle. For all experiments, the voltage at BDCC was set equal to the average intake manifold pressure obtained from a Data Instruments, model SA, absolute pressure sensor. This is the recommended procedure used by Ford Motor Company for low speed operation [12].
Before and after gathering all experimental data sets, the pressure transducer was calibrated with a dead weight tester to verify linearity and sensitivity, in bar per volt, over the pressure range of 0 to 40 bar.
Several authors have also addressed the subject of cylinder pressure data integrity, in particular peak pressure and phasing with respect to volume. Both Kenney et al. [12] and Lancaster et al. [13] recommend plotting motored log P vs. log V to verify that the compression and expansion strokes are linear and have polytropic exponents between
(1.30 1.42) and (1.33 1.45) respectively. Figure 2.3 shows a plot of volume vs. pressure on a logarithmic scale generated from the ensemble average pressure trace while motoring the engine at 1500 RPM with intake pressure of 0.5 bar. The polytropic exponents were
1.34 for compression and 1.42 for expansion which is within the expected range, and the sharpness of the point at which they meet is indicative of correct phasing with respect to volume. Figure 2.4 illustrates the effect of even a one degree shift in the phasing. These plots further confirm the accuracy of the test set-up because a crossover is seen near TDC when advanced, and the sharpness of the point is lost when retarded. A final verification of proper phasing is to check that the maximum pressure while motoring the engine occurs approximately 1 degree BTDC. For the test point discussed above, the average peak pressure location for 150 cycles was 179.3' ABDC.
In addition to phasing errors, Ford motor company also recommends periodically checking the transducer for thermal shock effects while firing. In a paper by Stein et al.
[14], thermal shock or strain was defined as a reduction of measured peak pressure and
25

IMEP due to cyclical exposure of piezoelectric transducers to high temperature combustion gases. This thermal strain was shown to cause a reduction in peak pressure of 1.5% at
1500rpm, WOT, while the effect was much smaller at part load.
To verify measured peak pressure, tests were performed at 0.5 bar and 0.8 bar using a fiber optic, Optrand Model C81255-SP, spark plug mounted transducer at the same time as the side mounted Kister 6051. Figure 2.5 shows a comparison of the two pressure traces for three consecutive cycles at 1500 rpm, 0.5 bar MAP. Excellent peak pressure agreement, less than 1% difference, was observed at both 0.5 bar and 0.8 bar. However, the accuracy of the Optrand spark plug transducer was poor during the intake and exhaust stroke, which prevented its use alone. It should also be reiterated that the highest load point observed in this study was 0.8 bar MAP, and a majority of the experimental points were taken at 0.5 bar MAP, where the effects of thermal strain should be small. The pressure transducer used was also recessed slightly from the head, as shown previously in figure 2.1. This should allow it to operate at cooler temperatures than a flush mounted transducer.
2.5 Cylinder Pressure Analysis
For each individual test point, two data files were collected containing 150 cycles of cylinder pressure each. These data files were processed using the MIT Burn Rate
Analysis code, which is based on a general heat release program originally developed by
Gatowski et al. [15] and Chun et al. [16] and later modified by Chueng and Heywood [17].
This program uses a simple one zone energy model which includes the effects of engine crevices and heat transfer to the cylinder walls. Gamma values for propane are fitted by linear functions of charge temperature during compression and expansion but are assumed to be constant during combustion, as set by Chueng. With these approximate gamma values, the rate of chemical energy release can be determined from the measured incylinder pressure with equation 2.3 shown below:
_ y -1 dV dO
y
~dO y -
1 dp
V dO
dO
dO (Eq2.3
26

Output of the bum analysis program includes peak pressure with location, IMEP, and total mass fraction burned for each individual cycle, as well as, summary statistics, ensemble pressure trace information, and mean polytropic exponents. For all the experimental operating conditions, the polytropic exponents, while firing, ranged from
(1.26-1.32) for compression and (1.30- 1.32) for expansion. This further confirmed the integrity of the pressure data according to Kenney et al. [12].
In addition to these parameters, equation 2.3 can be used to calculate the amount of fuel that must be burned at each crank angle to match the heat released. These values can then be normalized to the total mass fraction burned to give the following bum angles: 0-
2%, 0-10%, 0-50%, and 10-90%. Since the modeling analysis requires bum rate as an input, the parameters of the burn profile wiebe function are specified for a good match with the experimental bum angles. A sample plot of the bum angles from the baseline operating condition, and the corresponding wiebe function equation are shown in figure 2.6. A more detailed discussion of the Wiebe function will be held until the MIT cycle simulation is discussed next chapter.
2.6 Residual Fraction Predictions
Quader [18] and Aiman [19] both showed the dramatic effect of charge dilution on
NO formation in spark ignition engines. However, because sampling in-cylinder gases prior to the flame arrival is a difficult experiment, little data exists on residual fraction as a function of operating conditions. When this project was started, there was no published experimentally measured residual fraction data for a modem dual overhead cam four valve engine. However, one of the major works done on a two-valve engine was performed at
MIT by Galliot et al. [20], and later Fox et al. [21] developed a theoretical model for predicting residual fraction as a function of operating conditions. Since a four valve engine had never been studied, experiments were conducted following the method demonstrated
by Galliot in 1990.
Galliot's method involves using a Cambustion fast flame ionization detector (FFID) to measure in-cylinder hydrocarbon concentration just prior to spark discharge. A relative comparison to the HC concentration while motoring the engine is then made to determine
27

the residual fraction. Since this technique has been documented before in many papers
[20,22], only a brief description of the sampling method will be given here.
2.6.1 Residual Measurement with Atmospheric CP Chamber
Figure 2.7 shows a schematic of the sampling system used along with a table of sampling specifications. The FFID acts as a carbon atom counter that produces a voltage that depends upon both hydrocarbon concentration and the mass flow rate of the sampled gas. A continuous stream of gas was pulled from the cylinder using a small sampling probe, which was design to go through a Kister 6051 pressure transducer blank. This was done to avoid drilling an extra hole in the engine head. Since the cylinder pressure is fluctuating greatly throughout the cycle, the sample must be sent to an expansion tube, which is contained within a constant pressure (CP) chamber that is vented to the atmosphere. A fraction of the flow exiting the expansion tube is then pulled through a teepiece into the hydrogen-air flame (see figure 2.7 again). The pressure difference across the tee-piece is fixed at 0.14 bar. Therefore, the mass flow rate into the flame is held constant regardless of the driving pressure in the cylinder, and the output voltage is only proportional to the sample HC concentration.
An illustration of the three typical FFID signals needed for the residual fraction measurement along with corresponding pressure traces are shown in figure 2.8, and a discussion of each trace will now begin. Looking at the firing trace, it can be seen that once the cylinder pressure rises above CP chamber atmospheric pressure, the hydrocarbon concentration rises as fresh mixture is brought into the cylinder and mixes with the left over residual gases. The trace attempts to reach a plateau level before the flame reaches the tip of the sampling probe, causing the voltage to drop sharply back to zero. Because of the short plateau level, it was necessary to skip fire the engine or disable the spark once every 15 cycles to obtain a longer plateau level. Looking at the skip fired trace, it is seen that without the arrival of the flame, a longer plateau is observed until the cylinder pressure drops back below atmospheric pressure resulting in a back flow in the sampling line. This skip fired plateau level now represents the HC concentration of the cylinder gases with residual present.
28

The final piece needed for a residual fraction measurement is a calibration voltage level corresponding to the HC concentration of the cylinder gases with no residual fraction present. This is accomplished by motoring the engine for several minutes to allow all of the residual gas to be purged from the cylinder, leaving only a fresh mixture of propane and air. The calibration trace appears very similar to a skip fired trace. However, it now reaches a higher plateau since no residual gas is present. Finally, the molar residual fraction can be determined from the voltage difference between the calibration trace and the skip fired trace, according to equation 2.4 shown below. Galliot showed that this molar residual fraction is approximately equal to the residual mass fraction to within 4%.
Xresidual
-
Vmotored
-
Vskipfired
Vmotored (Eq. 2.4)
Using the method described above, experiments were performed while varying the intake manifold pressure, since MAP is the major parameter affecting residual fraction level. Four of the five load points were observed between 0.33 and 0.6 bar MAP, but the highest load point could not be taken because the single cylinder setup did not allow skip firing to occur above 0.6 bar. At each load point, 100 cycles of motoring data were recorded, and 300 cycles of skip firing data were collected. The experimental residual value was then determined from the average motoring plateau level and the average of the
20 skip fired cycles, according to equation 2.4. Table 2.3, shown below, summarizes the limited data set recorded under these sampling conditions. It should also be noted that day to day variation or repeatability of this data set was approximately +/- 15%.
Table 2.3: Experimentally Measured Residual Fraction with Atmospheric Sampling
MAP (bar)
Residual Fraction %
0.33
19.6
0.40 0.50
15.5 13.4
0.60
11.1
2.6.2 Residual Measurement with CP Chamber Held Under Vacuum
Earlier attempts at measuring residual fraction were performed with a different sampling system. For this data set, the CP chamber was held under vacuum at 0.46 bar and the volume of the CP chamber was increased by 1 liter to dampen out pressure fluctuations.
By keeping the CP chamber under vacuum, there will be forward flow from the cylinder
29

for a much longer portion of the cycle. This results in a longer plateau and skip firing was not deemed to be necessary at the time. The motoring calibration technique was kept the same.
Figure 2.9 shows the two typical FFID signals needed for the residual fraction measurement, without skip firing, along with corresponding pressure traces. Figure 2.10
shows the typical cycle by cycle variation observed with this technique for five consecutive cycles. From these two figures, it can be seen that the signals have the same general characteristics that the atmospheric sampling system produced for the firing case.
However the motored trace appears different because there is always a forward flow of gas from the cylinder. It should also be noted that since skip firing was not as necessary with this technique, a high load point could be obtained in addition to the four low points already observed above. Equation 2.4 can again be used to calculate residual using the average of all 300 fired cycles instead of the skip fired traces as before. Table 2.5 shown below lists the data set collected in this earlier attempt. The residual data had repeatability to within 15% with this technique also.
Table 2.4: Experimentally Measured Residual Fraction with CP Chamber held at 0.46 bar
MAP (bar)
Residual Fraction %
0.33
17.5
0.40
15.8
0.50
14.1
0.6
11.8
0.9
10.2
2.6.3 Verification and Extension of Limited Experimental Data
To verify the limited experimental data shown above for the load sweep, and to extend the data set for the equivalence ratio, EGR, and spark timing sweeps, a review of other papers containing measurements of residual fraction was performed. The load sweep will be discussed first.
One major work done to measure residual fraction as a function of load on a modem four valve engine was done by Miller et al. [8] at Ford Motor Company in 1999.
Miller took experimental measurements of residual fraction as a function of MAP, equivalence ratio, EGR, and spark timing using a fast sampling valve technique. The engine used in the study was a four cylinder two liter engine with a pentroof chamber, dual
30

overhead cams, and a 200 overlap period. Referring back to table 2.2, this is almost identical to the Nissan engine used in this study, so the results should be directly comparable. The Fox correlation based on the Galliot data set was adjusted to match the
Nissan four valve geometry and was also used for a comparison over the load sweep.
Equation 2.5 shown below summarizes the Fox correlation:
=
-
1.266 N
___.87
( pi -
.7rT:-i+0.6 i
+
pi-07
( e)
(Eq. 2.5)
By calculating the overlap factor for the Nissan engine, equation 2.5 can be used to predict residual fraction as a function of intake pressure.
Fox defined overlap factor according to equation 2.6 shown below:
OF =
D A + D Ae
V disp
(Eq. 2.6)
IV=EV with Ai- = Lj-dO and
IVO
EVC
Ae= fLe -dO
IV=EV
Since the displacement volume and the diameter of the valves is known, the only measurement needed was valve lift as a function of crank angle. A static measurement on the Nissan engine was performed for both the intake and the exhaust valves, and the results are shown in figure 2.11. Figure 2.12 is a zoom in on the overlap period to show explicitly how the two valve areas are calculated. For the Nissan engine the overlap factor was then calculated to be 0.22 '/m.
A comparison of the Ford data set, the fox correlation, and the two different experimental data sets measured on the Nissan engine is shown in Figure 2.13. Reasonable agreement was observed between experimental data sets and the Fox correlation. The one high load point observed begins to deviate from the expected trend, possibly due to experimental error or back pressure effects. For ease during the upcoming modeling analysis, whole number residual estimates were made for the five load points based on
Figure 2.13, and the final values are summarized below in table 2.5.
31

Table 2.5: Assumed Residual Mass Fraction Values Load Sweep
MAP (bar) 0.33 0.40
Residual Fraction % 19 16
0.50
14
0.6
12
0.8
10
EGR, equivalence ratio, and spark timing have all been shown to have only modest effects on residual gas fraction [23,8]. Since, Miller is the only author who studied a four valve engine, extrapolation of the residual fraction value at 0.5 bar MAP were based off of his data sets (see Figure 2.14). The residual values assumed for the remainder of the experimental matrix is summarized below in table 2.6 below. A sensitivity analysis will be performed in Chapter 5 to explore how residual fraction errors will affect NO predictions.
Table 2.6: Assumed Residual Mass Fraction Values PHI, EGR, and Spark Sweep
Equivalence Ratio
Residual Fraction %
E.G.R.
Residual Fraction %
1.252
15
0
Residual Fraction % 14
Spark Timing + 10
13.5
1.12 1.06-0.914 0.838 0.77-0.71
14.5 14 13.5 13
4 8 12 16
17.5 21 24 27
+5 MBT -5 -15 -20
13.5 14 14.5 14.5
2.7 NO Measurements
NOx emissions include both nitric oxide (NO) and nitrogen dioxide (NO
2
).
However in spark ignition engines, the ratio of N0
2
/NO is very small, and NO
2 is assumed to be negligible in comparison. Therefore, experimental measurements of NO concentration are only considered and modeled in this study. The most well established technique used for measuring NO concentration in exhaust gases is chemiluminescence.
With this technique, dried exhaust gas is mixed with a controlled flow of ozone which reacts with the NO in the sample to form NO
2
. The chemical reaction between NO and ozone then produces energy in the form of light proportional to the NO concentration, which can be amplified and measured with a photomultiplier.
Until recent years chemiluminescence could only be used for steady state measurements with a resolution time of several seconds. Cambustion has now developed a sampling system (Fast NOx meter, FNO) based on the FFID that can measure changes in
32

NO concentration with a response time of approximately 4 ms [24,25]. The Fast NO meter allows cycle by cycle measurements to be made during each exhaust event which lasts 27 ms at 1500 rpm. Figure 2.15 shows a schematic of the FNO meter sampling system, and the similarities to the FFID, shown previously in figure 2.9, are obvious. A sampling probe is attached to an expansion tube that is contained in a constant pressure chamber held under vacuum at 0.46 bar. A tee piece is then connected to the end of the expansion tube for taking a small fraction of the exhaust gases up into the reaction chamber. In the reaction chamber, the exhaust gases are mixed with a flow of ozone, and a fiber optic cable then picks up the light emission from the reaction and carries it to a remote photomultiplier.
For this experimental work, two different measurements were required, steady state average engine out NO concentration, and cycle by cycle exhaust port NO concentration.
The steady state measurement will be used to validate MIT's cycle simulation over the range of engine operation, and the exhaust port NO values will be used to investigate and understand the amount of NO cyclic variation. To remain consistent, the Fast NO meter was used for both of these measurements even though the extra speed is not a requirement for measuring steady state NO. The steady state measurements were taken from a 10 gallon damping taken located downstream of the exhaust port. The details of the exhaust port measurement technique and signal analysis will be saved for chapter 6, after the steady state results are discussed.
The same sampling probe was used for both measurement types. The probe had an overall length of 250 mm and an inside diameter of 0.6mm. A resistive heating unit was also used to keep the probe temperature at approximately 120 'C and to avoid water condensation. It should be noted that no sample drier is used in the Fast NO sampling system. Since water vapor adversely interferes with the NO and ozone chemical reaction light emission, Cambustion recommends a 0.5% increase in NO concentration for every
1% of water vapor present in the exhaust sample [26]. To estimate water vapor concentration as a function of operating conditions, a correlation from Heywoods text was employed [23]. Finally, a calibration of the Fast NO meter was performed outside the engine before and after gathering each data set, by introducing a 1000 ppm and then a 5000 ppm mixture of NO and nitrogen, to the front of the sampling probe. The instrument was
33

zeroed with a pure mixture of nitrogen. Negligible drift in the calibration and zero values was observed during all experiments.
Figures 2.16 and 2.17 show the four steady state NO sweeps of MAP, equivalence ratio, EGR, and spark timing. It should be re-emphasized that the baseline operating condition, 1500 rpm 0.5 bar MAP Stoichiometric MBT, appears in all of the experimental sweeps. A discussion of each sweep will occur along with model comparisons in Chapter 4. Appendix A contains a more detailed listing of experimental results in spreadsheet form.
34

e Kitler6051
3 mM
Piston #4
@ TDC
6
+---+
-
X ~ -
-
35

101
0-)
CL LO
10
.E10
N
03 n a)
:3 a)
0~ a)
~ 10~
U
101
102
Cylinder Volume (cc)
10
3
Figure 2.3: Motoring
1500rpm 0.5bar
1' Advanced Phasing 1 Retarded Phasing
101
(D
U
0,
I IV
102
Cylinder Volume (cc)
10 3
102
Cylinder Volume (cc)
Figure 2.4: Motoring Log P Log V 10 Error in Volume Phasing
1500rpm 0.5bar
10
36

25-
Kistler 6051 - Side Mounted
Fiber Ontic - Spark Mounted
0
I..
(0
(0 a)
L.
0.
I...
a)
V
C
L.
(U
.0
a)
15-
10-
0
5-
0
Li
U
500 1000 1500
I
'I
I
2.5:
.0
*0
U)
(U
M.
MI
100-
90-
80-
70-
60-
50-
30-
10-
0
0 n Experimental Burn Angles
Fitted Wiebe Function
I Vb =
1- exp[-9.2
0 j
50)
50 100 150
Oasp - Crank Angle after Spark (deg)
I
37

r Sample Tube I.D.
Tee Piece I.D.
I
0.254 mm
0.200 mm
122 mm Sample Tube Length
Mi
ampling
CP Bleed
@ T DC
CP Chamber
2.7:
8--
7-
0)
6--
CD
5-
--
3 -
L
-
-25
I
I a
I .
'Skip
-
Trace
- 15
-10
k 5
24
0 290
390
0
38

a -
-
>(
5
4 cc
2
66
--
25
-- 20
15
10
5
0
8
4--
2
-5
11
39
-- 30
-- 25'
15
C

-
-
10 -
8
6-
*
*0
0
*
.*
0
0
* e
0
0
0
0
S
S
Exhaust Valv e
i
580 680
0
0
0
U
U
*
*
C'.
U
U
U
U
U
U
Intake Valve
U
U
U
U
U
U
N
I
I
M
980 780 880
0.30 -
0.25 -
E.
EI
0.15 -
EVC
0.10 Ivc
0.05
Ai Ae
0.00
700 705 710 715 720 725 730
40

0.25 -a
0
0
L.
U-
0.15 --
0.10--
- - - - -
Fox Correlation
Ford Data Set - SAE 982562
.
A
CP Chamber ATM Pressure
CP Chamber under Vacuum
0.05
0.00
0.2 0.3
0.6 0.7 0.8 0.9 1
25
20
-
:2
(0
0
15
10
5
0
0
251
201+
15
(0
0 10 +
5
0
.
0.6 a
0 0
0 .8
I
1
PHI
*
1.2
I
*
1.4
I
10 20 30 40
Spark Advance (deg)
50
35
30
25
(0
0
20
15
-i
10
5
0 -I
0
0
0
0
6
I
5
I
10
% EGR
I
15 20 25
41

Reaction chamber
Ozone in
Sam
CP chambr CL D Remote
SamT pling Head
CP VAC
42

3000--
2500--
0 z
2000--
1500--
0
C
1000--
500
-
0.2
N
0
0.4
0
0
0.6
MAP (bar)
N
0.8
,2000
-
E
0-
1500
-
0 z
3 1000
-
0
U
No
0
0
E
.
500-
C
U
0
0
U
0.6 0.8 1.0
-
2000
0 z
E
1500
1000 -
O
)
500 -
C
0-
-
I
0
N
5
' '
I
'
I I
'
10 15 20
% EGR
3000
2500
0 z
2000
1500
0
(D
C
1000
500
0
-
0
"
U
E
0
E
10 15 20 25 30 35 40 45
Spark Timing
-
43

44

MODELING METHOD
3.1 Modeling Approach
In order to gain useful information from a comparison of experimental data and model output, it is necessary to describe in detail how the modeling results were generated.
One of the specific objectives of this study was to understand whether the primary difficulty in predicting engine out NO concentration is a thermodynamic problem or a chemistry problem. Therefore, a modeling approach was designed to investigate each of these independently. In particular, a great deal of effort was put into making accurate predictions of in-cylinder pressure and temperature before predictions of NO concentration were considered. Figure 3.1 visually describes the modeling approach used to generate comparisons of experimentally measured NO and model output.
Following the top half of the flow chart, the first box represents the variables that can easily be measured on the Nissan test engine. Through data analysis, an accurate picture of the thermodynamic state of the cylinder gases throughout the engine cycle can be gained and is represented by the variables shown in the second box. Following the bottom half of the flow chart, the first box represents the main inputs that are typically feed to an engine cycle simulation. Prior to making a meaningful prediction of NO concentration, one must compare how well the cycle simulation has done in modeling the thermodynamic state of the gases. In particular crevice, heat transfer, and valve flow sub-models must be properly calibrated to ensure that the correct composition and amount of mass is being considered and that the gases are being subjected to the correct pressure history. Finally, once the thermodynamic modeling is accurate, a meaningful prediction of NO can be made, and an analysis of the chemical kinetics can follow.
The modeling study performed here was done with a modified version of the MIT
Quasi-Dimensional Cycle Simulation that was originally developed and documented by
Poulos and Heywood in 1982 [5]. Because a significant number of changes have been made over time, a brief description of the simulation's current status and sub-models will now follow to aid the reader in understanding the modeling predictions shown later.
45

3.2 Cycle Simulation General Description
The MIT cycle simulation is a zero dimensional, two zone thermodynamic model that solves a set of twenty differential equations derived mainly from the conservation of mass, energy, and the ideal gas law for each of the four main engine processes: intake, compression, combustion/expansion, and exhaust. Chapter 14 of Heywoods text [23] gives a good general description of this type of spark ignition engine model. Figure 3.2 gives a visual description of how the cycle simulation is set up including a crevice volume. The simulation gets the name quasi- because it utilizes a geometry routine that can calculate the position of the flame front with respect to the combustion chamber, by assuming the flame develops as a sphere anchored at the spark plug electrode.
At each crank angle of the 720 degree cycle, the in-cylinder pressure, unburned gas temperature, and burned gas temperature are calculated and assumed to be spatially uniform. The unburned gases are assumed to be a uniform mixture of fuel, air, and residual, and their thermodynamic properties are calculated from polynomial curve fits to the JANAF tables. During combustion, the burned gas thermodynamic properties are determined using a method developed by Martin and Heywood [27]. This method, developed for saving numerical time, uses an approximation to equilibrium properties, and all of the modeling results shown in Chapter 4 were calculated using the Martin routine.
The effect of using Sandia's full equilibrium code, STANJAN, instead of this approximation will be discussed in Chapter 5. Finally, during the exhaust process, when the temperature drops below 1800K and dissociation is no longer important, the burned gases are again assumed to be a uniform mixture and thermodynamic properties are obtained from JANAF table curve fits.
Table 3.1 shown at the end of the chapter contains a list of the input variables required by the cycle simulation. It should be noted that the pressure and temperature at
must be specified to start the routine. The simulation then goes through the entire cycle and compares the state of the gases at the end of the cycle with the initial guess made.
If the values are not in agreement, the end state is used for a second iteration and so on until the cycle simulation converges on a steady state solution.
46

3.3 Valve Flow Sub-Model
Flow through the intake and exhaust valves is modeled as a one dimensional compressible flow according to equation 3.1 and 3.2, for choked flow, shown below.
S C
DA m -
,efP o P T 1/y
2 y
J
1
-
PTr( -
-
1/2 ri-
---
)0
(Eq. 3.1)
* C DA rejP0
1 1
M (R To)
1/ y
K
2
N ('~
2
(l
+
+
14 with Aref --
_____
(Eq. 3.2)
The mass flow is a function of the pressure ratio across the valve and the effective
(CD) discharge coefficient. Since the reference area used by the cycle simulation is a constant, based on the inner seat diameter of the valves, the discharge coefficient must be a function of crank angle. Figure 3.3 shows the assumed value of the discharge coefficient map as a function of crank angle used by the cycle simulation. This map was experimentally determined for a Ford engine, but it had to be used for this study because no discharge coefficient information was available from Nissan. This is one possible source of error in the cylinder charging process routine.
Also, since the pressure in the cylinder at IVO is approximately equal to the exhaust port pressure (-1.01bar), there is a large amount of blowback when the engine is throttled.
This means that residual gases will flow from the cylinder up into the intake manifold until the pressure in the cylinder drops below the intake manifold pressure. During this period, the model assumes that the residual gases are in a plug flow, and that no heat transfer or mixing takes place in the intake manifold. Therefore, all of the residual gases that flow into the intake manifold return to the cylinder at the same temperature, before any fresh mixture can enter. In reality, under highly throttled conditions, the blowback process would be a vigorous flow with a significant amount of mixing and heat transfer taking place. Those assumptions add another source of modeling error.
However, the focus of this thesis was not to improve the valve flow sub-model of the cycle simulation. So, rather than attempting to model the blowback process and calibrate the discharge coefficient map, it was decided to force the model to ingest the
47

correct amount of fuel, air, and residual. This procedure will be explained in detail when the model calibration procedure is discussed next chapter.
3.4 Combustion Sub-Model
Once the cylinder has been charged and compressed, the combustion process is started at the specified spark timing. At this time, the gases are divided into an unburned zone and a burned zone, with a fixed combined mass. The flame is assumed to propagate through the unburned gases as an infinitely thin sheet which leaves behind burned gases in chemical equilibrium. The transfer of mass from the unburned to the burned zone is controlled by the Wiebe function which is shown below in equation 3.3.
Xb = 1-exp
F
(-Ospark sur
(A Burn)
N + j
(Eq. 3.3)
The constant, a, exponent, m, and the burn duration are determined from the heat release analysis burn angles as was described in Chapter 2. By using the Wiebe function, burn rate is eliminated as a variable and possible source of modeling error.
3.5 Heat Transfer Sub-Model
During the entire cycle, convective heat transfer takes place between the cylinder gases and the combustion chamber component surfaces. As was shown in Table 3.1, the component temperatures are taken as inputs to the model and assumed to be fixed throughout the cycle. The final variable needed to model heat transfer is a heat transfer coefficient between the cylinder gases and the surfaces. Heat transfer coefficients were calculated from the well accepted Woshni Correlation, which does not require any turbulence model. The Woshni heat transfer coefficient is a function of the mean piston speed and the state of the cylinder gases according to equation 3.4 shown below.
hc(w
/ m2K) = Ci(3.26)Bore(m)
2
P,, (kpa)
0 " T(K)~
05 3
W(m / s)0. (Eq. 3.4)
W(m / s) = f
( MeanPistonSpeed , DisplacementVolume, Pvc, Tvc,Vic)
48

00
MI
Ihead
= 380 K
Burned gases
Tpion
K
The variable named C
1 is a calibration constant which will be used next chapter for making accurate predictions of in-cylinder pressure.
With the heat transfer coefficient known, the amount of heat lost to the component surfaces can easily be calculated according to equation
3.4 shown below:
Q=hAon,(Tzone- Tcom) (Eq. 3.4)
During intake, compression, and exhaust all of the gases are in one zone and assumed to be at a uniform temperature.
Therefore three components of heat loss are calculated for each different component.
During the combustion process, there are two zones, each at a uniform temperature, as is illustrated in the figure shown above. Since the flame is assumed to be a sphere anchored at the spark, the geometry routine is able to calculate the contact area between the two zones and the three components. Then, six components of heat loss are calculated. It should be noted that no heat is exchanged between the unburned and burned gases during combustion, and radiative effects are neglected.
3.6 Combustion Inefficiency Sub-Model
In a firing spark ignition engine, a small percentage of the mass trapped inside the cylinder will escape the normal combustion process either due to blowby (leakage past the piston and valves), flame quenching at the wall, or by being trapped in crevices. This absence of burnable mixture will cause a reduction in peak pressure and corresponding temperatures. Therefore, this combustion inefficiency must be modeled to make accurate predictions of engine out NO concentration.
Blowby in well designed modern engines is usually less than one percent of the inducted mixture, so crevices have a larger effect on in-cylinder pressure. Crevices are small volumes in the combustion chamber which the flame can not propagate into, such as the piston top land, head gasket area, and pressure transducer tap. The flow of cylinder
49

gases into and out of these crevice regions is a function of the cylinder pressure, as shown in equation 3.5 below:
-c
( V ) e(Eq.
3.5)
As the cylinder pressure increases, mass is packed into the crevice. After peak pressure, gas flows from the crevices back into the combustion chamber.
The total volume of all the known engine crevices is generally assumed to be a few percent of the clearance volume. However, since this trapped gas is in direct contact with the combustion chamber surfaces, crevice gas temperature is much lower than the average cylinder temperature during combustion. This means that a small crevice volume can hold as much as 6-8% of the in-cylinder mass at the time of peak pressure due to the density effect. In reality, the composition of the gas in the crevices is a mixture of both unburned and burned gas, with the relative percentage of each depending on in-cylinder motion and the location of the spark plug.
The model used for combustion inefficiency in this study was based on a crevice model initially developed by McGrath [6]. The McGrath thesis gives a complete derivation of the cycle simulation equations needed when using the crevice model.
However, mass initialization and accounting errors during combustion clouded the conclusions reached in that study. At this time, only the major assumptions of this submodel will be discussed. The size of the total crevice volume was set equal to 2.2% of the cylinder clearance volume (see model calibration section in Chapter 4) and was assumed to remain fixed for all operating conditions. The location of this crevice volume is assumed to be entirely around the piston top land. This will allow the geometry routine to easily determine where the flame is in relation to the crevice. The pressure in the crevice is the same as the cylinder pressure, and the temperature was assumed to remain constant regardless of the operating conditions as defined by equation 3.6 shown below.
2 Twaii (Eq. 3.6)
Several things about this simple model will lead to a large deviation from what actually occurs with real engine crevices. First, since the flame propagates as a perfect sphere and all of the volume is located at the piston top land, no burned mixture reaches
50

the crevice before peak pressure with a centrally located spark plug. After peak pressure, the flow will only be out of the crevice. Therefore, all of the gas in the crevice will remain unburned regardless of operating condition. Also the temperature of the crevice and thus the effective size, would fluctuate with operating conditions (e.g. higher loads with slightly higher component temperatures). Thus, it should be noted that this is not intended to be a crevice model but rather an overall combustion inefficiency model which accounts for the all of the effects (blowby, crevices, and flame quenching) that reduce cylinder pressure.
Also, the effect of varying the size of the crevice will be studied in the final sensitivity analysis presented in Chapter 5.
3.7 NO Formation Sub-model
There are essentially four formation mechanisms for NO in combustion processes: thermal, N20, prompt, and fuel bound. Since propane was used for all experiments in this study, there is no nitrogen in the fuel and fuel bound NO can be eliminated immediately.
Prompt NO refers to the formation of NO in the flame zone. However, in spark ignition engines, the combustion process occurs at high pressure, and the flame can be assumed to be thin. Therefore residence times within the flame zone will be short, and the contribution from the prompt mechanism can be assumed to be small in comparison to NO formed in the post flame gases. This leaves the thermal (extended Zeldovich) mechanism and the
N20 mechanism. One of the main objectives of this study was to determine whether the challenge of predicting NO was primarily a thermodynamic or chemistry problem. In light of the recent modeling work done by Ford [7,8], a more specific objective along these lines was to verify whether the extended Zeldovich mechanism alone was accurate enough to predict NO formation, once the thermodynamic routines were calibrated. For this reason only the thermal mechanism will be used in the modeling study, and a discussion of the
N20 mechanism will be held until the sensitivity analysis section in Chapter 5.
It is well established that the main pathway to NO formation in high temperature fuel-air combustion applications is by oxidation of atmospheric nitrogen in the post flame gases. Zeldovich was the first to study this process [2], which is commonly referred to as the thermal mechanism, and he proposed equations 3.6 and 3.7 as the principal reactions.
51

This reaction set was later extended by Lavoie, Keck, and Heywood with equation 3.8 [3] to give the complete extended Zeldovich mechanism shown below.
N+NO >N2+O (Eq. 3.6)
N+0
2
2
>NO+O
(Eq 3.7)
N+OH
3
>NO+H (Eq. 3.8)
The forward rate constants for each of these reactions have been measured experimentally, however the accuracy of the measurement is only known to within a factor of 2. The reaction rates used for the initial model assessment were taken from a recent study done by Miller and Bowman [28] and are listed below:
ki = 3.3E+12 T'2 k
2
= 6.4 E+9 T e3
160
/T k3= 3.8 E+13
From equations 3.6 through 3.8, expressions for the rate of change of NO and N can be made after calculating reverse reaction rates as well, and the complete derivation was shown in other works [6,23]. By making the assumption that N atom formation has reached steady state and assuming the radical pool in the post flame gases has reached chemical equilibrium, a single differential equation that describes the formation of NO by the thermal mechanism can be written as in equation 3.9 shown below.
d[NO] dt
([NO] / [NO]e) 2
}
(Eq. 3.9) with Ri ki[ NO]e[ N]e R2 = k2[02]e[N]e R3 = k3[OH]e[N]e
In the above expressions [] denotes the concentration of the species in moles per cubic centimeter, and the subscript e denotes the equilibrium concentration. Therefore the only thing needed for calculating NO formation with the Zeldovich mechanism is a temperature profile and equilibrium concentrations for the post flame gases.
3.8 Temperature Profiles within the NO Sub-model
There are three different ways in which the temperature profile to be used for equation 3.9 can be generated with MIT's cycle simulation. The first is that the burned zone temperature calculated from the energy equation during combustion can be used
52

directly without any adjustment. The second is to assume that the burned zone is adiabatic and surrounded by a thermal boundary layer whose thickness if a function of the calculated heat transfer. With this method, there are three zones set up within the cylinder, each at a uniform temperature and composition. The third possible method of calculating NO is to again assume there is a thermal boundary layer, but now the adiabatic core is layered to model the temperature stratification that has been shown to exist in the cylinder during combustion [3,9]. With this method, a new layer is formed at each crank angle, and it is assumed that no mixing or heat transfer takes place between subsequent layers. Figure 3.4a
shows the concept of a fully mixed adiabatic core with a mean temperature, and figure 3.4b
shows an unmixed adiabatic core which has many different layers that each have a unique temperature profile.
It is worth emphasizing that the choice of NO routine has no effect on the calculated pressure trace of the cycle simulation. The NO sub-model is called only during the combustion/expansion process and only after the pressure, heat transfer amount, and temperature of the unburned and burned zones are already calculated. Therefore, the temperature profile used for the NO routine results from a secondary calculation and has no effect on any of the performance calculations of the cycle simulation.
A more detailed discussion of how the different temperature profiles are generated will now follow. It is well established that during combustion, a thermal boundary layer is set up everywhere the burned gases come in contact with the combustion chamber surfaces.
So, for the purposes of this modeling study, a thermal boundary layer routine was always used and only the choice of whether the adiabatic core was fully mixed or layered remains.
3.8.1 Fully Mixed Temperature Profiles
The fully mixed model of the adiabatic core works as follows. At each crank angle during the combustion process, the new mass fraction burned, temperature of the burned and unburned gases, cylinder pressure, and heat loss from the burned zone is passed to the
NO routine. Using this information, the temperature characteristics of thermal boundary layer can be assumed by using equations 3.10 and 3.11 shown below:
53

Tb.!.
Tburned + Twau
(Eq. 3.10)
Tburned
(Eq. 3.11)
Thermodynamic properties of this boundary layer can be easily calculated using curve fits to JANAF tables, since the temperature is below 1800K and the composition is assumed to be frozen. The mass fraction of the boundary layer can now be calculated based on these properties and the total heat lost to the three cylinder components with equation 3.12 shown below:
X urned
Xb.A
[b!-
B dpb.1.
Apb.l. dTb.I.
T dpb.1. dpb.1.
X b..
(hburned
hb.i.)
(w/ A&B thermodynamic variables of the boundary layer)
Q burned mass
(Eq. 3.12)
Once the boundary layer size is known, the mass and enthalpy of the adiabatic core can easily be calculated by using equation 3.13 and the simple energy balance shown in equation 3.14. This enthalpy and the cylinder pressure then completely define the state of the adiabatic core and the temperature can be iterated for with the thermodynamic property routines.
ma.c. =Xbumed(mass) Xbl (mass) (Eq. 3.13) h-
=a mburned
(hurned
) -m
(hbi )
(Eq. 3.14)
The boundary layer mass fraction and adiabatic core temperature profiles generated by using equations 3.12 and 3.14 are illustrated in figure 3.5 for the baseline operating condition. To aid in understanding the figure, the burned gas temperature and burned mass fraction, from the main simulation, are also shown. As the burned mass fraction increases, the size of the boundary layer grows, and this drives the temperature of the adiabatic core above the burned zone temperature for the remainder of the combustion process. Equation
3.9 can now be used to calculate the amount of NO in the adiabatic core gases at each crank angle. Because of its much lower temperatures, NO is not formed in the boundary layer, but NO is transferred in as it grows according to equation 3.15 shown below.
54

d[iNO]b.l. dt
(Xbl [NO]ac )+(XbH [NOL)
Xb,
(Eq. 3.15)
With the NO concentration in the adiabatic core and boundary layer now known, the overall cylinder NO concentration can be calculated by mass weighting the zones as in equation 3.16. The NO profiles for the two different zones and the overall cylinder NO concentration are illustrated in Figure 3.6 for the baseline operating condition.
[NO]cyl =NOIL.c. a.c. +INObLMbl mac + mb.1. + m
(Eq. 3.16)
3.8.2 Unmixed/ Layered Temperature Profiles
When using the layered or unmixed model NO routine, the boundary layer analysis stays the same, but now the temperature in the adiabatic core is calculated as follows. For each crank angle during the combustion process, a new layer of the adiabatic core will be formed and followed through the expansion process. The initial temperature of the layer is calculated by assuming that the unburned mixture goes through a constant enthalpy combustion process. The thermodynamic routines are called to iterate for a burned gas temperature that meets the constraints of equation 3.17. Once the initial temperature is known, the layer is assumed to follow an isentropic compression and expansion process until the exhaust valve opens as given by equation 3.18.
T( j,1)= f (j, P~
, hunbuned, Compositionbuned) (Eq. 3.17) ybi-Ilybi
T(i
)i j, EVO = T(ji
L -
P)
.
(Eq. 3.18)
Therefore, a new adiabatic core layer is formed at each crank angle until the burned fraction reaches 1. For example, at the baseline operating conditions, j
= 49 different layers are formed each with a unique temperature profile. The temperature profile for the first, tenth, twentieth, thirtieth, and fortieth layers are shown in figure 3.7 to illustrate this layering routine at the baseline operating condition. Each of these temperature profiles can be fed to equation 3.9 to calculate the NO concentration of each layer. Figure 3.8 illustrates the resulting NO concentration profiles calculated from the above listed layer
55

temperature profiles. The first layer to burn has a frozen concentration of over 6000 ppm, while the fortieth layer to burn is below 200 ppm. This emphasizes the large amount of incylinder NO stratification that can exist. The first and tenth layers also show that NO concentrations are limited by the equilibrium concentration as indicated by the negative formation rates. To determine the overall adiabatic core NO concentration, the amount of mass that burned during each crank angle must be used to mass weight the results according to equation 3.19.
[NO]. = mass,[ NO]j massi (Eq. 3.19)
A comparison of the temperature profiles generated with a fully mixed and a unmixed layered model is shown in Figure 3.9. It should be noted that almost no mass is subjected to the first two layered temperature profiles, j=1 and 10, shown in the figure 3.9.
Therefore, even though the NO concentration in these early layers is very high, there contribution to the overall adiabatic core NO is limited. The same is true for the last profile,
= 40, which has very little NO. This leaves the third and fourth profiles which straddle the fully mixed temperature profile. It will be shown later in Chapter 4, that the layered model can predict, both, higher and lower overall NO concentrations, depending on operating conditions.
56

NO (ppm)
COMPARE--
--------
I
NO (ppm)
CYCLE
SIMULATION
NO (ppm)
9
-1

MAXITS
PSTART
TSTART
RPM
FUELTP
PHI
EGR
TEGR
TATM
=
=
=
=
=
=
=
=
5
1.015
900
1500
2
1.004
2.5
300
298
TFRESH
PIM
PATM
PEM
BORE
CLVTDC
TIVO
TIVC
TEVO
TEVC
TSPARK c1
TPSTON
THEAD
TCW
ILAYER
BLAYR
SPBURN
DTBRN
CONSPB
EXSPB
CREVICE air resid aexp resexp imf ieg o2 c3h8 n2 co2 co h2 no h2o
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
235
483
723
337
1.63
420
380
380
T/F
T/F
T/F
50
9.2
310
0.467
1.00
1.00
8.6
58.77
-13
3.6
T/F
T/F
T/F
0.189
0.14
T/F
T/F
0.162
0.0325
0.7522
0.0227
0.0002
0.0001
0.00017
0.0304
Maximum Number of Iterations Allowed
Pressure at Intake Valve Open (atm)
Temperature at Intake Valve Ogen (K)
Engine Speed in Revolutions 1er Minute
Fuel Tvoe Used - 2 for Progane
Fuel Air Equivalence Ratio
Percent Exhaust Gas Recirculation
Temperature of Exhaust Gas Recirculation (K)
Ambient Temperature (K)
Temoerature of the Fuel Air Mixture in Manifold (K)
Intake Manifold Pressure - MAP (ATM)
Ambient Pressure (atm)
Exhaust Manifold Pressure (atm)
Piston Bore (cm)
Cylinder Clearance Volume (CC)
Intake Valve Open (BTDCE)
Intake Valve Close (ATDCE)
Exhaust Valve Open (ATDCE)
Exhaust Valve Close (ATDCE)
Spark Timing (ATDCE)
Calibration Constant for Woshni Correlation
Temperature of Piston (K)
Temnerature of Engine Head (K)
Temierature of Cylinder Walls (K)
Trigger between Layered and Mixed Burned Zone
Trigger for Boundary Layer during NO calculation
Trigger for Burn Rate Model
Wiebe Function Burn Duration
Wiebe Function Constant
Wiebe Function Exoonent
Trigger for Crevice Sub-model
Trigger for Specified Charge
Trigger for Specified Charge
Exoerimentally Measured Air Flow (grams/cycle)
Experimentally Measured Residual Fraction
Trigger for using Stanian for Thermodynamic Routine
Trigger for using Stanian for Mole Fraction Routine
Intake Mixture Mole Fraction - Oxven
Intake Mixture Mole Fraction Prooane
Intake Mixture Mole Fraction - Nitrogen
Intake Mixture Mole Fraction
-
Carbon Dioxide
Intake Mixture Mole Fraction
-
Carbon Monoxide
Intake Mixture Mole Fraction
-
Hydrogen
Intake Mixture Mole Fraction
-
Nitric Oxide
Intake Mixture Mole Fraction
-
Water Vapor
58

0
0
0
I
/
/
~ --- .
- -
0
0
Me
-I
d
dm
d t
dm dt =
0 0 i
- m e
- m
0 w
0 c rev
0
W=pV
0
Q= hAAT p = pRT
59

0.6
0 0.5 -
0
0 0.3 -
0.1
Ir
0.0
I I
I
I
I
I
-15 85 185 285 385
585 685
K
(b) Unmixed/ Layered Adiabatic Core
60

cc
3000 -
2500 -
--
Adiabatic Core Temperature
- - - Burned Zone Temperature
.
A
Burned Mass Fraction - Xb
Boundary Layer
M.F. - Xb.1.
C)
0-
E
-
1.4
1.2
'S -
1500 eeteeeSe
1
eeeeeeeee
0.8
1000 -
0.6
0
50 -
AA AA A
A AAA AAA AAA AAA A
A
-0.2
AA 0 -
335 360 385 410 435 460 485
0
C
0
.o
Crank Angle (Spark = 335)
3.5
as
U-
0
0 z
E
0.
1600
I jV
800
I,
'
/ A
1
-
-
--
//
0
-
335 360 385
61

3000
-
2500 -
2000 -
Q
E.
1500
E
-
-
500 -
0-
335 360 385
0.
0.
10000
-
8000
-
C
6000
-
0
C
-
1st Element
1 Oth Element
20th Element
(U
30th Element
40th Element
0r-II-
337 357 377 397
62

3000 -
2500
-
1.8
- 1.6
a)
-
-
- 1.2
Lw
E
1500 -
U
-X
-1
0
C
-0.8
'
1000
- 0.6 L-
0
500 -
.
x
0
E
.IumI..E* I
335 360
I
385
I
I
I
- 0.4
-
63

64

CHAPTER
STEADY STATE MODELING RESULTS & DISCUSSION
4.1 Model Calibration Approach
The easiest way to motivate a discussion of model calibration is to examine the error between the uncalibrated model and the experimental data. Figure 4.1 shows a comparison of cycle simulation predictions of NO concentration and experimental data for the stoichiometric load sweep prior to calibration. Obviously, regardless of the NO temperature profile routine used (Fully Mixed or Layered), the modeling predictions of NO concentration are over a factor of two higher than the experimentally measured values.
Referring back to the modeling approach flow chart shown in figure 3.1, these modeling results were generated by using burn rate, MAP, engine geometry, and timing as inputs.
However, crevice effects were neglected, and no investigation of whether the model was accurately modeling the thermodynamic state of the cylinder gases was performed yet.
In modeling studies, gross indicated mean effective pressure, IMEP, is often chosen as an indicator of accurate thermodynamic predictions. Figure 4.2 shows a comparison of the uncalibrated model predictions of IMEP and the experimentally determined IMEP for the same load sweep, both calculated using equation 4.1 below. It can be seen that the modeling predictions of IMEP are higher at all load points, with percent differences shown below in Table 4.1. Therefore, a model calibration must be performed to correct the thermodynamic routines before an assessment of the NO predictions can follow.
BDCE
IMEP = I PdV (Eq. 4.1)
BDCC
Table 4.1: Comparison of Uncalibrated Model and Experimentally Determined IMEP
1500 rpm, = 1.0, Load Sweep
MAP (bar) 0.33
Model IMEP % Higher 19
0.4
23
0.5
26
0.6
30
0.8
33
Since all of the experimental points used in this study revolve around the baseline operating point (1500 rpm, stoichiometric, 0.5bar MAP), the calibration procedure will be
65

performed at this point initially. To further illustrate the large IMEP difference, figure 4.3
shows the uncalibrated model pressure trace, the experimental pressure trace, which is the ensemble average of 150 cycles of measured pressure, and a unit-less rate of volume change curve for the baseline operating condition. Two things should be noted from figure
4.3: the large over prediction of peak pressure and the higher pressures during the expansion stroke. Looking at the rate of change of volume curve, it is clear that IMEP will be much more sensitive to expansion stroke differences than peak pressure differences.
However, high peak pressures will lead to high peak temperatures and corresponding NO concentrations, whereas NO predictions are fairly insensitive to late expansion stroke errors because production of NO freezes well before EVO. Therefore, IMEP may be an informative variable for performance calculations, but peak pressure and its location are more critical variables for modeling NO formation. It is now important to understand what model deficiencies are causing the pressure trace differences. Inaccurate cylinder charging, underestimating heat transfer, and neglecting crevice effects are all possible reasons for higher predicted pressure. Each of these will now be discussed independently.
4.2 Valve Flow Calibration
Figure 4.4 shows a comparison between model predictions of inducted air and residual gas fraction and the experimentally measured values. Since the amount of fuel scales with inducted air, the total mass being considered by the cycle simulation is higher than the experiment for all operating conditions. The overall mixture composition is also off because the predictions of residual fraction are lower than experiment. The actual percent difference for amount of air and residual are shown in Table 4.2 below:
Table 4.2: Uncalibrated Model and Experimental Comparison of Air and Residual Fraction
1500 rpm, $ = 1.0, Load Sweep
MAP (bar)
Model Air % Higher
0.33
8
Model Residual % Lower 24
0.4
11
25
0.5
12
30
0.6
18
28
0.8
22
40
Both of these valve flow modeling errors will lead to a poor prediction of pressure and NO concentration. By inducting too much air, there will be more burnable mixture in the
66

cylinder to increase heat release and peak pressures. The mixture composition error caused
by lower residual fraction predictions, will have only a minor effect on pressure. However, lower residual fraction will decrease the heat capacity of the mixture allowing it to attain higher temperatures and NO when subjected to the same pressure trace.
Because a discharge coefficient map could not be obtained for the Nissan engine, and modeling heat transfer and mixing during the blowback process was neglected, the residual and total mass errors are to be expected. However, the purpose of this study was not to improve valve flow modeling. Thus, in order to eliminate total mass and mixture composition errors, the experimentally averaged inducted air mass and residual fraction approximations from Chapter 2 were added as inputs to the cycle simulation. At the end of each complete cycle calculation, the predicted inducted air and residual fraction were compared with the experimental inputs. If the two values were in disagreement by more than 0.5%, the MAP and amount of EGR were adjusted according to equations 4.2 and 4.3
shown below:
MAP =MAP -(AIRdel - AIRxp) (Eq. 4.2) new, oldmoe p
EGRnew EGRoli -
-
(Eq. 4.3)
Using this iterative method, the cycle simulation is able to converge on the correct total mass and mixture composition at all of the operating conditions. This allows the valve flow routine to be eliminated as a source of modeling error.
Appendix A contains a list of the model inputs for MAP and EGR, from equations
4.2 and 4.3 above, that had to be used to obtain the exact cylinder charge for each sweep.
At all operating conditions, the MAP was lowered by approximately ten percent, and the
EGR ranged from 0-3 percent of the total charge. Figure 4.5 shows the pressure trace for the baseline operating condition before and after the valve flow routine was forced to trap the correct charge. The peak pressure has been reduced by seven percent and the IMEP has dropped by nine percent, but a considerable error still exists.
Because the intake manifold pressure was lowered to make the inducted mass correct, IMEP values should be slightly lower than experiment even if the peak pressure period is accurate. One other assumption that should be re-emphasized is that the temperature of the incoming mixture, including the EGR, was assumed to be at 310K,
67

regardless of the operating conditions. This will lead to errors in the overall temperature of the charge at the time of spark and affect the combustion and expansion temperatures even if the pressure trace is accurate. The effect of assuming a higher temperature of the incoming mixture will be explored during the sensitivity analysis chapter later.
4.3 Heat Transfer and Crevice Calibration
Referring back to figure 4.5, the model still over predicts peak and expansion stroke pressures at the baseline operating condition. Both the heat transfer routine and the lack of a crevice routine were considered to be possible sources of further pressure differences. In a first law analysis, as more heat is removed from the cylinder gases, pressure will be reduced. The effect should be more dramatic during the later stages of combustion and expansion when a large percent of the cylinder gases have been burned and average cylinder temperatures rise above 2000 K. The crevice effect represents a reduction of burnable mixture that reduces the total amount of heat released and corresponding pressures during combustion. Thus, including a crevice routine should have the same effect on the baseline pressure trace that the charge correction above did.
Initially, adjustments to each routine were made independent of the other to see if crevice or heat transfer effects alone could explain the pressure differences that still exist even after the valve flow calibration. Figure 4.6 shows a pressure trace when the amount of heat transfer was doubled during combustion and expansion with no crevice volume, and a pressure trace when a crevice volume equal to 5% of the clearance volume was used with no heat transfer adjustment. These two sizes are extreme but were selected because the resulting pressure traces all had approximately the same IMEP. Figure 4.6 now illustrates that large pressure differences can exist, even though the IMEP is correct. It also shows that increasing heat transfer has little effect on peak pressure, but a larger effect on the expansion stroke pressure. Including the large crevice had the opposite effect as heat transfer, with a larger peak pressure effect but almost no change to expansion pressure slope.
It becomes clear that combining both a heat transfer increase and a smaller crevice volume could give a good match to the experimental pressure trace. Figure 4.7 shows an
68

almost exact match of peak pressure and good agreement during the expansion process.
This calibrated pressure trace results from using a crevice volume equal to 2.2 % of the clearance volume and increasing heat transfer by 70% from the time of spark until the exhaust valve opens. The calibrated pressure trace has an IMEP equal to 3.85 bar, which is slightly lower than the 3.9 bar experimental trace. However this is to be expected since the intake pressure was lowered for the model.
If the crevice volume is meant to model combustion inefficiency in the form of crevices, blowby, leakage, and flame quenching, then a 2.2% volume is not unreasonable.
The heat transfer increase is justifiable because the spherical flame assumption minimizes contact area between the burned zone and the walls for a centrally located spark plug, and radiation effects were neglected. The single cylinder engine set-up also operates with a full
4 cylinder cooling system, thus the engine would tend to operate cooler. Also, the original valve flow error of not predicting enough residual fraction indicates that more heat should be removed, since cooler end gas temperatures result in higher residual gas density.
Calibrating the heat transfer routine during other parts of the cycle was not considered, simply because NO concentration is frozen by the time the exhaust valve opens and the valve flow routine was forced to trap the correct charge. This leaves only a possible temperature error at the time of spark as was mentioned before.
Figure 4.8 shows the total amount of heat transferred from the cylinder gases to the combustion chamber surfaces for the load sweep before and after the 70% increase. The bar chart is expressed as the percent of the available fuel energy, which is equal to mass of the fuel times the heating value. The total heat transfer was increased by only 25% with a
70% increase during the combustion and expansion events, since no adjustment to the
Woshni coefficient was made during the intake, compression, or exhaust process. The overall amount of heat loss after calibration is typically less than 35% of the available fuel energy. This seems reasonable based on the limited experimental data available on measured heat loss. The same size crevice volume and the same 70% increase in heat transfer will be employed for all the operating conditions. To summarize, the valve flow routine was calibrated at each operating point, and a single calibration of the heat transfer and crevice routine was done only at 1500 rpm, 0.5bar, Phi=1.0, and assumed to be the
69

same for all the experimental points. Figures 4.9 and 4.10 show how effective the single point calibration was for the load and equivalence ratio sweep. The predictions of peak pressure and IMIEP were all within 6 percent of the experimental results. Location of peak pressure was also in agreement within 1 crank angle, which indicates that the bum rate was accurate for all test points. The spark and EGR sweeps both showed similar accuracy, with the details shown in Appendix A.
4.4 Calibration of Kinetic Routine
By guaranteeing that the mixture composition was correct at every operating point and by calibrating the heat transfer and crevice routines at the baseline point, the thermodynamic portion of the modeling process has been eliminated as a major source of error. With confidence in the temperature and pressure profiles, meaningful predictions of NO concentration can now be made, and the kinetic routine can be investigated. Figure 4.11
shows the steady state load sweep after the thermodynamic calibration; the NO profiles were generated with both fully mixed and layered adiabatic core temperature profiles.
Referring back to figure 4.1, the thermodynamic calibration has improved NO predictions considerably. Both of the curves appear to have the correct slope of the experimental data, with the layered adiabatic core predicting NO concentration approximately 20% lower than the fully mixed routine at all points. However, even the layered adiabatic core results are still higher than the experimental data.
Figure 4.12 shows the steady state equivalence ratio sweep with accurate thermodynamics. Again both routines are predicting NO levels higher than the experimental values, but the slope of the two curves is not as consistent as the load sweep.
It should be noted that the layered profile is again predicting NO values lower than the fully mixed routine near stoichiometric points. However, under extreme rich and lean conditions, the layered profile predicts higher NO values when compared to the fully mixed routine. This initial comparison of the two routines with the experimental data, indicates that the layered routine has the slope captured more accurately. Regardless of the slope difference though, these two initial sweep comparisons merit an investigation of the rate constants used to describe the Zeldovich mechanism, since the model predictions are
70

always higher than experiment. So, a more detailed discussion of the results will be held until the rate constants are adjusted.
It is well excepted that rate constants for the Zeldovich mechanism can be in error
by as much as a factor of two. Previous work done by McGrath [6] showed that NO concentrations are mainly a function of the first reaction rate constant, ki, which was taking to be 3.3E+12TO.
3
. The calibration procedure chosen for the rate constants was simple and consistent with the thermodynamic routine calibration: a single point calibration of k, was done only at the baseline point. Miller et al. [7,8] at Ford Motor Company also recommend this single point calibration of k
1 at the world wide map point, which is almost identical to the baseline point of this study. The layered routine will be used to set k, for an exact match with the experimentally measured NO concentration. At 1500 rpm, stoichiometric, 0.5 bar MAP, the measured NO level was 1579ppm, while the layered routine with k, equal to 3.3E+12T
0 3 predicts 2035ppm. The k, value had to be adjusted down to 1.66E+12 T
0 3 until the layered routine predicts 1577ppm. This corresponds to nearly a factor of two decrease in k, which is typically considered to be the accuracy with which rate constants can be determined. It should also be noted that experimentally determined rate constants are studied at atmospheric pressure and 1000K. This could lead to large discrepancies from engine conditions of interest which are typically 2300 K and 20 atm at 0.5 bar MAP. Therefore, it is not unreasonable to adjust the rate constant by this factor of two. A comparison of the calibrated k, value used in this study with those used
by the authors discussed in the previous work section of this thesis is shown in Table 4.3.
Table 4.3: Comparison of Rate Constants for Reaction 1, N + NO -+ N2 + 0
Authors - Affiliation
Ball, Stone, Collings - Cambridge
University
Year ref. # Rate Constant, ki
*
1999 10 1.60E+13
Miller, et al. - Ford Motor Company 1998 7,8 1.55E+1
2
T-3+/- 30%
**
Raine, Stone - University of Auckland 1995 9 1.6E+13 - 3.3E+12Tu
Miller and Bowman
McGrath - MIT
1989
1996
28
6
3.3E+12Tu
Bowman
This Study after Calibration
1976 29
2000
1.60E+13
1.66E+1
2
T
0
.
**
T 3 is equal to 7.94 @ 1000K, 10.2 @ 2300K
This applies to peak pressure of 15-20atm, typical of MAP = 0.5 bar
71

The above table indicates that at typical spark ignition engine combustion temperatures and pressures, the rate constant used in this study for an exact match at the baseline operating condition is in good agreement with other researchers. Miller and
Bowman's [28] most recent study in 1989 was purely a chemistry review and involved no engine specific analysis. Ford motor company also used the Miller and Bowman value but had to employ a factor of five reduction over the load sweep for good agreement with experiment. The value at half load conditions is shown in table 4.3. With the appropriate rate constant now being used, a modeling assessment of the four sweeps can be made.
4.5 Load Sweep Modeling Results
Figure 4.13 shows the steady state load sweep with k, = 1.66E+12T
0 3
used for both the fully mixed and layered adiabatic core. The experimental data shows the expected trend of increasing NO with engine load. As the intake manifold pressure increases, the amount of burnable fuel air mixture increases, and the corresponding heat release drives the increase of pressure, temperature, and NO concentration. The amount of residual gas present is also decreasing with load, and this will lead to higher NO as well. Since the rate constant was calibrated at 0.5 bar MAP, an exact match is seen at that operating condition.
However, now it becomes clear that the slope of the modeling predictions is not in line with the experimental data points. With the layered adiabatic core temperature profile, the model predicts 3% lower than experiment at 0.33 bar MAP and 12 % higher at 0.8bar. A similar but larger difference trend was also noted by Ford Motor Company [7,8], with a
25% over prediction at wide open throttle after calibrating the rate constant for an exact match at a mid load point.
It is important to note that the factor of two adjustment to k, used for an exact match at 0.5 bar, had a varying effect over the entire load sweep. Figure 4.14 illustrates the percent reduction in NO concentration after dropping k, from 3.3E+12T
0
.
3 down to
1.66E+12T
0 3
. As the engine load decreases, the sensitivity to the rate constant grows.
This is explainable since NO formation is limited by the equilibrium NO concentration as is shown in equation 4.4 below.
72

d[N dt
[NO]
[NO I
(Eq. 4.4)
When the rate constant is increased at higher loads, the ratio of NO concentration to the equilibrium level grows large and formation rates are slowed to limit the effect. Also, if the level ever exceeds the equilibrium concentration, the rate of NO production becomes negative and decomposition will drive NO formation back down until freezing occurs.
While at light loads, the NO concentration is held well below equilibrium for all layers, making rate constant adjustments more pronounced. During the sensitivity analysis section, this varying rate constant effect should be remembered when determining if any of the variables actually change the slope of the modeling prediction curve.
When using a fully mixed adiabatic core, the slope of the model predictions is even steeper. The difference between layered and fully mixed predictions ranges from 9% higher at 0.33 bar to 18% higher at 0.8 bar. The sensitivity to the factor of two rate constant adjustment with a fully mixed adiabatic core is very similar to the layered model shown in figure 4.14. Therefore, if the rate constant was further reduced to get an exact match with the fully mixed model at 0.5 bar, the deviation at high loads would be considerably higher than the 12% seen with layering. This may explain why the work presented by Ford, which did not use a layered adiabatic core, showed a larger error at higher loads than seen here. A more detailed investigation of the possible causes of this slope difference will be carried out in the sensitivity analysis section next chapter.
4.6 Equivalence Ratio Sweep Modeling Results
Figure 4.15 shows the steady state equivalence ratio sweep after the single point rate constant calibration. It is seen that NO concentration reaches a maximum at the slightly lean condition where temperatures are still high and oxygen is readily available.
As the engine is forced leaner, temperatures continue to fall and NO formation is limited even with the excess oxygen. Temperatures are highest under slightly rich conditions, however available oxygen, which is critical to NO formation, is limited as the fuel air ratio increases. Again, the exact match is seen at the stoichiometric point and only of the slope of the curves is important.
73

When using the layered adiabatic core temperature profile, the modeling accuracy is quite good for the entire range of equivalence ratios. The layered model predicts 11 % higher at the extreme lean condition, 0.713, but predicts 16% lower at the extreme rich condition of 1.25. The rest of the sweep is within 10% accuracy. It should also be noted that the two extreme points have experimental concentrations below 200 ppm. Here, neglecting more detailed chemistry and NO formed in the flame, could have a larger impact in terms of percent errors. Some of the modeling error could also be attributed to a lack of experimentally measured residual fraction at the very lean and rich points. From
Chapter 2, less residual than stoichiometric was assumed under lean operation and more while rich. If residual was allowed to remain fairly constant, modeling accuracy would be improved. Figure 4.16 again illustrates the percent reduction in NO before and after the factor of two rate constant adjustment over the equivalence ratio sweep. There is a distinct increase in rate constant sensitivity as the mixture becomes leaner. This is also explainable
by the ratio of NO concentration to equilibrium levels.
With a fully mixed adiabatic core routine, the slope of the sweep is too steep on both sides of stoichiometric. If the rate constant was adjusted to match at 0.5 bar MAP, the last two lean and rich operating conditions would be substantially lower than the experimental values. This would indicate that a fully mixed model does not accurately represent the contribution from the early and late burning elements. It is difficult to fully explain why the ratio of layered to fully mixed routine predictions varies widely with operating conditions. However, the results observed in this study are consistent with some of the previous work done with layered models. Figure 4.17 shows an equivalence ratio sweep comparison of this study and an early work done by Blumberg and Kummer [4]. A similar slope is seen with slight differences on the rich side. The study by Raine and Stone
[9] also showed an approximately 15% reduction with a layered model compared to fully mixed routine at stoichiometric conditions. However, neither of these two studies used a boundary layer analysis, combustion inefficiency routine, or experimentally determined bum rates.
74

4.7 Exhaust Gas Recirculation Sweep Modeling Results
Figure 4.18 shows the steady state EGR sweep after the model has been calibrated.
The experimental data shows that NO concentration is reduced as the incoming mixture becomes more diluted. This is explained by the fact that EGR increases the heat capacity of the cylinder contents, keeping cylinder temperatures down during combustion. An exact match with experiment is seen again at the baseline operating condition, but now the slope of the layered model predictions departs more significantly from the experimental data. At
8% EGR the error has reached 12%, and with 16% EGR the model is predicting concentrations 29% lower than experiment. Again, it should be noted that a fully mixed model has an even steeper slope than the layered routine. If calibrated at the 0% EGR point, the fully mixed error would be much larger at the high dilution points.
Several possible reasons could be given for why the error is larger as dilution increases. First, experimental error could have occurred while metering the flow of EGR into the cylinder with the critical flow set up. The orifices could only be calibrated with a bubble flow meter outside the engine, so as to avoid the fluctuating manifold conditions which are under vacuum. The EGR was also simulated with a mixture of nitrogen and carbon dioxide, which merely had the same heat capacity as natural EGR. Therefore, not as much water vapor was present in the cylinder as would be with natural EGR. This would be similar to lowering the humidity ratio of the incoming mixture which has been shown to decrease NO concentrations [4]. Since the thermodynamic and equilibrium concentration routines both assume natural EGR composition, modeling error could be introduced. These problems will be briefly explored in the sensitivity analysis section.
One final note is that the Zeldovich mechanism can only model post flame production, not flame NO, with a limited amount of chemistry considered. Thus as post flame production drops, say below 500 ppm, using the Zeldovich mechanism alone could be a poor assumption.
75

4.8 Spark Timing Sweep Modeling Results
Figure 4.19 shows the steady state spark timing sweep in comparison with the calibrated model. The experimental data represents the effect of maximum heat release location. As spark timing is retarded, a large portion of the heat release occurs later in the expansion stroke and peak pressure, temperature, and NO concentration all decrease sharply. When the timing is advanced, the heat release process is taking place in conjunction with the compression stroke. This will lead to higher observed pressures and corresponding NO concentrations.
With the exact match at MBT timing guaranteed, the layered model gives an exceptional match with experimental data over the range of timings. The error for all data points is less than 6%. Unlike the other sweeps, the fully mixed model appears to also have the correct slope. If the rate constant was further reduced at MBT, the fully mixed model could be used to accurately predict NO concentration as a function of spark timing.
This accuracy with spark timing is expected, since little changes in terms of the radical pool, cylinder charging process, and mixture composition when only the ignition point is varied.
76

5000 Fully Mixed Routine
- - - Layered Routine n Experimental Data
E
-
0 z
3000 -
0
C
.w
1000
-
,-U
-0 v
W
N
9 i i i
0.3
0.5 0.6 0.7 0.8 0.9
-
10
-
- -Uncalibrated
Model
* Experimental Data
8-
CU
.0
6 -
0
(0
U
E
0
0
0.3
0.5 0.6 0.7 0.8 0.9
-
77

30
25
CL
15
(I)
10-
5
IM rE
U
U
- - - - Uncalibrated Model m Experimental Data
* dV/dO
I I i I v
280 305 330 355 380
0.4
-
U
Z 0.3 -
U
U
0
0.1 -
Ii
U
U
0 I I
0.2 0.3 0.4 0.5 0.6 0.7 0.8
MAP (bar)
0.9
..
25 -
C
0
20 -
15 -
10 -
U
U
U
U
U
5 -
0
I I I
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
MAP (bar)
-
78

30 r
25 -
S15
=
=4.46
- Uncalibrated
,Correct Charge
A A
\ A Experimental Data
A A
AA\
A
10
5 --
A,
0
280 305 330 355 380
4.5:
25 -
IMEP = 4.46
L.
.0
a)
L..
U,
(0
0)
L..
0)
C
20 -
15
-
10 -
0
5 b
/A
AA
\
AAA A
A
----
2x Heat Transfer
Experimental Data
Correct Charge
5% Crevice
0
320 350 380 410
440
79

25 -
Final Calibration
A Experimental Data
cc
Cu
CL
.
U)
15 -
10 -
-
5 -
0
A
-t
I I I I I I I
280 305 330 355 380 405
-r
I
4-'i
0
35
0
E 30
25
Ii*
15
10
5-
0-
0
0.33
I I I
0.5
0.6
0.8
-I
-
80

8c 6-
Cl2
0 1-
0.2
U
'
0.4
'
0.6
'
0.8
40
cc
630-
Cu
0)
0~
u 20-
0.
Q~
10-
0
'
0.2
I
-
I
0.4
0.6
MAP (bar)
-
0.8
5-
.
Eu U
U
U
0
3-
2I
I
0.6 0.8 1.0
I
"25-
.0
a.
)
1 0
.
I
0.6 0.8
1.0
I
1.4
Ratio
81

O
0 z
C
4000
3500
3000
2500
2000
1500
1000
500
0 I
0.2
-
Fully Mixed A.C.
-Layered A.C.
* Experimental Data oo
-I
*0
=U
0.4
4-
E
0.6
MAP (bar)
U
0.8
3
-
-
0
0
3500
3000 -
0 z
0.
2500 -
2000 -
1500 -
1000 -
500 -
0
0.6
/
=U
U..\
Fully Mixed A.C.
Lavered A.C.
* E xperime ntal Data
0.8
1.0
Equivalence Ratio
1.2
1.4
0 .
3
-
-
82

3500
-
3000
-
E
C2500 -
0 z
C-
0 1500 -
C
1000 -
500
0 '
-
-Fully Mixed A.C.
- - Layered A.C.
N Experimental Data
0.
0.6
0.8
3
-
-
40 -
30 -
0 z
C~'I
20 +
10 -
0 -
0.33
i
0.4
i -H--
0.5
MAP (bar)
0.6
0.8
-- i
83

2500
TF
-
0 1000
-
-
-
-Fully
-
Mixed A.C.
A.C.
.
Experimental Data
0
0.6
0.8
1.0
-
-
45
0 z
30
0*
01 15
U
~1~~~~ I
'
I I
I
I
I
'
I
111
I I
0.72 0.77 0.84 0.91 0.96 1.01 1.06 1.12 1.25
I
Equivalence Ratio
84

(A) Blumberg and Kummer [4]
1500 rpm, 0.66 bar MAP
1.2
NOmixed 1
NOunmixed to0
I
-
.9
.7
I
.9 1.0 1-1 .2 1,3
FUEL-AIR EQUIVALENCE RATIO
1.1
1.3-
(B) This Study
1500 rpm, 0.5 bar MAP
1.2
0.9-
0.8
0.7 0.8 0.9 1.0 1.1 1.2 1.3
Equivalence Ratio
0 z
0
G)
C
2000 -
1800
-
1600
I
1400 -
1200 -
1000 -
800 -
600-
400
200 -
0
C
-
-Fully Mixed A.C.
- -Layered A.C.
.
Experimental Data
5
% EGR
10 15
kl=1.66E+12T03 -
-
85

3500
3000-
E
0- 2500 -
0
z 2000 F o 1500 +
5> 1000
-
C
-
0
10
-
Fully Mixed A.C.
A.C.
* Experimental Data
0-
000',
....
--
I
30
0
3 -
-
-
86

CHAPTER 5
SENSITIVITY ANALYSIS
Whenever model results are presented in comparison to experimental data, it is important to explore the limitations of the model and identify how key parameters can change model predictions. This section will attempt to investigate the impact of model assumptions on NO predictions (i.e. the completeness of the Zeldovich mechanism, replacing Martin's simplified thermodynamic routine with a full equilibrium code, NO formation starting from a non-zero level, and using a kinetically controlled radical pool).
In addition, a more general sensitivity analysis will be performed in which selected parameters are perturbed, and the corresponding change in NO predictions are tabulated.
Other than the N20 mechanism and residual NO concentration sections, five operating conditions were chosen for this sensitivity analysis: 0.3 bar, 0.5 bar, 0.8 bar, PHI = 0.912, and 8% EGR. The points were chosen to give a wide variety of different in-cylinder temperature and pressure histories. Once the sensitivity analysis is complete, conclusions on the steady state modeling results can then be made.
5.1 Effect of Adding N20 Mechanism
In all of the modeling results shown so far, only the three reaction Zeldovich mechanism was used to describe the NO formation process. This was done to determine its accuracy alone over the operating range. However, under highly dilute conditions with excess air or exhaust gas recirculation, other researchers [3,6,7] have suggested that another chemical pathway to NO, called the N20 mechanism, can become important. The set of equations below show the mechanism used to describe the formation of N20 and additional NO in the post flame gases.
87

N
2
0+H >N
2
+ OH.......k
4
= 3.OE +13exp(-5350 / T)
N
2
0+0->N I
2
+02.........k
5
= 3.2E + 12 exp(-18900 /T)
N
2
0+0 -N0N NO...k
6
= 3.2E + 12 exp(-18900 T)
N
2
0+N
2
7 >2N
2
+0......k
7
= 1.E +15exp(-30500 / T)
By employing a steady state assumption for N20, as was done with the N atom in the extended Zeldovich mechanism, a single equation can still be used to describe the rate of
NO formation in each layer of or the fully mixed adiabatic core as in equation 5.1 below.
The radical pool is again assumed to be in equilibrium.
d[NO]
_ dt
dt
2R
6
{1 -([NO] /
[NO]e )2
/' (R
4
+ R
5
R,) (
R
4
= k4[ N
2
O]e[H]e R
5
R
6
= k
5
[N
2
0]ie[O]e R
7
= k,[N20]e[N2 ]e
The same temperature profiles that were used to generate the results presented last chapter can now be used with equation 5.1 to evaluate the amount of additional NO formed through the N20 mechanism. Figure 5.1 shows the steady state equivalence ratio sweep with the predictions from the Zeldovich mechanism shown last chapter and the new curve after adding the N20 effect. On the rich side of the sweep, the predictions seem to be relatively unaffected. However, near stoichiometric and leaner, the predictions are noticeably increased by the N20 mechanism as was expected. To make a fair estimate of what actual increase has occurred, the same rate constant calibration procedure must be applied at the baseline operating condition. At 0.5bar, the NO prediction was increased from 1577 ppm up to 1622 ppm. Thus, k
1 will now have to be reduced to maintain the exact match. Recalling figure 4.15 , the lean side was shown to be more sensitive to rate constant adjustments than stoichiometric. Therefore, the N20 mechanism effect may not be as large as it appears in figure 5.1.
Figure 5.2 shows the readjusted N20 curve with k
1 now equal to 1.5E+12TO. 3 for an exact match at the baseline point. The rich side still remains unaffected, but now the lean side is showing only modest increases. Figure 5.3 shows the percent increase seen by adding the N20 mechanism for all the sweeps after the re-calibration of k 1
. The maximum
88

increase is 14% at the extreme lean condition. The EGR and load sweep both showed a negligible amount of change after re-calibration of k1.
With the Zeldovich mechanism alone, the lean side was already over predicting NO concentration. Now the predictions will be even higher. In addition to a possible residual error that was mentioned earlier, it now becomes clear that a combination of the fully mixed and layered routines may be a better solution. Recalling the Chapter 4 analysis that showed a fully mixed model would under predict lean operation, if the number of layers were cut down to allow some mixing, a more accurate prediction may result.
5.2 Effect of Considering the Residual NO Concentration
When modeling the NO formation process, the amount of NO present in the cylinder due to residual gases is neglected, and the NO concentration starts from zero when using equation 3.5 or 5.1. Because residual fraction varied in this study from 10 to 19 %, there will always be a small amount of NO in the cylinder before the spark, depending upon the previous cycles NO level. As was discussed earlier, residual fraction is mainly a function of MAP, with fixed valve timing, and figure 5.4 shows the experimentally determined residual fraction and average exhausted NO concentration. By assuming the residual gases have an average level of NO present, a multiplication of the two experiment curves will give an estimate of the amount of NO present in the cylinder prior to spark discharge. With the temperature and NO stratification effect already demonstrated with the layered model, it is probably not a fair assumption to assume the residual gases have the average NO level. However, the solid line in figure 5.4 describes the relative NO concentration in the unburned mixture with this assumption.
By adding the unburned mixture NO concentration to the input of the cycle simulation, a new model prediction curve can be generated for the load sweep. Figure 5.5 shows the percent increase observed at each operating conditions. At 0.33bar, where residual fraction levels were 19%, almost all of the residual NO concentration shows up in the exhaust. However at 0.8 bar, where residual was 10%, less than one third was still present in the exhaust. This is again explainable by the level of NO being formed in the cylinder in relationship to the equilibrium concentrations. At light load the ratio of
89

NO/NOeq at time of peak pressures and temperatures with a fully mixed analysis is quite small, ~ 0.1, while the ratio at 0.8bar is near 0.6. Therefore forward rates at low loads will not be affected by the equilibrium level and most of the residual gas NO is merely added the amount formed during combustion. As NO concentrations near equilibrium at high loads, the forward rate is limited, making the starting level less important.
Looking at the load sweep predictions presented from last chapter in figure 5.6a, the residual NO effect looks encouraging. However, remembering the rate constant sensitivity curve, figure 5.6b, it is seen that a ki reduction to remove the rise of 8% at 0.5 bar will not drop the NO level at 0.8bar by the same 8% which would help correct the slope. Rather, the decrease would only be around 4%. Similarly, the 0.33 bar level would be decreased
by over 12% with an 8% adjustment at half load. Therefore, the residual NO effect will not significantly change the slope of the modeling curve, because the sensitivity to rate constant adjustments and the residual NO level follow the same trend. It is assumed that the remaining sweeps would also by only slightly changed. However, later in chapter six when cycle by cycle NO is considered, the amount of left over NO should be remembered.
5.3 Upgrade to Full Equilibrium Calculation
As was discussed in Chapter 3, at each crank angle of the engine cycle, the simulation calls upon a sub-routine for a complete description of the thermodynamic state of the cylinder gases (enthalpy, gamma, density, and rates of change of these variables with respect to pressure and temperature). For unburned mixtures and burned gases below
1800K, it is safe to assume the mixture is frozen. With this assumption and by considering only major species (no dissociation), the composition of the mixture can be determined from stoichiometry, and then properties are determined from JANAF tables.
During combustion and expansion, when burned gases are well above 1800K, a significant amount of dissociation occurs. By assuming the combustion products are in chemical equilibrium, a standard equilibrium code such as STANJAN can be used to calculate the thermodynamic properties at any given temperature and pressure. This calculation can be time consuming, so an approximation developed by Martin and
Heywood [27] is used by the cycle simulation to approximate the effects of dissociation on
90

thermodynamic properties. An analysis was performed at the five sensitivity analysis points to compare predictions of burned zone temperature and properties from STANJAN and the Martin routine.
Several things should be noted about this analysis. First, STANJAN did not have a direct variable for the rate of change of enthalpy with respect to pressure, dh/dP, so a one sided direct derivative was calculated with an additional call. Rates of change of density,
dp/dT and dp/dP, had to be calculated in the same manor. Also, STANJAN was not used for the NO sub-routine which generates an adiabatic core temperature profile after the burned zone temperature is known. Therefore no direct comparison was made for NO calculations during this initial investigation. Using the full equilibrium code just for calculating burned zone temperature increased run times by over a factor of ten. Some evidence of a significant error due to the Martin routine will be needed to merit the use of
STANJAN elsewhere in the cycle simulation.
Figure 5.7 shows the burned zone temperature predictions generated using the two different property routines. The temperature profile while using STANJAN was approximately 0.6% higher than the Martin routine up until NO formation would become frozen. The difference between the two at maximum temperature was 14 K. Cylinder pressure predictions showed negligible differences with the two different routines. Figures
5.8 and 5.9 show the predictions of specific heat, ratio of specific heats, enthalpy, and density for the baseline conditions during the entire combustion and expansion process.
Gamma differences were less than one percent, so this confirms the layered routine which is based on isentropic compression and expansion. Specific heat and enthalpy differences were small throughout the cycle, and density predictions were essentially identical.
Similar trends were observed at 0.33 bar, 0.8bar, 8% EGR, and lean operating conditions, with burned zone temperature profiles always within 0.6% agreement. This slight error could be attributed to the direct derivative calculations. Because STANJAN was in such good agreement with the cycle simulation, further work to upgrade the NO adiabatic core temperature routines was deemed unnecessary.
91

5.4 Kinetically Controlled Radical Pool Investigation
As was discussed in Chapter 3, during the thermodynamic and NO formation routines, the post flame gases are assumed to instantly reach a chemical equilibrium composition. By making this assumption and a steady state assumption for the N atom, the calculation of rate limited NO is easily written with a single differential equation. For a comparison, Sandia's SENKIN [30] chemical kinetics package was employed to calculate
NO formation through the extended Zeldovich with a kinetically controlled radical pool.
This will allow an assessment of the error introduced by the equilibrium radical pool assumption to be made.
To set up the analysis for SENKIN, a detailed propane oxidation mechanism proposed by Daguat et al. [31] was used which completely describes 68 species of the C-H-
O system. The three reactions of the Zeldovich mechanism with the calibrated rate constants from Chapter 4 were added to this mechanism. It should be noted that no additional nitrogen flame chemistry was considered in the mechanism used. This analysis was conducted to test the NO formation profile from the extended Zeldovich mechanism with a kinetically controlled C-H-O species radical pool during and after the flame oxidation process.
SENKIN requires an input file with the details of the unburned mixture composition, and the temperature and pressure history as a function of time. A fully mixed model is not applicable for this type of analysis, since the adiabatic core continually has unburned mass being added and mixed with the burned gases. However, the layered routine sets up nicely for a comparison, since each individual layer is treated separately and has an isentropic pressure and temperature history. Figure 5.10 again shows the baseline operating condition temperature profiles from the layered adiabatic core routine. Three representative layers were selected from figures like 5.10 for each of the five sensitivity analysis points and sent to the SENKIN package along with the corresponding time step
(crank angle) and cylinder pressure. For example, at 0.5 bar, the
2 0 th, 2 5
th and
3 5 th layer temperature profiles where chosen since over 75% of the mass bums in between layers 20 and 35.
92

The last unknown needed for the SENKIN analysis was a temperature profile for the gases during the first few time steps. This is during the fuel breakdown period when unburned mixture is converted to burned products. In reality, the unburned mixture would go through a constant enthalpy fuel oxidation process at near constant pressure. A brief investigation of the temperature profile that would result from a constant enthalpy reaction was performed. Figure 5.11 shows that the rise from unburned temperature to burned temperature, which could represent the flame, occurred over a time step of around two crank angles and that little NO forms through the Zeldovich mechanism during this process.
Therefore, for a simple analysis, the unburned gases were assumed to experience a step change from the unburned mixture temperature up to the initial burned zone temperature over a two crank angle (0.22ms) time step. For example, looking at figure
5.10 again, the input file for the
2 0 th element would have a specified temperature of 750 K at time zero and a temperature of 2421 K at 0.22ms. Then, the remainder of the time steps would follow the circle marked profile. If the unburned mixture was started at 2421 K at time zero, NO formation rates would have a large initial spike. This would be due to high radical concentrations combining with the unrealistically high temperatures during the fuel breakdown period, which would not be correct. By starting at the unburned temperature of
750K, NO formation rates were small until all of the hydrocarbon species were oxidized which agrees with the constant enthalpy calculation. This allowed a fair comparison with the cycle simulation NO profiles for the remainder of the process.
Figures 5.12 through 5.16 show the three selected representative layer NO profiles for each of the five sensitivity analysis points. It can be seen that the SENKIN analysis and the cycle simulation both predict similar NO formation profiles for all the points. Though, the amount of error is varying with operating conditions. Table 5.1 summarizes the percent difference between the two methods for the three layers at each condition.
93

Table 5.1: Percent Difference between SENKIN and Simulation Predicted Layer NO Profiles
Operating Condition
0.3 bar - Stoich
0.5 bar - Stoich
0.8 bar - Stoich
0.5 bar - Lean - 0.91
0.5 bar 12 % EGR
Representative Layer first second third fourth
4.8% -1.3% 0.4%
-
5.9% 2.8% 2.7%
7.9% 7.3% 9.2% -
-
17.3% 19.8% 21.7% 23.4%
4.4% 1.8% -1.5% -
Several trends should be noted from table 5.1. Since each layer analysis with
SENKIN is quite time consuming, NO predictions could not be made for an entire cycle.
However, if these layers are truly representative, the general trend in what the NO prediction would be can be inferred. The cycle simulation routine shows its largest over prediction, 25%, under lean operation. The load sweep showed an increasing amount of over prediction, but the differences were all under 10%. The EGR point showed a slightly lower over prediction than the baseline operating condition. (These small errors are attributed to slight differences in radical concentration, either due to calculation error or non-equilibrium effects. The varying amount of error with operating conditions my be explained by the radical error combining with varying temperatures and excess oxygen.)
Nonetheless, all three of these trends were noted during the comparison with experimental data in chapter 4. Correcting these three trends would move model predictions closer to experimental data points. This brief analysis is encouraging and merits further work with kinetic calculations, and this SENKIN analysis method would allow more detailed nitrogen chemistry to be investigated. It may also be useful to develop a cycle simulation that fully utilizes both SENKIN and STANJAN to calculate the NO routine temperature profiles and formation rates for the entire combustion expansion process, if numerical time is not a concern.
94

5.5 General Sensitivity Analysis
It is often helpful to see how slight changes, to variables that are expected to be important and parameters where uncertainty exists, will affect the model NO predictions.
Tables 5.2 and 5.3, shown on the following two pages, detail this type of analysis applied to the five different sensitivity analysis points. The excess of load points was chosen for the purpose of exploring the mismatch in the modeling slope and for the upcoming cycle
by cycle analysis section. A spark timing point was neglected because of the absence of any substantial modeling error for that sweep. The tables represent the amount of forced change given to a input parameter and a corresponding delta NO value listed in ppm and percent change. A listing of the models corresponding IMEP prediction was also included to demonstrate how the parameter adjustment would affect the thermodynamic calibration from Chapter 4. All of the delta NO values are in reference to the final modeling comparison figures shown last chapter (figures 4.13, 4.15, 4.18, 4.19) which were after the calibration of k, = 1.66E+12T
0 3
. A large number of the variables were considered with a layered adiabatic core since that was the main focus of the modeling results section.
However, a few selected variables are also shown while using a fully mixed routine to generate the adiabatic core temperature profile. The layered model analysis results will be discussed first.
5.5.1 Variables of Uncertainty Residual, Heat Transfer, Effective Crevice Size
The discussion of Table 5.2 and 5.3 will begin with the first three variables, residual fraction, heat transfer increase, and effective crevice volume, which were pivotal in making accurate predictions of cylinder pressure and temperature during the calibration of the models thermodynamic routine. For the entire load sweep, it is seen that residual fraction is the most important of these three. Though, a ten percent change in residual had a decreasing effect as the manifold pressure increased. This may be simply explained because the residual changed from 19% at 0.33bar down to 10% at 0.8bar, making the absolute change greater at low load. Regardless, the sensitivity to residual fraction remains substantial for all of the operating conditions shown, because of its large effect on the thermal capacity of the unburned mixture and resulting burned gas temperature. This large
95

Operating Condition
Final Layered Model
Residual +10%
Residual -10%
Heat Transfer Mult. +10%
Heat Transfer Mult. -10%
Crevice Volume +10%
Crevice Volume -10%
Delta Burn +50
Delta Burn -50
Fuel Air Ratio +1%
Fuel Air Ratio -1%
Spark Timing ADV 20
Spark Timing RET 20
Intake Mixture Temp (+200)
Rate Consant 1 +50%
Rate Consant 1 -50%
Adiabatic Core Temp +1%
Adiabatic Core Temp -1%
Mole Fraction Routine
5.2:
-
0.3 bar MAP - PHI = 1.0
706ppm 240 IMEP
IMEP Delta NO % Change
235 -178 -25.21
244
236
245
240
241
243
235
201
-33
71
-4
15
-139
174
28.47
-4.67
10.06
-0.57
2.12
-19.69
24.65
240
238
237
243
-13
25
117
-81
-1.84
3.54
16.57
-11.47
237
241
240
240
240
242
77
224
-281
198
-165
-12
10.91
31.73
-39.80
28.05
-23.37
-1.70
393
383
387
391
378
385
384
381
389
0.5 bar MAP - PHI = 1.0
1577ppm 385 IMEP
IMEP Delta NO % Change
380 -218 -13.82
390
378
245
-79
15.54
-5.01
83
-27
40
-237
261
-70
87
143
-139
5.26
-1.71
2.54
-15.03
16.55
-4.44
5.52
9.07
-8.81
382
385
385
386
385
385
128
280
-513
332
-298
-30
8.12
17.76
-32.53
21.05
-18.90
-1.90
0.8 bar MAP
2649ppm -
- PHI = 1.0
628 IMEP
IMEP Delta NO % Change
622
634
625
-194
178
-82
-7.32
6.72
-3.10
641
624
631
644
624
29
-29
21
-285
263
1.09
-1.09
0.79
-10.76
9.93
628
618
622
642
-174
123
153
-143
-6.57
4.64
5.78
-5.40
622
628
628
628
628
639
123
229
-553
360
-356
-81
4.64
8.64
-20.88
13.59
-13.44
-3.06
Final Fully Mixed Model
Delta Burn +50
Delta Burn -50
Adiabatic Core Temp +1%
776ppm 240 IMEP 1858ppm 385 IMEP 3259ppm 628 IMEP
IMEP Delta NO % Change IMEP Delta NO % Change IMEP Delta NO % Change
243 -80 -10.32 390 -187 -10.06 644 -209 -6.41
235
240
86
259
11.10
33.42
377
386
159
479
8.56
25.78
625
628
170
463
5.22
14.21
Experimental NO = 734ppm Experimental NO = 158 1ppm Experimental NO = 232 1ppm

5.3:
-
Operating Condition
Final Model Predictions
Residual +10%
Residual -10%
Heat Transfer Mult. +10%
Heat Transfer Mult. -10%
Crevice Volume +10%
Crevice Volume -10%
Delta Burn +50
Delta Burn -50
Fuel Air Ratio +1%
Fuel Air Ratio -1%
Spark Timing ADV 2*
Spark Timing RET 20
Intake Mixture Temp (+200)
Rate Consant 1 +50%
Rate Consant 1 -50%
Adiabatic Core Temp +1%
Adiabatic Core Temp -1%
Mole Fraction Routine
346
366
357
360
364
346
355
361
354
363
354
359
359
359
359
359
0.5 bar MAP -
1731ppm -
PHI = 0.914
359 IMEP
IMEP Delta NO % Change
355 -256 -14.79
363 279 16.12
-103
85
-44
8
-330
371
-68
42
214
-215
166
429
-667
376
-339
-69
-5.95
4.91
-2.54
0.46
-19.06
19.13
-3.93
2.43
12.36
-12.42
9.59
24.78
-38.53
21.72
-19.58
-3.99
0.5 bar MAP
700ppm -
397
EGR = 8%
393 IMEP
IMEP Delta NO % Change
384 -98 -14.00
91 13.00
380
401
-33
43
-4.71
6.14
-1.57
392
395
399
379
382
391
389
398
387
393
393
393
393
392
-11
4
-156
199
-56
32
100
-99
53
184
-278
186
-158
-21
0.57
-22.29
29.14
-8.00
4.57
14.29
-14.14
7.57
26.29
-39.71
26.57
-22.57
-3.00
Final Model Predictions
Delta Burn -50
Delta Burn +50
Adiabatic Core Temp +1%
1731ppm 359 IMEP
IMEP Delta NO % Change
345 216 11.14
363
359
-256
530
-13.20
27.33
700ppm 393 IMEP
IMEP Delta NO % Change
379 -96 -13.71
398 109 15.96
393 242 35.43
Experimental NO = 1649ppm Experimental NO = 803ppm

effect on NO also comes with very little change in predicted IMEP, meaning the pressure trace was relatively unaffected by the residual perturbation. For the mid load lean and
EGR points a ten percent error in residual fraction could lead to an approximate 15% error in NO predictions. Since very little experimental data is available for residual fraction, a
10% error in the assumed values used is not unreasonable.
The heat transfer multiplier, used to increase the amount of heat loss to the walls during the combustion and expansion process, affected NO predictions less than half as much as residual fraction for the same ten percent change at all operating conditions. A ten percent change in the size of the crevice used to model combustion inefficiency had a negligible effect on NO predictions for all cases. This variation in crevice size is assumed to also verify the insensitivity of NO to the assumed temperature of the crevice volume.
However, neglecting the combustion inefficiency effect altogether would cause large modeling errors as was demonstrated earlier.
5.5.2 Input Parameter Perturbation
The following section will explore the sensitivity of NO to parameters which were assumed to be known with certainty and used as model inputs (Burn Rate, Fuel Air
Equivalence Ratio, Spark Timing, and Fresh Mixture Temperature). Burn rate was varied
by changing the delta burn parameter of the Wiebe Function to simulate a 5 degree slower and faster burning cycle. This will essentially keep the 0-2% bum angle the same while extending or reducing the 0-50 and 0-90 bum angles. Faster burning cycles will shift the heat release earlier in the cycle increasing cylinder pressure and temperature, and slow burns will have the opposite effect. The layered model has a large amount of sensitivity to these modest changes for all of the operating conditions, with a maximum delta NO change of over twenty percent at the light load and EGR points. However, these slight changes in burn rate do have a large effect on the location of peak pressure. So, it is unlikely that any of the experimentally determined bum rates used to generate the modeling results could have been in error by +/- five degrees without this being detected easily.
Fuel air ratio changes of one percent had only moderate effects on the predicted NO values overall. A 1% error though is reasonable and often accepted to be the level of
98

accuracy for the Horiba Meter used. Also, unlike the other parameters investigated thus far, NO predictions show their largest sensitivity, over 6 %, to the amount of fuel at the
high load point and EGR point. This is interesting mainly because these tables were generated in hope of identifying why high load points are over predicted and EGR points are under predicted. Sensitivity to fuel air ratio may also be important when trying to understand how mixture non-uniformity will affect NO predictions. All model results are generated under the assumption that the charge composition is uniform throughout the cylinder.
Spark timing variation was investigated mainly because from cycle to cycle the initial kernel development often varies, causing an effective degree or two shift in 0-2% burn rate. The sensitivity results are obvious based on the spark timing sweep discussed earlier. Also, as with adjusting the burn rate duration parameter, these errors are easily detected because of the shift in peak pressure location they cause. Regardless, it is important to show both of these sensitivities to demonstrate that accurate burn rate data is critical to NO predictions.
Finally, the assumed incoming fresh mixture temperature was increased from 310K up to 330K. Since intake temperature has a large affect on the amount of trapped mass, the routine to guarantee accurate cylinder charging was used with this parameter. This was done to effectively increase the temperature of the unburned mixture prior to spark discharge with out changing the total mass or composition. The NO sensitivity to this 20 degree increase was less than 10 percent for all operating conditions and had a diminishing affect at high load. This is an important variable, however, because the thermal environment of the intake manifold will vary with operating conditions, and the amount of heat absorbed by the incoming mixture would fluctuate.
Rate constant effects were again explored because the percent change curves shown earlier were developed while going from 3.3E+12T
3 down to 1.66E+12TO.3. A plus or minus fifty percent value was employed to the calibrated value of 1.66, and the expected trend was again observed. Unlike earlier observations, decreasing k, further below the already calibrated value showed less variation with load than before, since NO levels are
99

now further below equilibrium values at the time of peak pressure and temperature. This could make effects such as residual NO slightly more of a factor.
5.5.3 NO Sub-model Variables
The final two parameters adjusted relate to the temperature profile generated for the adiabatic core and the mole fraction routine. Since STANJAN had been implemented to investigate the thermodynamic routines, it was easy to validate MIT's simplified equilibrium concentration routine as well. When species concentrations are required for the Zeldovich reactions, the MIT cycle simulation uses a code which takes pressure and temperature as inputs and returns burned gas composition, considering only fourteen chemical species. Therefore, in place of this routine, STANJAN was called. NO showed very little sensitivity, less than 4 %, to the equilibrium mole fraction routine used for all conditions. This is interesting for the EGR point since direct mole fractions could be input to STANJAN that matched the artificial EGR used, instead of the original cycle simulation's natural EGR with water vapor.
The adiabatic core temperature parameter represents a direct 1% increase to the temperature of each layer in the adiabatic core during the combustion and expansion process. This shows the dramatic effect that an approximate 25 degree error in the peak temperature prediction can have on NO concentrations. A one percent change in adiabatic core temperature caused predictions to be off by over twenty percent at all points other than the high load condition. Therefore, it seems unreasonable to expect a model with this level of sophistication to predict NO concentration with more accuracy an what was demonstrated earlier. The room for error is just too large. This also makes the STANJAN section more of a concern, since a 0.6 % temperature error can be so important.
5.5.4 Fully Mixed Sensitivity
Even though the layered routine was chosen to be the focus of this study, the fully mixed NO sensitivity to burn rate was investigated for the purpose of later discussion. It is seen that when subjected to the same amount of burn duration variation, +/- five degrees, a fully mixed routine shows only half the sensitivity as with a layered model. This is too be
100

expected since a fully mixed routine is essentially removing the contribution of early and late elements, unlike the layered model which will pick up and amplify slight changes in the bum angle development. This point will be discussed again in the cycle by cycle analysis section next chapter. The fully mixed model is also shown to be equally sensitive to even a one percent error in adiabatic core temperature.
5.6 Summary and Conclusions Steady State Modeling
A modeling calibration methodology was developed to ensure accurate thermodynamic representation of the cylinder gases throughout the cycle. The first variable considered was the total mass and mixture composition in the cylinder at the time the intake valve closes. Since it would not be meaningful to compare NO results from a model with poor residual and total mass predictions, the MIT cycle simulation was forced to trapped the correct charge for all experimental points considered. This resulted in having to drop the intake manifold pressure slightly to reduce total mass and inputting a small amount of EGR to match the amount of expected residual gas fraction.
However, predicting accurate cylinder pressure required more than just correct mass. A single point calibration of the heat transfer routine was done at the baseline operating condition. It was concluded that a 70% increase to the Woshni correlation, from the time of spark through EVO, gave good pressure agreement during the expansion stroke.
Taking in to account that not all of the inducted mass contributes to the heat release analysis was also deemed necessary for improving model pressure predictions. To handle this, a combustion inefficiency sub-model, based on a simple crevice analysis, was employed to get an accurate match of peak pressures. A comparison of model and experimental pressure data showed that the single baseline point calibration of heat transfer and combustion inefficiency was effective over the operating range considered. Finally, since bum rate was used as an input, peak pressure location agreed well with experiment and modeling error was reduced.
After the thermodynamic routines were calibrated in the manor described above,
NO predictions were investigated. It was concluded that using a Zeldovich mechanism reaction one rate constant of 1.66E+12TO.
3 gave an exact match with experimental data at
101

the baseline operating condition. Using this same rate constant, the extended Zeldovich mechanism alone was shown to be able to predict NO concentration as a function of engine load, fuel air equivalence ratio, and spark timing to within 15% accuracy. It was also concluded that, for all experimental sweeps, using a layered adiabatic core with a boundary layer improved the slope of modeling predictions in comparison with a fully mixed model.
A great deal of accuracy would be lost without using a layered routine to model in-cylinder temperature stratification. An attempt was made at modeling the effects of EGR on NO concentration by using a simulated mixture of nitrogen and carbon dioxide. The accuracy of the calibrated model with the Zeldovich mechanism was shown to deviate from experiment by close to 30%.
A detailed sensitivity analysis was carried out to explore the key factors in NO formation and limitations of both the cycle simulation itself and the extended Zeldovich mechanism. Increasing the amount NO chemistry considered, by adding the N20 mechanism, was shown to have a negligible effect on the load, spark, and EGR sweeps and only modest increases under lean operation. The simplified thermodynamic routine developed by Martin and Heywood to approximate equilibrium properties was verified by
STANJAN over the operating range studied. The effects of considering the left over residual NO concentration was concluded to have only a minor effect on the load sweep, but could be important for cycle by cycle modeling. Using a detailed kinetic calculation instead of a simple equilibrium radical pool analysis was shown to improve the slope of the
Zeldovich mechanism predictions at the cost of more computational time. Based on the input/unknown parameter sensitivity analysis tables, model predictions of NO concentration were demonstrated to be most sensitive to residual gas fraction and bum rate.
NO predictions were also concluded to be only slightly affected by small changes in the amount of heat transfer and effective size of crevice volume used.
For practical applications, this investigation showed that model areas that require the most attention to detail are the valve flow routines and the burn rate model. With confidence in these two areas, a single point calibration of combustion inefficiency effects and heat transfer was concluded to be adequate for making good predictions of cylinder pressure and its phasing. The final conclusion reached was that predicting NO
102

concentration is mainly a thermodynamic problem, and the extended Zeldovich mechanism alone with a layered adiabatic core and boundary layer can be used to make predictions of
NO concentration over a wide variety of operating conditions with reasonable accuracy.
103

1800
1600
0.
0.
0 z
1000
0
800
C
W
600
'
I'
I'
I
0
I
0.6
W I
0.8
/
/'
Im
1.0
I
N20 Me chanism
-
h Mech.
* Experirr ental Data
i
I
-
3
-
1500 rpm -
0-
0 z
0
1800
1600
1000
800
600
0
-I
0.6
I
/
'
Adjusted N20 Mech.
-
-Zeldovich Mech.
' Experimental Data
I I
0.8
1.0
-
0
.
3
-
1500 rpm -
104

1.30 -
1.150
0
N)
1.15
0
1.00
A
A
A
A
A
A
A a A
-C)
0
N
1.000 a
U z
C)4 0.85
0.70
i i i i 1
0.60 0.80 1.00 1.20 1.40
Equivalence Ratio
0
C'J z
0.850
0.33bar
0% EGR
* Load Sweep
* EGR Sweep
£
+
0.8bar
16% EGR
5.3:
All Sweeps
2500 -
A
0
0
2000 -
0
1500 -
0 z
1000
-
500-
0-
I
0.3
.
A
0
0 Exp. Engine Out NO
-- NO in Unburned Mixture
& Exp. Residual n
0
A
E
A
I
0.5
0.7
MAP (bar)
A
-20
.16
14
L
10
88
0.9
5.4:
-
105

-
0 z
7@
10
0
15 +
10 +
0)
5 +
0±4-
6
0
C;
5.5:
-
3500-
3000-
2500 --
(a)Steady State Load Sweep
---
Layered AC.
*Experimental Data
(b) Sensitivity to Rate Constant k1
40-
30-
0 1500-
1000
500
0
0.2 meCU
/~elwt
7-dV
0.4 0.6
MAP (bar)
0.8
10_
0.33 0.4 0.5
MAP (bar)
0.6 0.8
5.6:
106

3000
m Stanjan Equilibrium Code
Martin Simulation Routine
0.
E
2500
-
C)
0
N
-
-
~--
1500
335 350 365 380 395
5.7:
-
m Specific Heat Cp - Stanjan
Specific Heat Cp - Martin
2.3E+07 _
2.1 E+07 -
0D
1.9E+07 -
*1)
CD
1.7E+07 -
0.
(0
1.5E+07 -
1.3E+07 '
1500
'
2000
Temperature (K)
'
2500
1.28 -
1.26
Mu
E
E
1.24
1.22
CU
1.2
1.18
1.16
1.14
1500
-U
-U
-U
-
U.
_ Gamma
-
Stanjan
-Gamma - Martin
2000
Temperature (k)
2500
5.8:
-
-
-
107

0-
V
C)
6.OE+09
4.OE+09
-
2.OE+09
0.OE+00
-2.0E+09
-
'
-4.OE+09
-6.OE+09
-8.OE+09
-1.0E+10
1500
+
Enthalpy - Stanjan
Enthalpy - Martin
2000 2500
Temperature (K)
3.OE-03 -
2.5E-03 -
(D
2.OE-03 -
0
'a
1.5E-03
(D
1.OE-03
-
-
5.OE-04
-
-~
0.OE+00
1500
+ density - STANJAN
density - Martin
-/
2000
Temperature (K)
2500
-
-
-
3000
2500
2000 -
.
20th Element
CL
E.
11
-
500 -
335 360 385
Crank Angle (Spark = 337)
108

3000 --
Temperature
- - - - NO concentration
2500
--
2000 -
L
E
CD
:-
1 500 4
1000 x
500 -
*1
0
1 2 3 4
Time (Arbitrary CA @ 1500rpm)
5
= 850K
1500 rpm, Stoichiometric
2800 -
,:~:--
- - Layer Temperature
.
Senkin -NO
Simulation-NO
7000
6000
E
5000 o
.i
0
E
1600 -
800 -
3000 C
0
0
z
1000
0
0 !
337 387
1500 rpm, MAP=0.5bar, Stoichiometric
109

2800
2400
Q
2000
1600
E c-
(D
1200
800
400
0
-
330
- - - Layer Temperature
.
Senkin -NO
Simulation - NO
- 3500
- 3000 E
CL
ME
00
0
1500 C
-
1000 z
350 370 390 410
Crank Angle (Spark = 330)
430 450
0
1500 rpm, MAP=0.3bar, Stoichiometric
-- --
Layer Temperature
2800-
2400
U
Senkin - NO
Simulation - NO
8000
-7000
E
C-
.2
2000
-
-
E
-
6000 C
.
CL
- 5000
0
3000 0
800-
9
* mesmmmmmummm
I I -
-
- 1000
0
-
362 382
0
1500 rpm, MAP=0.8bar, Stoichiometric
110

.
(D
E
2800 -
e: ---
- - - Layer Temperature
.
Senkin NO
-: Simulation NO
8000
7000
--
- 6000
-
~.--
-
5000
0
1600 -
...
-
800 -
.....
.,,
-
- 3000 0
0
Eu....
.
.
-
1000 %
E..
0
-
' '
' ' -
-0
337 357 377 397
1500 rpm, MAP=0.5bar, PHI=0.914
2800 -
- - - Layer Temperature
.
Senkin - NO
Simulation - NO
- 2000
- 1800
2400
- - ----
2000
-
L
1 600
E
0
1 200
800
mU
""" ""ME "
-
1600 E
CL
C.
: q
-
- 1400 r
0
1200
- 1000
C
-j ca
0
-800 0
0
- 600 z
400
400
0 '
328 348
.
368 388 408
Crank Angle (Spark = 328)
-
428
-200
448
0
1500 rpm, EGR=12%
111

CHAPTER 6
CYCLE BY CYCLE NO VARIATION DURING STEADY STATE
OPERATION
Up to this point, all of the NO data shown was from time averaged steady state operation results. This is very useful for understanding general trends in NO formation, and modeling this process only requires an estimation of average input parameters, such as air flow, equivalence ratio, residual fraction, burn rate, and so on. However, even while operating at a steady state point, it is well established that cycle by cycle variation of the above listed parameters can be substantial [11]. From chapter 5, it was also clear that slight variations to these input parameters can cause large changes in model predictions for engine out NO. Thus, it would be interesting to explore the amount of cycle by cycle variation that occurs while operating at a steady state point, and see if what was learned during chapter 4 could be used to explain cyclic trends in NO concentration.
6.1 Fast NO Meter Notes
As was discussed in Chapter 2, Cambustion has developed a fast NO detector which is capable of measuring NO concentration with a response time of approximately 4 ms. With each exhaust event lasting 27 ms, the device can be used to sample gas from the exhaust port and analyze engine out NO on a cycle by cycle basis. Figure 6.1 re-illustrates the fast NO sampling system as well as the sampling location used to gather the upcoming experimental data. A sample probe with a length of 250 mm and inside diameter of 0.6 mm was placed in the exhaust port of the firing cylinder approximately 10 cm downstream of the valves, just after the port septum, as marked by the X in figure 6. 1b.
By estimating the volume in the port between the tip of the probe and the valves, a calculation of the amount of exhaust gas mass lying ahead of the probe can be made using the ideal gas law. At an assumed exhaust temperature of 1000 K, there is approximately
0.05 g of exhaust ahead of the probe; this will be used in an upcoming illustration. The reader is referenced back to chapter 2 for how the fast NO meter, shown in figure 6.1a,

operates. For our purposes later, it is also important to note that, it takes approximately 32
CA at 1500 rpm for exhaust gas to travel from the tip of the probe up into the reaction chamber for analysis. This value was calculated with the flow modeling software provided
by Cambustion [26], SATFLAP 3, based on the size of sampling probes and system operating pressures.
For the upcoming analysis, 150 cycles of data were processed, containing exhaust port NO readings, cylinder pressure, and fuel air equivalence ratio, at each of the operating points in the four sweeps discussed in the steady state modeling chapter.
6.2 Signal Characteristics During Lean Operation PHI = 0.91
When using a fast response emission analyzer, considerable effort must be made to understand and interpret the signal as a function of operating conditions. Figure 6.2 shows the fast NO output signal along with cylinder pressure for five consecutive cycles while firing the engine under steady state conditions with PHI = 0.914, MAP = 0.5 bar. The steady state average NO concentration for this operating condition was 1694ppm.
Following the figure from left to right, it can be seen that while the first cycle is in the compression and combustion process, the fast NO signal remains steady at around
1500ppm, since there is no motion in the exhaust port. At this time, the detector is sampling the left over exhaust from the previous cycle. Late in the expansion stroke, when the exhaust valves open, fresh exhaust is released into the port pushing the previous cycle along, causing the NO concentration to rise sharply. (The first cycle shown is the fastest burning of the five and has the highest peak pressure, thus the high level of NO is understandable.) The signal then has an initial spike before leveling off at its new closed exhaust valve level. The signal is again steady until the exhaust valves open from the second cycle shown. Now there is a rapid drop in the signal from 2200 ppm back down to
1200ppm, because the second cycle was a very slow bum with a low peak pressure.
The amount of observed variation seen here is quite substantial and illustrates the importance of accurately modeling the burn rate and peak pressure when making predictions of NO. In figure 6.2, the NO level seems to scale well with the observed peak
114

pressure of the cycles. Other researchers have looked at making the correlation of NO with peak pressure for steady state operation [32] and recently for exhaust port readings [10].
Figure 6.2 offers an excellent opportunity to see if the steady state model we have discussed up to this point can predict the large amount of variation seen on a cycle by cycle basis. By processing this data set with the MIT heat release analysis program, the 0-2, 0-
10, and 0-90 percent burn angles can be determined for the five cycles shown. The Wiebe function can again be defined for each cycle and fed to the simulation, while leaving the remainder of the inputs at the same values used for accurately modeling the steady state ensemble results. Figure 6.3 shows modeling predictions of cylinder pressure and cyclic
NO level in comparison to experiment. Simply by changing the bum rate alone, the layered model is able to predict peak cylinder pressure to within 4% and location of peak pressure to within 1 degree. The wide variation in NO concentration is also captured quite well. However, displaying the over 3400 cycles of data collected in this manor would not be practical or informative, a methodology must be developed to tag an EVO period NO concentration level to each cycle. Then this value can be filed with heat release analysis information and interpreted in a broader sense.
6.3 Processing the Fast NO Exhaust Data
The first step in generating a cycle resolved value for NO concentration is understanding the appearance of the signals and how the signal varies with operating conditions. To accomplish this, many cycles from the five sensitivity analysis points (0.3 bar, 0.8 bar, 0.5 bar, PHI = 0.91, and 10% EGR) were analyzed individually. The first thing noted was that the three operating conditions that all have MAP = 0.5 bar had signals with similar characteristics. This is easily explainable since the exhaust flow event is mainly a function of the engine load. Therefore, the analysis will proceed while considering only three conditions: 0.3 bar, 0.8 bar, and 0.5 bar lean.
Figure 6.4 shows one complete representative cycle from each of three load points along with corresponding pressure traces. This figure will be discussed extensively in the
115

next several pages, and several things should be noted. First, the exhaust valve closed period (0-303CA) has a stable level for all three operating conditions, since the exhaust in the port is stagnant and mixed during this time. Secondly, the three traces all take a different amount of time to peak or bottom out denoted by the dashed line. This response time delay can be attributed to two areas of the sampling set-up: mixing in the exhaust port and mixing in the reaction chamber. Since the blowdown event is smaller at 0.5 bar, more time would be available for mixing in the port and the previous cycle NO concentration may have a larger impact on the new level than at 0.8 bar, where the old exhaust gases will be purged quickly during blowdown. Looking at the trace labeled PHI=0.91, one could speculate that at 360' CA new exhaust is entering the port at a concentration level of around 1000 ppm. However, due to mixing effects in the port and reaction chamber, the signal takes over 1600 CA to reach its steady level and no longer be influenced by the 2200 ppm previous cycle exhaust.
One final note from figure 6.4, all three traces have a distinct delay from the time the exhaust valve opens to the point at which the NO signal starts to change from the closed valve level to the new cycle NO concentration. This delay is marked by the arrows and labeled in crank angles after EVO. This signal delay of 65' CA for 0.5/0.8 bar and
180' CA for 0.33 bar has two sources: time for the new exhaust to reach the front of the probe and time for the new gases to flow through the sampling system, 32' CA. Since the sampling system pressures are fixed for all operating conditions, the 32' CA shift will remain constant. However, the time for new exhaust to travel from the cylinder to the tip of the probe requires a more detailed analysis.
6.3.2 Plug Flow Modeling of the Exhaust Event
Figure 6.5 shows the predicted exhaust mass flow rate and integrated total mass exhausted, taken from cycle simulation output generated during the steady state analysis.
As the intake manifold pressure increases, the initial blowdown process becomes larger.
This is expected since the ratio of cylinder pressure to exhaust manifold pressure, at EVO, varies from near 1 at 0.33 bar to approximately 4 at 0.8 bar as was shown in figure 6.4.
The displacement part of the exhaust flow profile remains consistent regardless of load.
116

Because modeling of mixing phenomena in the exhaust port of a firing engine was beyond the scope of this study, the flow through the port will be modeled as a plug flow.
Therefore, when the exhaust valves open, it is assumed that the new exhaust will push the previous cycles exhaust gas away with little mixing. With this assumption, the time it takes the new cycle exhaust to reach the tip of the sampling probe can be calculated. Since the mass ahead of the probe was estimated at 0.05 g, figure 6.5 marks the added delay due to this effect. For 0.5 bar and 0.8 bar the port is filled in about 35 CA. While at 0.33 bar the delay is over 150 CA since mass flow is initially from the exhaust port back into the cylinder at EVO. By adding this to the 32 CA transit time, the total delay seen in figure 6.4
is closely approximated.
6.4 Three Different Methods of Determining a Cycle Resolved NO Value
With the signal appearance now better understood, three different ways of calculating the cycle resolved NO concentration from the data in figure 6.4 can be considered. The simplest method would be to time average the NO level over the exhaust valve closed period, and attribute this concentration level to the previous cycles pressure history. At light loads this may be an acceptable assumption since the mass remaining ahead of the probe will represent a large fraction of the total exhaust. The next method considered was to assume a constant delay, equal to the transit time through the system and additional time to fill the port, as described above. Then the signal can be time averaged over the shifted exhaust valve open period. In figure 6.4, this would be equivalent to averaging the NO concentration for 0.8 bar, or 0.5 bar lean, from 3700 to 6100 CA. The final method would be to assume the exhaust is a plug flow with the same constant delay.
Then the exhaust profiles from figure 6.5 can be used to mass weight the NO trace between
370 and 610 CA. To determine which of these methods would be the most accurate, it is important to consider more than one representative trace, to understand whether the analysis used above can be applied to all the cycles collected.
117

6.4.1 Analysis of Three Methods for MAP = 0.5 bar, PHI = 0.91
Figure 6.6 now shows five consecutive cycles at the 0.5 bar lean operating conditions used in figures 6.2 and 6.3. The cycle used in figure 6.4 is the last cycle in this plot, represented by the square data points. For discussion purposes, the mass flow rate profile from the cycle simulation is also shown along with the highest and lowest pressure traces.
The first thing noted is that the 65' CA delay time is consistent for all of the cycles.
The signal remains steady during the exhaust valve closed period and then begins to change at approximately 370' CA. The fastest burning cycle, 1, has the highest peak pressure and the highest NO concentration level. Cycle number 5 represents the lowest peak pressure and NO level. The response time or time it takes each cycle to reach a maximum or minimum level is similar for each trace. Assuming new exhaust reaches the front of the probe at 370' CA, it takes between 1200 and 180' CA for each trace to reach a maximum or minimum. Using the mass flow curve, both time and mass average NO values can be calculated between 370' and 610' CA. It should be noted that the same cycle simulation mass flow rate profile was used for all the cycles at any one operating condition. (Cyclic variation in the cylinder pressure trace will change the mass flow diagram slightly.
However, investigation showed, this effect was negligible.) Table 6.1 shows a comparison of cycle resolved NO values calculated with the three different approaches.
Table 6.1: Comparison of Different Methods for Calculating a Cycle Resolved NO Value
1500
MAP = 0.5 bar, PHI = 0.914
Cycle # Mass Avg. Time Avg. EVC
1 2236 2230 2129
2
3
4
5
1266
1558
2066
1352
1315
1589
1962
1474
1214
1372
2239
991
For the all the cycles, the mass averaging and time averaging method predict NO levels within 10% of each other. However, for the last three cycles noticeable differences exist between the EVC value and the others, with a maximum of 27 % for cycle number 5.
Cycle five shows a gradual drop from 2200 down to 990 ppm over the first 180' CA.
118

During this period, its unclear how much of the signal's appearance is due to mixing with the previous cycle in the port and reaction chamber and how much is due to NO stratification effects. If it was a stratification effect, it would be expected to be more consistent, such as the first gases through the valve are always higher or lower than the remainder of the cycle. Because it appears to be a mixing effect, the cycle resolved NO level will be biased by the previous cycle when using the mass and time averaging methods.
6.4.2 Analysis of Three Methods for MAP = 0.8 bar, Stoichiometric
Figure 6.7 shows five consecutive cycles for the 0.8 bar operating conditions plotted with the same format. Again, all the cycles begin responding to the fresh exhaust event at approximately the same location of near 360' CA. The fastest burning cycle is again producing the most NO, but the slowest burning cycle does not produce the least NO anymore. The appearance of the traces overall is much different than the 0.5 bar lean condition shown before. The response time or time to reach a maximum or minimum has also shortened.
Four of the traces show an initial dip followed by a substantial rebound back up before the exhaust valve closes, as in cycles 1 and 3. This could be due to a stratification effect in the exhaust or possibly due to flow disturbance in the sampling system due to the vigorous blowdown period. If it is truly due to stratification, the importance of mass weighting the signals now becomes clear. The large mass flow during the blowdown process corresponds well with the initial dips and most be taken into account when determining the cycle resolved NO concentration. Table 6.2 again shows the difference in calculated NO values with the three different approaches. Mass and time averaged values again agree quite well for all the cycles. For the two cycles that showed the largest initial dip, the EVC value is significantly higher than the other two methods.
119

Table 6.2: Comparison of Different Methods for Calculating a Cycle
1500 rpm, MAP = 0.8 bar, PI = 1.0
Cycle # Mass Avg. Time Avg. EVC
1
2
3
4
5
1718
2158
1684
2061
2606
1806
2116
1772
2112
2639
2285
2076
2158
2181
2673
Resolved NO Value
6.4.3 Analysis of Three Methods for MAP = 0.33 bar, Stoichiometric
For completeness, figure 6.8 shows five consecutive traces for the lowest load considered in this study, MAP = 0.33 bar. However this operating condition showed similar characteristics and does not exhibit any features that have not been discussed.
Appendix B of this thesis contains a summary of the bum rate analysis output and the three different calculated NO values for all the operating conditions considered, and the reader should review this information now before continuing. The format shows the first and last cycle details, along with average, standard deviation, and coefficient of variation values for the 150 cycle data sets. The coefficient of variation, COV, is defined according to
Heywood [23] as the standard deviation of a variable over the mean value. A discussion of the four different sweeps will now be given.
6.5 Load Sweep Cycle by Cycle Variation
Figure 6.9 shows the amount of cyclic variation observed over the entire stoichiometric load sweep. The 150 cycles of mass weighted NO data are plotted against the observed peak pressure of each cycle for the five different load points. Several interesting trends should be noted. At low loads, the cycle resolved NO value corresponds almost linearly with peak pressure. It is expected that if the pressure trace is modeled accurately at low loads, the scatter, at any given peak pressure, could be explained by slight variations in residual fraction, fuel/air ratio, or overall crevice effect. However, as the load increases, the amount of scatter in the data appears to be growing and no clear correlation with peak pressure can be observed. This raises the question: Why, with essential the same pressure trace, does engine out NO concentration vary greatly at high load? When the NO cyclic data was plotted against IMEP or representative bum angles, the scatter showed
120

similar trends. Therefore, peak pressure was determined to be the most informative and will continue to be used. Appendix B, table B.1, shows the average heat release information and a summary of the three different ways of calculating the cycle resolved NO value for the five different load points. The information in this table will now be used to examine the spread in the data at high load.
Figure 6.10 shows the effect of using time averaging and exhaust valve closed period NO calculations instead of mass weighting. Even though, individual cycles were shown earlier to be sensitive to the method used, the overall scatter and 150 cycle average value remained almost unchanged. This is seen in the last three columns of table B.1.
Therefore the high load trend is not attributable to how the cycle resolved NO value is being determined. Since mass weighting most accurately represents what is physically happening in the port, all upcoming NO results will be presented with this method.
The scatter at high load expressed in terms of COV of NO about the single point mean is changing very little. However, along any given line of peak pressure, the amount of variation is increasing with load. The COV of IMEP and the 10 90% burn angles, from table B.1, both remain small over the entire sweep. Thus, nothing about the combustion process changes can explain the added scatter. The only variable that shows increasing variation with load is the fuel/air equivalence ratio. Referring back to table 5.1 and 5.2 from chapter 5, high load operation also showed the least amount of sensitivity to residual, heat transfer, crevice volume, and burn rate. Thus, the sensitivity analysis does not explain the added variation. Again, the only variable that showed larger sensitivity at high loads was fuel/air ratio.
One possible explanation for the larger high load scatter is that the exhaust port NO value was calculated from a single point measurement. The larger scatter could simply be due to error introduced by the single measurement location, becoming more significant as mass flow rates past the probe are increasing. The overshoot and recovery effect demonstrated in figure 6.7, would support this explanation. If point measurements were taken from several different locations in the port simultaneously, a better understanding of
NO stratification and mixing would be developed.
121

Another explanation could be that mixture non-uniformity effects are growing with intake pressure. The mass closest to the spark plug which bums early in the cycle could be varying from rich to lean on a cyclic basis. If the first few layers to bum were slightly lean, the NO concentration could be maximized. With the opposite being true, if the first few layers bum under rich conditions. The question that remains is why would the mixture be less uniform at high load conditions than light loads. The first thing that would support this explanation would be the increasing amount of fuel/air ratio variation with load noted in table B.1. One standard deviation at 0.8 bar was shown to exceed one percent of the average value. From the sensitivity analysis, table 5.1, a one percent change in fuel air ratio could cause NO prediction errors of approximately 6%. It could be speculated that early burning layers could be several percent lean or rich and drive the NO concentration swings. The second thing that would support this explanation is that the blowback of residual into the intake during the valve overlap period is decreasing with load. Therefore, at MAP = 0.8 bar, the amount of mixing in the manifold would be less vigorous than the low load conditions. The propane was continually fed into the manifold rather than being injected cyclically, and this would also support the need for blowback mixing.
6.6 Cycle by Cycle NO Variation EGR, Equivalence Ratio, and Spark Sweeps
Figures 6.11 through 6.14 describe the amount of cyclic variation seen while operating the engine at 1500 rpm, 0.5 bar, and varying the fuel/air equivalence ratio and the amount of EGR. Figure 6.12 was used to make the lean side of the sweep clearer and shows only the first two lean points. Tables B.2 through B.4 contain a summary of NO values and combustion variables along with statistics for each sweep. Again, the NO values calculated with the three different approaches showed negligible differences for average NO value and experimental scatter.
6.6.1 EGR Sweep
Looking at the EGR sweep in figure 6.11 first, each level of dilution appears to follow a linear correlation with peak pressure, and the fit gets better as the EGR level rises.
This information is detailed in table 6.3 along with the slope of each linear fit and the 150
122

cycle average NO value. The range of peak pressure variation remains fairly consistent between 17 and 24 bar, and the NO sensitivity to peak pressure appears to decrease with higher dilution levels. However, when looking at the slope of the linear fit, the average NO level must be considered. On a percent basis, the NO sensitivity to peak pressure changes is actually growing with increasing amounts of EGR. From the layered modeling approach applied earlier, NO concentration should roughly scale with peak pressure and corresponding temperature as shown in equation 6.1 below. Therefore, as burned gas gamma increases, the sensitivity to peak pressure should increase slightly. Exhaust gas recirculation will cause slight increases in burned gas gamma based on output from the cycle simulation thermodynamic routines so the data follows the expected trend.
(7
Y-1/y
NO~ T max spark
(Eq.6.1)
The heat release analysis from appendix B showed that combustion variability, in terms of 10-90 percent bum angle and IMIEP, remained similar for all the levels of EGR considered in this study. The increasing COV of NO about the mean value is explained by the increasing sensitivity to peak pressure fluctuations and the slight increase in range of peak pressure observed. It can be assumed that, for each operating condition, all the cycles with same peak pressure would have a similar mean NO formation profile. Then, the spread about the linear fit would have to be explained by variations in the amount of fuel, residual, and mixture non-uniformity from cycle to cycle.
Table 6.3: Linear Correlation Analysis Results for Cyclic NO Concentration and Peak Pressure
Map = 0.5 bar
Operating Condition
Stoichiometric
4% EGR
8% EGR
12% EGR
16% EGR lean - 0.956 lean - 0.914 lean - 0.838 rich - 1.06 rich 1.12 rich - 1.21
Linear Fit Equation Goodness of Fit 150 Cycle Average NO y = mx+b R2 - Value Concentration (ppm)
NO = 95(P.P.) - 388 0.45 1622
NO = 82(P.P.) - 594 0.51 1161
NO = 66(P.P.) - 580
NO = 64(P.P.) 792
NO = 42(P.P.) 527
NO =136(P.P.) - 1050
0.61
0.73
0.74
0.71
805
530
340
1709
NO = 169(P.P.) - 1695 0.8 1741
NO = 179(P.P.) - 2355 0.74 1154
NO = 75(P.P.) - 466
NO = 44(P.P.) 331
NO = 14(P.P.) - 133
0.24
0.24
0.32
1167
644
178
123

6.6.2 Lean Operation
Figure 6.12 illustrates that as the mixture becomes slightly lean, the NO concentration correlates well with the measured peak pressure. Figure 6.13 then shows the remainder of the lean operation data with only the linear fit lines included from the first two points. Appendix B, table B.3 again shows the combustion information for the entire lean sweep.
The first thing noted from these two figures is that a linear fit is considerably better for the first three lean points than under stoichiometric conditions as detailed above in table
6.1. NO sensitivity to changes in peak pressure is shown to be increasing as the mixture becomes leaner. The same trend was demonstrated in the Chapter 5 sensitivity analysis.
This can again be explained by using equation 6.1 and noting that burned gas gamma will increase on both side of stoichiometric. Below PHI of 0.838, the combustion variability begins to grow substantially, and a linear fit is no longer appropriate. Here, the COV of
NO about the mean is over 35%. This should again demonstrate why expecting steady state modeling results to be better than about 15% is unrealistic.
6.6.3 Rich Operation and Spark Sweep
Figure 6.14 and table B.4 give the details of the cyclic variation of NO under rich operating conditions. As the mixture becomes rich, the combustion stability improves slightly and the peak pressure variation is reduced. The linear fit is no longer as good a correlation as it was on the lean side. Since burned gas gamma will increase as the mixture goes rich, sensitivity to bum rate and peak pressure is also expected to increase.
Finally, figure 6.15 shows the cyclic variation of NO with peak pressure over the stoichiometric, 0.5 bar spark sweep. The main thing to note is that the overall curve becomes non-linear as the timing is advanced. This can be explained by the fact that NO formation rates are non-linear with temperature. At retarded timings, the amount of peak pressure variation is large and the data sets appear similar to those discussed above. As the timing is advanced, peak pressure location nears TDC and the variation of peak pressure itself is reduced. On a percent basis, though, the amount of scatter at any given peak pressure value in the spark sweep is remaining fairly constant.
124

6.7 Cycle by Cycle Burn Rate Modeling
Since the amount of NO cyclic variation has been presented, it would be interesting to see how well the steady state model used in Chapter 4 captures the slope of NO variation with peak pressure. Figure 6.3 showed that, for the lean operating condition of PFH=0.914,
MAP = 0.5 bar, the steady state model was capable of following the NO variation by simply matching the bum rate alone for each cycle. No adjustment to the residual fraction, fuel air ratio, or temperature at the time of spark was considered. Now, this same burn rate modeling approach will be applied to fifteen cycles from each of the five sensitivity analysis points considered earlier, (0.5 bar, 0.33 bar, 0.8 bar stoichiometric and 0.5 bar
10% EGR and PHI = 0.914). Rather than simply taking 15 consecutive cycles at random from each operating condition, five cycles around the mean bum rate, the five slowest burning cycles, and the five fastest burning cycles were chosen at each condition. This will test whether the layered model can capture the slope of the NO variation with peak pressure. However, there are just too many unknown variables (residual fraction variation, fuel variation, NO present in the residual variation, temperature at time of spark, and mixture non-uniformity) to make a realistic, meaningful cycle resolved prediction without making too many assumptions. For this reason, the slope of modeling predictions will only be discussed in this section.
6.7.1 Load Sweep CBC Modeling
Figure 6.16 illustrates the 15 cycles chosen at the each load condition. The modeling curve is not new, rather it is from the steady state modeling section shown earlier in Chapter 4. It is displayed here for the purpose of illustrating the room for error when making predictions of average NO concentration at any single operating condition. At
MAP = 0.5 bar and 0.33 bar, the NO concentration level of each peak pressure group showed less than +/-10 % variation. Therefore, if the slope of the data with peak pressure could be modeled, it is reasonable to attribute the variation to slight differences in residual, fuel, or temperature at spark, based on the Chapter 5 sensitivity analysis. However at 0.8 bar, the five slowest burning cycles, with peak pressure of approximately 30 bar, show NO
125

concentrations ranging from 1500 ppm to nearly 3000 ppm. It does not seem reasonable that this large variation is due simply to residual, or fuel. It is clear that either the model used is missing some key physics at high load (such as mixture non-uniformity), or measurement error due to mixing effects is driving the scatter.
Figure 6.17 shows model predictions of peak pressure and NO concentration when only bum rate is fed to the cycle simulation. All the model points are generated using the starting temperature, pressure, EGR amount, and MAP that resulted in the assumed steady state levels of residual fraction and total mass. Thus, the state of the mixture prior to spark was identical for all fifteen cycles at each operating condition. The Wiebe function parameters were then defined based on the experimentally derived burn angles of each cycle, and this was the only thing changed in the input file. For 0.33 bar and 0.5 bar, the layered model was capable of predicting the changes in peak pressure and its location quite well simply by changing bum rate. However, the model predicts higher sensitivity to peak pressure than the experimental data showed based on the slope of the two different data sets. The largest difference between model and experiment was 25% for both the high pressure group at 0.33 bar and the low pressure group at 0.5 bar. From Chapter 5, table
5.2, a fully mixed model was shown to be approximately half as sensitive to burn rate changes as the layered routine. Accounting for mixing between layers, by decreasing the amount of new layers formed, could improve the slope of the NO predictions with peak pressure.
Several things should be noted before discussing the high load condition. First, from the steady state results summarized in table A. 1, the model was already predicting average concentration 12% higher than experiment. Thus, the model is expected to be making higher predictions in figure 6.17. The calibrated model was also under predicting pressure slightly, and this can be seen at all of the peak pressure groups. For the high pressure group, the bum rate of all five cycles was very similar and only one modeling point was needed. The slope of the modeling predictions is close to the experimental trend.
However, the extreme variation of NO concentration in the low pressure group is difficult to explain with anything explored in the sensitivity analysis section.
126

6.7.2 Lean and EGR Points CBC Modeling
Figure 6.18 shows the last two operating conditions considered in the sensitivity analysis section. Again the slope of the model predictions is steeper than the experiment data for both operating conditions. For the lean condition, at high peak pressures, model predictions of NO are with 5% accuracy even though the peak pressure is slightly low.
This was also seen in the first modeling example, figure 6.2. By again returning to figures
6.4 and 6.6 now, some of this slope difference at the lean condition may be explainable.
The PHI=0.91 trace showed exhaust port NO profiles with slow response times, and the previous cycles NO concentration may be affecting the cycle resolved value. The number
5 trace in figure 6.6 and table 6.1 illustrates this point by showing that if the exhaust valve closed level was used, cycle resolve NO would be much lower. Regardless, burn rate appears to explain most of the cycle by cyclic variation observed for peak pressure and NO scatter can be explained with the sensitivity analysis, other than at the high load condition.
6.8 Observations and Recommendations
Three different techniques were considered for obtaining a cycle resolved value of
NO concentration. For all of the individual cycles collected, less than 10% difference was observed between time averaging and mass weighting the fast NO signal during the exhaust open event. This is in agreement with the conclusion reached by Ball and Stone
[10]. Differences between mass weighting and exhaust valve closed period NO concentration were as large as 40% for certain cycles. However, overall the three techniques showed a similar amount of scatter at each operating condition.
A large amount of NO cyclic variation was observed at each operating point. This re-emphasizes the importance of accurately modeling the experimental burn rate. Cyclic
NO concentration correlates almost linearly with peak pressure for low load, EGR, and lean engine conditions. A given delta change in peak pressure resulted in a larger percent difference in NO concentration (larger sensitivity to peak pressure) on both sides of stoichiometric and with EGR. This trend was attributed to increasing burned gas gamma under these conditions. The amount of NO variation at any given line of peak pressure was shown to increase with engine load. The variation at low and mid load could be attributed
127

to slight variations in mixture composition and starting temperature differences according to the sensitivity analysis results. However, the larger amount of scatter at a given high load peak pressure was attributed to increasing variation in fuel air equivalence ratio and possible mixture non-uniformity effects.
The brief burn rate modeling analysis with a layered model showed the cycle simulation to be overly sensitive to bum rate changes as indicated by a steeper slope of NO with peak pressure. Since a fully mixed model was demonstrated to be half as sensitive to burn rate, it was concluded that limiting the number of layers used could improve the slope of predictions. The large amount of high load scatter merits that future modeling efforts should investigate the effects of mixture non-uniformity on NO predictions.
Any future work with the fast NO meter must focus on developing a better understanding of the response time of the instrument and modeling the mixing process in the port and reaction chamber. Sampling probes should be placed in several different location of the exhaust port to ensure error is not introduced by a single point measurement. This could reduce the amount of experimental scatter, by de-coupling the current cycle of exhaust from the previous NO level.
128

Reactic chamber
....
A t_
Ozone in
Sample
IDOptic-fibre
CP chamber CLD Remote
Sampling Head
CP VAC
Gaw mput
MCU
X ~t
-32 CA at 1500rpm
-
0.05g @
2800
25
0
0
--
15
10
800
0
0
'
I
1000
3000
I
5
0
-
=
-
= 0.913
129

2800
-
0 z
.
1600
x
800
0
1
0
A
Model NO and Pressure Prediction
Experiment NO and Pressure
-AA-A
25
I ~~
1000
1
I
I
I ti kmm,\0
3000
--
-
10
5
0
0
-
-
=
-
= 0.913
3500
3000 -
EVO
- 35
MAP = 0.8 bar
.
PHI= 0.91 x MAP = 0.33 bar 30
0
0
LU
-25A c-
E
2500 -
I a a
2000 20
1500
-0
150-
-
1000
-
"xXXXxx : r 10 .S
bb CA Dei1ay
500
(XXXXXX xxx x
1CXXXXXXXXAX
180 CA Delay
-5
0I -0
0 60 120 180 240 300 360 420 480 540 600 660 720
Crank Angle Acquired
-
130

0.007
0.006
3 0.005
0
0.004
0.003
MAP = 0.8 bar
------- MAP = 0.5 bar
A
MAP = 0.33 bar
0.35
0.3
0.25
0.2
0.15 Wj -
2 0.002
0.001
SA
A
'A- 0.1
0.05
-0.001
483 523 563 603
Mass Level to Fill Port"A
643 683 723
-0.05
Crank Angle (EVO - EVC)
6.5:
-
0
@
3000 --
2500
1
NO-1 - - -NO-2 25 x NO-3
A
NO-4
NO-5
A
20.
-
1500 -x
A
15 x x x x x X x x x x x xo 10
w 500 Mass Flow 50
0
0 60 120 180 240 300 360 420 480 540 600 660 720
0
-
=
-
=
131

3000
-
-NO-1 x NO-3
.
NO-5
- -NO-2
A NO-4
40
35
0 z
E_ 2500
CL
2000
01
(L
X
3
L
X X
30
25
20 g
X~ X 15 )
=~ 1500 wU
10
EVO
5 ass Flow -
1000
0 60 120180240300360420480540600660720
0
Crank Angle Acquired
-
=
-
1000r~f~
20V
800kAAAA
0 z
E
0.
600-
0
400-
3
x
NO-1 -
A
NO-2
NO-4 A hAA l NO-5 IN
X x
N
AA' AA& 10
-
12 $
C
8AQ
X w
200-
4
'fss Flow
0
0 60 120 180 240 300 360 420 480 540 600 660 720
0
Crank Angle Acquired
-
=
-
132

3500
3000
0 z
E
2500
0
1500
0X
W1000
500 -
0
10
-
0.5 bar
0.6 ba p
-- a
-
A
A AA
A~ AA
AL
0.8 bar
A
A
A LA AA
" AA
#4tA AA
AA
AL A
~
A
P
UN-
15 i I I I
25 30
I
35
-
-
-
-
(a) Time Averaging
3500
-
3000
-
A
A A
A
2500
AA
0
2000U.
A
A&
1500-
(0 xL 1000
-
500-
0i
10 15 20 25 i
30 35 40
Measured Peak Pressure (bar)
3500 -F
(b) Exhaust Valve
Closed Period Averaging
A
A
A
0-
3000
-
0 z
.W
2500
-
2000Al
A A
0
M~
1500-
Ak
U
1000-
4-
(0
500
A
0
I I
I I
I I
I
10 15 20 25 30 35 40
Measured Peak Pressure (bar)
-
-
-
133

2000
-
1750 -
* 0% EGR x
1500
-
0 z
4-0
0
.
1250 -
1000
-
750
-
4-
500
250 x x
A
4% EGR
A
.
x
8% EGR
12% EGR
* 16% EGR
0-V
10 15 20 25
30
-
-
-
-
(a) PHI = 0.956
(b) PHI = 0.914
2500--
0
E o. 2000 --
0
1000--
4.1
2500
-
E
2000
-
0.
0 z 1500-
0
O 1000-
* U
U
U
500--
0-
I I
| _ I_
10 12.5 15 17.5 20 22.5 25
Measured Peak Pressure (bar)
X
500w
0 1 1 i I
i
10 12.5 15 17.5 20 22.5 25
Measured Peak Pressure (bar)
-
-
= 0.5 bar -
-
134

2500
T
PHI = 0.914
PHI = 0.956
I-
E
0.
0
2000
1500
0
1000
J PHI = 0.838
m a 4 .PHI =0.773
w
500
_ vPHI=0.715
0
10 12.5
15 17.5
20
Measured Peak Pressure (bar)
22.5
-
25
-
-
= 0.5 bar -
-
1750
PHHI = 1.00
1500
0 z
1250
0
1000
750--
II
500 -x
--...
AX3& X PHI = 1.06
PHI = 1.12
250-PHI = 1.21
AA,
0 --
10 15
25
30
-
-
= 0.5 bar -
-
135

3500
0 z
3000
E.
2500
2000
1500
x wL
1000
500 + 4
RET - 5,10 deg
:. 1
MBT
ADV + 5,10,15,20 deg
W
I-- U
0
10 15 20 25
30
-
-
-
3500
0 z
E
0.
3000
2500
-
0M
XI
-c
1500
-
1000
500 -
0
0
------- Steady State Model - Layered A.C.
* MAP = 0.8 bar
* MAP = 0.5 bar
*
MAP = 0.33 bar
E
*R
0
S
S
-Is
S
U
U
U
U a:
~i.
0.'..
-0
A&
10
i
30
E i
-
-
136

3500
3000
-
-
500
0
5
0
0
0
2500
0.
0 z
V-
0 a.
2000
1500 c x wU
1000
A
A
A
A
It
A 0
0
0
0
0
I r
A
A
A
OFFu
10 15 20 25 i
30
Peak Pressure (bar)
0
35 40 45
-
-
-
(a) PHI = 0.914
0 z
CLu x w
3000
-
E
0.
2500
-
1500
0
0
0 N o *
1250
0
2000 -
1500 -
1000 -
U
011
0.
z
0
1000
750
CL
500
U
"
U.
MJ 13
500 250
0I
15 17.5 20 22.5 25
Peak Pressure (bar)
0 i i i i
15 17.5 20 22.5 25 i
Peak Pressure (bar)
-
-
137

138

BIBLIOGRAPHY pi www.epa.gov
[2] Zeldovich, J., "The Oxidation of Nitrogen in Combustion and Explosions," Acat
Physiochem, URSS, Vol2l, pp. 577, 1946.
[3] Lavoie,G. A., Heywood, J.B., and Keck, J.C. "Experimental and Theoretical
Investigation of Nitric Oxide Formation in Internal Combustion Engines,"
Combustion Science and Technology, Vol. 1, pp. 313-326. 1970.
[4] Blumberg, P. and Kummer, J.T., "Prediction of NO Formation in Spark-Ignited
Engines- An Analysis of Methods and Control," Combustion Science and
Technology, Vol. 4, pp.73-75, 1971.
[5] Poulos S.G. and Heywood, J.B., "The Effect of Chamber Geometry on Spark-
Ignition Combustion," SAE Paper 830334, 1983.
[6] McGrath, P.J. "Assessment and Improvement of Current Computational Spark
Ignition Engine NOx Formation Models for Auto Industry Development and
Design Use," M.E. Thesis, Massachusetts Institute of Technology, 1982.
[7] Miller, R., Davis, G., Lavoie, G., Newman, C., and Gardner, T., "A Super Extended
Zeldovich Mechanism for NOx Modeling and Engine Calibration" SAE Paper
980781, 1998
[8] Miller, R., Russ, S., Weaver, C., Davis, G., Lavoie, G., Newman, C., and Kaiser,
E., "Comparison of Analytically and Experimentally Obtained Residual Fractions and NOx Emissions in Spark Ignited Engines," SAE Paper 982562, 1998
[9] Raine, R.R., Stone, C.R., "Modeling of Nitric Oxide Formation in Spark Ignition
Engines with a Multizone Burned Gas," Combustion and Flame, Vol. 102, pp. 241-
255, 1995.
[10] Ball J.K., Stone, C.R., Collings, N., "Cycle-by-cycle modelling of NO formation and comparison with experimental data," Proc Instn Mech Engrs, Vol. 213 Part D, pp. 175-189, 1999.
[11] Hinze P.C., "Cycle to Cycle Combustion Variations in a Spark Ignition Engine
Operating at Idle," PhD. Thesis, Massachusetts Institute of Technology, 1997
[12] Kenney T., Fader, H., Fenderson, A., Gardner, T., Keeble, B., Kwapis, J., Meyer,
D., Morris, G., Rehagan, L., Shearer, S., Stein, R., Tobis, B., Tuggle, G., Wagner,
T., Wernette, B., "Acquisition and Analysis of Cylinder Pressure Data
Recommended Procedures" Ford Manual, 1992.
139

[13] Lancaster, D., Krieger, R.B., Lienesch, J.H. "Measurements and Analysis of Engine
Pressure Data," SAE Paper 750026, 1975.
[14] Stein, R.A., Mencik, D.Z., Warren, C.C., "Effect of Thermal Strain on
Measurement of Cylinder Pressure," SAE Paper 870455, 1975.
[15] Gatowski, J.A., Balles, E.N., Chun, K.M., Nelson, F.E., Ekchian, J.A., Heywood,
J.B., "Heat Release Analysis of Engine Pressure Data" SAE Paper 841359, 1984.
[16] Chun, K.M., and Heywood J.B., "Estimating Heat-Release and Mass of Mixture
Burned from Spark Ignition Engine Pressure Data," Combustion Science and
Technology, Vol. 54, pp. 133-143, 1987.
[17] Cheung, H.M. and Heywood J.B., "Evaluation of a One-Zone Burn Rate Analysis
Procedure Using Production SI Engine Pressure Data," SAE Paper 932749, 1993.
[18] Quader, A.A., "Why Intake Charge Dilution Decreases Nitric Oxide Emission from
Spark Ignition Engines," SAE Paper 710009, 1971.
[19] Aiman, W.R., "Engine Speed and Load Effects on Charge Dilution and Nitric
Oxide Emission," SAE Paper 720256, 1972.
[20] Galliot, F., Cheng, W., Cheng, C., Sztenderowicz, M., Heywood, J., Collings, N.,
"In-cylinder Measurements of Residual Gas Concentration in a Spark Ignition
Engine," SAE Paper 900485, 1990.
[21] Fox, J.W., Cheng, W.K., and Heywood, J.B., "A Model for Predicting Residual
Gas Fraction in Spark Ignition Engines," SAE Paper 931025, 1993.
[22] Cheng W.K., Galliot, F., Collings, N., "On the Time Delay in Continuous Incylinder Sampling from IC Engines," SAE Paper 890579, 1989.
[23] Heywood, J.B., Internal Combustion Engine Fundamentals, McGraw-Hill Book
Co., New York, 1988.
[24] Reavell, K., Collings, N., Peckham, M., Hands, T., "Simultaneous Fast Response
NO and HC Measurements from a Spark Ignition Engine," SAE Paper 971610,
1997.
[25] Peckham, M., Hands, T., Burrell, J., Collings, N., Schurov, S., " Real Time Incylinder and Exhaust NO Measurements in a Production SI Engine," SAE Paper
980400, 1998.
140

[26] "fNOx400 High Frequency Response NO Detector User Manual," Cambustion
Ltd., 1992.
[27] Martin, M.K., and Heywood, J.B., "Approximate Relationships for the
Thermodynamic Properties of Hydrocarbon-Air Combustion Products,"
Combustion Science and Technology, Vol.15, pp. 1-9, 1976.
[28] Miller, J.A. and Bowman, C.T., Prog. Ener. Combust. Sci., Vol. 15, pp. 287-338,
1989.
[29] Bowman C.T., "Kinetics of Pollutant Formation and Destruction in Combustion,"
Prog. Ener. Combust. Sci., Vol.1, pp. 33-45, 1975.
[30] Lutz., A., Kee, R., Miller, J., "SENKIN: A Fortran Program for Predicting
Homogeneous Gas Phase Chemical Kinetics with Sensitivity Analysis," SAND87-
8248 Sandia National Laboratories Unlimited Release, 1988.
[31] Dagaut, P., Cathonnet, M., and Boettner, J., "Kinetic Modeling of Propane
Oxidation and Pyrolysis," International Journal of Chemical Kinetics, Vol. 24, pp.
813-837, 1992.
[32] Kalghatgi, G.T., "Effects of Combustion Chamber Deposits, Compression Ratio and Combustion Chamber Design on Power and Emissions in Spark Ignition
Engines," SAE Paper 972886, 1997.
141

142

-
File
Name
3bar12
4bar12
5barl2
6bar12
8bar12
MAP Spark Gross Theta Max Average Air in Exhaust NO Coolant Residual Fast NO Fast NO bar
0
BTC IMEP Pmax Pressure phi g/cyc Temp C volts Temp Exper. ppm corrected
0.33 30 2.48 12.90 13.79 1.003 0.1176 498 1.09 75 19 685.5346 733.522
0. 4
0.5
0.6
0.8
24
23
21
18
3.08
3.91
4.78
6.58
14.67
13.45
13.93
14.10
EXPERIMENTAL RESULTS
16.25
21.09
25.46
34.82
1.000 0.1464
1.003 0.189
1.002 0.2256
0.999 0.3064
510
548
566
586
1.59
2.33
2.67
3.45
75
75
75
75
16
14
12
10
1000 1070
1477.987 1581.447
1679.245 1796.792
2169.811 2321.698
MODEL RESULTS
File UNCal UNCal UNCal Calibrat Calibrat Calibrat Uncal NO Uncal NO CAL NO CAL NO FINAL NO FINAL NO
Name Residual Air IMEP IMEP Pmax opmax MIX LAYER MIX LAYER MIX LAYER
Sbar12
4bar12
14.50 0.1275
11.98 0.1631
2.96
3.8
5bar12 9.80 0.2108 4.92
6bar12 8.62 0.2662 6.23
8barl2 5.97 0.3724 8.76
2.4
2.99
3.86
4.62
6.28
13.23
16.01
21.03
24.84
33.83
13
14
14
14
13
2979
3545
3871
3996
4035
2303
2854
3287
3554
3807
1283
1909
2559
3094
3620
1055
1561
2025
2465
2997
776
1228
1858
2466
3259
706
1086
1578
2018
2649
MODEL INPUTS
File Wiebe Constants Calibrat Calibrat
Name CONSPB EXSPB DTBRN EGR
3bar12 9 3.6 61 2.55
MAP
0.3195
4bar12 bar12
6bar12
8barl2
9
9.2
9
9
3.6
3.6
3.6
3.6
53
50
47
42
2.05 0.375
2.76 0.4635
2.2
2.4
0.535
0.7
z

-
File
Name rich56 rich34 richl2
MAP bar
0.5
0.5
0.5 st12 lean12
0.5
0.5 lean34 0.5 lean56 lean78
0.5
0.5 lean9l0 0.5
Spark
*BTC
22
21
22
23
23
25
29
33
37
Gross Theta (atc Max
IMEP
Average Air in Exhaust NO Coolant Residual Fast NO Fast NO
Pmax Pressure PHI g/cyc Temp C volts Temp Exper. ppm corrected
3.93
3.97
3.99
3.94
3.79
3.73
12.39
13.01
13.16
13.56
13.95
13.16
22.33
22.06
21.95
21.11
20.26
20.35
1.252
1.121
1.058
1.005
0.956
0.914
0.189
0.189
0.189
0.189
0.189
501
530
544
550
544
529
0.26
0.94
1.68
2.35
2.52
2.53
77
77
77
77
77
15
14.5
14
14
14
14
113.522
174&.9686o
591.195 632.5786
1056.604 1130.566
1477.987 1581.447
1584.906 1687.925
1591.195 1694.623
3.52
3.36
3.11
12.80
12.95
13.30
EXPERIMENTAL RESULTS
19.39
18.26
16.12
0.838
0.773
0.715
0.189
0.189
0.189
498
475
460
1.62
0.75
0.21
77
77
77
13.5 1018.868 1080
13.5 471.6981 497.6415
13 132.0755 139.3396
MODEL RESULTS
File
Name Residual rich56 rich34 richl2 st12 lean12 lean34
UNCal
10.00
9.87
9.75
9.80
9.83
10.04
UNCal
Air
0.2129
0.2137
0.2143
0.2108
0.2147
0.2150
UNCal Calibrat Calibrat Calibrat Uncal NO Uncal NO CAL
IMEP IMEP Pmax
NO CAL NO FINAL NO FINAL NO opmax MIX LAYER MIX LAYER MIX LAYER
4.83 3.7 22.1 12 391 348 122 173 69 146
4.93 3.75 21.62 13 1302 1129 776 683 561 588
4.98 3.83 21.69 13 2399 2100 1662 1346 1267 1126
4.92 3.86 21.03 14 3871 3287 2559 2025 1858 1578
4.85 3.75 20.27 14 5088 4135 3069 2393 2077 1758
4.66 3.59 20.11 13 5622 4453 3044 2460 1939 1731 lean56 10.45 0.2154 4.35 3.31 18.88 13 4524 3592 2106 1838 1197 1193 lean78 10.73 0.2155 4.06 3.16 17.42 14 2228 1977 881 852 462 512
Iean9l0 10.92 0.2163 3.78 2.98 15.6 16 730 713 283 281 145 156
MODEL INPUTS
File Wiebe Constants
Name CONSPB EXSPB DTBRN rich56 9 3.4 47 rich34 9 3.4 47 rich12 9 3.5 48 st12 9.2 3.6 50 lean12 9.5 3.6 52 lean34 9.5 3.6 54 lean56 9 3.6 60 lean78 12 lean9l0 16
3.7
5
73
87
Calibrat Calibrat
EGR MAP
3.4 0.467
2.95 0.468
2.55
2.76
2.56
2.27
0.465
0.4635
0.463
0.461
1.35 0.459
1
0.2
0.457
0.45

-
I-lie
Name ret12 ret34 mbt12 adv12 ad\64 ad\66 adv78 bar
0.5
*BTC
13
0.5 -18
0.5
0.5
53
28
0.5
0.5
2:3
3
-r
0.5 24%
3.78
3.88
3.91
3.85
3.72
3.69
3.57
24.50
18.50
13.97
8.93
5.56
3.61
1.58
15.04
18.25
20.73
23.43
24.83
25.68
26.45
1.003
1.000
1.003
1.003
1.002
1.002
1.002
0.189
0.189
0.189
0.189
0.189
0.189
0.189
576
563
549
541
535
531
526
1.29
1.87
2.31
2.91
3.29
3.66
3.99
77
77
77
78
78
78
78
13.5
811.32081868.1132
13.5 1176.101 1258.428
14 1477.987 1581.447
14 1830.189 1958.302
14 2069.182 2214.025
14.5 2301.887 2463.019
14.5 2509.434 2685.094
File UN~aI UN~aI
Name Residual ret12 ret34
9.80 mbtl2 adv12 adB4 ad%66 adv78
Air
0.2108
'
MODEL RESULTS
UNCaI Calibrat Calibrat Calibrat Uncal NO Uncal NO CAL NO CAL NO FINAL NO FINAL NO
I
IMEP
492
IMEP
3.91
3.93
3.86
Pmax
16.33
18.45
21.02
Opmax
MIX LAYER MIX LAYER MIX LAYER
23 1866 1377 1233 918
18 2260 1712 1529 1219
14 3842 3282 2602 2059 1841 1579
3.75 22.66 9 2895 2358 2264 1938
3.61
3.47
3.27
23.92
24.61
25.43
7
5 1
3 1
File Wiebe Constants
Name CONSPB EXSPB DTBRN ret12 9 3.6 51 ret34 mbt12 adv12 ad84 ad66 adv78
9
9.2
9
9.6
9.7
9.8
3.6
3.6
3.6
3.7
3.8
4
51
50
51
53
56
57
MODEL INPUTS
Calibrat Calibrat
EGR MAP
2.86 0.4635
0.4635
2.7
2.76
2.5
0.4635
0.4635
0.4635
2.4
2.65
2.57
0.4635
0.4635
1
_
3219
3364
3597
2638
2795
3082
2612
2831
3161
2247
2454
2781

-
File
Name base12
512
1012
MAP bar
0.5
0.517
0.533
0.55
'
Spark
*BTC
23
27
29
32
*
Gross
'
IMEP P max
3.89
3.94
3.97
3.99
4.03
13.80
12.88
13.08
13.47
Max I Average
I
Pressure phi
20.67 1.003
21.21 1.004
20.88 1.005
20.46
20.73
1.001
1.004
14 in txnaust
9/cyc Temp C
0.189 549
0.189 538
I
0.189 532
0.189
0.189
520
517
NU ppm
2.28
1.71
1.19
0.78
0.49
Uooiani i
Temp I
77
77
77
77 r. ppm j 1477.987
1075.472
748.4277
S490.566
308.1761
Fast NO corrected
1581.447
1150.755
800.8176
[524.9057
329.7484
1512
2012 0.568 35 12.95
77
MODEL RESULTS
File UNCal UNCal UNCal Calibrat Calibrat Calibrat Uncal NO Uncal NO CAL
Name Residual Air IMEP IMEP Pmax
NO CAL NO FINAL NO FINAL NO opmax MIX LAYER MIX LAYER MIX LAYER base12 9.80 0.2108 492 3.86 21.02 14 3842 3282 2602 2059 1841 1579
512 1
3.89 21.84 13 1821 1527 1160 1106
1012 3.93 21.88 13 1123 1004 681 700
1512
3.96 21.2 14 621 592 349 384
2012 1 4.01 21.34 13 356 373 194 231
MODEL INPUTS
File Wiebe Constants Calibrat Calibrat
Name CONSPB EXSPB DTBRN EGR MAP base12 9.2 3.6 50 2.76 0.4635
512 9.8 4 54 6.9 0.482
1012 9.8 4 58 10.6 0.5
1512 9.8 4 65 14.2 0.515
2012 9.8 4 70 17.51 0.532

-
Cycle # P.P. CA P.P.
1 14.77 190
150 14.89
Average 13.80
1 -STD
C.O.V.
0.94
6.82
192
194
2
1
MAP = 0.33 bar
IMEP Lambda 0-2 0-20 0-50 0-90 a10-90 xbmax NO
EVC NO time NO mass
2.40 1.000 16.6 22.6 30.9 39.2 16.6 0.91 810 801 833
2.48 0.990 16.3 22.9 32.4 39.9 17.0 0.94 807 813 812
2.48 0.996 16.9 23.7 34.5 43.7 20.0 0.94 735 731 737
0.03 0.007 1.6 1.9 2.5 3.2 1.9 0.02 107 100 116
1.38 0.687 9.7 7.9 7.1 7.3 9.6 1.67 14 14 16
Cycle # P.P. CA P.P.
1 15.65 197
150 16.71 195
Average 16.25 196
1 - STD 0.94
C.O.V. 5.79
2
1
MAP = 0.4 bar
IMEP Lambda 0-2 0-20 0-50 0-90 al0-90 xbmax NO - EVC NO - time NO - mass
3.09 0.990 14.4 20.5 31.6 40.5 20.0 0.94 976 988 959
3.05 1.000
3.08 1.000
0.04 0.007
12.9
14.2
1.3
19.5
20.2
1.5
29.4
30.4
1.9
37.9
39.2
2.5
18.4
18.9
1.8
0.94
0.94
0.02
1082
1067
135
1086
1060
118
1069
1059
129
1.28 0.709 9.4 7.6 6.4 6.4 9.6 1.91 13 11 12
Cycle # P.P. CA P.P.
1 20.64 195
150 20.64 196
Average 21.23 194
1 - STD 1.37
C.O.V. 6.47
2
1
MAP = 0.5 bar
IMEP Lambda 0-2 0-20 0-50 0-90 al 0-90 xbmax NO - EVC NO - time NO - mass
3.91 1.000 12.3 17.9 27.7 37.7 19.9 0.91 1728 1806 1850
3.92 1.010
1.14 0.899
13.6
3.94 0.994 13.2
0.04 0.009 1.5
11.3
19.8
19.0
1.9
10.0
30.3
28.3
2.4
8.5
37.7
37.1
3.0
8.2
17.9
18.2
2.0
11.1
0.92
0.93
0.01
1.55
1595
1568
218
14
1618
1600
191
12
1669
1622
194
12
Cycle # P.P. CA P.P.
1 23.79 196
150 27.00 193
Average 25.61 195
1 -STD
C.O.V.
1.40
5.46
2
1
MAP = 0.6 bar
IMEP Lambda 0-2 0-20 0-50 0-90 a10-90 xbmax NO - EVC NO - time NO - mass
4.74 1.020 12.0 18.4 28.1 37.7 19.3 0.94 1458 1467 1467
4.80
4.78
1.000
0.999
0.06 0.011
1.27 1.070
10.9
12.2
1.3
10.3
16.3
17.9
1.7
9.5
24.4 32.4
26.7 34.9
2.0
7.7
2.3
6.7
16.1
17.0
1.6
9.3
0.95
0.95
0.01
1.55
1662
1758
282
16
1979
1811
255
14
2011
1821
256
14
Cycle # P.P. CA P.P.
1 37.13 192
150 33.51 197
Average 35.12 195
1 - STD 2.05
C.O.V. 5.84
2
1
MAP
= 0.8 bar
IMEP Lambda 0-2 0-20 0-50 0-90 a10-90 xbmax NO - EVC NO - time NO - mass
6.54 1.000 10.0
6.40 0.990 11.8
6.60 1.001 10.9
0.10 0.013 1.2
1.50 1.279 11.1
14.7
18.1
16.7
1.7
10.3
21.6 29.6
26.4 33.9
25.2
2.2
8.7
32.7
2.4
7.5
14.9
15.8
16.0
1.4
8.9
1.00
0.99
2149
1897
1.01 2390
0.02 1_354
1.72 15
1951
1717
2328
333
14
1886
1668
2329
347
15

00
-
Cycle # P.P. CA P.P. IMEP Lambda 0-2
1 20.64 195 3.91 1.000 12.3
150 20.64
Average 21.23
STD 1.37
196
194
2
3.92
3.94
0.04
1.010
0.994
0.009
13.6
13.2
1.5
6.47 1 1.14 0.899 11.3
0-20
17.9
0% EGR
0-50
27.7
19.8
19.0
1.9
10.0
30.3
28.3
2.4
8.5
0-90 a10-90 xbmax NO
-
EVC NO
time NO
mass
37.7 19.9 0.91 1728 1806 1850
37.7
37.1
17.9
18.2
0.92
0.93
1595
1568
1618
1600
1669
1622
3.0 2.0 0.01 218 191 194
8.2 11.1 1.55 14 12 12
Cycle # P.P. CA P.P. IMEP Lambda 0-2
1 22.40 192 3.91 1.000 15.4
150 18.23
Average 21.16
1 - STD 1.45
6.85
196
194
2
1
3.86
3.91
0.05
1.28
1.000
0.996
0.009
16.2
15.4
1.7
0.943 11.1
Cycle #
1
150
P.P. CA P.P.
21.41 193
21.85
Average 20.85
192
194
1 - STD 1.61
7.70
3
1
IMEP Lambda 0-2
3.97 0.990 18.2
3.97
3.97
0.05
1.30
1.000
0.994
0.008
17.2
16.9
2.1
0.774 12.5
Cycle # P.P. CA P.P. IMEP Lambda 0-2
1 20.66 196 4.01 1.000 19.0
150 20.90
Average 20.50
195
194
1 - STD 1.64 2
4.01
3.98
0.06
1.010
0.998
0.008
19.9
19.0
1.9
8.01 1 1.55 0.793 10.0
Cycle # P.P. CA P.P. IMEP Lambda 0-2
1 20.31 192 3.91 1.000 18.3
150 17.86
Average 20.76
1 - STD 1.79
198
194
3
3.98 0.990
4.03
0.06
0.995
0.007
20.4
21.2
2.4
8.63 1 1.58 0.725 11.2
0-20
21.5
22.6
21.7
2.0
4% EGR
0-50
30.4
35.5
31.7
2.6
9.4 8.3
0-90 a10-90 xbmax NO EVC NO time NO mass
39.3 17.7 0.92 1177 1216 1195
49.0 26.4 0.93 961 1051 1061
42.1
3.5
20.4
2.5
0.93
0.01
1145
189
1170
169
1161
169
8.3 12.4 1.46 16 14 15
0-20
8% EGR
0-50
23.9
23.0
23.5
2.5
33.6
32.4
34.2
3.0
10.81 8.8
0-90 a10-90 xbmax NO - EVC NO - time NO - mass
44.0
44.2
20.0
21.2
0.92
0.94
988
603
932
618
924
611
46.3
3.9
22.9
2.8
0.94
0.01
790
157
814
140
805
137
8.3 12.0 1.47 20 17 17
0-20
27.0
27.4
26.3
2.3
12% EGR
0-50
39.0
38.4
37.8
3.0
8.9 7.8
0-90 a10-90 xbmax NO EVC NO time NO mass
49.8 22.8 0.93 522 509 501
50.4 23.0 0.93 575 557 551
51.8 25.6 0.93 519 538 530
4.4 3.3 0.02 131 128 124
8.5 12.7 1.67 25 24 23
0-20
26.1
16% EGR
0-50
39.3
45.6 28.6
28.6
2.7
40.6
3.3
9.3 8.2
0-90 a10-90 xbmax NO
-
EVC NO
time NO
mass
58.3 32.2 0.90 239 285 284
60.8 32.2 0.93 240 267 266
55.7
4.6
27.1
3.4
0.93
0.02
329
89
343
89
340
87
8.2 12.4 1.65 27 26 26

-
Cycle # P.P. CA P.P. IMEP Lambda 0-2
1 21.47 193 3.82 1.040 12.0
150 17.47 199 3.74 1.050 15.5
Average 20.19 195 3.78 1.048 13.5
1 - STD 1.40 2 0.05 0.009 1.6
6.92 1 1.44 0.903 12.1
Cycle # P.P. CA P.P. IMEP Lambda 0-2
1 19.44 193 3.53 1.200 16.0
150 21.08
Average 19.62
1 -STD 1.32
1 6.75
190
193
2
1
3.44
3.52
1.190
1.195
14.4
16.6
0.05 0.008
1.43 0.676
1.8
11.1
PHI = 0.956
0-20
17.5
0-50
26.3
22.3
19.3
2.0
10.2
33.4
29.0
2.5
8.5
0-90 a10-90 xbmax NO EVC NO time NO mass
35.8 18.3 0.93 1938 1934 1973
44.9 22.6 0.93 1447 1420 1427
38.4
3.3
19.1
2.2
0.92
0.02
1679
256
1697
221
1709
227
8.5 11.7 1.64 15 13 13
PHI = 0.914
Cycle # P.P. CA P.P. IMEP Lambda 0-2
1 22.16 192 3.75 1.100 12.2
150 20.03
Average 20.35
194
194
3.70
3.73
1.100
1.095
14.5
14.5
1 - STD 1.35
6.62
2
1
0.06
1.56
0.009 1.7
0-20
17.3
20.6
20.4
2.1
0.831 12.0 10.3
0-50
26.6
30.4
30.2
2.4
8.1
0-90 a10-90 xbmax NO - EVC NO - time NO - mass
36.3 18.9 0.95 1919 1909 1932
42.2 21.6 0.94 1390 1559 1492
40.2 19.8 0.94 1697 1731 1741
3.1
7.7
2.2
11.3
0.02
1.68
295
17
246
14
254
15
PHI = 0.838
0-20 0-50
22.7
20.5
23.1
2.1
9.1
33.2
29.6
33.6
2.5
7.5
0-90 al0-90 xbmax NO - EVC NO - time NO - mass
46.7 24.0 0.96 979 1226 1180
42.2
44.8
21.8
21.8
0.94
0.95
1692
1117
1728
1146
1754
1154
3.3 2.4 0.01 314 260 276
7.3 11.2 1.56 28 23 24
Cycle # P.P. CA P.P. IMEP Lambda 0-2
1 18.85 193 3.39 1.290 19.5
150 21.45 190 3.41 1.300 18.0
Average 18.25 194 3.36 1.294 19.8
1- STD 1.71 2 0.06 0.009 2.3
1 9.37 1 1.83 0.715 11.4
PHI = 0.773
0-20
26.6
23.9
26.9
0-50
37.8
33.2
38.5
2.5
9.3
3.3
8.6
0-90 al0-90 xbmax NO - EVC NO - time NO - mass
51.1 24.4 0.98 541 523 543
44.7 20.9 1.02 830 763 811
51.7 24.8 0.97 479 497 504
5.0 3.4 0.02 229 175 191
9.6 13.9 1.93 48 35 38
PHI = 0.715
Cycle # P.P. CA P.P. IMEP Lambda 0-2
1 16.39
150 20.30
Average 16.03
1 - STD 2.10
13.09
195
190
195
3
1
3.22
3.19
3.15
0.10
3.06
0-20 0-50
1.390 25.3 33.0 44.9
1.390 23.2
1.398 24.0
29.2
31.8
37.5
45.3
0.012 2.9 3.2 4.6
0.877 12.0 10.1 10.1
0-90 al0-90 xbmax NO - EVC NO - time NO - mass
60.6 27.5 0.99 149 125 140
48.7
62.2
19.4
30.4
0.98
0.98
317
131
270
143
337
149
7.8
12.6
5.5
18.0
0.02
2.18
108
82
85
60
97
65

-
Cycle # P.P. CA P.P. IMEP Lambda 0-2
1 20.64 195 3.91 1.000 12.3
150 20.64
Average 21.23
196
194
3.92
3.94
1.010
0.994
13.6
13.2
1 - STD 1.37
6.47
2
1
0.04 0.009
1.14
1.5
0-20
17.9
19.8
19.0
1.9
0.899 11.3 10.0
PHI = 1.0
0-50
27.7
30.3
28.3
2.4
8.5
0-90 a10-90 xbmax NO - EVC NO - time NO - mass
37.7 19.9 0.91 1728 1806 1850
37.7 17.9 0.92 1595 1618 1669
37.1
3.0
8.2
18.2
2.0
11.1
0.93
0.01
1.55
1568
218
14
1600
191
12
1622
194
12
Cycle # P.P. CA P.P. IMEP Lambda 0-2
1 22.58 191 3.95 0.930 9.7
150 21.20
Average 21.92
1 - STD 1.10
196
194
2
4.04
3.97
0.03
0.940
0.945
0.007
14.7
12.2
1.4
5.02 1 0.75 0.782 11.7
Cycle # P.P. CA P.P.
1 22.97 191
150 19.04 200
Average 22.05
1 STD 1.13
5.12
194
2
1
IMEP Lambda 0-2
3.97 0.890 10.9
3.96 0.890 13.6
3.97
0.03
0.892
0.007
11.3
1.3
0.71 0.754 11.4
PHI = 1.058
0-20
14.3
20.7
17.7
1.7
9.8
0-50
23.5
29.7
26.7
2.0
7.3
0-90 a10-90 xbmax NO - EVC NO - time NO - mass
33.7 19.4 0.92 1128 1128 1145
37.4
35.0
16.7
17.3
0.96
0.93
1003
1128
964
1146
978
1167
2.3
6.6
1.6
9.5
0.01
1.12
183
16
163
14
167
14
PHI
=
1.121
0-20
15.7
20.3
16.8
1.7
9.9
0-50
23.4
30.6
25.8
1.9
7.6
0-90 a10-90 xbmax NO - EVC NO - time NO - mass
32.0 16.4 0.94 729 667 684
39.5
33.7
2.3
19.2
16.9
1.5
0.93
0.93
0.01
395
627
122
428
630
101
430
644
102
6.7 8.9 1.17 19 16 16
Cycle # P.P. CA P.P.
1 21.28 194
150 23.98
Average 22.32
1 - STD 1.07
4.80
190
193
2
1
IMEP
3.92
3.90
Lambda
0.800
0.800
0-2
13.1
8.9
PHI = 1.252
0-20 0-50 0-90 al0-90 xbmax NO - EVC NO - time NO - mass
19.8 28.0 35.6 15.8 0.93 174 164 171
14.3
3.93 0.799 11.3 17.0
0.04 0.005 1.4 1.8
0.96 0.647 12.5 10.5
22.7
25.9
1.9
7.3
30.0
33.7
2.1
6.2
15.8
16.7
1.4
8.5
0.94
0.94
0.01
1.26
177
171
31
18
193
173
25
15
196
178
27
15