Politecnico di Milano

School of Industrial and Information Engineering

Master of Science in Energy Engineering
Energy Department

CFD Study on Spark Assisted

Compression Ignition Combustion for a
Natural Gas Powered Heavy Duty
Engine

Supervisor: Prof. Tommaso LUCCHINI

Master thesis by:

Federico PESSAH Matr. 883247

Academic Year 2018-2019

Contents

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . II

1 Internal combustion engines 1

1.1 Heavy duty spark ignition engines . . . . . . . . . . . . . . . . . 9
1.2 Flame propagation combustion . . . . . . . . . . . . . . . . . . . 13
1.3 Knock and spontaneous combustion . . . . . . . . . . . . . . . . 15
1.4 Fuel and pollution: Natural gas . . . . . . . . . . . . . . . . . . . 24
1.5 Numerical simulations . . . . . . . . . . . . . . . . . . . . . . . . 29

2 Numerical solvers 33
2.1 Solver description . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2 Combustion modelling . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3 Spark ignition combustion . . . . . . . . . . . . . . . . . . . . . . 48
2.4 Spontaneous combustion . . . . . . . . . . . . . . . . . . . . . . . 50

3 Spark ignition combustion 55

3.1 Spark advance sweep for engine optimization . . . . . . . . . . . 68
3.2 Constant torque sweep . . . . . . . . . . . . . . . . . . . . . . . . 74
3.3 Constant speed sweep . . . . . . . . . . . . . . . . . . . . . . . . 79
3.4 Engine map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

4 Knock prediction 87
4.1 Knock experimental validation . . . . . . . . . . . . . . . . . . . 93
4.2 Natural gas knock . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.3 Boundary conditions and temperature wall function analysis . . . 108

5 SACI engine development 117

5.1 Spontaneous ignition effects on engine performances . . . . . . . 119
5.2 Wall temperature, spark timing and load sweeps . . . . . . . . . 125
5.3 Piston geometry influence . . . . . . . . . . . . . . . . . . . . . . 137
5.4 Lean SACI combustion . . . . . . . . . . . . . . . . . . . . . . . . 149
5.5 SACI engine map . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
5.6 Pollutants analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 164

I
II
List of Figures

1.1 Otto thermodynamic cycle . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Realistic pressure-volume curve . . . . . . . . . . . . . . . . . . . 2
1.3 Motored and fired pressure curves . . . . . . . . . . . . . . . . . 3
1.4 Combustion lasting . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Otto cycle efficiency plot with compression ratio . . . . . . . . . 4
1.6 Diesel thermodynamic cycle . . . . . . . . . . . . . . . . . . . . . 5
1.7 Diesel engine combustion . . . . . . . . . . . . . . . . . . . . . . 6
1.8 Turbocharged spark ignition engine at 4 cylinder scheme . . . . . 7
1.9 6 cylinders engine . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.10 Downsizing concept . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.11 Charge motion different kinds . . . . . . . . . . . . . . . . . . . . 10
1.12 Diesel converted to natural gas piston design . . . . . . . . . . . 10
1.13 Flame developments in the two different combustion chamber
compared . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.14 Different piston turbulent kinetic energy . . . . . . . . . . . . . . 12
1.15 Different piston pressure curves . . . . . . . . . . . . . . . . . . . 12
1.16 Heat transfer development comparison between investigated pis-
ton shapes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.17 Laminar flame velocities of different fuels at different equivalence
ratios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.18 Knock illustration . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.19 Autoignition delay . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.20 Spark ignition combustion temperature chart highlighting differ-
ences that can be present in the whole domain considered . . . . 17
1.21 Pressure curve in knock presence . . . . . . . . . . . . . . . . . . 18
1.22 HCCI: Homogeneous charge compression ignition combustion . . 18
1.23 HCCI pollution chart, highlighting the advantages of a lean com-
pression ignition combustion and differences with a diesel one . . 19
1.24 CAI: controlled autoignition combustion . . . . . . . . . . . . . . 20
1.25 Rapid compression and expansion machine . . . . . . . . . . . . 20
1.26 RCEM Blower adopted to generate turbulence . . . . . . . . . . 21
1.27 SACI experimental setups . . . . . . . . . . . . . . . . . . . . . . 21
1.28 Diesel pilot injection experimental setups . . . . . . . . . . . . . 22
1.29 Experimental spark assisted efficiency . . . . . . . . . . . . . . . 22
1.30 Experimental diesel pilot efficiency . . . . . . . . . . . . . . . . . 23
1.31 Methane first oxidation reaction . . . . . . . . . . . . . . . . . . 24
1.32 Diesel particulate filter . . . . . . . . . . . . . . . . . . . . . . . . 25
1.33 Urea injection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.34 Catalytic converter for SI engine . . . . . . . . . . . . . . . . . . 25

III
1.35 Catalytic conversion efficiency . . . . . . . . . . . . . . . . . . . . 26
1.36 Natural gas composition ranges . . . . . . . . . . . . . . . . . . . 27
1.37 Consulente energia natural gas compositions . . . . . . . . . . . . 28
1.38 Unirc lecture natural gas compositions . . . . . . . . . . . . . . . 28
1.39 SNAM natural gas compositions . . . . . . . . . . . . . . . . . . 28
1.40 Gasdyn interface . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.41 3D simulation example . . . . . . . . . . . . . . . . . . . . . . . . 30
1.42 OpenFOAM mesh example . . . . . . . . . . . . . . . . . . . . . 31
1.43 OpenFOAM mesh top view . . . . . . . . . . . . . . . . . . . . . 31
1.44 Autodesk Inventor interface . . . . . . . . . . . . . . . . . . . . . 32

2.1 CFD gas exchange example . . . . . . . . . . . . . . . . . . . . . 33

2.2 Cell equiangular skew . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3 Cell aspect ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4 Mesh orthogonal quality . . . . . . . . . . . . . . . . . . . . . . . 36
2.5 Mesh size ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.6 Mesh motion layers removal example . . . . . . . . . . . . . . . . 37
2.7 Specific quantity flows for each cell . . . . . . . . . . . . . . . . . 38
2.8 Cell references . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.9 Courant number representation . . . . . . . . . . . . . . . . . . . 41
2.10 Turbulent field with eddies of different size . . . . . . . . . . . . 42
2.11 Energy cascade . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.12 Numerical simulation differences . . . . . . . . . . . . . . . . . . 44
2.13 Reynolds average . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.14 Flame interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.15 Chemical table algorithm description . . . . . . . . . . . . . . . . 51
2.16 Table generation code organization . . . . . . . . . . . . . . . . . 52
2.17 Table-solver interaction . . . . . . . . . . . . . . . . . . . . . . . 53

3.1 CNG-heavy duty engine at top dead centre . . . . . . . . . . . . 55

3.2 Whole cylinder volume . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3 Engine map of given experimental points . . . . . . . . . . . . . . 57
3.4 Xi equilibrium sweep . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.5 1000-1710 validation: pressure curve comparison . . . . . . . . . 59
3.6 1000-1710 validation: wall heat transfer comparison . . . . . . . 60
3.7 1000-1710 validation: cumulative heat release comparison be-
tween simulated and experimental data . . . . . . . . . . . . . . 60
3.8 1000-1710 validation: apparent and rate of heat release compar-
ison between simulated and experimental data . . . . . . . . . . 61
3.9 1200-855 validation: pressure curve comparison . . . . . . . . . . 62
3.10 1000x1710 validation: wall heat transfer comparison . . . . . . . 63
3.11 1200-855 validation: cumulative heat release comparison . . . . . 63
3.12 1200-855 validation: apparent and rate of heat release comparison 64
3.13 1800-1650 validation: pressure curve comparison . . . . . . . . . 64
3.14 1800-1650 validation: wall heat transfer comparison . . . . . . . 65
3.15 1800-1650 validation: cumulative heat release comparison . . . . 66
3.16 1800-1650 validation: apparent and rate of heat release compar-
ison between simulated and experimental data . . . . . . . . . . 66
3.17 Gross indicated work error in constant torque sweep . . . . . . . 67
3.18 Gross indicated work error in 1200 rpm sweep . . . . . . . . . . . 67

IV
3.19 spark timing sweep for 1200-421 working point . . . . . . . . . . 69
3.20 PV curve for 1200-421 working point . . . . . . . . . . . . . . . . 69
3.21 Spark advance sweep: cylinder temperatures . . . . . . . . . . . . 70
3.22 Spark advance sweep: wall heat transfer . . . . . . . . . . . . . . 70
3.23 Spark advance sweep: cumulative heat release . . . . . . . . . . . 71
3.24 Flame interface of low load engine at (from left to right) 26, 24
and 20 of spark advance . . . . . . . . . . . . . . . . . . . . . . . 71
3.25 1200-421 efficiency curve with respect to spark timing variations 72
3.26 Gross indicated work with respect to spark timing variations . . 72
3.27 Spark timing sweep cumulative wall heat transfer curve . . . . . 73
3.28 1200-421 spark timing sweep: pressure and temperature maxima 73
3.29 Constant torque sweep . . . . . . . . . . . . . . . . . . . . . . . . 74
3.30 Gross indicated work at constant torque . . . . . . . . . . . . . . 74
3.31 Cumulative heat released curves at constant torque . . . . . . . . 75
3.32 Optimized pressure curves at constant torque . . . . . . . . . . . 75
3.33 Cumulative heat transfer curves at constant torque . . . . . . . . 76
3.34 Temperature curves in constant torque sweep . . . . . . . . . . . 76
3.35 Wall heat transfer curves in constant torque sweep . . . . . . . . 77
3.36 turbulent kinetic energy in constant torque sweep . . . . . . . . . 77
3.37 Efficiencies in constant torque sweep . . . . . . . . . . . . . . . . 78
3.38 Optimum spark timing in constant torque sweep . . . . . . . . . 78
3.39 Load sweep illustration . . . . . . . . . . . . . . . . . . . . . . . . 79
3.40 Load sweep pressure curves . . . . . . . . . . . . . . . . . . . . . 79
3.41 Load sweep cumulative heat release curves . . . . . . . . . . . . . 80
3.42 Gross indicated work histogram . . . . . . . . . . . . . . . . . . . 80
3.43 Constant regime temperature curves . . . . . . . . . . . . . . . . 81
3.44 Constant regime wall heat transfer curves . . . . . . . . . . . . . 81
3.45 Efficiency at constant regime . . . . . . . . . . . . . . . . . . . . 82
3.46 Optimum spark timing at constant regime . . . . . . . . . . . . . 82
3.47 Pressure maximum values at constant regime . . . . . . . . . . . 83
3.48 CFD and optimized engine map comparison . . . . . . . . . . . . 83
3.49 CFD working points efficiency engine map . . . . . . . . . . . . . 84
3.50 CFD efficiency engine map after optimization . . . . . . . . . . . 84
3.51 Experimental working points spark timing map . . . . . . . . . . 85
3.52 CFD optimized spark timing engine map . . . . . . . . . . . . . . 85

4.1 Engine flame front and spontaneous ignition visualization . . . . 87

4.2 Autoignition propagation in 2 crank angle degrees . . . . . . . . 88
4.3 Pressure curves when engine is powered with isooctane as fuel . . 88
4.4 Pressure rise curves when engine is powered with isooctane as fuel 89
4.5 Heat release curves when engine is powered with isooctane as fuel 90
4.6 Autoignition timing when engine is powered with isooctane as
fuel, -12 spark timing case . . . . . . . . . . . . . . . . . . . . . . 90
4.7 Cfresh source term when engine is powered with isooctane as fuel 91
4.8 Cfresh source term in a not knocking CFD simulation . . . . . . 91
4.9 Temperature curves when engine is powered with isooctane as fuel 92
4.10 Wall heat transfer when engine is powered with isooctane as fuel 92
4.11 Main specific quantities normalized . . . . . . . . . . . . . . . . . 93
4.12 Paper experimental engine setup . . . . . . . . . . . . . . . . . . 94
4.13 Experimental and simulated engine tuning . . . . . . . . . . . . . 95

V
4.14 Reference natural gas composition . . . . . . . . . . . . . . . . . 96
4.15 Added hydrocarbons ranges . . . . . . . . . . . . . . . . . . . . . 96
4.16 Natural gas composition with ethane addition . . . . . . . . . . . 97
4.17 Natural gas composition with propane addition . . . . . . . . . . 97
4.18 Natural gas composition with butane addition . . . . . . . . . . . 97
4.19 Pressure curves used for sensor signal conversion . . . . . . . . . 98
4.20 CFD simulation of voltage output sensor with ethane addition . 98
4.21 CFD simulation of voltage output sensor with propane addition . 99
4.22 CFD simulation of voltage output sensor with butane addition . 99
4.23 Combustion velocity fuel comparison . . . . . . . . . . . . . . . . 100
4.24 Pressure curve comparison between different natural gas compo-
sition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.25 Cumulative heat release comparison between different natural gas
composition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.26 Compression ignition progress variable of natural gas considered 102
4.27 Natural gas knock region . . . . . . . . . . . . . . . . . . . . . . 103
4.28 Pressure curves of knocking natural gas fuelled engine . . . . . . 103
4.29 Pressure rise curves of knocking natural gas fuelled engine . . . . 104
4.30 Unburnt fresh temperature curves of knocking natural gas fuelled
engine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.31 Unburnt fresh temperature values in the whole cylinder . . . . . 105
4.32 Autoignition strength in constant torque sweep . . . . . . . . . . 106
4.33 Rate of heat release of knocking natural gas fuelled engine . . . . 106
4.34 Wall heat transfer of knocking natural gas fuelled engine . . . . . 107
4.35 Turbulent kinetic energy of knocking natural gas fuelled engine . 107
4.36 Natural gas knocking areas . . . . . . . . . . . . . . . . . . . . . 108
4.37 Cylinder walls names . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.38 Autoignition principle if crevice walls are considered adiabatic . . 109
4.39 ”Adiabatic liner” wall heat transfer comparison . . . . . . . . . . 110
4.40 ”Adiabatic piston” wall heat transfer comparison . . . . . . . . . 110
4.41 Same case of before but crevice wall are not considered adiabatic 111
4.42 Head heat transfer in adiabatic walls boundary condition com-
parison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.43 Piston heat transfer in adiabatic walls boundary condition com-
parison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.44 Liner heat transfer in adiabatic walls boundary condition com-
parison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.45 Crevice temperatures when different boundary conditions are ap-
plied . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.46 Head heat transfer wall functions comparison . . . . . . . . . . . 113
4.47 Piston heat transfer wall functions comparison . . . . . . . . . . 114
4.48 Liner heat transfer wall functions comparison . . . . . . . . . . . 114
4.49 ”Adiabatic piston” heat transfer wall functions comparison . . . 114
4.50 ”Adiabatic liner” heat transfer wall functions comparison . . . . 115
4.51 Simulated knock when Huh Chang wall function is used . . . . . 115
4.52 Total heat transfer wall functions comparison . . . . . . . . . . . 116

5.1 Combustion chamber shapes at top dead centre . . . . . . . . . . 117

5.2 Volume variation with compression ratio . . . . . . . . . . . . . . 118
5.3 Mass of fuel injected in different compression ratio engines . . . . 118

VI
5.4 Autodesk inventor design render, with 15 as CR . . . . . . . . . 119
5.5 Revisited engine map, showing SACI points target . . . . . . . . 119
5.6 Pressure curves obtained by both solvers in the same initial con-
ditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
5.7 Cumulative heat release curves obtained by both solvers in the
same initial conditions . . . . . . . . . . . . . . . . . . . . . . . . 120
5.8 Pressure rise curves obtained by both solvers in the same initial
conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.9 Gross indicated work curves obtained by both solvers in the same
initial conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.10 Instantaneous gross indicated work curves obtained by both solvers
in the same initial conditions . . . . . . . . . . . . . . . . . . . . 122
5.11 Wall heat transfer curves obtained by both solvers in the same
initial conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.12 Temperature curves obtained by both solvers in the same initial
conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.13 Autoignition propagation in CR 15 engine; 13 deg after TDC
leftwards, 14 deg after TDC rightwards . . . . . . . . . . . . . . 123
5.14 Autoignition principle in CR 15 engine . . . . . . . . . . . . . . . 124
5.15 Pressure-volume curves obtained by both solvers in the same ini-
tial conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.16 Autoignition anticipation when engine walls are at higher tem-
peratures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.17 Heat losses curves when engine walls are at higher temperatures 126
5.18 Instantaneous gross indicated work curves when engine walls are
at higher temperatures . . . . . . . . . . . . . . . . . . . . . . . . 126
5.19 Cumulative heat release in high compression ratio spark advance
sweep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.20 Pressure and temperature curves in high compression ratio spark
advance sweep . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.21 Instantaneous gross indicated work in high compression ratio
spark advance sweep . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.22 Flame front and cylinder pressure in an entire flame propagation
combustion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.23 Flame front and cylinder temperature in an entire flame propa-
gation combustion . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.24 Pressure gradient at compression ignition origin, 8 deg after TDC 129
5.25 Pressure gradient immediately after compression ignition is com-
pleted, 9 deg after TDC . . . . . . . . . . . . . . . . . . . . . . . 130
5.26 Pressure gradient after wave has propagated inside the combus-
tion chamber, 11 deg after TDC . . . . . . . . . . . . . . . . . . 130
5.27 Turbulence intensity comparison between before (left) and after
(right) autoignition . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.28 Efficiency curve with spark timing in a heavy knock engine . . . 131
5.29 Autoignited mixture fraction in spark timing sweep. AI indica-
tor is the ratio between autoignited mixture mass and the whole
mixture mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.30 Pressure curve in load sweep, in knock presence . . . . . . . . . . 133
5.31 Temperature curve in load sweep, in knock presence . . . . . . . 133
5.32 Wall heat transfer in load sweep, in knock presence . . . . . . . . 134

VII
5.33 Instantaneous gross indicated work in load sweep, in knock presence134
5.34 Efficiency curve in load sweep, in knock presence . . . . . . . . . 135
5.35 Autoignition principles in full load condition . . . . . . . . . . . . 135
5.36 Pressure maximum values in load sweep, in knock condition . . . 136
5.37 Turbulent kinetic energy values in full load condition . . . . . . . 136
5.38 Piston geometry comparison at top dead centre with respective
names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
5.39 ”Turb” piston geometry at top dead centre . . . . . . . . . . . . 137
5.40 3D render of ”Eng” combustion chamber . . . . . . . . . . . . . . 138
5.41 Pressure curves associated to different piston geometries . . . . . 138
5.42 Flame front visualization in CFD domain associated to different
piston geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.43 Turbulence intensity associated to different piston geometries . . 139
5.44 ”Turb” pressure curves in spark timing sweep . . . . . . . . . . . 140
5.45 ”Turb” heat release curves in spark timing sweep . . . . . . . . . 140
5.46 Compression ignition evolution between 11 deg and 14 deg in
”Turb” geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.47 Compression ignition evolution between 11 deg and 14 deg in
”Eng” geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
5.48 ”Turb” turbulent kinetic energy curves in spark timing sweep . . 142
5.49 ”Turb” wall heat transfer curves in spark timing sweep . . . . . . 143
5.50 ”Turb” cumulative wall heat transfer curves in spark timing sweep143
5.51 ”Turb” efficiency curve in spark timing sweep . . . . . . . . . . . 144
5.52 ”Turb” pressure rise curves in spark timing sweep . . . . . . . . . 144
5.53 Geometry comparison pressure curves . . . . . . . . . . . . . . . 145
5.54 Geometry comparison turbulent kinetic energy curves . . . . . . 146
5.55 Geometry comparison cumulative heat release curves . . . . . . . 147
5.56 Geometry comparison wall heat transfer curves . . . . . . . . . . 147
5.57 Wall per wall heat transfer geometry comparison . . . . . . . . . 148
5.58 Gross indicated work geometry comparison, with corresponding
thermal efficiency values reported in percentage . . . . . . . . . . 149
5.59 Pressure curves when engine is running in lean conditions . . . . 150
5.60 Pressure curves spark timing sweep in lean conditions . . . . . . 150
5.61 Temperature curves spark timing sweep in lean conditions . . . . 151
5.62 Cumulative heat release curves spark timing sweep in lean con-
ditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.63 Cumulative wall heat transfer curves spark timing sweep in lean
conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5.64 Instantaneous gross indicated work curves spark timing sweep in
lean conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5.65 Gross indicated work curves spark timing sweep in lean conditions153
5.66 Pressure gradient in the CFD domain after lean compression ig-
nition occurred . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.67 Pressure rise comparison between stoichiometric and lean mix-
tures conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
5.68 Rate of heat release comparison between stoichiometric and lean
mixtures conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 154
5.69 Engine efficiency of lean CNG-heavy duty piston shape in spark
timing sweep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.70 Pressure curves of lean ”Turb” piston shape in spark timing sweep156

VIII
5.71 Efficiency curves of lean ”Turb” piston shape in spark timing sweep156
5.72 Heat Losses curves of lean ”Turb” piston shape in spark timing
sweep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
5.73 Turbulent kinetic energy of lean ”Turb” piston shape in CFD
domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
5.74 Main piston shape and equivalence ratio comparison in heat re-
lease maximum values . . . . . . . . . . . . . . . . . . . . . . . . 158
5.75 Main piston shape and equivalence ratio comparison in pressure
rise maximum values . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.76 Flame propagation when an equivalence ratio of 0.7 is adopted,
showing that the minimum b value can not be considered 0 . . . 160
5.77 Flame extinction when an equivalence ratio of 0.6 is adopted . . 160
5.78 Efficiency SACI engine map . . . . . . . . . . . . . . . . . . . . . 161
5.79 Efficiency curve for a SACI engine in load sweep . . . . . . . . . 161
5.80 Spark timing SACI engine map . . . . . . . . . . . . . . . . . . . 162
5.81 Spark timing in load sweep at 1200 rpm . . . . . . . . . . . . . . 162
5.82 Pressure rise SACI engine map . . . . . . . . . . . . . . . . . . . 163
5.83 CO production in 1200-half load working point comparison. φ =
0.8 SACI reported in green, while stoichiometric flame propaga-
tion in red . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
5.84 CO2 production in 1200-half load working point comparison. φ
= 0.8 SACI reported in green, while stoichiometric flame propa-
gation in red . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.85 unburnt hydrocarbons after the whole combustion process in 1200-
half load working point comparison. φ = 0.8 SACI reported in
green, while stoichiometric flame propagation in red . . . . . . . 165
5.86 Cell temperatures in 1200-half load working point comparison, at
the same engine time. φ = 0.8 SACI reported leftwards, while
stoichiometric flame propagation rightwards . . . . . . . . . . . . 166
5.87 CO production in 1600-half load working point comparison. φ =
0.8 SACI reported in green, while stoichiometric flame propaga-
tion in red . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5.88 CO2 production in 1600-half load working point comparison. φ
= 0.8 SACI reported in green, while stoichiometric flame propa-
gation in red . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5.89 Unburnt hydrocarbons after the whole combustion process in
1600-half load working point comparison. φ = 0.8 SACI reported
in green, while stoichiometric flame propagation in red . . . . . . 167
5.90 CO production in 1200-full load working point comparison. Both
power cycles performed through entire flame propagation . . . . 168
5.91 CO2 production in 1200-full load working point comparison. Both
power cycles performed through entire flame propagation . . . . 168
5.92 Unburnt hydrocarbons after the whole combustion process in
1200-full load working point comparison. Both power cycles per-
formed through entire flame propagation . . . . . . . . . . . . . . 169

IX
List of Tables

1.1 Natural gas main hydrocarbons . . . . . . . . . . . . . . . . . . . 27

1.2 Most common inerts present in natural gas . . . . . . . . . . . . 27
1.3 SNAM North Europe natural gas composition in mass percentages 29

3.1 CNG-heavy duty main geometrical parameters . . . . . . . . . . 56

3.2 Initial available pressure, temperatures and spark timing for every
experimental working point represented in CFD map . . . . . . . 58
3.3 Comparison between computed and experimental data . . . . . . 62

4.1 Comparison between paper engine and CNG-heavy duty one . . 94

4.2 Natural gas composition in mass percentages . . . . . . . . . . . 102

5.1 SACI and Xi solvers comparison . . . . . . . . . . . . . . . . . . 124

5.2 Wall temperature values . . . . . . . . . . . . . . . . . . . . . . . 125
5.3 Standard and high wall temperature performance comparison . . 126
5.4 Spark timing sweep performance table . . . . . . . . . . . . . . . 132
5.5 Efficiencies comparison of different piston geometries, with the
same initial conditions and spark timing . . . . . . . . . . . . . . 139
5.6 Simulation summary for ”Turb” spark timing sweep . . . . . . . 145
5.7 Simulation summary for ”Eng” spark timing sweep . . . . . . . . 145
5.8 ”Turb” piston shape lean spark timing sweep . . . . . . . . . . . 157
5.9 ”Eng” piston shape lean spark timing sweep . . . . . . . . . . . . 158
5.10 Differences in gross indicated work between starting flame prop-
agation combustion and obtained spark assisted one, with differ-
ences of thermal efficiencies reported rightwards . . . . . . . . . . 163
List of acronyms

AHRR Apparent Heat Release Rate

AKR Anti Knocking Regulation
BDC Bottom Dead Centre
BDS Backward Difference Scheme
CAD Crank Angle Degrees
CDS Central Difference Scheme
CFD Computational Fluid Dynamics
CPU Central Processing Unit
DI Direct Injection
DPF Diesel Particulate Filter
ECU Electronic Control Unit
EVO Exhaust Valve Opening
GIW Gross Indicated Work
HCCI Homogeneous Charge Compression Ignition
ICE Internal Combustion Engine
ID Ignition Delay
IVC Intake Valve Closing
LHV Lower Heating Value
MON Motored Octane Number
ODE Ordinary Differential Equation
ON Octane Number
OpenFOAM Open source Field Operation And Manipulation
PCCI Premixed Charge Compression Ignition
PSR Perfect Stirred Reactor

XIII
PV Progress Variable
RoHR Rate of Heat Released

RON Research Octane Number

SACI Spark Assisted Compression Ignition
SCR Selective Catalytic Reduction

SI Spark Ignition
SOC Start Of Combustion
SOI Start Of Injection
TDC Top Dead Centre

XIV
Sommario

Al giorno d’oggi avere motori puliti e in grado di emettere poca anidride carbon-
ica è di fondamentale importanza per la salvaguardia e la tutela dell’ambiente.
In quest’ottica, il gas naturale è una tipologia di carburante molto interessante
in quanto, essendo in condizioni atmosferiche allo stato gassoso, non forma al-
cun tipo di polvere sottile e, in aggiunta, a causa del basso rapporto carbonio-
idrogeno (C/H) permette di avere una minor produzione di CO2 rispetto alla
combustione di una pari quantità di benzina o gasolio. A causa dell’elevato nu-
mero di ottani di questo tipo di carburante, il ciclo termodinamico scelto come
riferimento è il ciclo ”Otto”, che tradizionalmente comporta l’installazione di
una candela necessaria per rilasciare una quantità di energia minima in grado
di controllare la combustione vera e propria, creando un fronte di fiamma. Tut-
tavia, la fiamma del gas naturale non è abbastanza veloce da poter considerare
la combustione come istantanea (assunta teoricamente dal ciclo Otto) e per cui,
se si considerano motori a carico pesante in cui un singolo cilindro può avere
un volume massimo maggiore di 2 litri, velocizzare il processo di combustione
del gas risulta fondamentale per lo sviluppo dei motori a combustione interna.
Per fare ciò, modalità di combustione alternativa sono state considerate, come
la più famosa HCCI (in cui il rapporto di compressione è talmente alto da far
accendere spontaneamente tutta la carica all’interno del volume di controllo) o
la più moderna combustione SACI (in cui una parte della miscela brucia per
propagazione di fiamma e una minore reagisce per compressione). La seconda
risulta più fattibile dato che è grazie alla presenza della candela è possibile con-
trollare maggiormente la combustione e avere minore variabilità tra due diversi
cicli di funzionamento. Tuttavia anche quest’ultima ipotesi non è di semplice
realizzazione, in quanto è possibile che la combustione per compressione generi
onde di pressione talmente importanti da diminuire l’efficienza termica del mo-
tore e, nel peggiore dei casi, danneggiare la macchina stessa, risultando quindi
detonazione. In questo lavoro di tesi viene quindi analizzata la differenza tra
il concetto di detonazione e quello di accensione spontanea e, utilizzando le
considerazioni raccolte, viene successivamente sviluppato un motore SACI (ad
accensione assistita) in grado di sfruttare l’autoaccensione per velocizzare il pro-
cesso di combustione e quindi migliorare le prestazioni del motore stesso. Nel
corso di questo lavoro quindi sono stati ottenuti due diversi criteri di proget-
tazione che permettono di sfruttare l’accensione spontanea al fine di ridurre il
consumo di carburante: un primo prende in esame la forma del pistone, un
secondo analizza l’eventuale eccesso d’aria presente in camera di combustione.
Attraverso simulazioni numeriche fluido-dinamiche, le prestazioni e le emissioni
del motore ottenuto sono state confrontate con quelle di uno tradizionale ad
accensione comandata, riscontrando apprezzabili vantaggi garantiti da questa

XV
nuova modalità di combustione soprattutto nelle condizioni a basso e medio
carico.

XVI
Abstract

Nowadays to have clean engines able to reduce both carbon dioxide and pollu-
tants emissions is of fundamental importance for safeguarding and protecting
the environment. For this reason, natural gas is a very interesting kind of fuel,
since it does not form any particle matter as it is already gaseous in atmospheric
conditions, and the low carbon-hydrogen ratio (C/H) leads to a reduced CO2
production with respect to the combustion performed with the same amount of
gasoline or diesel. Due to the high octane number of this kind of fuel, the ther-
modynamic cycle chosen as a reference is the ”Otto cycle”, which traditionally
considers the installation of a spark-plug needed to release a minimum energy
quantity to control the combustion, creating a real flame front. In any case, the
natural gas flame front velocity is not high enough to consider the combustion
as instantaneous (which is the one theoretically assumed by the Otto cycle) and,
if heavy duty engines which can have a displacement per cylinder over than 2
liters are considered, to speed up the combustion process results fundamental
for the development of internal combustion engines. To do so, different and
innovative combustion modes have been investigated, as the most famous HCCI
(where compression ratio adopted is so high to have a complete compression ig-
nition of the whole charge inside the control volume) or the more modern SACI
one (where a part of the mixture burns due to flame propagation and the other
one gets ignited by compression). The second one results more feasible since,
thanks to the presence of the spark-plug, it is possible to have a better control
on the combustion and as a consequence cycle-variability is reduced. In any
case, even this last one is difficult to realize, as it is possible that compression
ignition combustion generates such intense pressure waves to decrease thermal
efficiency of the engine and, in the worst of the cases, to damage the mechanical
structure of the machine, hence resulting into knock. In this thesis work, the
difference between the concept of knock and the one of spontaneous ignition
has been analysed and, using collected considerations, a SACI (spark assisted
compression ignition) engine has been developed in such a way it can exploit
autoignition phenomenon to speed up the whole combustion process and to im-
prove engine performances. During this work, two different design criteria which
permits a good exploitation of partial spontaneous ignition have been derived:
the first one considers the shape of the piston, while the second one analyses the
possible air excess present in the combustion chamber. Through fluid dynam-
ics numerical simulations, performances and emissions of the engine obtained
have been compared with the ones of a traditional spark ignition heavy duty
machine, highlighting the appreciable advantages of this innovative combustion
mode especially in low and half load conditions.

XVII
Chapter 1

Internal combustion engines

Internal combustion engine is an energy production technology used especially

in transportation sector. Automotive industry is very important for worldwide
economy, such that in 2014 89.75 millions of vehicles were globally produced,
and over 80 millions with a thermal engine installed. The main conceptual
difference between an internal combustion engine and a lot of other energy pro-
duction devices consists in having compression, combustion and expansion in
the very same volume. As a consequence, they can reach very high powers in
small volumes, making the technology suitable for both very small applications
(chainsaw for instance) and heavy duty ones (tracks, ships or energy produc-
tion).
Internal combustion engines can be classified in a lot of different ways: number
of strokes per thermodynamic cycle, air feeding, fuel injection are just some of
the criteria that people can use. In this introduction section, an analysis has
been performed starting from the most known thermodynamic cycles and they
can be practically realized, then moving more specifically to combustion and
ICE possible future development.
Most known thermodynamic cycles describing engine working principle are Otto
and Diesel ones.

Otto cycle

Figure 1.1: Otto thermodynamic cycle

1
In figure 1, ideal Otto cycle has been illustrated both on pressure-volume and
temperature-entropy plane. First phase is adiabatic compression, correspond-
ing to piston rise from bottom dead centre (v1 ) to top dead centre (v2 ); then
an ideally constant volume combustion is performed at top dead centre, ris-
ing cylinder temperature to extremely high values. Afterwards, working fluid
is expanded, and finally pressure and temperature initial values are restored
once cycle is completed. Gas exchange cycle is displayed leftwards to point 1 in
pressure-volume plane.
This leads to huge power densities but losses such as thermal ones can be-
come important: ideally, compression and expansion are adiabatic, but since
in practice cylinder mixture temperature can reach and overcome 2200 K it is
impossible to assume adiabatic walls; also ideal combustion is isochoric (at con-
stant volume), drawn by a vertical line in pressure-volume plane. To assume a
constant volume combustion means that fuel heat release is instantaneous; but
even if flame velocities are quite high, combustion process needs time to develop,
and this time is not negligible if compared to compression/expansion stroke du-
ration: at 1200 rpm, a stroke requires 25 ms to be completed, and combustion
duration is in the order of magnitude of 5 ms. These two main kinds of losses
are the reason why, when pressure volume curve is actually plotted, chart is
quite different with respect to the ideal cycle.

Figure 1.2: Realistic pressure-volume curve

In figure 2, two areas are highlighted: A and B. Area A represents com-

pression, combustion and expansion processes, and it is a more realistic rep-
resentation of the ideal working cycle illustrated in figure 1: combustion can
not be assumed instantaneous, and process representation ends with a rounded
pressure maximum. Area B instead represents the gas exchange cycle. As far as
engine adopted is a four stroke engine, it needs two shaft rotations to complete
a full thermodynamic cycle. With respect to two stroke engines, this design
choice aims to maximize thermal efficiency because exiting of burnt mixture
from combustion chamber and following engine feeding improves significantly,
but as a drawback power reduces because there is just one expansion stroke on

2
four. Therefore, area A is clockwise oriented, able to represent some useful work;
area B can be counterclockwise oriented, requiring energy to be completed.
In modern Otto cycle engines, combustion starts through spark light ignition
from a spark plug, a device installed in combustion chamber releasing a small en-
ergy quantity to fuel-air compressed mixture. To have a good control of ignition
timing, an electronic control unit is used: spark timing is always anticipated with
respect to top dead centre (point where cylinder volume is minimum), searching
for a compromise between a low compression work and maximizing expansion
one, due to combustion non instantaneous lasting.

Figure 1.3: Motored and fired pressure curves

In figure 3, motored and fired pressure curves are plotted with respect to the
crank angle. This x-axis represents piston position, and it assumes the value
of 0◦ in correspondence of top dead centre, and the value of 180◦ (positive or
negative) at the bottom dead centre. It is also known as engine time, since it
can be directly related to physical time once engine rotational speed is known.
Motored curve stands for cylinder pressure curve when combustion does not take
place: what the engine spends for compression, is given back to expansion and
gross indicated work is ideally zero. Fired curve is pressure curve in presence of
combustion, which is asymmetrical, otherwise it would be impossible to obtain
any useful work. Yellow star stands for spark timing, the exact moment when
spark light is igniting the mixture. In this figure it is possible to notice a delay
between mixture ignition and effective pressure rise, but this kind of delay is al-
ways present. Anticipating spark timing, compression losses rises because there
is a pressure increase before top dead centre, decreasing pressure-volume area;
also, higher pressures are reached, and rightward (with respect to 0◦ ) pressure
curve underlying area increases. Numbers above represent combustion phases,
as ignition delay (1◦ ), combustion development (2◦ ) and both combustion com-
pletion and expansion (3◦ ). Figure below reports some flame images different
combustion moments, to emphasize its not-ideal lasting.

3
 Figure 1.4: Combustion lasting

Efficiency of an ideal Otto cycle depends on compression ratio, defined as

the ratio between the minimum and maximum cylinder volumes.
Vmax
compressionRatio = CR = (1.1)
Vmin
1
ηth = 1 − (1.2)
CRγ−1
cp
γ= (1.3)
cv

Figure 1.5: Otto cycle efficiency plot with compression ratio

Therefore, ideally thermodynamic efficiency depends just on equations re-

ported above. In practice, spark ignition commercial engines have a compression
ratio that can vary between 8 and 11, depending on the fuel they are feeded
with, on the kind of injection (port or direct) used and also on the design appli-
cation. What is limiting compression ratio increase is possible knock presence.
Knock is defined as ”abnormal combustion”, meaning that unburnt fresh mix-
ture reaches such high pressures and temperatures during combustion process
that it autoignites. Autoignition in fact is nothing but completing combustion
chemical reactions without a real flame propagation, and energy that for flame
development is given by spark light, for knock it is given by high pressures and

4
temperatures. This is why controlling spark timing is so important: by an an-
ticipation, higher cylinder pressures and temperatures are reached, but also it
makes more probable autoignition presence.

Diesel cycle

Diesel cycle has been idealized by Rudolf Diesel in 1893.

Figure 1.6: Diesel thermodynamic cycle

As illustrated in figure 6, it is very similar to Otto cycle, since both of

them are involving an adiabatic compression, an adiabatic expansion and a
gas exchange cycle in four strokes engines. The real point of difference is that
in the Otto cycle a constant volume combustion is performed, in the Diesel
one combustion develops at constant pressure, hence line connecting point 2
to point 3 is horizontal. This difference in combustion process leads to both
thermodynamic and engineering differences between the technologies. Starting
from thermodynamic ideal efficiency, in equations 4 and 5 ηth of both cycles is
computed:
1
ηth = 1 − Ottocycle (1.4)
CRγ−1
1 h 1 rγ − 1 i
c
ηth = 1 − · Dieselcycle (1.5)
CRγ−1 γ rc − 1
Expressions illustrated above are very similar, with a multiplicator function of
Cut-off ratio present in Diesel cycle thermodynamic efficiency. Cut-off ratio is
defined as the ratio of the volume once combustion is completed (v3 ) with the
top dead centre one (v2 ), where combustion is starting.
Expression in the square brackets in equation 5 is always higher than 1, hence
Diesel cycle efficiency is lower with respect to Otto cycle one for the same
compression ratio. The point of strength of diesel engine is that they can adopt
an higher compression ratio with respect to Otto cycle ones: in spark ignition
engines, compression ratio is limited to avoid knock presence, assuming values
between 8 and 10 in gasoline fuelled passenger cars; in compression ignition
diesel engines, this limit is not present and they assume values between 16
and 18 in diesel fuelled passenger cars. Combustion in Diesel cycle engines is
spontaneous, meaning there is not a clear flame propagation as happens in spark
ignition engines, but as soon as the fuel evaporates and mixes up with air, it
ignites by compression.

5
 Figure 1.7: Diesel engine combustion

Activation energy necessary for combustion depends on the kind of fuel used.
In spark ignition engines, autoignition is something unwanted, hence a suitable
fuel must have a high autoignition delay, such as gasoline or natural gas. In
Diesel engines, combustion is performed by compression ignition, hence a suit-
able fuel must have a very low autoignition delay. This is one reason why diesel
fuel and gasoline are so different between each other: the first one is quite heavy,
but as soon as it is injected in the cylinder it has to ignite due to high pressures
and temperatures; the second is more volatile, but much more resistant to com-
pression ignition. In addition, the higher compression ratio in Diesel engines is
contributing to reduce diesel fuel ignition delay. Diesel fuel must be injected
directly in the cylinder at very high injection pressures (100 MPa as order of
magnitude) to have a good spray brake up and an almost instantaneous com-
bustion. As shown in figure 7, in a compression ignition combustion there is not
a clear flame front propagating but burnt region volume depends on how fuel
spray evaporates and mixes up with air. Since there is not a flame propagation,
there is no need to have a stoichiometric (or slightly lean) air-fuel ratio as in
spark ignition combustion: compression ignition engines work with high air ex-
cess, which is important to reduce heat losses, increasing engine performances.
In addition, it is an advantage for the gas exchange cycle too, because there is
no need to install a throttle valve, hence pressure losses are reduced.
As a disadvantage, Diesel engines need to be more robust to stand higher stresses
due to compression ignition combustion. They are heavier and, to have good
spray characteristics, fuel injection system is much more complex with respect
to gasoline direct injection ones.
Diesel engines usually run at lower rotational speeds, because the fuel needs
time to ignite by compression. As a consequence, a Diesel engine have more
torque with respect to an Otto one (at the same power), since it presents max-
imum torque point at lower rotational speed. Having a high torque improves
driveability whenever the engine is installed in a passenger car or in a heavy
duty vehicle, since they can rapidly react if resistant forces increase.
As a final consideration, as explained in [2], Diesel engines are more suitable

6
when power required is high (such as tracks, ships or heavy passenger cars),
while spark ignition engines when specific power is lower (light passenger cars).

ICE layout and importance of turbocharging

Figure 1.8: Turbocharged spark ignition engine at 4 cylinder scheme

In figure 8, a modern spark ignition engine layout is illustrated. Air mass

cylinder feeding is regulated through a throttle valve, drawn exactly at intake
manifold inlet. Closer the throttle valve, lower cylinder pressure inlet and hence
working fluid mass; fuel injection is electronically controlled by ECU. In a natu-
rally aspirated internal combustion engine (not having both compressor and tur-
bocharger), maximum cylinder intake pressure is the atmospheric one, slightly
higher if pressure waves are well exploited. At partial load, breathing pressure
is lower with respect to the atmospheric one, leading to the presence of coun-
terclockwise B area illustrated in Figure 2. If throttle valve is further closed,
work required for engine breathing increases, penalizing engine fuel consump-
tion. Eliminating throttle valve is a design target for future internal combustion
engines.
Torque can be evaluated using following expression:

ρair V QHV
Tb = ηb · λv · (1.6)
α 2πcycle

Two kinds of efficiencies are present: volumetric (λv ) and global (ηb ) efficiency.
Volumetric one is representing engine ability to evacuate burnt already exploited
gasses to be feeded with fresh charge; global efficiency is related to compression,
combustion and expansion processes. QHV is fuel heating value, while cycle
represents number of rotations needed to complete a full thermodynamic cycle
(1 in two stroke engines, 2 in four stroke engines).
ρair V /α represents fuel mass inside the cylinder, since α is defined as air and
fuel mass ratio. For what appears by mathematical formula, increasing air
density or cylinder volume leads to no change if losses are not considered. This
is the reason why main strategy to have higher naturally aspirated engine power
is increasing cylinder displacement, since atmospheric air pressure and density

7
values can not be overcome. This inevitably leads to design bigger cylinders or
adopting more than one.

Figure 1.9: 6 cylinders engine

Implying bigger cylinders, wall surface increases considerably and it is not

good for thermal losses. Adopting more than one cylinder (keeping the same
displacement) allows both to reduce total heat transfer area and (in spark igni-
tion engines) average flame path in every cylinder, enhancing mixture autoigni-
tion resistance. In modern context where reducing pollutant or greenhouse gas
emissions is very important, installing higher engine power sacrificing its effi-
ciency is not good both by an environmental and an economical point of view.
Therefore, to increase both power and efficiency, a turbocharging system is in-
stalled in a lot of modern commercial internal combustion engines: burnt gasses
(which have a lot of energy content due to high pressures and temperatures)
are further expanded through a Francis turbine that drives a centrifugal com-
pressor. Compressor is placed before intake manifold (as figure 8 shows), such
that higher intake valve closing pressures can be reached. By a thermodynamic
point of view, this leads to higher thermal inertia due to more mass entered in
the cylinder, decreasing losses importance. In addition, if pressure values over-
come atmospheric one, difference between breathing pressure and atmospheric
one becomes positive, making area B shown in figure 2 clockwise oriented, rep-
resenting some more useful work. This is the reason why in the context of
internal combustion engines the word ”downsizing” has been used, to indicate
this tendency to build very small engines with high boosting to reduce CO2 and
pollutant emissions. This design choice had some very appreciable results when
NEDC (New European Driving Cycle) was reproduced in test bench, but when
actual ordinary engine usage and aging were considered, thermal efficiencies had
not the predicted benefits. Therefore, at the state of the art it is more common
to hear about ”rightsizing”, a displacement reduction to have good test bench
results but not so extreme to compromise engine durability or efficiency stability
in different working points.
Common maximum boost pressures reached are in the order of magnitude of
2-2.5 bar in commercial passenger cars or heavy duty vehicles, implying a den-
sity increase. Intercoolers (working with ambient air as coolant fluid) are placed
after compressor to reduce temperature increasing in compression process, to

8
have air density benefits and reducing compression work.

Figure 1.10: Downsizing concept

Increasing compression ratio is of key importance for modern internal com-

bustion engine efficiency. In spark ignition context, it is limited to avoid knock
presence, that is mechanically dangerous for the engine. In Diesel one, compres-
sion ratio has a commercial range of 16-20, also depending on their application.
In modern history, different solutions were studied to increase CR, such as direct
fuel injection in combustion chamber, which is a more expensive with respect
to port fuel one but leading to appreciable efficiency benefits since knock point
is delayed. Another interesting solution is Atkinson cycle, a slightly variation
of the Otto one.

1.1 Heavy duty spark ignition engines

This thesis work is focused on a natural gas heavy duty engine. In a commer-
cial environment, reduction of fuel consumption is an absolute priority by an
economical point of view, and, at the same time, emission standards must be
fulfilled. A lot of track application engines are diesel fuelled, in order to adopt
high compression ratio and reduce fuel consumption. At the state of the art,
engine designers are searching for both new technologies able to improve fuel
combustion and different kinds of fuel to reduce pollutant emissions. Natural
gas can be a very interesting option for future heavy duty vehicles, as far as it
is an already gaseous fuel (no soots are formed) and it is less taxed with respect
to other hydrocarbons refinery products. Therefore, a lot of engine models have
been converted from a compression ignition diesel design to a spark ignition
natural gas one, through spark plug installation and changing of aftertreatment
system.

9
 Figure 1.11: Charge motion different kinds

In figure 11 different kinds of charge motions are illustrated. In Diesel en-

gines, swirl is very important to spread diesel fuel inside the combustion cham-
ber. As far as there is not a real flame but just heat release, diesel fuel injection
is at high pressure and directly inside the cylinder. Swirl production therefore
is determinant for a good fuel combustion. At contrary, spark ignition engines
have a flame propagation, and they can be easily assumed as homogeneous
charge (especially when port fuel injection is used). Most important charge mo-
tion is tumble, since it is able to create turbulences to increase flame velocity.
Design criteria can vary a lot between compression ignition engines (where a lot
of space is needed to spread fuel along combustion chamber) and spark ignition
ones, and shapes can be very different. Another difference consists in cylinder
head: diesel engines run lean, hence feeding the cylinder is not the highest prior-
ity; otherwise, spark ignition engines run stoichiometric, and a good air feeding
can improve both power and efficiency of engine itself.

Figure 1.12: Diesel converted to natural gas piston design

In [”Development of a Natural Gas Engine with Diesel Engine-like Efficiency

Using Computational Fluid Dynamics” published by Ahmed Abdul Moiz, Zainal
Abidin, Robert Mitchell, and Michael Kocsis in the Southwest Research Insti-
tute in 2019], conversion of diesel heavy duty engine into a natural gas one
is analysed. Main modifications consist in replacing injection system with an

10
ignition one through a spark plug installation in the combustion chamber; con-
verting common rail into a natural gas injection system; changing piston shape
for compression ratio, which is not a gasoline engine one but it can take benefits
of higher natural gas octane number. Increasing tumble fluid motion can be a
target of intake port modification.
In this study, both an exhaust gas recirculation (EGR) analysis and a computa-
tional fluid dynamics (CFD) analysis were carried out for the two piston shape
illustrated in figure 12. ”Stock piston” geometry has an higher squish area ra-
tio (defined as ratio of area increasing squish fluid motion and the global one,
projected on the horizontal plane) that is going to enhance turbulent kinetic
energy for a faster flame propagation. ”SwR1” piston (the second one) has a
lower squish area but also lower wall surface, to contain heat transfer losses. In
CFD simulation, RANS κ − ε model has been used.

Figure 1.13: Flame developments in the two different combustion chamber com-
pared

As figure 13 illustrates, flame development is very different between the two

geometries, with upwards ”stock Piston” presenting a faster flame with respect
to downward ”SwRIv1 Piston”. Especially regarding turbulent kinetic energy,
”stock piston” top plot is significantly higher with respect to all the others,
having more turbulence where the flame is going to start. To match pressure
curve, spark timing must be anticipated in the second ”SwRIv1” geometry,
because of more important ignition delay due to lack of turbulence. If results
at close pressure maximum values are compared, second downward geometry
”SwRiv1” has a better efficiency with respect of high squish ratio one, because
of lower wall heat transfer losses.

11
 Figure 1.14: Different piston turbulent kinetic energy

In figure 14, turbulent kinetic energy of different piston geometries is plotted.

Turbulence intensity can not be well represented as an unique value, as far as
it can vary a lot depending on cylinder regions. What is important is having
turbulence where flame needs to propagate, to increase flame velocity (which,
in presence of turbulence gets quite higher with respect to laminar flame one).
Squish area presence play a very important role in it.

Figure 1.15: Different piston pressure curves

Despite appearances, ”SwRiv1” piston shape runs with an anticipated spark

timing with respect to ”Stock” one, because of higher ignition delay. Pressure
maximum values of the two simulated curves are very close, with differences
present only in flame propagation instants. Cumulative wall heat transfer in
figure 19 is very different, and it is due especially on piston area variations. In
the end, second ”SwRiv1” geometry presents higher efficiency even if there is a
slower flame, since lower heat transfer maximizes useful work once combustion

12
is completed.

Figure 1.16: Heat transfer development comparison between investigated piston

shapes

After such analysis, the paper continues to analyse exhaust gas recirculation
rate and effects on combustion, showing an efficiency improvement when EGR
is adopted. This section is not reported in this introduction chapter as long
as this thesis work does not deal with exhaust gas recirculation. In any case,
conversion between diesel engine and natural gas spark ignition one is not trivial,
and performances can vary a lot depending on piston geometry adopted.

1.2 Flame propagation combustion

Combustion is both a physical and chemical process through that fuel can oxidise
and then release energy. As shown in previous figures, it is characterized by
three different phases: ignition delay, turbulent flame propagation and then
combustion completes in external cylinder regions. Velocities of these reactions
influence engine performances, and they can be measured in different ways:
controlling velocity of reactants consumption
controlling velocity of oxidation products formation
controlling velocity of heat released by reactions
The third criterium is the one most able to summarize the whole combustion
process. Since reactions are performing in gas phase, combustion velocity is
maximum when fuel-air mixture is homogeneous, since there are not delays
correlated to physical fuel evaporation or its mixing with the oxidant. Velocity
of these chemical reactions depends on reactants concentration, but especially
on temperature. Therefore, it is possible to model reaction velocity wr through
an Arrhenius correlation:
dc −Ea
wr = = Cpn exp( ) (1.7)
dt RT

13
Where C, n and Ea are constants. Ea is representing activation energy needed
to overcome the energy barrier and hence to break intermolecular bounds of
the fuel, starting combustion process. This activation energy is determined
by slowest steps of oxidation chain reaction (the so called rate determining
steps) and it varies significantly with the fuel considered. In particular, reaction
velocity increases exponentially with temperature. A normal spark ignition
combustion is starting from spark plug location and then it propagates to the
remaining part of the cylinder, radially to most external regions. Therefore, it is
possible to distinguish an area separating combustion products to fresh charge,
and flame front is then recognized. The flame front is composed by two different
regions: a preheating region, where fresh charge is just heated up but not enough
to fully perform combustion reactions; a second one where exothermic chemical
reactions associated to combustion are completed. It is possible to define a
laminar flame velocity wcl as the attitude of a certain air-fuel mixture to burn
more or less in a fast way. Therefore, flame velocity is proportional to reaction
velocity wr , but it will decrease with thermal diffusivity of the mixture itself.
s
√ λwr
wcl ∝ χwr = (1.8)
cp · ρ

The higher the thermal diffusivity (χ) of the mixture, the more energy is needed
to heat up a mixture portion, hence the slower the flame. In fact, it is possible
also to define flame thickness, which is increasing with thermal diffusivity:
r
χ χ
Sf l ∝ ∝ (1.9)
wcl wr

Flame velocity is maximum when mixture is slightly rich. There is a range

within flame velocity variations are low (0.1 to 0.2 m/s as order of magnitude),
but external to that regions it is possible to have flame extinction because there
is not enough fuel or oxidant.

Figure 1.17: Laminar flame velocities of different fuels at different equivalence

ratios

14
 In figure 17 some flame velocities of different fuels such as methane or gaso-
line are plotted. Methane maximum flame velocity value is very close to stoi-
chiometric air-fuel ratio. It is good behaving in lean mixture conditions (with
φ lower than 1), but bad performing when mixture is rich. Gasoline and isooc-
tane maximum values are more stable, with maximum flame velocity values in
rich mixture conditions (φ equal to 1.2), but when engine is running lean the
flame is significantly slowed down. In particular, flame velocity varies with fresh
mixture temperatures and pressures, as explained by equation below:
T 2 p
wcl = wcl0 · ( ) · ( )−0.25 (1.10)
T0 p0

With wcl0 reference flame velocity is indicated, and with T0 and p0 reference
temperature and pressure conditions corresponding to wcl0 . This experimental
formula is good to perform a comparison between different thermodynamic con-
ditions: a temperature increase speeds up the flame, since chemical reactions are
faster and mixture thermal diffusivity increases also; a pressure increase has a
negative effect on flame velocity, since it is not appreciably influencing chemical
reactions but energy and mass transport are reduced.
Turbulence contributes to speed up the flame. Defining turbulence intensity u’
as mean square value of velocity fluctuations, it is possible to relate turbulent
flame velocity wct to intensity of these fluctuations, through the expression

wct = wcl + awcl (u0 /wcl )b (1.11)

From a physic point of view, turbulent micro vortices of a comparable scale with
flame front intensify energy and mass transfer, hence increasing wct . Constants
a and b are experimentally determined. A b typical value reported in [2] is 0.7,
highlighting the almost linear influence of turbulence intensity on flame velocity.

1.3 Knock and spontaneous combustion

Combustion in spark ignition engines is considered abnormal when it is not
caused by flame propagation from spark plug. When high local pressures and
temperatures are reached, a portion of fresh charge can autoignite, releasing
huge amount of heat. When it happens, two different flame fronts coexist,
affecting engine performances.

Figure 1.18: Knock illustration

In figure 18 knock is illustrated: some fresh hydrocarbons ignite sponta-

neously far from burnt gas region. Therefore, two different flame fronts can be

15
recognized in the cylinder: a first one related to normal combustion, a second
one starting from knock origin. This kind of heat release can generate dan-
gerous pressure waves, propagating in the whole combustion chamber. When
knock occurs, a metallic noise can be heard, and it is originated by high fre-
quency pressure oscillations. Knock is mechanically dangerous for the engine.
Even autoignition can be modeled through an Arrhenius correlation: it is in-
fluenced by cylinder pressure, temperature and mixture fraction, but also by
time available for reactions to perform. It is possible to define an autoignition
delay τa , as the time interval between the moment when mixture reaches certain
pressure and temperature conditions and the moment in which an appreciably
heat release by autoignition is present.

Ea
τa = Ap−n exp( ) (1.12)
RT
Autoignition delay expression and reaction velocity ones are very similar: they
present the very same coefficients, but signs of exponential functions are op-
posites. Introducing autoignition delay concept allows to avoid modelling all
chemical reactions associated to combustion, providing a very practical time
value. It measures the time duration of the whole intermediate process leading
to mixture autoignition.

Figure 1.19: Autoignition delay

As displayed in figure 19, after a certain time (τa ), there is a considerable

pressure rise because of heat released by autoignited mixture. In this field,
methane and gasoline behave differently. In air-gasoline mixture, the whole au-
toignition delay is divided in two smaller parts: after a first delay τ1 , a cold
flame (with low heat release) appears and cylinder pressure slightly increases;
after a second delay τ2 , fuel oxidation performs completely and a huge amount
of heat is released. Since this thesis work is focused on natural gas combustion,
just an unique autoignition delay τa is considered, as in the air-methane mixture
case.
Pressure in spark ignition engines is almost homogeneous, hence pressure rise
due to flame propagation burnt gases is also involving fresh mixture. For what
concerns temperature, it can not be homogeneous because flame front clearly
divides burnt mixture region from unburnt one. In burnt mixture region, very
high temperatures (over than 2500 K) are reached; in the unburnt mixture re-
gion, temperature rise is due to compression and high temperature gradient
because of combustion, hence huge temperature differences are present in the

16
cylinder. Because of a combination of both the effects, unburnt fresh tempera-
ture maximum value is reached slightly after top dead centre, before the whole
mixture is considerably expanded.
As a generic rule, the more spark timing is anticipated, the more severe ther-
modynamic conditions are reached, making knock presence more probable.

Figure 1.20: Spark ignition combustion temperature chart highlighting differ-

ences that can be present in the whole domain considered

Knock is defined as ”spontaneous ignition abnormal combustion” because

it is very violent causing huge pressure rise. When it appears, pressure waves
inside combustion chamber propagates and get reflected by cylinder walls, caus-
ing oscillatory pressure rise behavior. In real engine installation it is detected
by an accelerometer, and as a corrective action it delays spark timing for the
following cycle. There are two kinds of knock: light knock and heavy knock.
Light knock can be present in a lot of cycles, leading to not strong pressure rise;
heavy knock is something to avoid otherwise engine walls (specially piston) risk
to be damaged.
Knock strength and importance are not depending just on pressure and tem-
perature, but also on mixture fraction. In this natural gas powered engine case,
fuel-air mixture is homogeneous and once a fresh mixture portion is going to
burn spontaneously, it is dropping all the other. If charge is not homogeneous,
reaction rate is not uniform inside the combustion chamber and it can enhance
or delay knock presence depending on local mixture fraction values.

17
 Figure 1.21: Pressure curve in knock presence

In any case, homogeneous charge compression ignition combustion is very

fast. If sonic waves are avoided, engine could theoretically take benefits of this
combustion kind. This is the main concept behind new and innovative com-
bustion modes, even to have a better exploitation of fuel properties. As it will
be defined in section 1.4, octane number is indicating autoignition resistance of
the fuel: diesel one is between 15-20 because spontaneous ignition is something
wanted when it is used; commercial gasoline one is 95 (or 98 sometimes); natural
gas one can vary a lot, but it is in range 120-130. Having a lot knock resistant
fuel spontaneously ignited allows to increase compression ratio in an appreciable
way, to increase efficiency, but in stoichiometric homogeneous mixture context
heat release would be huge, and pressure waves would mechanically destroy the
engine itself.
HCCI (homogeneous charge compression ignition) combustion is something that
has been investigated a lot for developing future internal combustion engines.
High octane number fuel (gasoline or natural gas) is compressed to sponta-
neously ignite, in similar way of what has been described above.

Figure 1.22: HCCI: Homogeneous charge compression ignition combustion

As figure 23 shows, conceptually HCCI is something in between a diesel

engine (where no spark plug is needed) and a gasoline engine (since a spark
ignition fuel is used), and the result is a compression ignition of fuel air mixture.
No flame is present or propagates, but this kind of technology offers a chance
to have fast combustion. For a conceptual point of view, it is something similar
to knock because it is also causing very high heat release in a very short time,
but this kind of heat release is controlled by changing air-fuel mass ratio: HCCI

18
engine runs lean, meaning inserting more air mass than the reaction would need.
Equivalence value φ is defined as:

mair,stoichiometric
φ= (1.13)
mair

meaning that φ = 1 when air mass is exactly stoichiometric one, lower than one
if there is any air excess. In this case, HCCI runs lean with φ between 0.4-0.5,
in order to have a lower heat release spread in the combustion chamber and
avoid huge pressure sonic waves. Another great advantage in running in lean
condition is that lower maximum cylinder temperatures are reached. As figure
23 illustrates, low equivalence ratio and maximum temperature values can avoid
NOx production and oxidise CO formed in combustion process.

Figure 1.23: HCCI pollution chart, highlighting the advantages of a lean com-
pression ignition combustion and differences with a diesel one

The huge drawback for HCCI engines is combustion control: in a spark

ignition engine, combustion can be started by spark plug electronic control; in a
diesel compression ignition engine, fuel octane number is so low that as soon as
liquid fuel evaporates, it autoignites. In this new combustion concept, it is very
hard to predict exactly when combustion is going to take place, making engine
control very difficult. To limit these difficulties (and to be sure that no misfire
is going to present), compression ratio could be increased, but then mechanical
stresses on engine itself would become huge and structurally unbearable.
These are the reasons why innovative combustion concept has been changed,
passing from a HCCI engine (where just compression ignition is present) to
a controlled autoignition engine, where a spark plug can control autoignition
timing.

19
 Figure 1.24: CAI: controlled autoignition combustion

Main combustion ignition mode is always compression ignition one, but in

this configuration there is a device able to control it and decide when it is taking
place. In figure 24, both spark plug and fuel injector are displayed, to illustrate
working principle of spark assisted compression ignition (SACI) engine; but it is
not the only way to control spontaneous ignition of high octane number fuel, as
far as also dual fuel technology is at the state of the art investigated: it consists
into injecting few amount of diesel fuel in homogeneous high octane number
charge combustion chamber to start autoignition. Once diesel evaporates, com-
bustion is going to begin and then involving the whole charge.
Some both experimental and CFD papers have the purpose to investigate these
new combustion modes, as far as they can be promising for increasing combus-
tion efficiency and reducing pollutant emissions at the same time. In [11], a
comparison between spark assisted compression ignition technology and a dual
fuel diesel-natural gas one is performed. A rapid compression and expansion ma-
chine was used to simulate just compression and expansion stroke, in order not
to damage the engine if knock occurs. They consider a very high compression
ratio (20:1) and very low engine speed (800 rpm) for heavy duty application.

Figure 1.25: Rapid compression and expansion machine

20
 Since just two strokes are experimentally performed and there is not a con-
stant engine rotational speed, swirl motion must be generated though a blower,
connected with both intake gasses and exhaust ones.

Figure 1.26: RCEM Blower adopted to generate turbulence

The very same rapid compression and expansion machine has been used
for spark assisted compression ignition combustion and diesel pilot compression
ignition combustion, with diesel pilot fuel mass corresponding to 2% of homo-
geneous mixture low heating value. Some different investigations are carried,
as shown in tables below, where different air excess ratio or piston heads were
considered.

Figure 1.27: SACI experimental setups

21
 Figure 1.28: Diesel pilot injection experimental setups

Therefore, experiments were performed to investigate both pressure rise and

efficiency. Since compression ratio is very high (20:1 is commercially unrealistic
for high octane number fuel at the state of the art), pressure and heat release
curves are not discussed in this introduction chapter, but efficiency plot to
understand if these new combustion modes have some future potential.

Figure 1.29: Experimental spark assisted efficiency

For what concerns spark assisted, combustion efficiency reaches and over-
comes 40% at very low engine speed. Maximum value is reached in presence of
stoichiometric air-fuel ratio and 20% of exhaust gas recirculation. When just
air excess (λ = 1.3, defined as ratio of actual fuel air ratio with stoichiometric
one) is considered, misfire is present since experimental mixture is too lean to
burn, probably also explained by lack of turbulence in experimental machine.

22
 Figure 1.30: Experimental diesel pilot efficiency

When diesel pilot injection is considered, efficiency gets quite higher with
respect to spark advance. Diesel pilot injection is very close to top dead cen-
tre, enhancing how fast its ignition taking place; in spark assisted efficiencies
reported in figure 30, spark advance needed to be anticipated up or even further
of -40◦ (ATDC: after top dead centre), meaning spark assisted fuel air mixture
is very difficult to be ignited in experimental engine considered.
This thesis work is focused on spark assisted compression ignition engine, in or-
der to verify its feasibility and efficiency improvements. Engine is a heavy duty
diesel converted with a compression ratio of 11.7, similar to the one analysed
in [12]. Before starting with engine analysis and investigation, fuel has been
discussed, since natural gas has not an octane number or heating value imposed
for commercial reasons as gasoline, but its properties can change depending on
its composition.

23
1.4 Fuel and pollution: Natural gas
Hydrocarbon combustion can be described by the following chemical reaction:
3n + 1
Cn H2n+2 + O2 − > nCO2 + (n + 1)H2 O (1.14)
2
This is valid for every hydrocarbon combustion (diesel, gasoline, natural gas)
at its completeness. On the rightward member, just combustion products are
present but not any pollutants, as far as they are a product of uncomplete
reactions. In fact, global chemical reaction associated to combustion is the
results of a multiplicity of lower ones having different reaction rate and taking
place in different condition. For instance, first methane oxidation step can be
summarized by following figure.

Figure 1.31: Methane first oxidation reaction

Different kinds of products can appear during these processes, as a result

of incomplete combustion. Most of them later oxidize once combustion is com-
pleted, in order to form just CO2 and water; but this is not always possible,
because of very fast pressure and temperature variation at which an internal
combustion engine is exposed to.
Principal internal combustion engine pollutants are:

CO, formed in stoichiometric or reach combustion due oxygen lack

soot, present when liquid or solid fuels are used

NOx, produced at very high temperature in oxygen abundance

Unburnt hydrocarbons, always present in different percentages

Quantities and kinds of pollutions production strongly depends on thermo-

dynamic cycle and fuel used: diesel engines usually have high soot and high
NOx emissions because of difficult fuel evaporation and very high temperature
reached, but CO produced in combustion process oxidizes in very lean environ-
ment; spark ignition gasoline engine produces all the three kinds of pollutant in
different quantities, since it is high volatility fuel but working at stoichiometric
charge.
A common strategy to reduce internal combustion engine pollutant emissions in
real drive behavior is installing an after-treatment system, therefore something
able to recombine the three kinds of pollutants to form N2 , CO2 and H2 O.

24
 Figure 1.32: Diesel particulate filter

Figure 1.33: Urea injection

In diesel engines usually both a diesel particulate filter (DPF) and urea
injection system are installed. The first one aims to filter soot particles, in such
a way that they are trapped in the filter itself and not released in atmosphere;
the second one makes selective catalytic reaction happen, a chemical reaction
that presents ammonia and NOx as reactants and N2 and water as products.
These systems development contributed in reducing diesel engine pollutants,
even if they (specially DPF) introduce a pressure loss in correspondence of
exhaust pipe.

Figure 1.34: Catalytic converter for SI engine

For what concerns spark ignition engine, aftertreatment system is very dif-
ferent. A particulate filter is present again in gasoline fuelled ones, even if it
is smaller because generally they produce a lower soot quantity; to make pol-
lutants chemically react, catalytic converter (working with noble materials) is

25
placed in aftertreatment system. This chemical device needs CO, NOx and
unburnt hydrocarbons to work, to recombine as much as possible.

Figure 1.35: Catalytic conversion efficiency

This concept is explained in the conversion efficiency plot, in figure 35: for
very rich mixtures, NOx are almost all converted (also because a low quantity
is produced) but hydrocarbons and CO conversion efficiency is very low; for
very lean mixtures, conversion efficiency of carbon monoxide and hydrocarbons
is quite high also because they are oxidizing with the remaining oxygen, but
NOx can not be converted. Therefore, SI engines are constrained to run in
stoichiometric mixture condition if three way catalyst is installed, to maximize
its conversion efficiency. Fuel consumption increases because efficiency is higher
when engine is running slightly lean, but pollutant emissions increases signifi-
cantly.
Natural gas is a very promising fuel for future internal combustion engines, such
as it is an already gaseous fuel and no soots are produced in combustion process.
In any case, a catalytic converter is installed to convert carbon monoxide and
nitrogen oxides, working with unburnt hydrocarbons from combustion chamber.
These aftertreatment systems (specially spark ignition ones) introduce a small
pressure loss having an effect on cylinder performance, hence target for future
internal combustion engine is to perform an always cleaner combustion. In
these terms, developing natural gas fuelled spark assisted compression ignition
or diesel pilot compression ignition can be a future opportunity for automotive
industry.
A natural gas composition analysis has been carried out since it can have a
strong influence on fuel properties and octane number. Octane number is an
autoignition indicator, assuming the value of 100 if the fuel behaves as isooctane,
higher if it is more knock resistant. Natural gas usually has an octane number
between 120-130, and it can change between one composition and another.

26
 Hydrocarbon Ch. formula MW [g/mol] LHV [MJ/kg] Octane number
methane CH4 16,04 50,0 130
ethane C2 H6 30,07 47,622 108
propane C3 H8 44,1 46,35 103
butane C4 H10 58,12 45,75 91

Table 1.1: Natural gas main hydrocarbons

As showed by the table, in natural gas both ”light hydrocarbons” (such as

methane and ethane) and ”heavy hydrocarbons” (such as propane and butane)
are present, and it can be distinguished dry natural gas (poor in heavy hydro-
carbons) from wet natural gas (rich in heavy hydrocarbons). Methane is a very
knock resistant fuel, but butane and propane are not: as a consequence, their
presence can lead the mixture to ignite spontaneously or even knock. Octane
number is an experimental indicator (it can even be distinguished between re-
search and motored octane number depending on testing fuel temperature), it
can not be arithmetically computed but an experimental or simulation setup
is necessary. Especially in homogeneous charge, as soon as a small portion au-
toignites, it will easily drop all the others, no matter if spontaneous ignition
principle is physically given by propane or butane contribution. Hydrocarbons
heavier than butane are uncommon in commercial natural gas.
Since it is already in gaseous phase, also nitrogen and carbon dioxide can be
present, which are inerts in combustion process, not reacting.

Inert Chemical formula Molecular weight [g/mol] LHV [MJ/kg]

nitrogen N2 28 0
carbon dioxide CO2 48 0

Table 1.2: Most common inerts present in natural gas

Inerts presence tends to decrease fuel octane number as far as oxygen chem-
ical activity decreases, and they give no heating value contribution. There are
not any commercial rules constraining natural gas composition for any applica-
tion, hence a composition review has been taken.

Figure 1.36: Natural gas composition ranges

27
 Figure 1.37: Consulente energia natural gas compositions

Figure 1.38: Unirc lecture natural gas compositions

Figure 1.39: SNAM natural gas compositions

A lot of different natural gas composition sources were considered, and all
agree that Italian (sometimes indicated as ”Nazionale”) natural gas is reacher in
methane with respect to the others. Compositions are given in volume or mole
percentages, and when converted into mass ones heavier hydrocarbon values rise
more than lighter ones. For combustion investigation, natural gas composition
must be fixed, and then any variation is addressed to sensitivity analysis. There-
fore, composition chosen was Nord European one provided by SNAM source,
as far as it is quite rich in heavy hydrocarbons and both propane and butane
percentages are provided.
Natural gas composition has been converted into mass percentages, and then
following values have been derived.

28
 Chemical formula mass percentage
methane CH4 81.9%
ethane C 2 H6 7.65%
propane C 3 H8 2.27%
butane C4 H10 1.22%
nitrogen N2 3.77%
carbon dioxide CO2 3.19%

Table 1.3: SNAM North Europe natural gas composition in mass percentages

As shown in table above, chosen natural gas composition is quite rich in

methane, but also propane, ethane and butane mass percentages are important.
Taking in consideration a lower octane number fuel for SACI development leads
to have a higher compression ignition control, despite flame propagation when
compression ignition is not feasible or unwanted. Inerts are something less than
7% of the whole mixture.
In the whole thesis work, a multiplicity of natural gas composition is going to
be considered for instance to validate solver used or even to provide an engine
performance sensitivity analysis. Target is, as explained before, increasing en-
gine efficiency and at the same time reducing pollutant emissions.
In the end, natural gas can be considered an easy fuel to be stored, for engine
conversion from diesel commercial ones and easy for refuelling since fuel station
are currently available.

1.5 Numerical simulations

For internal combustion engines design, numerical simulations are very useful
since a lot of conditions can be tested with much lower economical efforts and
without risking of damaging an experimental engine. CFD, computational fluid
dynamics, is used to predict flows behavior in the whole engine, from breathing
pipes to exhaust ones, including combustion chamber. In any case, modelling
flow ducts is very different with respect to simulating what happens in combus-
tion chamber, hence two different tools are used:
- zero or one dimensional software for flow management
- three dimensional software for combustion chamber design

Figure 1.40: Gasdyn interface

Figure 40 reports an example of a mono-dimensional fluid dynamics code. Il-

lustrated interface is from Gasdyn R software, that can describe how inlet and

29
outlet flows are managed between different pipes or in presence of some devices,
such that compressor, turbine or even silencer. It can compute global engine
performances, but it can not model the combustion chamber. It is not a station-
ary fluid dynamics model, considering compressible flows neither adiabatic or
isentropic. Mass, momentum and energy conservation equations are written in
just one dimension, implying a much lower computational time needed if com-
pared to a 3D CFD software. To model friction forces acting on the fluid is very
important, since it can not be properly considered viscous because this assump-
tion would imply a two-dimensional velocity distribution, but tangential friction
stress is represented by following equation, function of Reynolds number:

1 0.25
τw = f ρu2 with f= k 5.74 2
(1.15)
2 [log10 ( 3.7D+ Re0.9 )]

These kinds of codes are very appropriate to give good fluid pressure and
temperature values almost everywhere, but they can not consider properly tur-
bulence or other 2D or 3D phenomena, that in combustion are quite relevant.
This is the reason why when designing combustion chamber and engine ge-
ometry a 3D fluid dynamics code must be used. Software chosen has been
OpenFOAM R , because it is an open source and easy modifiable for author
preferences.

Figure 1.41: 3D simulation example

OpenFOAM is a C++ based open source software. At the state of the

art, version 6 is available for free download, but version used is 2.2.x owned
by Politecnico di Milano, since it presents some useful modifications. LibICE,
developed by Politecnico di Milano in Energy department, has some supplemen-
tary codes and utilities used to realize this thesis work.
CFD simulations can require huge computational time, especially if valve mo-
tion (opening and closing) and all strokes are both modelled and simulated. To
make numerical simulations also suitable for commercial laptops, not the whole
cycle is simulated, but just compression, combustion and expansion.

30
 Figure 1.42: OpenFOAM mesh example

Figure 42 illustrates an OpenFOAM case opened in ParaView R interface.

Cylinder volume is divided in a lot of smaller ones (over 6000) composing mesh,
which has to be numerically suitable for the case studied, not presenting any
huge discontinuities, non orthogonalities or some more characteristics compro-
mising mesh quality. Simulations starts from -175◦ (intake valve closing) ending
at 124◦ (exhaust valve opening), with variable time steps to decrease computa-
tional time. Not the whole combustion chamber is modelled: as figure 43 shows
from the top view, just 2◦ of arc of the whole cylinder geometry is meshed,
imposing radial symmetry for what concerns the rest of the volume.

Figure 1.43: OpenFOAM mesh top view

Figure above has been taken from Mesh generator for Diesel Combustion
Chamber geometries developed in Politecnico di Milano in the Energy Depart-
ment, by Professors Augusto Della Torre and Tommaso Lucchini. This has been
an useful tool for mesh editing, in order to simulate case in OpenFOAM envi-
ronment.
Last software used is a CAD (computer aided design) one, Autodesk Inventor R
(student license), which has been essential for geometry design.

31
 Figure 1.44: Autodesk Inventor interface

Using this approach is possible to design a completely different engine ge-

ometry through Autodesk Inventor, editing the mesh using Automatic Mesh
Generator and simulating the case in OpenFOAM environment. To begin, a
numerical code able to consider both flame propagation and compression igni-
tion combustion needs to be edited.

32
Chapter 2

Numerical solvers

Computational fluid dynamics (also known with the acronym CFD) can be a
very important design tool for fluid machines. Thanks to very powerful com-
putational resources available nowadays, very complex systems can be entirely
simulated with excellent results. In this way, it can significantly reduce eco-
nomical efforts in the design process, especially for internal combustion engines
where some extreme conditions have to be investigated. Testing the engine can
imply the risk of damaging it, while building a computational simulation no
risk to compromise any useful machine is present. Investigating modern com-
bustion modes, heavy knock or too severe pressure rise can occur, leading to
concrete risk of damage any experimental setup. Another great benefit of CFD
is the possibility to build in-house codes dedicated to different research targets,
hence perfectly suitable to investigated application. A physic and engineering
knowledge of the flow studied is needed even when writing the code, as far as
choosing a wrong turbulence model or numerical scheme can compromise sim-
ulation results. For all this reasons, OpenFOAM R has been chosen to build
the CFD code, since it is an open-source software with many pre-implemented
capabilities.

Figure 2.1: CFD gas exchange example

CFD domain is the region being simulated. In this thesis work (just focused

33
on power cycle), domain of interest is the whole cylinder, divided in different
finite volumes by different cells. With the term ”mesh” the whole set of cells
is indicated. Mesh has already to take into account how the solution should be
like, since it needs to be refined in presence of critical areas, while it can be
courser elsewhere. Boundaries separate mesh to the surrounding environment,
which can actually influence fluid physical quantities if the system can not be
considered isolated but just closed: this in internal combustion engine is the
case of wall heat transfer, heat losses decreasing engine performances. Therefore,
some boundary conditions need to be defined, in order not to model also external
regions to determine energy transfer with the investigated system. Especially
when complex geometries are composing the domain, mesh represents a first
description of the domain itself. In this thesis work some CAD geometries are
going to be simulated, hence a different mesh is generated for any interested
geometry. As reported in [14], ”the main role of the volume mesh is to capture
the 3D geometry. Cell should not overlap and completely fill computational
domain. A priori knowledge of solution is useful in mesh generation process, in
order to locate high resolution zones to capture critical parts, such as shocks,
boundary layer”.
Mesh grid is considered structured if it is very regular, unstructured if it is
identified by irregular connectivity. Mesh used is going to be hybrid, presenting
zones where grid is very regular and some others where it is very refined, such
that solver is able to consider fuel injection or sonic waves with higher accuracy.
Some indicators are commonly used in CFD environment, helping to have a
good mesh quality:
Cell equiangular skew :
Cell symmetry is investigated. It is defined taking into account minimum
cell angle and maximum one. If cell geometry is equiangular (perfect
square for instance), θmax and θmin are equivalent and skewness is 0, as
optimal case; if it is not, this indicator will get closer to 1. The lower it
is, the better for CFD calculations. Defining θe as angle for equiangular
cell (60◦ in case of triangular cell, 90◦ in case of quadrangular cell and so
on), mesh skewness is evaluated as expression below:
θmax − θe θe − θmin
Skewness = M AX[ , ] (2.1)
180◦ − θe θe

Figure 2.2: Cell equiangular skew

Cell aspect ratio :

Aspect ratio is very similar to equiangular skew, but referred to cell bound-
aries length. It is defined differently depending if mesh considered is trian-
gular or at least quadrangular, and in second case it is the ratio between

34
 the longest boundary side with the shortest one. Best aspect ratio is 1;
the higher it is, the worse for mesh quality. Expressions for aspect ratio
computation are the following ones:

R
AR = f T riangular (2.2)
r

max(Sidelength )
AR = others (2.3)
min(Sidelength )

In triangular aspect ratio expression, R is circumscribed circumference

radius, while r is the inscribed circumference one. In other cell geometries,
aspect ratio is defined as the ratio between longest and shortest cell side
length.

Figure 2.3: Cell aspect ratio

Mesh orthogonal quality :

It is a mesh orientation indicator. If every cell surface normal is pointed
to the centre of all the neighboring ones, it assumes the value of 1 as
optimal case. If cells are not oriented one to each other, mesh orthogonal
quality is decreasing and it can cause some numerical problems during
CFD calculations. Hence, defining fi as vector connecting centroid of
reference cell and centroid of one of its faces, and ci as vector connecting
centroids of two different cells, orthogonal quality is computed as reported
by expressions below.

OrthogonalQuality = M IN (EQ1; EQ2) (2.4)

Ai ∗ fi
EQ1 = (2.5)
|Ai | ∗ |fi |

Ai ∗ ci
EQ2 = (2.6)
|Ai | ∗ |ci |

35
 Figure 2.4: Mesh orthogonal quality

Mesh size ratio :

This last mesh indicator is to ensure that every cell side length do not
vary in a huge way if compared with neighboring ones. In any case, this is
not going to have any important effects in this context where, especially
in squish area, flow is oriented mainly to cylinder centre. Defining as
∆xM AX and ∆xM IN two different cells length in one chosen direction,
size ratio is computed as:

∆xM AX
SR = −1 (2.7)
∆xM IN

Figure 2.5: Mesh size ratio

First simulations are going to increase fluid flow physic knowledge, especially
if it has been just hypothesized a priori when generating the mesh. Therefore,
an iterative procedure is commonly applied to improve mesh quality, since after
some simulations regions of interest are more clearly defined.

Mesh Motion

There are some applications where domain geometry moves during the calcula-
tion. This is the case of internal combustion engines, where the piston is rising
from bottom dead centre to top dead centre. Moving deforming mesh algorithm
will allow the domain to change its shape during the simulation and preserve
its validity, but structure of internal mesh needs to remain unchanged. There

36
are different ways to adapt the mesh to domain geometry variations. ”Shape
change” is the most intuitive one, since mesh in boundary layers change their
shapes during calculations. This is not very flexible, since it is well performing
when changes are very low, but if boundaries deformations are extreme high
discretization errors can arise. In internal combustion engines, boundaries are
fixed except for the piston which is vertically moving in just one direction; hence,
adding or removing computational cells to accommodate boundary deformation
is an easier and more suitable mesh motion strategy.

Figure 2.6: Mesh motion layers removal example

As shown in figure 50, mesh structure remains the same, but number of
mesh layers change considerably. There are some cases where both kinds of
described mesh motion can be used, if for instance also valve opening or closing
is modelled. In this thesis work, since it is just focused on power cycle, mesh
layers are just added or removed and no valve motion is simulated.
If all these mesh quality indicators are globally analysed, the more the mesh is
uniform, the lower numerical issues can arise because of a poor mesh quality.
In the context of in-cylinder CFD simulations, a compromise needs to be found
between a good mesh suitable for engine geometry and computational time, as
far as all equations that will be described are solved for every cell at every time
step.

Transport equations

Transport equations of different quantities maintain actually a very similar

structure, hence before going deep in every physical detail, a general analy-
sis has been carried with a specific quantity φ. This φ has not any physical
meaning, but it is useful for a numerical analysis that can explain pressure,
temperature, density or any other fluid dynamics quantity equation.

37
 Figure 2.7: Specific quantity flows for each cell

For a generic domain Ω, φ can change its value only if fluxes and/or source
terms are present:
Z Z Z
∂
φdΩ = − F luxes dS + SourceT erms dΩ (2.8)
∂t Ω Surf Ω

Fluxes can be two kinds of: convective if interested quantity φ is transported

by the fluid: Fc = φU ; diffusive if interested quantity φ is transported by its
gradient: Fd = −k∇φ.
For what concerns source terms, they can be surface related such as wall heat
transfer, or volumetric such as heat released by combustion in certain cells.
Every surface term must be integrated along respective cell face, while volume
ones along the whole cell volume. Hence, the previous equation becomes
Z Z Z Z Z
∂
φdΩ + φU dS = k∇φdS + QV dΩ + Qsurf dS (2.9)
∂t Ω Surf Surf Ω Surf

Time derivative and convective flux terms are usually reported in the first equa-
tion member, while diffusive flux, surface source and volumetric source terms
are usually reported on the right side of the equation. Since some of the integrals
must be performed in volume domain and some others in surface domain, Gauss
divergence theorem has been applied to convert surface integrals into volume
ones. Therefore, the following equations are derived, with equation (25) being
the integral form of conservation equation and equation (26) the differential
(and most known) form of conservation equation.
Z Z Z Z Z
∂
φdΩ + ∇ · (φU )dΩ = ∇ · (k∇φ)dΩ + QV dΩ + ∇(Qsurf )dΩ
∂t Ω Ω Ω Ω Ω
(2.10)
∂φ
+ ∇ · (φU ) = ∇ · (k∇φ) + QV + ∇Qsurf (2.11)
∂t
Substituting φ with interested physical quantities and defining properly con-
stants and source terms, conservation equations used in CFD are derived.
Discretization
To make equations numerically manageable by a computer, a domain discretiza-
tion is necessary, hence the mesh. Every cell is recognized by OpenFOAM as
a small volume portion, with its well defined faces. For any specific quantity,
every cell has a φ value ideally located in the cell centre, and it is represented by

38
different colours when simulation results are displayed; but to evaluate fluxes,
surface values are needed for each cell face. Some discretization schemes and
criteria need to be introduced, hence every term of transport equation is anal-
ysed separately.
Cell interested by equation is indicated as ”P”, then neighbouring cells such
as northwards (”N”), eastwards (”E”), southwards (”S”) and westwards (”W”)
can be easily recognized, as shown in figure 52.

Figure 2.8: Cell references

Time derivative term is referred to central cell value. Euler forward scheme
is used to discretize this term, and it is the most intuitive choice since time is
just moving forward. Therefore, φtp is the value in the cell P of the previous
iteration (known), while φt+∆t
p is the one at the current time step, real unknown
of the equation. Time derivative can hence be written as:

φt+∆t − φtp
Z
∂φ p
dΩ = · Ωcell (2.12)
Ω ∂t ∆t

Subscript ”p” is referred to cell position, superscripts ”t” and ”t + ∆t” to the
different time instants considered.
Volumetric source term si discretized depending on the nature of the source.
Assuming for simplicity a constant intensive source term σ, it has just to be
multiplied by cell volume.
Z
σdΩ = σ · Ωcell (2.13)
Ω

When considering flux terms, just applying an Euler discretization scheme can
be not enough. A different flux term must be evaluated for every face of P,
hence different φi (ideally located in each face centre) need to be computed to
perform convection term discretization.
I NX
f aces
φU dS = (φi U ∗ Si ) (2.14)
Surf i

39
A method to compute surface value φi must be defined. Central difference
scheme is a second order method, easy to implement with high precision. If φN
is referred to north cell and φn corresponds to face shared by cells P and N,
applying central difference scheme φn is computed as in the equation (30)

φn = fx · φP + (1 − fx ) · φN (2.15)

Where fx is a value between 0 and 1 taking into account distances between cells
centres and faces.
nN
fx = (2.16)
PN
As it can be deduced by equations above, φn is not influenced by flow direction,
hence it assumes the same value if the flow is oriented northward or southward.
This can result in a problem when high velocity flows are considered, because
the scheme is not able to recognise direction of the flow or even strength of
convection relative to diffusion, leading to instability when Peclet number is
high.
ρu
Pe = (2.17)
k/δx
As described in the equation above, Peclet number is a measure of the rela-
tive strengths of convection and diffusion. To apply central difference scheme,
Peclet number must be less than 2, and it is not the case when dealing with
in-cylindrical simulations.
Therefore, Gauss linear upwind scheme has been used, since it is a stable second
order scheme.

φn = φP + λP −N (φS − φP ) if u · nn > 0 (2.18)

φn = φN + λN −N N (φN N − φN ) if u · nn < 0 (2.19)

With ”NN” cell northwards to north cell is indicated. λ considers distances
between interested cells, so it is defined by faces and cells centroids position.
xn − xP xn − xN
λP −N = or λN −N N = (2.20)
xS − xP xN N − xN
Using this approach, φn can assume different values depending if the flow is
oriented to north direction or to south one. Transportiveness property is always
guaranteed and scheme is hence stable. Discretization formula of the convection
term finally results:
Z
if u · nn > 0 φU dS = (φP + λP −N (φS − φP ))(U ∗ Sn ) (2.21)
Nsurf
Z
if u · nn < 0 φU dS = (φN + λN −N N (φN N − φN ))(U ∗ Sn ) (2.22)
Nsurf

These formulae have to be applied for every direction.

Diffusion term is easier to be discretized, since φ gradients can be evaluated
from different cells centre values.
fX
aces
φi − φP
I
k∇φ · dS = (k · · Si ) (2.23)
Surf i
d

40
With d as the distance between cell centres.
Applying different discretization schemes for every equation term, equations can
be handled by a computational machine and then they can be solved.

Solving equations

After writing transport equations for every cell, a linear system is obtained:

[A][φ] = [R] (2.24)

Where [A] is the matrix coefficient, [φ] value of φp of every cell and [R] is the
right hand side. [A] is potentially very big, since it is a square matrix of order
NxN, where N is the number of the cells composing the CFD domain. Therefore
inverting [A] matrix to solve the equation would be computationally expensive,
even occupying a lot of memory, and it is not practically possible to have an
exact solution of the system.
At contrary of directs methods, iterative procedures are not so computationally
demanding, and they can be managed by a computational machine in a more
reasonable time. It works by starting with a guessed solution to be improved
during the procedure. Guessed solution is usually the one obtained when solving
previous time step, and then a relative tolerance has been imposed to define
convergency.
φi,n+1 − φi,n
RelT oll = (2.25)
φi,n
With lowercase ”n” indicating number of iterations performed. If this relative
tolerance is less than a pre-chosen value (10−9 for pressure equation, 10−8 for
all the others) equation is considered solved, and a solution very closed to the
exact one is obtained.

Courant number

Courant number is a very important indicator for numerical convergence: it is

defined as the ratio between fluid velocity and time step as numerator and mesh
length as denominator.
u∆t
Co = (2.26)
∆x

Figure 2.9: Courant number representation

41
 This non-dimensional number is essential to define or correct time step before
starting the simulation: on the numerator length covered by the fluid in one time
step is represented. By a Lagrangian point of view, a Courant number equal
to one means that every fluid particle runs through one mesh size. If Courant
number is higher, space covered by a fluid particle is higher than a mesh size. If
it happens, a fluid particle can bypass neighbouring cells, causing instability and
divergence of the equations when bypassed cell is considered. Fluid velocity u
can not be imposed a priori since it can be the result of previous iterations; mesh
length ∆x and time step ∆t can be numerically imposed or modified, matching
convergence criteria. The courser the mesh, the larger the time steps can be
imposed, both actions decreasing computational time despite result accuracy;
finer the mesh, lower time steps need to be imposed and then computational
time can increase significantly. This numerical behavior must be considered
even in mesh editing process.

Turbulence

Turbulence is a very complex phenomenon to be modelled, as far as it is both

random and cahotic. It is made by highly unstable vortexes, called eddies.

Figure 2.10: Turbulent field with eddies of different size

Eddies of different size have different characteristic velocities and timescales,

as far as they tend to dissipate their own kinetic energy. Reynolds number,
ratio between fluid inertia forces and viscous ones, is very important in CFD
environment and its formulation can be applied also on different eddy sizes:

ul
Re = (2.27)
ν
When Reynolds is 1 inertial forces are completely dissipated by viscous ones.
This happens only if eddy considered is very small, having very few kinetic
energy. Kolmogorov scale η is consequently defined, corresponding to smallest
eddy present in the fluid, the one that has a Reynolds number equal to unity.
For Kolmogorov scale eddies, they dissipate all the kinetic energy they have

42
with a consequent temperature increase.
Kolmogorov introduced the energy cascade concept, assuming that bigger ed-
dies transfer their kinetic energy to smaller ones. This introduces the energy
dissipation rate ε, a rate of energy transfer between different length scale eddies.

Figure 2.11: Energy cascade

Biggest eddy size is known as integral length, and it depends on environmen-

tal geometry. Figure above highlights that energy dissipation rate ε is related
to eddy size, and how large scale eddies have much more energy than smaller
scale ones. Solving a turbulence problem means solving length, time and veloc-
ity scales for each eddy, which is possible but extremely expensive in terms of
computational time: DNS (direct numerical simulation) has hence the purpose
to compute these scales eddy per eddy, but it is not used for industrial applica-
tions because of too huge computational and memory efforts required.
A turbulence model is able to compute turbulent kinetic energy region by re-
gion, but not to distinguish different eddies, and then neither their oscillatory
behavior, but they present huge computational advantages with respect to direct
numerical simulations. RANS approach (Reynolds Average Numerical Simula-
tion) is hence the most used one for engineering design purposes, as far as it
is the best compromise between accuracy and computational time available at
the state of the art. There are a lot of RANS turbulence models implemented
in OpenFOAM or in other CFD programs, but they are all based on defining
a turbulent viscosity, function of turbulent kinetic energy and turbulent dissi-
pation rate. Eddies velocity components are not axis per axis computed, then
when RANS turbulence intensity is displayed in 3D mesh there is not any vortex
representation, but just their kinetic energy predicted. LES (large eddy simula-
tion) can be a very interesting solution for CFD, as far as it solves numerically
just large eddies length scales (that have higher kinetic energy), while smaller
ones are solved using a RANS approach. At the state of the art, LES is very
interesting for fluid physical analysis, but in more complex scenarios such as

43
internal combustion engine combustion mode design it is prohibitive for com-
mercial purposes.

Figure 2.12: Numerical simulation differences

Looking at figure 56 it is possible to appreciate differences between a DNS

or a LES simulation and a RANS one: in the two upwards, the vortex is well
described, while in the last one just different kinetic energy values are predicted.
Reynolds average concept needs hence to be introduced. If velocity over time is
plotted, resulting chart is going to be the combination of two different contri-
butions:
a stable velocity given by designed engine operation
a fluctuating term given by turbulence, which varies cycle per cycle
In the figure 57 displayed below, velocity is composed by an average value (0
in the case shown) and some fluctuations caused by turbulence presence. Big,
intermediate and small eddies can hence be recognized looking at amplitudes
and frequencies of velocity oscillations. In any case, fluctuation behavior is
unpredictable because they can change velocity initial condition from cycle to
cycle.

Figure 2.13: Reynolds average

44
 By a mathematical formulation, this kind of analysis is not suitable just for
velocities, but also for any specific quantity φ.
0
φ=φ+φ (2.28)
0
where φ is the average component and φ the fluctuating one. If average of the
whole quantity φ is performed, what results is:

φ = φ + φ0 = φ + φ0 = φ + 0 (2.29)

While, as predictable, average quantity (φ) and average of its average (φ) are
the same value, φ0 is null by a statistical point of view since there is no reason
why fluctuations should always increase or decrease φ global value. Therefore,
all quantities fluctuating average are considered null a priori.
In first appearance, change introduced by Reynolds average is just conceptual,
but if two different quantities such as φ and ξ are multiplied (as it happens in
convective flux term), what results is the following equation:

φ ∗ ξ = (φ + φ0 ) ∗ (ξ + ξ 0 ) = φ ∗ ξ + φξ 0 + ξφ0 + φ0 ξ 0 (2.30)

φ ∗ ξ = φ ∗ ξ + φ0 ξ 0 (2.31)
While average fluctuation terms are null if considered alone, when two of them
are multiplied their influence can not be neglected. By a mathematical point of
0 0
view, it is easier to understand if φ and ξ had instant per instant the very same
fluctuating component: if there is a positive contribution, multiplication result
stays positive; if there is a negative contribution, multiplication result becomes
positive. Hence there is a term adding in conservation equation, as far as both
φ and U are composed by an averaged term and a fluctuating one. Conservation
equation becomes:
∂φ
+ ∇ · (φU ) = ∇ · (k∇φ) + QV + ∇ · Qsurf − ∇ · (φ0 U 0 ) (2.32)
∂t
Modelling turbulence in RANS approach is determining ∇ · (φ0 U 0 ) value. From
this point ahead, if fluctuated component is considered, apex is reported; if
averaged quantity is considered, overline is not reported.

2.1 Solver description

In the first subsection, CFD principles and transport equations have been de-
scribed, while in this one conservation equations as written in the solver are
analysed. Source terms can be defined differently depending on solver engi-
neering purpose, while time derivative, convection and diffusion terms are very
similar. Combustion modelling in a CFD code is not as easy as other fluid dy-
namic quantities, hence it has been dealt in two subsections apart, one dedicated
to spark ignition combustion, another one dedicated to spontaneous ignition
combustion.
mass conservation equation
∂ρ
+ ∇ · (ρ~u) = Ṡevap (2.33)
∂t

45
momentum equation
∂ρ~u
+ ∇ · (ρ~u~u) = ρg + µ∆~u + Ṡturb + ṠU,spray (2.34)
∂t

mixture fraction equation

∂ρZ µt
+ ∇ · (ρZ~u) − ∇ · ((µ + ) · ∇Z) = ṠZ (2.35)
∂t Sct

mixture fraction traces equation

∂ρZt µt
+ ∇ · (ρZt ~u) − ∇ · (µ + ∇Zt ) = ṠZt (2.36)
∂t Sct

These are the first equations to be solved. By the source term related to fuel
evaporation (Ṡevap ) and the one to momentum exchange of liquid droplets
(ṠU,spray ), the solver can consider also spray injection and liquid fuel evapo-
ration. In this case, since the engine is working with an already gaseous fuel
and with no direct injection, these two mass and momentum source terms will
be null. Since velocity is a vector, momentum equations to be solved in CFD
simulation are three, one for each axis.
For what concerns turbulence source term Ṡturb , the two equations κ-ε tur-
bulence model has been used. It is based on Boussinesq’s hypothesis, which
assumes an analogy between Reynolds stress tensor (the one related to turbu-
lence) and mean strain one.
∂U
Ṡturb,xy = −ρu0x u0y = µt (2.37)
∂y
Hence a turbulent viscosity µt is introduced and by determining this µt , mo-
mentum source term is derived. Turbulent viscosity is not function of any axial
coordinate. Different RANS models use different equations to compute µt .
In k-ε one, turbulent viscosity is function of turbulent kinetic energy and tur-
bulent dissipation rate.
k2
µt = ρCµt (2.38)
ε
As reported, µt depends on both turbulent kinetic energy (κ) and its dissipation
rate (ε). The solver is modelling all kinds of turbulence, from big eddies to small
ones, and large eddy simulation (LES) is never applied. Therefore, the solver
needs one equation for κ and one equation for  to compute turbulent viscosity:
κ: turbulent kinetic energy equation
∂ρk µt
+ ∇ · (ρ~uk) = ∇ · (µ + ∇k) + ρε + 2µt Sij Sij (2.39)
∂t σk

ε: turbulent kinetic energy dissipation rate equation

∂ρε µt ε
+ ∇ · (ρ~uε) = ∇ · [(µ + )∇ε] + (Cε1 Pk + Cε2 ρε) (2.40)
∂t σε k
costants such as Cε1 , Cε2 and Pk are calibrated on the specific application

46
About turbulent kinetic energy equation, the last term is equal to laminar kinetic
energy destruction term, but opposite in terms of sign; this means that if an
increasing of turbulent kinetic energy is wanted, fluid laminar kinetic energy
(K) must be destroyed. This is going to have some relevance in the end of this
work.
This kind of model is one of the most used in literature because it is both
reliable and stable for many kinds of fluid motion, and also not requiring huge
computational time with respect to others available at the moment.
After solving combustion equations (which are described in a separated sub-
section), there is a composition change of species inside the cylinder because of
chemical reaction associated to combustion. Hence, two enthalpy equations are
solved, one for global enthalpy h, the other one for unburnt gasses enthalpy hu :
enthalpy conservation equation
∂ρh Dp
+ ∇ · (ρ~uh) = ∇2 αef f h + + Ṡspray (2.42)
∂t Dt
unburned enthalpy conservation equation
∂ρhu ρ Dp ρ
+ ∇ · ρ(~uhu ) = ∇2 αef f hu + + Ṡh ,evap (2.43)
∂t ρu Dt ρu u
From these two equations, bunt and unburnt temperatures are derived. So also
energy equilibrium is reached and time step solver algorithm ends here.

2.2 Combustion modelling

As commonly done by CFD researchers, combustion chamber gasses are dis-
tinguished into burnt gasses and unburnt gasses, so by a computational point
of view there will be the ”burnt” and ”unburnt” temperature as well as the
”burnt” and ”unburnt” enthalpy.

burnt gasses: Tb and hb

unburnt gasses: Tu and hu
global quantities: T and h
With global quantities as the weighted average of the previous two. Modelling
combustion results in a quite difficult computational aspect, since a lot of re-
actions are involved. Methane combustion can be summarized in the following
chemical reaction:
CH4 + 2O2 − > 2H2 O + CO2 (2.44)
But these reported above are just the global reactants and products of the re-
action itself. Considering very low time steps, reactions involved are many (of
the order of magnitude of hundreds or thousands), and hence impossible to be
modelled in acceptable computational time nowadays.
In the introduction section, a global combustion reaction velocity and an au-
toignition delay were introduced using Arrhenius correlation, which was a prac-
tical value taking into account chemical reaction kinetics without model it. In
CFD solver, a progress variable has been introduced, indicating (as suggested
by the name) the progress of combustion process. Therefore, a variable c has
been defined such that:

47
if combustion has not already taken place, c = 0

if combustion is fully completed, c = 1

c
0 → 1

It is defined in such a way to have a monotonous trend as reactions proceed.

Progress variable must be connected to a physical quantity that is monotonic
with combustion progress, assuming a minimum value before combustion starts,
and having its maximum when combustion is fully completed. Therefore, choices
available are the following:

- species formation enthalpy at a reference temperature of 298 K

- sensible enthalpy of the mixture h = f(T, t)

- linear combination of fuel mass fraction

Species formation enthalpy is the chosen one for this thesis work. Enthalpy
values are not in the range [0;1] as progress variable has been defined, hence c
has been normalized (cnormalized ) in the following way:

h − hmin
cnormalized = (2.45)
hmax − hmin

Using equation 60, progress variable becomes both monotonous and in range
[0;1].
CFD system is a 5-species system formed by fuel, O2 , CO2 , H2 O and N2 : the
increasing of the progress variable is associated to an increasing of combustion
products, so CO2 and H2 O.
In the end, progress variable c has both a chemical and a fluid dynamic meaning,
because it is an indicator of reaction progress and of heat released by combustion
itself at the same time (once enthalpy balance has been performed), and it is
also a way to impose consistency between these two aspects.
Since two kinds of combustion are present, the solver needs two different progress
variables to establish the ongoing of both processes. Therefore, the solver defines
a priori:

c as an indicator of spark-ignition combustion

cf resh as an indicator of compression ignition combustion

This solver algorithm gives the possibility to model them in a completely differ-
ent way without generating inconsistencies or superpositions between the two.

2.3 Spark ignition combustion

Spark ignition combustion is modelled through the 2-equations Weller combus-
tion model.

48
 Figure 2.14: Flame interface

As described by the image above, there is a clear flame interface dividing

burnt gas region and unburnt one: in the burnt gas region the progress variable
is equal to 1; in the unburnt gas region the progress variable is equal to 0. In the
end, the flame represents chemical reactions associated to combustion, but in
this way it is just the flame to be modelled, not chemical reactions themselves.
To have numerical and computational advantages, equations are written respect
to a regress variable b:
b = 1 − cN ormalized (2.46)
∇b = −∇cN ormalized (2.47)
Another quantity to be introduced is the wrinkle factor Ξ, which in Weller’s
work has been derived by the ratio of the flame area per unit volume (Af ) and
the flame area per unit volume projected into the mean propagation direction
(As ). Starting with this first definition, the wrinkle factor can be reconducted
to the ratio of turbulent (Sturb ) and laminar (Slaminar ) flame velocities, as it
has been shown in [”The Development of a New Flame Area Combustion Model
Using Conditional Averaging”, published by H.G. Weller in 1993]:

Af Sturbulent
Ξ= = (2.48)
As Slaminar
Indicating with S the corresponding flame velocity.
In flame propagation combustion, turbulent flame velocity is much higher with
respect to laminar one, meaning having high Ξ values is something improving
combustion process.
Through mathematical analysis, two equations are derived:
∂ρb
+ ∇ · (ρ~ub) − ∇ · (µt ∇b) = ρ~uSu Ξ|∇b| (2.49)
∂t
∂Ξ 1 ∇|∇b|
+ Us ∇Ξ + DΞ = G + Ξn̂∇Ut n̂ − n̂∇Ut n̂ + Ξ(Ut − Us ) (2.50)
∂t Ξ |∇b|
In this two equation model, two important things must be highlighted: first,
the source term of the regress variable is proportional to flame velocity SI Ξ,
meaning that the flame propagation is the source of the chemical reaction and
heat released associated to combustion; second, the flame velocity is computed

49
in every iteration and in every cell, hence varying region by region.
This way spark ignition combustion is fully modeled, and solver betaFlameletX-
iEngineDyMFoam is entirely described. To consider mixture autoignition, chem-
ical reaction model must be added to the solver, having hence the betaFlamelet-
SACIXiEngineDyMFoam solver.

2.4 Spontaneous combustion

To model autoignition, no flame must be considered. While spark ignition

progress variable c is fully depending on flame propagation, the introduced
progress variable cf resh is representing chemical oxidation of fuel particles. In
homogenous mixture context, this kind of combustion is going to be very rough
and violent, since chemical activity of the fresh mixture is almost homogenous.
As before, the progress variable cf resh is the formation enthalpy of the chemical
species at reference temperature of 298 K; but this kind of progress variable acts
only on fresh mixture, so regions not already reached by the flame (not burnt
regions). The fresh progress variable equation is the following one:

∂ρcf resh
+ ∇ · (ρ~ucf resh ) + µ∇cf resh = ρċf resh · b (2.51)
∂t

This kind of equation presents on the left side time-derivative, convection and
diffusion term as every conservation equation; on the right side the source term
related of progress variable increasing, composed by:

a term proportional to chemical reactions velocity (ċf resh )

a term connecting spark ignition combustion to spontaneous one (b)

Term b is added to cf resh source term equation because burnt mixture (having
b = 0) has already reacted, and it can not give any further heat release contri-
bution.
The solver needs to have velocity of chemical reactions of the fuel-air mixture.
This chemical reaction is depending on both fluid dynamics quantities such as
pressure, temperature and mixture fraction of every cell, but also on chemical
reaction mechanism of the fuel itself. As anticipated, even considering a fuel
composed just by one kind of hydrocarbon (methane for instance) to solve the
whole chain reaction is too heavy in computational terms. This is the reason
why for this equation, another open-source software has been used: Cantera R .
This software allows to build a pre-made chemical kinetics table before the sim-
ulation, which, once pressures, temperatures and mixture fractions are given as
inputs, it can return source term ċf resh as an output.

50
 Figure 2.15: Chemical table algorithm description

As shown by figure 59, a set of multiple pressures, temperatures, mixture

fractions and EGR are considered in table generation phase.

NP SRtable = nEGR ∗ np ∗ nTu ∗ nZ (2.52)

Of course meaningful ranges need to be defined. Equation (67) shows how con-
sidering very wide ranges of pressures or temperatures can affect size of the
table, hence time needed for generation. PSR stands for perfectly stirred reac-
tor, as far as homogeneous mixture is considered. What is important is defining
correctly mixture fraction, pressures and temperature ranges, while EGR are
not considered in this analysis.
Pressures are going to assume a very wide range since flame propagation com-
bustion is going to give a significant contribution before spontaneous ignition
can occur. Therefore, pressure maximum considered is 180 bar, as it has been
assumed maximum engine pressure. Unburnt fresh temperature range is also go-
ing to be wide since temperature gradient caused by spark ignition combustion
is going to influence unburnt fresh temperature, and so 1400 K is considered as
maximum temperature. It is almost unrealistic to assume presence of such high
temperature values in volumes not already reached by combustion but applying
a very wide range is just compromising table generation computational time,
while choosing a too small range can compromise quality of results obtained.
In the end, mixture fraction considered goes from very lean ones (Z = 0.1) to
stoichiometric one (Z = 1), and no rich mixture condition is tabulated.
System is composed by five different species by a numerical point of view: CO2 ,
H2 O, N2 , O2 and fuel. To compute molar fractions, atoms conservation equation

51
and absolute mixture enthalpy equations are imposed.
P
nvs Pns
i=1 xv,i n C,i = xi nC,i
Pi=1


Pnvs ns
 i=1 xv,i nH,i = i=1 xi nH,i


 Pnvs Pns
i=1 xv,i nO,i = xi nO,i (2.53)
 P nvs Pi=1
ns
i=1 xv,i nN,i = i=1 xi nN,i




Pnvs x w H (T, p) = Pns x w H (T, p)

i=1 v,i i i i=1 i i i

Adopting virtual species composition is an approach able to significantly reduce

memory demanded for table generation. Code organization is well described by
the following flow diagram.

Figure 2.16: Table generation code organization

As figure 60 shows, in code implementation there is also the possibility to

evaluate nitrogen oxides emission, but they are not included as far as NOx

52
transport equation is not available. End time values is a vector providing time
at which integration is stopped, in case combustion does not complete for con-
sidered pressure, temperature and mixture fraction values, and it presents when
chemical reaction reaches asymptotical equilibrium.
When dealing with CFD simulation integration, what is important is cf resh
˙ re-
sulting by fluid dynamics cell conditions and what kind of species (combination
of C, N, O and H) are generated by combustion presence. In this case, pressure,
temperature and mixture fractions are input, and what has solver as an output
is pvS, the progress variable rise.

Figure 2.17: Table-solver interaction

pvS = f (p, T uf resh , Z) = ċf resh (2.54)

Figure 62 shows how table and code interact between each other. In this com-
pression ignition analysis, a multiplicity of fuels is considered, hence a different
kinetics table is generated for every fuel. If equivalence values range is correctly
defined, tables do not need to be changed if lean and stoichiometric mixture
engines (running with the same fuel) are compared.

53
54
Chapter 3

Spark ignition combustion

In this section, a compressed natural gas fuelled engine for heavy duty ap-
plication (CNG-heavy duty) is analysed. Geometry of combustion chamber is
entirely known, and mesh is already generated. Some real engine working points
are known in terms of pressure, gross indicated work, heat release and wall heat
transfer, hence what is performed is a solver parameters tuning so that solver
itself can represent experimental working point as precisely as possible, then an
optimization of these working points in the engine map has been performed.

Figure 3.1: CNG-heavy duty engine at top dead centre

In figure 62 the given mesh is displayed: it is a hybrid grid, made up by blocks

of structured mesh in the whole combustion chamber. Just one cylinder of the
whole engine is simulated, and no valve motion is both modeled or considered.
Central spark plug is installed, and in the mesh it is represented at the point of
intersection between the revolve axis and the cylinder head. Walls represented
in the mesh are three: head (at the top of the mesh), piston (at the bottom of
the mesh) and liner (lateral part of the cylinder, drawn rightwards). Cylinder
walls are also characterized by a higher mesh density with respect to the rest of
the volume, to represent correctly boundary layers, which have a very important
part in determining heat transfer losses. Leftwards, cylinder axis with respect
to symmetry is imposed. The whole volume at cylinder top dead centre is
represented by Figure 64.

55
 Figure 3.2: Whole cylinder volume

stroke 0.135 m swirlRPMratio 1.5

bore 0.15 m swirlAxis (0 0 1)
conRodLength 0.230 swirlProfile 10−5
IVC -175◦ uprimeUpRatio 0.7
EVO 124◦ lintBoreRatio 0.017

Table 3.1: CNG-heavy duty main geometrical parameters

Mesh motion is performed by adding or removing mesh layers depending on

piston position, hence number of cells varies hugely between top dead centre
(TDC) and bottom dead centre (BDC). At top dead centre (where number of
mesh cells is minimum), cells are 8659, almost all of them hexahedral. Open-
FOAM is a 3D simulation solver, so cells must be defined in all three axial
directions: to not solve equations on y-axis, just 1 cell is present along that
direction, in order to have no specific quantity flow along that axis. Mesh has
been generated to take into account fuel injection, and this is the reason why
there is such a different orientation of some cells in proximity of revolve axis;
average orthogonal quality is 15.08, which is not compromising simulation re-
sults. In internal combustion engines flow preferential direction can change a
lot because off turbulence and/or pressure gradients introduced by combustion.
Maximum aspect ratio is 14.0, and maximum skewness is 3.3897.
The whole domain is not big, as far as it is just a cylinder with 0.135 m of bore;
cell areas are hence very small (minimum one is 2.0 · 10−10 ) and solving conser-
vation equations does not require a lot of iterations to reach tolerance imposed.
Time steps are defined in engine time (crank angle) and three different values
are used:

∆θ = 0.125◦ between -175◦ and -30◦

∆θ = 0.005◦ between -30◦ and 40◦

∆θ = 0.025◦ between 40◦ and 124◦

These ∆θ values lead to have 18526 different time steps between intake valve
closing (IVC) and exhaust valve opening (EVO), with a lot of computational
time demanded; this is necessary to have acceptable Courant numbers, because
if wider time steps are imposed, simulation numerically diverges. Simulation

56
duration is in the order of magnitude of 6 hours per processor, but it can vary
significantly on computer used.
Since it is a half cycle computation (including just one compression and ex-
pansion stroke), turbulence must be set a priori: swirl axis (0 0 1) coincides
with cylinder one, while its intensity can be changed increasing or decreasing
swirlRPMratio value. ”uprimeUpRatio” and ”lintBoreRatio” are there to define
turbulence intensity and its dissipation rate, as illustrated in the table at the
top of the page.

Figure 3.3: Engine map of given experimental points

In figure 64 engine map is showing all different experimental points avail-

able. Starting from reliable experimental data is very important to verify CFD
solver validity. Cylinder mixture pressure and temperature and cylinder walls
temperatures as well are known for each simulation starting point (-175◦ ).
In the engine map, two constant regime sweeps (at 1200 rpm and 1600 rpm)
and one constant torque one at 1650 Nm are displayed. Maximum IVC working
pressure is 2.44 bar, corresponding to 1200 rpm full load condition, and it is the
maximum torque working point experimentally available. Most known regime
is 1200 rpm, where six distributed points from 17% load (0.58 bar at IVC) to
100% load (2.44 bar at IVC) are given. Constant torque sweep is also composed
by six different points from 1000 rpm to 1900 rpm.
The very first target is to reproduce this experimental working points in a CFD
context, so simulations are set with the parameters illustrated in the tables be-
low.

1200-421 1200-855 1200-1270 1200-2030 1200-2440

Spark time -24◦ -20◦ -14◦ -9◦ -6◦
IVC pressure 0.587 bar 0.958 bar 1.32 bar 2.02 bar 2.46 bar
IVC temperature 363 K 360 K 361 K 362 K 362 K
Twall head 447 K 480 K 499 K 523 K 532 K
Twall liner 393 K 405 K 413 K 428 K 435 K
Twall piston 483 K 527 K 551 K 581 K 591 K

57
 1000-1710 1200-1670 1400-1670 1600-1680 1800-1650
Spark time -9.75◦ -11◦ -12◦ -13◦ -14◦
IVC pressure 1.72 bar 1.72 bar 1.73 bar 1.77 bar 1.77 bar
IVC temperature 358 K 362 K 364 K 368 K 372 K
Twall head 508 K 510 K 523 K 533 K 540 K
Twall liner 420 K 420 K 424 K 427 K 430 K
Twall piston 570 K 563 K 572 K 578 K 580 K

1600-839 1600-1290 1600-2140 1900-1630

Spark time -16 -14 -11 -15
IVC pressure 0.982 bar 1.37 bar 2.21 bar 1.75 bar
IVC temperature 371 K 368 K 369 K 375 K
Twall head 481 K 513 K 555 K 550 K
Twall liner 407 K 418 K 437 K 432 K
Twall piston 518 K 556 K 602 K 589 K

Table 3.2: Initial available pressure, temperatures and spark timing for every
experimental working point represented in CFD map

Working points name is derived by the combination of engine speed and

torque: for instance, 1000-1710 means 1710 Nm of torque at 1000 rpm.
A spark ignition solver validation needs to be performed, in order to verify that
it is able to reproduce correctly flame propagation combustion, both in terms
of flame velocity and heat release. Spark plug strength is kept constant (value
of 3), while its energy release duration is kept fixed at 1.25 ms, but changing
in terms of engine time. To control spark ignition combustion flame velocity,
Ξcoef f equilibrium coefficient has been modified.
r
u0
Ξ = 1 + Ξcoef f Rη (3.1)
Su

Figure 3.4: Xi equilibrium sweep

58
 The higher Ξcoef f , the higher flame velocity is going to be. By the expression
above, it is possible to appreciate also turbulence influence in increasing flame
combustion velocity. In figure 65 reported below pressure curves in function of
Ξcoef f are represented: with no spark plug energy release modification, both
pressure maximum values and pressure underlying areas change a lot, having
important effects on gross indicated work. Defining the correct coefficient is
essential for having a good solver accuracy result. In figure 65, range consid-
ered for Ξcoef f is 0.5 to 0.7. Lowest regime point (1000-1710) has been chosen
for a first Ξcoef f validation, resulting that value of 0.5 is the one best fitting
experimental results, as shown by both figures 66 and 69. To perform this kind
of analysis, fuel chosen is a natural gas composed by 100% of methane. Vary-
ing natural gas used as a fuel can lead to significant changes for what regards
autoignition, but flame properties are very similar.

Figure 3.5: 1000-1710 validation: pressure curve comparison

Two pressure curves plotted in figure 66 are very close and similar between
each other. If focus is pointed on differences, simulated pressure curve has a
lower ignition delay with respect to experimental one, detaching a little bit too
early from motored curve behavior. Combustion velocity is well represented
since in combustion development phase the curves are almost parallel. Pressure
maximum values are very similar, but in computed curve it is slightly antici-
pated. In the second part of expansion phase (once combustion is completed)
simulated curve is slightly above the experimental one.

59
 Figure 3.6: 1000-1710 validation: wall heat transfer comparison

Figure 3.7: 1000-1710 validation: cumulative heat release comparison between

simulated and experimental data

60
 As shown in figure 67, wall heat transfer curve is not as well reproduced as
the pressure one. In compression stroke, wall heat transfer has very low values,
hence also absolute errors are low. When combustion is taking place, wall heat
transfer increases in an appreciable way since curves get closer. Minimum of
curves are comparable, with simulated one slightly anticipated and higher in
terms of magnitude, but when combustion is completed and burnt mixture is
expanded, computed values are highly overestimated with respect to experimen-
tal ones, contributing in the slight pressure overestimation of the final part of
expansion stroke.
In figure 68 cumulative heat release curves are compared. X-axis scale is
different with respect to previous figures, since there is no reason to plot heat
release curve where combustion is not occurring. The lower ignition delay can
be detected also in this chart, as far as simulated cumulative heat release curve
detaches earlier from 0 value with respect to experimental one. During com-
bustion process, slopes of the two plotted curves are very similar, indicating
that combustion velocity is well represented in CFD environment; in any case,
maximum values are different, and at the end of the half cycle fuel-air mixture
heat release is overestimated.

Figure 3.8: 1000-1710 validation: apparent and rate of heat release comparison
between simulated and experimental data

From figure 69 it is possible to perform a comparison of both apparent and

rate of heat release, plotted over engine time. In both cases, curves rise from
0 value is anticipated because of a lower ignition delay, but then behaviors
are very similar up to curves maximum values. In apparent heat release rate,
maximum is underestimated, while in rate of heat release maximum is slightly
overestimated, but in both cases they are anticipated with respect to experi-
mental ones. After the maximum values are reached, simulated curves are less
inclined with respect to experimental values, increasing curves underlying area,
hence cumulative heat release.
Using a python post-processing script, main differences between simulated en-
gine case and experimental data have been summarized in the table below. As
suggested by already illustrated figures, main differences are about cumulative
wall heat transfer values, which is overestimated of more than 25%. More im-

61
portant data such as gross indicated work, pressure maximum value and its
engine time location are very similar. In table 3.3 main computed parameters
such as gross indicated work (giw) and maximum pressure timing are compared
to the experimental data available.

giw heat released heat transfer Pmax θ @ Pmax

computed 3512 J 9471 J -1141 J 92.38 bar 19◦
experimental 3333 J 8994 J -1567 J 97.7 bar 20◦
error 5.38% 5.03% -27.2% -1.41% -1.0◦

Table 3.3: Comparison between computed and experimental data

This procedure has been repeated also for all other engine experimental
working points available. To report all pressure, wall heat transfer and the
three heat release curves comparison would be too repetitive, so two different
critical points have been chosen: 1200-855 because of the low load, and 1800-
1650 because of the high regime.

Figure 3.9: 1200-855 validation: pressure curve comparison

Figure 70 displays pressure curve comparison at low load (35%) and 1200
rpm engine speed. With respect to previous pressure curve comparison in figure
66, here differences between values are more evident. Spark timing of this engine
working point is -20◦ (as illustrated by the table), much more anticipated with
respect to previous case; hence here ignition delay reduction is having more
influence, and simulated pressure curve reaches a higher maximum value and
it is also anticipated of few degrees. In any case, shape of both simulated
and experimental curves are very similar between each others, suggesting that
delaying spark plug energy release more similar curves can be obtained.

62
 Figure 3.10: 1000x1710 validation: wall heat transfer comparison

In figure 71 wall heat transfer curves are compared. In this case, simulated
and experimental values get more similar with respect to 1000-1710 working
point case, as far as curves are coincident during combustion development, but
after global minimum simulated wall heat transfer behavior is consistent with
the previous working point. In this case, wall heat transfer minimum is un-
derestimated, and its timing is getting closer with respect to experimental one.
When combustion is almost completed, computed curve returns to be overes-
timated, but if compared with previous working point cumulative wall heat
transfer difference decreases, especially in relative terms.

Figure 3.11: 1200-855 validation: cumulative heat release comparison

63
 As figure 72 illustrates, differences when cumulative heat release are com-
pared gets huge. Ignition delay underestimation plays a very important role,
as far as simulated and experimental curves are clearly detached. Maximum
cumulative heat release is overestimated. Cumulative heat release slopes in
combustion development are very similar, suggesting once again to delay mix-
ture ignition to get a better matching.

Figure 3.12: 1200-855 validation: apparent and rate of heat release comparison

In the end, figure 73 confirms what has already been described. Shape of
both heat release curves are very close, with both computed and experimental
curves presenting a similar rate of heat release maximum value. Their timing
is significantly anticipated, resulting in a too early combustion with respect to
what results from experiments.

Figure 3.13: 1800-1650 validation: pressure curve comparison

64
 Considering both figure 74 and figure 66, 1000-1710 and 1800-1650 working
point validation can be compared. In this high regime working point, pressure
curves are very similar. There is also in this case an ignition delay reduction but
having no significant effect in combustion developing phase. Simulated pressure
curve maximum value is slightly both lower and anticipated, suggesting to delay
mixture ignition for correcting maximum timing, but this kind of action would
reduce its absolute value. In any case, curve underlying areas are very similar,
since pressure gain present in computed curve in ignition phase is balanced by
a lower pressure maximum value reached.

Figure 3.14: 1800-1650 validation: wall heat transfer comparison

In this case, as shown in figure 75, wall heat transfer values are very different
between the plotted curves. From top dead centre ahead, wall heat transfer
curve is underestimated in every crank angle value, resulting in a cumulative
wall heat transfer much higher with respect to experimental one. Also, wall
heat transfer minimum timing is underestimated.

65
 Figure 3.15: 1800-1650 validation: cumulative heat release comparison

In figure 76 cumulative heat released comparison between computed and

experimental data is performed. Curves in this case are coincident for almost
all the cycle, leading to a good simulation of combustion process, both in terms
of shape of the curves and cumulative heat release final values. Differences arise
between 20◦ and 40◦ crank angle, in last combustion phase, but they are not
compromising cumulative value reached at the end of expansion stroke.

Figure 3.16: 1800-1650 validation: apparent and rate of heat release comparison
between simulated and experimental data

To conclude this engine validation process, also rates of heat released are
shown in figure 77. Curves are again consistent between each other, presenting
similar shapes but with different maximum values and timing. Regarding rate of
heat release, maximum value is comparable with simulated one, which is slightly

66
anticipated; regarding apparent rate of heat release, maximum value is both
anticipated and highly underestimated. In combustion final phase, both heat
release curves are less vertical with respect to experimental ones, in consistency
with previous cases.
What can be concluded by this analysis is that solver is very good in pressure
curve representation, that can be adjusted anticipating or delaying spark timing
if needed. At high load and engine velocities, heat release is well reproduced, and
ignition delay is less influencing solver results. Wall heat transfer computation
is the point of weakness of the solver: for every simulation performed, it is
overestimated, especially when low loads are considered.
To conclude this validation sub-section, Gross indicated work error is analysed,
and an error indicator must be defined:
GIWcomputed − GIWexperimental
ErrorGIW = (3.2)
GIWexperimental
Therefore, an error analysis of the constant torque sweep and of the constant
engine velocity sweep has been performed.

Figure 3.17: Gross indicated work error in constant torque sweep

As shown by figure 78, error in the worst condition slightly overcomes 5%,
getting lower at higher engine velocities.

Figure 3.18: Gross indicated work error in 1200 rpm sweep

While in figure 79 error in constant engine velocity has a more varied be-
havior. As discussed in 1200-855 validation point, fast increasing of the fired

67
pressure curve is influencing simulation results. To have more accurate com-
puted data and reducing gross indicated work errors, for low load points it is
enough to delay spark timing of few crank angle degrees, to have a better match-
ing of the pressure curve.
In the end, the solver is reproducing almost all the cases in a good way, espe-
cially when engine is running between 1200 Nm to 1800 Nm of torque. When
very low load is simulated, computed curves are anticipated with respect to ex-
perimental ones, but shapes are not too much different. When very high regimes
(1900 rpm) are simulated, pressure and heat release maximum timings are very
similar to experimental ones, but their magnitude values are quite lower; even
in the worst case, the difference between CFD and experimental pressure maxi-
mum value is -8.33%. It is possible to conclude that the solver is can reproduce
empirical data with an acceptable error for every engine working point given.
The choice of a proper χcoef f value has been fundamental for this validation.

3.1 Spark advance sweep for engine optimiza-

tion
Given experimental points are good to validate the solver and calibrate its set-
tings, but they are not very representative of engine performance: a correct
spark advance timing is crucial to improve gross indicated work and reduce fuel
consumption. In addition, spark timing variation is an easy and reliable control
strategy for the engine itself, such that a spark advance delay is important to
avoid knock whenever it occurs. At the same time, turbulence or natural gas
composition are not that easy to be controlled, hence these kinds of solver input
variation are considered a good sensitivity analysis as cycle variability.
Efficiency is the main parameter for thermal engines. It is defined as the ratio
between gross indicated work and heat released by the fuel.

grossIndicatedW ork GIW

η= = (3.3)
cumulativeHeatReleased cumRoHR

GIW = W orkexpansion − W orkcompression (3.4)

cumRoHR = ṁf uel · LHVf uel (3.5)

In order to understand how spark timing can be so relevant, the working point
1200-421 has been taken as example; then a constant regime analysis and con-
stant torque analysis are performed.
1200-421 is the lowest torque point available in the whole engine map: it corre-
sponds to 17% of load, with 421 Nm of torque. The engine is running in stoichio-
metric air-fuel ratio and natural gas considered is composed just by methane,
meaning knock can not be present (since a compression ratio of 11.7 is adopted).

lowest spark advance: -11◦ → -26◦ highest spark advance

Range imposed is very wide, in order to consider both maximum values but also
maximum efficiency working point stability.

68
 Figure 3.19: spark timing sweep for 1200-421 working point

Pressure curve underlying area is considerably wider when spark timing is

anticipated, and also pressure maximum value increases considerably, from 30
bar to over 40 bar. Not all the underlying area is corresponding to useful work:
from -175◦ (intake valve closing, IVC) to 0◦ (top dead centre, TDC) piston is
compressing air-fuel mixture, which is a ”cost” in terms of performance, as in
all others thermal machines. From 0◦ to 124◦ (exhaust valve opening, EVO)
expansion stroke is plotted, and that area region needs to be maximized. In any
case, p-θ curve is good to analyze pressure behavior with respect to the time,
but it is not strictly representing work done by power cycle; pressure-volume
curve is more connected to real work done by cylinder itself, and comparisons
with the ideal Otto cycle are easier to be discussed.

Figure 3.20: PV curve for 1200-421 working point

Otto theoretical combustion is a constant volume combustion, ideally a ver-

tical line in correspondence of top dead centre. The earlier the spark advance,
the more ideal the combustion process reaching also higher pressure and tem-
perature values. But also, compression work and heat transfer losses increase
anticipating spark timing, so the best compromise between higher combustion
ideality and minimization of compression and thermal losses is needed.

69
 Figure 3.21: Spark advance sweep: cylinder temperatures

In figure 82, cylinder temperatures are plotted. Advancing spark timing,

temperature maximum values do not change considerably. High cylinder tem-
peratures increase wall heat transfer losses, hence the curves crosses during the
expansion stroke. At the end of the expansion stroke, most advanced condition
is the one presenting lower temperature values at exhaust valve opening because
of the higher cumulative wall heat transfer, even if maximum value is slightly
higher.

Figure 3.22: Spark advance sweep: wall heat transfer

Figure 83 illustrates how heat losses are conditioned by spark timing, both
in terms of heat transfer minimum and cumulative values. Plotted curves are
very similar, suggesting that a delayed spark timing can reduce heat losses,
which contribute to decrease temperatures. Heat losses are quite important for
internal combustion engines because that temperature reduction decreases work
done especially in the second part of expansion stroke, when combustion can be
considered completed (around 40◦ ).

70
 Figure 3.23: Spark advance sweep: cumulative heat release

Heat release curves (plotted in figure 84) have the same maximum value,
since it is determined on the quantity of fuel present in the cylinder at intake
valve closing. Load can be changed by modifying initial cylinder pressure but,
as far as it is a constant load analysis, cumulative heat release curves are very
similar between each other. Differences start to arise just when flame inter-
face reaches squish area: there combustion proceeds slowly because of the low
space available. Normalized c can reach and overcome 99.98%, meaning that
combustion is considered completed.

Figure 3.24: Flame interface of low load engine at (from left to right) 26, 24
and 20 of spark advance

In figure 85, flame interface of different spark timing cases is compared at

10◦ after top dead centre to explain cumulative heat release behavior illustrated
in figure 84: most delayed spark timing ignition has an higher flame surface
available, hence the higher heat release rise in that piston position.
Combustion efficiency and gross indicated work are influenced by all these ther-
modynamic quantities. Spark timing able to maximize efficiency is -19◦ . This
thesis work is just focused on power cycle, hence any other considerations such
as having a good turbine pressure-temperature inlet for turbocharging or valve
timing variations are considered out of this work.

71
Returning to combustion analysis, maximum efficiency point is very stable, with
few variations per few degrees of spark timing changing, leading to a good flex-
ibility of the engine itself.

Figure 3.25: 1200-421 efficiency curve with respect to spark timing variations

In figure 86 efficiency curve with respect to spark timing is plotted. Max-

imum efficiency for this working point is 38.34%, but it is very stable since
advancing and delaying of 1◦ spark timing leads respectively to 38.33% and
38.32% efficiency values, so very low variations.

Figure 3.26: Gross indicated work with respect to spark timing variations

While in figure 87 gross indicated work is represented. As far as cumulative

heat release does not change, gross indicated work curve has the very same shape
of efficiency one if plotted with respect to spark timing axis. GIW varies from
1200 J to 1170 J in the worst case, proving again the high efficiency stability.

72
 Figure 3.27: Spark timing sweep cumulative wall heat transfer curve

In figure 88 cumulative heat transfer to the wall is shown. It is the only

solver quantity having a strictly monotone trend, also meaning that advancing
spark timing heat losses will always increase.

Figure 3.28: 1200-421 spark timing sweep: pressure and temperature maxima

In conclusion of this analysis, figure 89 reports cylinder pressure and tem-

perature maximum values. They both have an almost linear behavior, with
more maximum pressure relative variation with respect to maximum tempera-
ture one. Pressure maximum values can vary from 30 bar at very delayed spark
timing to 45 bar at very anticipated spark timing, meaning the 50% of relative
variation is present. At the same time, temperature (weighted averaged) maxi-
mum values are almost constant, with a very few increasing when early ignition
is adopted.

73
3.2 Constant torque sweep
By the engine map, it is possible to recognize a constant torque sweep at almost
1700 Nm. Engine has a minimum rotational speed of 1000 rpm and a maximum
one of 1900 rpm.

Figure 3.29: Constant torque sweep

Engine time is referred to piston position in compression and expansion

stroke, but if rotational speed doubles, stroke duration gets halved. As a con-
sequence, what is expected is no flame velocity variations in terms of time (in
seconds), but determinant variations in terms of engine time (in crank angle).
Also, electric spark light delay, fixed at the value 0.001 s, is going to be more in-
fluencing at high engine speed with respect to lower ones. In this sub-section, all
points have already been optimized with respect to spark timing, as previously
described. No turbulence or other numerical parameter has been changed.

Figure 3.30: Gross indicated work at constant torque

As described in engine presentation, for this analysis (and mainly in the

rest of this thesis work) load is determined by fixing cylinder pressure at IVC.
Since it is a constant torque analysis, gross indicated work (shown in figure 91)

74
is more a constrain than a parameter to be post processed, in fact values are
almost coincident. Once optimization is performed, gross indicated work from
power cycle varies depending on which working point was closer to optimization
in original settings.

Figure 3.31: Cumulative heat released curves at constant torque

By heat release curves illustrated in figure 92, low engine speeds need a
delayed spark timing to reach the optimum, and it can be seen by a delayed
appearing of low regime curves; but combustion needs less crank angle degrees
to be completed, and a curves intertwinement around 10◦ after top dead centre
is present. Even if in terms of seconds combustion is requiring almost the same
time, in terms of engine time (crank angle) combustion is slower, leading to a
less ideal one.

Figure 3.32: Optimized pressure curves at constant torque

Pressure curves displayed in figure 93 are very similar, even if engine rota-
tional speed changes. Mixture ignition needs to be anticipated when engine is
rotating faster, and pressure maximum values are getting lower, but changes in
optimized pressure curves are not very important.

75
 Figure 3.33: Cumulative heat transfer curves at constant torque

In cumulative wall heat transfer curves reported in figure 94, it is possible

to understand the real advantage of having higher engine speeds: heat losses
reduction. If engine is rotating faster, exhaust gasses have less time available
to exchange heat with cylinder wall, resulting in a more ideal compression and
expansion. This is possible to be observed also in temperature curves, where
temperature maximum values are almost the same, but at the end of expansion
stroke burnt gasses are hotter in higher engine speed cases.

Figure 3.34: Temperature curves in constant torque sweep

Figure 95 reports temperature curves in this constant torque sweep. Curves

are almost coincident, but there are two areas where some important differences
can be detected: when mixture starts to ignite due to different spark timings
imposed and temperature final values, due to lower wall heat transfer at high
engine rotational speeds. Even if temperature maximum values are very close,
end ones can change of more than 60 K between slowest engine rotation (1000
rpm) and fastest available one (1900 rpm).

76
 Figure 3.35: Wall heat transfer curves in constant torque sweep

Figure 96 can show how wall heat transfer curves have a very similar shape,
enhancing how faster rotational speed configurations have lower heat losses along
the whole cycle.
Another huge difference between load and regime variation is the fluid turbulent
kinetic energy. Swirl is initialized by imposing a ”Swirl-RPM ratio”, meaning
that keeping constant this ratio, fluid motion becomes more important at high
engine speeds. Turbulent kinetic energy curves in this analysis almost uniformly
increase to higher values.

Figure 3.36: turbulent kinetic energy in constant torque sweep

Therefore, turbulent kinetic energy of different working points is reported in

figure 97, evidencing both how turbulence increases with rotational speeds, but
also how they dissipate during compression stroke. At the end of high engine
speed curve, some vibrations are present. This kind of vibrations are the results
of a combination of multiple factors such as combustion behavior, pressure and
temperature condition but mainly piston geometry. In any case, they are going
to be discussed later in this thesis work. In this section, what is important is
to evidence that the more turbulence is present in the engine, the faster the
combustion (which is also explained in solver section, were b-Ξ equations are
discussed).

77
 Figure 3.37: Efficiencies in constant torque sweep

In efficiency terms, as shown in figure 98, a maximum value is present at

1600 rpm, which represents the best trade off between a faster combustion and
lower heat losses. In any case, figure displaying efficiency is very zoomed, since
the whole vertical axis varies from 40.3% to 40.8%.

Figure 3.38: Optimum spark timing in constant torque sweep

Even if η curve is quite flat, spark timing needs to be changed considerably

inside this sweep, as shown in figure 99. In this way it is possible to get the
already represented pressure curves to optimize different power cycles.

78
3.3 Constant speed sweep
In this subsection, engine speed is kept constant and different loads are sim-
ulated. Chosen regime is 1200 rpm, since there are six experimental points
available for that condition, and a spark timing optimization for all the loads
has been performed with criteria described in subsection 3.1. Other parameters
such as turbulence are not changed.

Figure 3.39: Load sweep illustration

In a stoichiometric mixture engine, having higher load means rising intake

valve closing pressure, hence quantity of fuel inside the cylinder. In consistency,
what is expected is a graphical curve enlargement of pressure, GIW, and heat
released.

Figure 3.40: Load sweep pressure curves

Figure 101 illustrates different pressure curves when different loads are com-
puted. Shape of the curves is very similar, with an almost coincident maximum
value timing.

79
 Figure 3.41: Load sweep cumulative heat release curves

The same behavior of pressures curves can be seen in cumulative heat release
ones, displayed in figure 102. They move to always higher values, accordingly
to the higher mass of fuel inside the cylinder.

Figure 3.42: Gross indicated work histogram

As expected, in figure 103 gross indicated work increasing when higher loads
are required to the engine is shown. Maximum intake valve closing pressure is
2.4 bar, value taken from experimental data.
What can be interesting of this analysis is verifying the existence of an optimum
spark timing or efficiency variation and what it is due to. In terms of temper-
ature, load doesn’t influence so much maximum values since engine is running
stoichiometric, but temperature reduction once combustion is completed.

80
 Figure 3.43: Constant regime temperature curves

Figure 104 illustrates the behavior described above: shape of the curves is
very similar, with very close maximum values because engine is running sto-
ichiometric. Air-fuel mixture mass inside the cylinder is not the same, and
hence neither thermal inertia: at low load, heat losses are relatively more im-
portant, and it can explain temperature differences starting to arise even before
combustion and increasing once combustion is completed.

Figure 3.44: Constant regime wall heat transfer curves

Wall heat transfer curves displayed in figure 105 confirm behavior described.
They are not proportional to air-fuel mixture mass, and proportionally heat
losses are more important when load is lower.

81
 Figure 3.45: Efficiency at constant regime

What finally results, as illustrated in figure 106, is a high load efficiency

higher with respect to low load one. Efficiency variations are huge if compared
to constant torque case, from 38.5% in lowest load case to 41.0% in highest load
one, over than 6% of relative variation. Since heat losses play a very important
role in that, this behavior explains also why turbocharging is so important for
modern internal combustion engines, since higher load can be reached without
increasing cylinder surface, hence reducing heat losses.

Figure 3.46: Optimum spark timing at constant regime

This efficiency-load monotonic behavior stands also because there is no

knocking issue since compression ratio is not that high and fuel considered is
fully methane. The reduction of wall heat transfer importance is reflected in
optimum spark timing ongoing, represented in figure 107: since at high load
cylinder gasses have more thermal inertia, pressure curves can be raised up
with lower drawbacks; optimum spark advance passes from -19◦ in lowest load
condition to -23◦ in highest load condition.

82
 Figure 3.47: Pressure maximum values at constant regime

In the end, maximum pressure reached is a little bit lower than 180 bar, as
shown by figure 108. Full load at 1200 rpm is the only working point available to
reach such a high pressure, hence this is going to be considered in spark assisted
compression ignition engine development, as this pressure maximum values is
very closed to supposed mechanical limit.

3.4 Engine map

Finally, with no change to engine geometry or any fluid dynamics settings, en-
gine map efficiency is significantly growing, evidencing the importance of spark
timing control.

Figure 3.48: CFD and optimized engine map comparison

Engine points location do not change very much, in fact at first sight the
two maps showed above very similar. Once also torque and efficiency numerical
values are observed, it is possible to notice some significant improvements to
the work done by the cylinder.

83
 Figure 3.49: CFD working points efficiency engine map

Figure 3.50: CFD efficiency engine map after optimization

As shown in figure 111, most significant improvements are at high load where
experimental spark timing is very delayed, while it needs to be anticipated to
improve performances. At this low compression ratio (in natural gas engine),
knock or maximum pressure values do not correspond to a real issue to be
managed. Regarding to engine efficiency values, spark ignition engine is better
behaving at high load with respect to low load working points. Therefore, at
low load efficiency must be improved in order to have a significant reduction of
fuel consumption.

84
 Figure 3.51: Experimental working points spark timing map

Figure 3.52: CFD optimized spark timing engine map

In figures illustrated above it is possible to compare the experimentally given

spark timings (in figure 112) and the optimized CFD derived ones (in figure 113).
Differences get higher at high load, corresponding also to condition where effi-
ciency improvement is maximum. At low load spark timings are very similar,
hence there is no significant efficiency improvement. Fuel used in CFD simu-
lation is composed by 100% of methane, hence knock can limit spark timing
anticipation in no way.

85
86
Chapter 4

Knock prediction

In spark ignition engine context, knock is an abnormal combustion occurring

when some cylinder regions are exposed to such high pressures and temperatures
that fresh air-fuel mixture burns spontaneously. It is very violent and dangerous
for the engine, and it is strongly unwanted. In this section solver behavior in
knock presence is analyzed, also to understand peculiarities of this phenomenon.
Spontaneous ignition combustion is strongly influenced by the kind of fuel used:
octane number is a knock tendency indicator, defined such that the higher
it is, the more resistant the fuel will be to knock. In the solver algorithm,
determining reaction velocities in function of local pressures, temperatures and
mixture fraction is left to Cantera table generation, hence a different table for
every kind of natural gas composition must be created. In order to let the
engine knock, just for a part of this section fuel used is isooctane, which has an
octane number of 100 by definition. Natural gas one can vary within a range of
120-130, so in this simulation set very violent knocks should take place.

Figure 4.1: Engine flame front and spontaneous ignition visualization

Combustion chamber used is the one described in validation process; no

change has been made. Therefore, compression ratio is fixed at 11.7 and work-
ing point chosen for this analysis is the full load one at 1000 rpm (1000-1710),
which should be the one presenting the highest knock intensity because of the

87
low rotational speed. In natural gas port fuel injection engines, mixture can
easily be assumed homogeneous, and the same assumption has been made also
for these isooctane fuelled cases. Imposing a homogeneous mixture fraction
means having no differences in Z values, hence once a part of engine knocks,
also all the remaining unburnt ones are supposed to. In figure 4.1, it is possi-
ble to notice that flame is propagating from spark plug location, while knock
starts from cylinder extremities. Once the very first fresh mixture region ignites
spontaneously, in 2-3◦ combustion is going to be fully completed.

Figure 4.2: Autoignition propagation in 2 crank angle degrees

Figure 4.2 is displaying what has been textually described. Engine time
difference between most anticipated cylinder image (leftward) and most delayed
one (rightward) is exactly 2◦ of crank angle. In such a short time, all the unburnt
fresh mixture is supposed to release heat, with a consequent pressure rise and
sonic wave propagation.

Figure 4.3: Pressure curves when engine is powered with isooctane as fuel

To make a deep analysis on how knock presents and how it can be con-
trolled, a spark advance sweep as illustrated in figure 4.3 has been performed.
In displayed pressure curves it is easily possible to detect knock: there is an
almost vertical pressure rise when it takes place, causing even some vibrations
afterward. Magnitude of this increase can vary a lot, depending on quantity of
autoignited air-fuel mixture and when (in terms of engine time) it is occurring.
Theoretically, pressure curves in knocking cycles should present two different
maximum values: a ”rounded” one because of flame propagation combustion
and a more vertical one because of compression ignition combustion. This sec-
ond one moves depending on where autoignition takes place: if it happens after
flame propagation pressure maximum value, the two both can be distinguished;

88
if knock is very anticipated, there is just compression ignition maximum visi-
ble. Another very important thing to be remarked is the importance of spark
timing for knock control: always regarding figure 4.3, it is possible to notice
how maximum values increase anticipating spark timing. Spontaneous ignition
in fact is enhanced when severe pressure and temperature conditions inside the
cylinder are present. Even if delaying spark timing logically air-fuel mixture
should have more time available to react, reaching high cylinder temperature
values is definitely more influencing, meaning that the more spark is advanced,
the more knock is enhanced.
By previous sections, two mechanical engine limits regarding pressure values
were supposed: an absolute maximum of 180 bar and a pressure rise one of
15-18 bar/deg. Pressure rise in post-processing CFD context has been defined
as:
∂p pθ+∆θ − pθ
pressureRise = = (4.1)
∂θ ∆θ
Due to high heat release presence (and consequent violent fluid dynamics quan-
tities increase) time step in combustion range needs to be reduced: in previous
section, ∆θ chosen between -30◦ and 40◦ was 0.005◦ , while in this one it has been
reduced up to 0.0025◦ (the halved value), otherwise Courant number becomes
huge and divergence when equations are solved arises.

Figure 4.4: Pressure rise curves when engine is powered with isooctane as fuel

In figure 4.4 pressure rise curves have been displayed. In every knocking cy-
cle, they report two maximum values: a rounded (and so smoother) one because
of flame propagation combustion, and a very vertical one due to compression
ignition combustion consistently with what shown in figure 4.3. This second
one can become really huge with values of even 60-70 bar/deg depending on
quantity of autoignited air-fuel mixture mass. After this maximum, some pres-
sure rise vibrations are present due to sonic waves propagating and rebounding
in the whole cylinder, mechanically heavy for the engine resistance. The solver
is not able to reproduce these waves in a proper way, because of the high time
step used and the turbulence model chosen. For a more physical description of
them, a LES could be more appropriate but, as written in introduction section,
it is at the state of the art too expensive in computational terms to be used for
this thesis work targets.

89
The only not knocking case can be easily recognized because it has just one pres-
sure rise maximum, occurring when spark timing is very delayed since engine is
designed to work with natural gas.

Figure 4.5: Heat release curves when engine is powered with isooctane as fuel

As shown in figure 4.5, also apparent heat release has a similar behavior
with respect to pressure rise: it presents two maximum values, one from spark
ignition combustion (hence function of flame properties such as front velocity
and surface), the other from compression ignition combustion. Even in this
plot, autoignition maximum value can be much higher with respect to flame
propagation one, and close to that region curves are as inclined as pressure rise
ones, highlighting the direct relation between heat released by air-fuel mixture
and pressure increase. Some negative values (low in terms of magnitude) can
arise in apparent heat release CFD data, but these are small numerical issues
with no physical meaning due to very violent heat release rise. Even in this
time, the only not knocking cycle has just one heat release maximum value
since knock is not present.

Figure 4.6: Autoignition timing when engine is powered with isooctane as fuel,
-12 spark timing case

In figure 4.6 a comparison between fraction burnt and maximum c fresh

ongoing has been performed. Spontaneous combustion starts when normalized

90
cf resh reaches the value of 1; afterwards, it involves the whole unburnt fresh
mixture remaining. In solver code, even if spontaneous ignition arises, flame
continues to propagate with a different velocity because of higher local pres-
sures, temperatures and even turbulent kinetic energy caused by compression
ignition; but mixture can not release any further heat because it has already
reacted in cf resh equation, and combustion products are computed in ”Updat-
ing composition” script. Normalized cf resh graph is very similar to a Heaviside
function, from 2% to 100% in less than 1◦ crank angle.
PvS is the source term for cf resh equation, and its behavior can explain such a
vertical progress variable increase.

Figure 4.7: Cfresh source term when engine is powered with isooctane as fuel

In figure 4.7 source term of cf resh equation is plotted. Velocity of homoge-

neous mixture compression ignition combustion can be easily seen also by pvS
values: when knock occurs, maximum pvS increases from 4 to 5 orders of mag-
nitude with respect to a not knocking case. This is numerically explaining why
progress variable and heat release have this ongoing once knock occurs, and why
spontaneous ignition combustion is so fast.

Figure 4.8: Cfresh source term in a not knocking CFD simulation

If figures 4.7 and 4.8 are compared, before knock occurrence pvS are very
similar between the different charts. Once knock presents, cf resh source term

91
increases vertically, homogeneously in almost all the unburnt regions. This
behavior can change if mixture can not be considered homogeneous anymore,
but since in this thesis work just natural gas port fuel injection has been taken
into account, Z values have no differences inside the whole CFD domain.

Figure 4.9: Temperature curves when engine is powered with isooctane as fuel

For what concerns temperatures, as shown in figure 4.9, autoignition is con-

tributing to rise up weighted average values considerably: maximum reached
is around 2700 K while, when knock is not occurring, it is from 200 to 300 K
lower. For what regards heat transfer to the walls, it is certainly influenced by
cylinder temperatures, but there is a more important contribution for its rise.

Figure 4.10: Wall heat transfer when engine is powered with isooctane as fuel

As figure 4.10 illustrates, wall heat transfer curves minimum values have too
huge changes to be explained with just a temperature increase; as it will be
discussed in section 5, pressure waves and fluid kinetic energy release plays a
very important role in determining heat losses. Of course, this heat transfer
behavior is influenced by knock presence, but specific quantities interested are
going to be analyzed when engine is fuelled with natural gas to provide a more
complete explanation. In this case, autoignition is so violent that in previous
figure it is difficult to distinguish heat losses increase due to flame propagation
combustion from the contribution given by knock.

92
 Figure 4.11: Main specific quantities normalized

Figure 4.11 provides an idea of autoignition effects on pressures and temper-

atures, focusing on rise speed. It is possible to appreciate how in much less than
1◦ from the first autoignition appearance pressure and temperature increase of
almost 8%, resulting in the huge pressure rise previously displayed. Crank angle
distance between first autoignited mixture particle and pressure/temperature
maximum values can depend on fuel used and even on the piston shape.
Even if the focus in this subsection has been pointed on spark advance, engine
speed is actually playing a very important role for autoignition: the faster the
engine, the lower the time available for the mixture to react. This sub-section
has been useful to introduce heavy knock and to explain how it can be detected;
from this point ahead, natural gas has been used again to provide a deeper and
more meaningful knock and autoignition description.

4.1 Knock experimental validation

In this particular study, it is not enough to validate just spark ignition combus-
tion since it is not the only heat release source available. Compression ignition
combustion is considered by the solver as something detached, included in a sep-
arated equation. In addition, natural gas can be an extremely octane number
variable fuel, depending on which kinds of hydrocarbons are present. To be sure
that the proposed solver is suitable for spark assisted combustion, also a knock
validation is necessary, and to have some reliable data an experimental study is
needed. This subsection is fully dedicated to comparison between solver results
and a paper published in 2012 by Jiri Vavra, Michal Takats, Vojtech Klir and
Marcel Skarohlid of the Czech Technical University, named ”Influence of Nat-
ural Gas Composition on Turbocharged Stoichiometric SI Engine Performance
[10].
In this experimental study, knock tendency of a compressed natural gas engine
has been analyzed by including some additives to a reference fuel composi-
tion. Engine has hence been tested with a multiplicity of different fuels, in
order to detect if knock is occurring or not. If the solver betaFlameletSACIX-
iEngineDyMFoam can reproduce results comparable with experimental values,
validation can be considered completed and the solver able to be used for de-
sign purposes. What has been looked for is a result consistency between CFD

93
simulation and researcher’s experimental paper: a perfect spontaneous ignition
representation can not be reached. In this paper, Czech Technical University
researchers wanted to investigate knock presence and performance sensitivity
of an experimental engine to the fuel. Layout of the turbocharged compressed
natural gas spark ignition engine is shown in figure 4.12.

Figure 4.12: Paper experimental engine setup

Air fuel ratio is stoichiometric (φ = 1) and compression ratio is 12:1, very

similar to the CNG-heavy duty one (11.7:1). Other geometrical parameters such
as bore or stroke are given, and in the following table a comparison between
the two is presented. What is not described in the publication is engine piston
geometry, which has a very important role in enhancing or avoiding autoignition.

paper engine CNG-heavy duty

compression ratio 12:1 11.7:1
max boost pressure 2.4 bar 2.44 bar
bore 102 mm 135 mm
stroke 120 mm 150 mm
IVC -125◦ -175◦
EVO 123◦ 124◦
φ 1 1

Table 4.1: Comparison between paper engine and CNG-heavy duty one

Boost pressure and exhaust valve opening are very similar between the two
engines. CNG-heavy duty is definitely bigger, since both bore and stroke values
are higher. Adopting a bigger engine is also a knock-promoting factor: if flame
velocity between the two cases is going to be very similar, flame path is predicted
to be higher in a bigger engine, hence more time available for the fresh mixture
to autoignite. As written before, experimental combustion chamber geometry
is not known and this is a source of results discrepancy, since parameters like

94
squish area ratio or piston and head shapes can influence knock presence.
To have a major data consistency, CFD compression ratio of CNG-heavy duty
engine has been increased up to 12:1 and turbulence has been reduced. This
numerical change is just for having more similar pressure curves between the
two engines, using ones provided by experimental paper. Compression ratio
changing procedure adopted is described in the following section.

Figure 4.13: Experimental and simulated engine tuning

Figure 4.13 illustrates a comparison between experimental and simulated

engines pressure curves. As one can see, they are not coincident and there
are considerable differences in both compression and expansion stokes. In any
case, pressure maximum value and its timing (illustrated in crank angles) are
very similar, and no further change has been imposed to the CNG-heavy duty
one. Spark timing in CFD simulations is set at -21◦ , the same declared in the
paper. Experimental engine can not reach very high knock pressures to preserve
mechanical integrability of the engine itself: this is why it is equipped with an
electronic control unit (ECU) able to delay spark timing whenever spontaneous
ignition presents. To determine not just knock presence but also its intensity,
a Bosh AKR sensor (knock recognition system) is installed in the engine: this
sensor through a Fourier transform can convert knock intensity (in terms of
pressure increasing) into a voltage output. As anticipated in the introduction
section, knock can be easily detected by the presence of pressure waves, which
have a much higher frequency than rotational engine one; through the Fourier
transform, the sensor can distinguish knock caused pressure rise from the flame
propagation one. An AKR 2V threshold is imposed to distinguish heavy from
light knock, hence if sensor voltage output is higher than 2 V it is considered
heavy, otherwise it is light.
To perform a fuel sensitivity analysis, a reference natural gas composition must
be fixed.

95
 Figure 4.14: Reference natural gas composition

Figure 4.14 describes the fuel that has been taken as reference, provided by
Czech grid operator data and reported in the experimental paper. Numbers in
the table represent hydrocarbons and inerts volume percentages, so when con-
verted into mass ones they need to be multiplied with the respective molecular
weight. In order to let the engine knock, some heavy hydrocarbons are added to
the fuel, and then a multiplicity of natural gas composition of different octane
numbers is obtained.

Figure 4.15: Added hydrocarbons ranges

In figure 4.15 ethane (C2 H6 ), propane (C3 H8 ), butane (C4 H10 ) and hydro-
gen (H2 ) additions to reference natural gas composition are shown. Data as in
the figure before are reported in volumetric fractions.
In order to summarize all kinds of fuel used in this investigation, figures be-
low display composition of all kinds of natural gas used, converted into mass
percentages. Standard deviations of reference natural gas composition have not
been considered.

96
 Figure 4.16: Natural gas composition with ethane addition

Figure 4.16 illustrates fuel composition in mass percentages when 5.5% and
12% in volumetric fraction of ethane is added to reference natural gas.

Figure 4.17: Natural gas composition with propane addition

Figure 4.17 illustrates fuel composition in mass percentages when 1% and 3%

in volumetric fraction of propane is added to reference natural gas. Additions are
considered separately, hence first column reporting reference fuel composition
is the same in all the three illustrated tables.

Figure 4.18: Natural gas composition with butane addition

Finally, figure 4.18 illustrates fuel composition in mass percentages when 1%

and 2% in volumetric fraction of butane is added to reference natural gas.
Equivalence between AKR knock voltage output and pressure increasing just
due to compression ignition has been determined, exploiting pressure curves
reported in the paper. The sensor is not having as output all kinds of pressure
variation, but just the high frequency ones caused by pressure waves inside the
cylinder and consequent values oscillation. Result obtained is 2.16 ”knocking
bar” per volt.

97
 Figure 4.19: Pressure curves used for sensor signal conversion

In figure 4.19 different pressure curves are represented and AKR sensor volt-
age output is reported. Associating pressure rise due to spontaneous ignition
(recognizable from the almost instantaneous heat release) to the reported volt-
age outputs, it is possible to derive the previously described 2.16 bar/V value.
Distinction between flame propagation and compression ignition pressure rise
in solver results has been made in the same way hence, applying the conversion
value, a simulated sensor output can be obtained.
What results by this analysis is a very good knock representation when a con-
siderable amount of ethane, propane or butane is added to reference fuel compo-
sition. In the other cases, CFD represented knock is quite heavy with respect to
experimental one, probably because of the higher average flame path due to the
bigger engine adopted. In any case, even when few amounts of hydrocarbons
are added, simulated AKR voltage output does not overcome the value of 2 V,
which is the threshold between light and heavy knock.
This accordance of results can lead to a good solver behavior, but it can be
considered also a confirmation that the two combustion equations (c and cf resh
transport ones) involved in the solver do not interfere between each others in
an undesired way.

Figure 4.20: CFD simulation of voltage output sensor with ethane addition

As shown in figure 4.20, simulated and paper values are similar. Light knock

98
is presenting when 5.5% in volume of ethane is added to the reference fuel, and
results are getting similar once higher quantity of ethane is added.

Figure 4.21: CFD simulation of voltage output sensor with propane addition

And when propane is added, as shown in figure 4.21, solver results are con-
sistent if compared to ethane addition. In any case, if just 1% in volume of
propane is added, light knock occurs; if the addition increases up to 3%, knock
turns to be heavy.

Figure 4.22: CFD simulation of voltage output sensor with butane addition

And results when butane has been added to natural gas are consistent with
previous two already discussed cases. As figure 4.22 confirms, CFD and exper-
imental paper knock results are similar when few amounts of hydrocarbons are
added, while they get more different as fuel becomes heavier. As previously
written, experimental engine combustion chamber geometry is not known, but
also bore and stroke of the two are different; this can be a cause of inconsistency
between CFD simulation and experimental reported results. What has to be
highlighted is that whenever knock is described as light in the paper, also the
one predicted in CFD simulation is; if knock detected by AKR sensor is heavy,
also the computed one is.

99
4.2 Natural gas knock

In the introduction section, some physical flame and chemical behaviors have
been discussed, focusing on the importance of flame development and how au-
toignition can be modeled defining a chemical reaction delay time. Passing from
an isooctane to a natural gas knock study, two main things must be highlighted:
isooctane flame is very fast and, because of the high temperatures reached for
its octane number, autoignition is going to be heavy and to considerably speed
up the final part of combustion.

Figure 4.23: Combustion velocity fuel comparison

Adopting a compression ratio of 11.7 is prohibitive of an isooctane fuelled en-

gine running with stoichiometric homogeneous charge, as knock intensity showed
in the first subsection. From now on, the focus is pointed on natural gas and
how its composition can influence engine knocking tendency.
Natural gas is not an easy manageable fuel for internal combustion engines be-
cause, as explained in introduction section, its composition can vary significantly
region by region, influencing its chemical properties. Always in the introduction
section, a reference natural gas derived from North Europe data composition
has been chosen (displayed further on in table 4.2), presenting almost 7% of
inert gases. If compared to methane as it can be see by figures below (4.24
and 4.25), when knock is not occurring, pressure and heat release curves of two
different natural gas are very similar. Spark timings used in this section are the
ones displayed in figure 3.52, as they maximize working points efficiency.

100
Figure 4.24: Pressure curve comparison between different natural gas composi-
tion

Figure 4.25: Cumulative heat release comparison between different natural gas
composition

Lower heating values of the hydrocarbons present in natural gas are very
similar, hence they do not lead to any appreciable difference in pressure or heat
release curves: if two natural gas with both no inerts and no knock presence are
compared, resulting curves will be coincident. This is not entirely valid if some
inerts are present in the gas composition: fuel lower heating value decrease as
figure 4.25, hence maximum pressure in figure 4.24 is affected. But as it can be
deduced by heat release curves, the flame propagation phases (ignition delay,
combustion development and its completing) do not vary appreciably, and no
visible difference is present in specific flame propagation quantities.
Another very important characteristic of SNAM North Europe natural gas fu-
elled case is the complete absence of spontaneous ignition: no rough increase
in both heat release and pressure curves occurs. In consistency, if normalized
cf resh value is plotted (as in figure 4.26), it does not overcome the value of 0.5%,
leading to a negligible heat release.

101
Figure 4.26: Compression ignition progress variable of natural gas considered

This kind of consideration is numerically important, because it is showing

that there is no interaction between c and cf resh equation implemented in the
solver, meaning that flame propagation and its heat release are not affected by
the addition of a different progress variable. To have some autoignited mixture
fraction, fuel composition needs to be changed in a heavier one, with no inerts to
enhance knock presence hence richer in propane and butane. Therefore, a very
heavy natural gas composition is introduced, as illustrated in the table below.

SNAM heavy Natural Gas

Methane 81.88 % 84 %
Ethane 7.65% 8%
Propane 2.28% 6%
Butane 1.23% 2%
Nitrogen 3.77%
Carbon Dioxide 3.19%

Table 4.2: Natural gas composition in mass percentages

In table 4.2, also reference ”SNAM” natural gas is reported in order to per-
form a comparison between the two. In the introduction section, time influence
on compression ignition occurring has been emphasized, as one of the main phe-
nomenon parameters. To verify its influence, a constant torque sweep has been
performed with the ”heavy Natural Gas” just edited, in order to understand
how knock behaves in a natural gas fuelled engine and how it is influenced by
combustion duration.

102
 Figure 4.27: Natural gas knock region

As shown by figure 4.27 that catches the 1200 rpm case at 20◦ after top dead
centre, knocking region is not changed since it is always interesting the most
external cylinder region not already reached by the flame front. Natural gas
compression ignition is very violent, with an almost instantaneous pressure rise
as happened for isooctane. All the cases considered (from 1000 to 1900 rpm)
are simulated in correspondence of spark timing optimization point, as derived
in section 3.1 and illustrated in figure 3.52 of section 3.4. The slower natural gas
flame velocity is of course a parameter increasing spontaneous ignition occurring
probability, even if pressure and temperature rises are going to be smoother. At
contrary, engine cooling and heat transfer to the wall are decreasing combus-
tion chemical reaction velocity, but in this study engine cooling system is not
investigated.

Figure 4.28: Pressure curves of knocking natural gas fuelled engine

As it can be seen by figure 4.28, even if a heavy fuel is used, spontaneous

ignition is very delayed: both rounded (due to flame propagation) and vertical
(due to compression ignition combustion) pressure maximum values are clearly
visible, with the first one higher than the second in all the computed cases.
Low engine speeds present a more severe knock with respect to the higher ones;
1000 and 1200 rpm pressure curves are almost coincident, while from 1400 to
1900 rpm compression ignition pressure increase is slightly visible. With such a
delayed knock, pressure underlying area gain because of a faster combustion

103
is very small, and this is a very important consideration for spark assisted
compression ignition engine development.

Figure 4.29: Pressure rise curves of knocking natural gas fuelled engine

Pressure rise curves displayed in figure 4.29 are very consistent with what
happened when isooctane was used as fuel. Even if absolute pressure increase
is not that evident, pressure rise maximum value in 1000 and 1200 rpm cases
is comparable with supposed engine limit. When rotational speed gets higher,
knock still occurs but its magnitude is very low: at 1400, 1600 and 1800 rpm it
is present but it is so delayed to be hardly recognized, while at 1900 rpm it is
more visible having a pressure rise maximum value of 2 bar/deg. Then time has
a strong influence on autoignition presence and strength, but also temperature
is a very important parameter to be considered.

Figure 4.30: Unburnt fresh temperature curves of knocking natural gas fuelled
engine

Unburnt fresh gases are that portion of the air-fuel mixture which is not
already reached by the flame front, hence not any chemical reaction has been
performed. Other unburnt gases such as nitrogen are homogeneously distributed
in the whole combustion chamber, but they have not the possibility to perform
any wanted combustion reaction. Fuel air mixture located in the external region
have the potentiality to autoignite before flame propagation, hence this unburnt
fresh temperature chart is referred to the most external part of the cylinder.

104
”TuFresh” is also the input necessary to obtain corresponding pvS, source term
of spontaneous ignition progress variable. Fresh unburnt mixture temperature
increases because of compression and of the gradient originated from the flame,
but its values are far from the over 2200 K of the burnt gas region. Because
of the lower time available for heat transfer, when rotational speed increases
unburnt gas temperature gets always higher, and cylinder conditions become
more severe for knock.

Figure 4.31: Unburnt fresh temperature values in the whole cylinder

In addition, by figure 4.31 displayed above, TuFresh (as ”unburnt mixture

fresh temperature”) values are almost uniform in the whole cylinder, with the
only exception for regions very close to cylinder walls, strongly influenced by
heat losses. In the introduction section it has been reported that combustion
reaction velocity and autoignition delay have a square dependency on temper-
ature, as in equation 1.10. This behavior seems to confirm that experimental
correlation, since 1900 rpm case presents an appreciable pressure rise if com-
pared to some lower engine speed cases, and at the same time it is the one with
the highest TuFresh values.
Therefore, the histogram reported below shows how autoignition strength varies
between the different rotational speeds, as it has been defined as the portion of
volume interested by compression ignition combustion.

105
 Figure 4.32: Autoignition strength in constant torque sweep

This not continuous knock behavior with engine speed is obtained because
1◦ of ∆θ in spark timing optimization phase has been set. Rate of heat release
(figure 4.33) is another good indicator for compression ignition; as happened for
isooctane knock, it presents two maximum values, in perfect consistency with
autoignition strength histogram and pressure rise curves.

Figure 4.33: Rate of heat release of knocking natural gas fuelled engine

If a further focus is pointed in unburnt fresh temperature curves of figure

4.30, one can see a little detach between heavier knock occurring curves and
the others just after autoignition. Because of the geometry of engine simulated,
knock takes place in squish area region, which is distant for the flame to be
reached, hence compression ignition should contribute in an increasing of both
kinds of temperatures values (weighted averaged and unburnt fresh one). But
wall heat transfer is hugely affected by knock presence, as shown in figure 4.34
displayed below.

106
 Figure 4.34: Wall heat transfer of knocking natural gas fuelled engine

Especially in 1000 and 1200 rpm heat transfer curve, a very vertical heat loss
increasing is present. This is actually the result of the pressure waves generated:
they propagate in the whole combustion chamber and rebound in correspon-
dence of cylinder walls, and both these phenomena increase fluid kinetic energy
of the already burnt mixture inside the cylinder.

Figure 4.35: Turbulent kinetic energy of knocking natural gas fuelled engine

As predicted, turbulent kinetic energy has a very similar behavior to wall

heat transfer in knock correspondence, providing a reliable explanation of such
a temperature behavior.
Solver in the end can represent all main knock characteristics, from rapid pres-
sure and temperature increase to the presence of waves inside the CFD domain.
To complete this knock study, fuel composition has been changed in order to
consider a wide multiplicity of fuels; according to what has been described in
section dedicated to natural gas, huge variations of ethane, propane and bu-
tane mass percentages were considered. Therefore, the figure below highlights a
”knocking area” (fuel composition such that knock occurs) and a ”not knocking
area” (knock safe natural gas composition) for the engine considered at 1000
rpm, in optimal efficiency case. on x axis, mass percentage of propane; on y
axis, mass percentage of ethane. Therefore, methane composition of natural gas

107
simulated is:

%methane = 1 − %ethane − %propane − %butane (4.2)

with butane amount fixed per plane and no inerts considered. If the natural gas
is richer in heavy hydrocarbon composition it is much more knock inclined fuel;
otherwise, it is a good octane number fuel if methane composition gets higher.

Figure 4.36: Natural gas knocking areas

From figures above, it is possible to notice how engine is knocking only

if very heavy fuels are used. Inerts were not considered in this section, and
their presence is surely increasing fuel octane number. In any case, natural gas
composition used for this knock analysis is very unlikely, as the brief research
in different fuel composition shows. If a spark assisted compression ignition is
wanted, engine designers must increase the compression ratio in order to have
more spontaneous ignition strength and stability. Even by figures illustrated
in this section engine performances and their spark timing optimization are
influenced considerably by the composition of natural gas used, hence another
good target of this study is to perform a sensitivity analysis addressed to natural
gas composition, to quantify its influence on engine performances.

4.3 Boundary conditions and temperature wall

function analysis

As introduced in section 2, modeling heat transfer to the walls of an internal

combustion engine is not trivial, hence boundary conditions and wall functions
must be imposed in order not to include in the CFD domain regions external
to the combustion chamber. As illustrated in table 3.2 of section 3, a different
wall temperature is known for each wall and working point, while cylinder ones
are a result of conservation equations solution. A first focus is pointed on kinds
of boundary conditions chosen for each wall, while in the second part of this
subsection a comparison between Angelberger and Huh Chang wall functions
has been performed.

108
 Figure 4.37: Cylinder walls names

In figure 4.37 different cylinder walls are illustrated, as well as how they are
called. Head, piston and liner are the main ones most conditioning heat transfer
values, while ”adiabatic piston” and ”adiabatic liner” are closing crevice region.
While head and piston areas are fixed within power cycle, liner surface changes,
presenting a maximum value in correspondence of bottom dead centre and a
minimum one at the top dead centre. Adiabatic walls are so called because it is
possible to assume no heat transfer in a complete spark ignition combustion, but
this could be a considerable mistake when knock is considered. Two different
walls boundary condition can be imposed to the solver:
- zeroGradient which assumes the selected wall adiabatic
- fixedValue which keeps the selected wall temperature constant in the cycle
For head, piston and liner the ”fixed value” boundary condition has been ap-
plied, selecting temperature values illustrated in table 3.2. As the name sug-
gests, for adiabatic walls ”zero gradient” boundary condition can be chosen,
but a comparison between the two has been performed to verify if it influences
knock presence. As previously discussed, when these kinds of walls are consid-
ered adiabatic, knock begins very close to crevice region, as illustrated in figure
4.38.

Figure 4.38: Autoignition principle if crevice walls are considered adiabatic

109
 Spontaneous ignition hence starts between adiabatic piston and adiabatic
liner, to then propagate inside the combustion chamber. If these two walls
can exchange heat (then a ”fixedValue” boundary condition is imposed to the
solver), heat losses in correspondence of that region is very low, if compared to
minimum heat transfer values.

Figure 4.39: ”Adiabatic liner” wall heat transfer comparison

Figure 4.40: ”Adiabatic piston” wall heat transfer comparison

Figures 4.39 and 4.40 display heat losses values through these walls, applying
to both adiabatic walls a ”fixedValue” boundary condition, imposing the liner
temperature (around 440 K) to ”adiabatic liner”, and the piston temperature
(around 530 K) to ”adiabatic piston”. Minimum of global heat transfer value
was in the order of magnitude of -30 J/deg, and values reported in both figure
4.39 and 4.40 are very low. When zeroGradient boundary condition has been
set, no heat transfer can occur a priori; when a fixedValue boundary condition
is chosen, wall temperatures are constant and heat transfer coefficient αt is
determined through selected wall function.

110
Figure 4.41: Same case of before but crevice wall are not considered adiabatic

In any case, this very low heat values difference lead to a very important
consideration: when zeroGradient boundary condition has been applied, knock
occurs as shown in figure 4.38; when fixedValue boundary condition has been
imposed, no knock presence is detected as figure 4.41. This result entirely
depends on heat transferred occurring in these walls: if also head, piston and
liner ones are compared, curves are perfectly coincident but in knock presence.

Figure 4.42: Head heat transfer in adiabatic walls boundary condition compar-
ison

Figure 4.43: Piston heat transfer in adiabatic walls boundary condition com-
parison

111
Figure 4.44: Liner heat transfer in adiabatic walls boundary condition compar-
ison

As shown by figure 4.42, 4.43 and 4.44, curves are perfectly coincident with
just a liner heat transfer detach because knock occurs very close to that region.
Surely even in knock absence using fixedValue boundary condition contributes
to increase heat transfer computed, but not in such a way to compensate ex-
perimental and simulated value difference detected when validation has been
performed.

Figure 4.45: Crevice temperatures when different boundary conditions are ap-
plied

In figure 4.45 there is a CFD visual representation of that heat losses effects.
Both ”adiabatic walls” heat transfer minimum values summed together are be-
tween -0.8 J/deg and -0.9 J/deg, but if focus is pointed on local temperatures
it is possible to see how different they are: when zeroGradient boundary condi-
tion is applied, temperatures are in a range of 950 - 1100 K, while in fixedValue
BC case they are few hundreds of kelvin lower. Hence in leftwards case knock
starts at the very bottom of the region illustrated, which in more realistic rep-
resentations it is cooled down by walls presence. In conclusion, if zeroGradient
boundary condition is applied, simulated knock is very delayed (over 30◦ after
top dead centre) because that region is very difficult to be reached by flame
front, but it is very unlikely because it is actually cooled by walls presence.

112
”FixedValue” boundary condition offers a much more realistic simulation, and
it is the only one applied from section 5 on.
As anticipated at the beginning of this subsection, also a wall function compar-
ison has been performed. Angelberger WF is the one used so far. Huh Chang
wall function has been developed to have a more realistic representation of heat
transfer in HCCI combustion, where influence of pressure waves is huge. Target
of different wall functions is computing αt , heat transfer coefficient used to de-
termine heat losses. Huh Chang wall function is composed by one steady term
and a time-dependent transient term, as explained in [18].

N
k X
q̇wall,HC = (Tm −Tl )+k Φn [(An +Bn )cos(nωt)−(An +Bn )sin(nωt)] (4.3)
l n=1

The presence of a time dependent term makes this formula better behaving in
correspondence of huge pressure waves inside the cylinder. HCCI combustion is
considered too violent and very difficult to control, hence huge pressure waves
must be considered in heat transfer formula.
Angelberger expression is much closer to standard wall functions ones:

ρcp uτ T ln( TWTall )

q̇wall,AN = (4.4)
2.075ln(y + ) + 3.9

As already remarked, this last wall function is the one used in section 3 where
engine validation has been performed. For a brief remind, almost all computed
heat transfer curves were underestimated, especially in cumulative terms. Min-
imum values were close to the experimental ones when engine regime was not
high, and even its timing was almost correct. Curves detached after they reached
minimum values, causing cumulative heat transfer differences of more than 25%
in the worst cases. With this comparison, the author wants to verify that the
used wall function is the best one fitting the experimental data, since, as many
case in CFD, there is not an ”a-priori” best one to be used.

Figure 4.46: Head heat transfer wall functions comparison

113
 Figure 4.47: Piston heat transfer wall functions comparison

Figure 4.48: Liner heat transfer wall functions comparison

As it can be seen by figures 4.46, 4.47 and 4.48, Huh Chang is further un-
derestimating heat transfer in correspondence of the three main cylinder walls.
In compression stroke and in combustion developing phase, all the three derived
curves are almost coincident, but after combustion development the detach is
clear with an always lower heat losses estimation of the Huh Chang wall function
with respect to the Angelberger one.

Figure 4.49: ”Adiabatic piston” heat transfer wall functions comparison

114
 Figure 4.50: ”Adiabatic liner” heat transfer wall functions comparison

While main combustion chamber heat losses are hugely underestimated by

Huh Chang wall function, figures 4.49 and 4.50 show that heat losses curves of
crevice region walls are almost coincident. But in the end this wall function
used in HCCI engine simulation is considerably underestimating heat losses. As
happened when the two boundary conditions were compared, this heat losses
underestimation provokes a knocking region.

Figure 4.51: Simulated knock when Huh Chang wall function is used

As illustrated by figure 4.51 displayed above, simulated knock occurs when

Huh Chang wall function is used, even if ”fixedValue” boundary condition is
imposed to adiabatic walls. In any case, this kind of knock is different with
respect to the one obtained before: in the previous case, it started from crevice
region to then propagate inside the combustion chamber, but in this one it is
originated in correspondence of both piston and head walls. Mixture in crevice
region does not chemically react because it is cooled down by ”adiabatic walls”,
presenting also very low turbulence values. Even if this kind of knock is origi-
nated by a heat transfer underestimation, this analysis provides a better idea of
where it starts and how it develops: knock obtained in ”zeroGradient” boundary
condition case is just numerical, starting in a region where spontaneous ignition
presence is almost impossible; this kind of knock, even if it is caused by a huge
heat losses underestimation, has a much more realistic point of origin to then
propagate inside the combustion chamber, as the others did.
In conclusion, from this section several useful results have been obtained: solver
including knock has been validated using an experimental publication, knock as

115
represented in the solver has been analyzed identifying its most important phys-
ical and numerical peculiarities, influence of natural gas composition has been
investigating and, in the end, importance of wall heat transfer influence of knock
has been highlighted. From now on, Angelberger wall function has been used to
not have a further heat transfer underestimation with respect to experimental
working point data of section 3, and fixedValue boundary condition has been
imposed to avoid numerical knock. Figure 4.52 briefly summarizes different wall
function heat transfer values obtained in this last subsection analysis.

Figure 4.52: Total heat transfer wall functions comparison

For a spark assisted compression ignition (SACI) engine development, a

stable spontaneous ignition must be achieved, without sacrificing accuracy of
solver results. As a consequence, engine compression ratio must be increased, in
order to reach more severe pressure and temperature values and have a stable
and reliable natural gas compression ignition, leading to engine performance
advantages.

116
Chapter 5

SACI engine development

In previous section CNG-heavy duty engine could knock only if powered with
very heavy natural gas, unlikely to be found in ordinary fuel stations. To develop
a SACI (spark assisted compression ignition) combustion, spontaneous ignition
must be reliable and under control, not depending on kind of natural gas used.
Therefore, compression ratio has been increased to have higher pressures and
temperatures and then enhance fresh air-fuel mixture autoignition. This leads
to a redesign of the combustion chamber, performed in such a way to increase
the compression ratio adopted but to not introduce significant changes on piston
shape.

Figure 5.1: Combustion chamber shapes at top dead centre

As it is possible to see in figure 5.1 where original design and the new one are
compared, volume at top dead centre of increased compression ratio chamber is
far way lower. Parameters such as bore and stroke are not changed, so this kind
of modification practically implies just a substitution of the original piston with
a new one. As a consequence, even head cylinder shape remains the same, while
the size of the crevice region has been kept as similar as possible between edited
designs. A multiplicity of engines working with different compression ratios has
been investigated, all of them obtained with a volume reduction that maintains
constant during the whole cycle, as shown in figure 5.2. By both figures 5.2
and 5.3, considering the same working points (characterized by initial cylinder
pressure and temperatures values) a lower mass of fuel is injected inside the
cylinder.

117
 Figure 5.2: Volume variation with compression ratio

Figure 5.3: Mass of fuel injected in different compression ratio engines

This leads to a first advantage in terms of efficiency: as shown in figure

above, for the same initial pressure and temperature values, in the CR 15 case
fuel amount is more than 2% lower with respect to the reference one at CR 11.7.
In addition, since a better thermal efficiency is predicted, even gross indicated
work should increase, leading to another efficiency benefit. Also knock must be
taken into account: in a spark assisted engine, it needs to be controlled to be
exploited for a more ideal constant volume combustion. At this purpose, two
different solvers are used in this section:
- betaFlameletXiEngineDyMFoam: including b equation
- betaFlameletSACIXiEngineDyMFoam: including b and cf resh equations
As briefly described, the just ”Xi” solver can not consider autoignition, and
as a consequence it is possible to simulate very extreme compression ratio en-
gines without any knock presence. ”SACI” solver instead includes both flame
propagation combustion and chemical reaction delay time, computing also heat
released by a possible mixture autoignition. A comparison between these two
solver results is very important to isolate combustion mode influence from all
the other parameters that affect the power cycle, and to have clearer advantages
and disadvantages concerning just spontaneous ignition.
To conclude this last section introduction, in figure 5.4 a render image of the
combustion chamber adopting a compression ratio of 15 designed using the soft-
ware Autodesk Inventor is illustrated. It is possible to see how important the

118
squish region is to generate turbulences inside the whole cylinder, as much as
combustion volume is far way lower with respect to 11.7 CR one. To have such
a small space for combustion leads to a delayed optimum spark timing, because
average flame path length is lower and it needs lower time to be covered.

Figure 5.4: Autodesk inventor design render, with 15 as CR

5.1 Spontaneous ignition effects on engine per-

formances
As already proved, spontaneous ignition can be very rough and violent, stressing
in a dangerous way mechanical components of the engine. Due to the almost
instantaneous heat release, it is possible to reach very high cylinder pressures
and, if bad controlled, it can turn into heavy knock. For this reason, not all
the working points in the engine map can work using a SACI combustion mode,
but to protect engine mechanical structure two different pressure limits are
introduced: an absolute maximum value slightly above 180 bar, and a pressure
rise not to overcome 18 bar/deg.
Therefore not all the working points in the engine map are suitable for a spark
assisted compression ignition combustion, but just the ones at half load or lower.
In the figure below engine map is revisited, with SACI combustion target points
drawn in green, while entirely flame propagation combustion ones in red. Low
load thermal efficiency were the most penalized in the optimization section. In
addition, at high loads knock is more likely to occur, hence to be able to perform
both kinds of combustion for the same regime (as it happens for 1200 and
1600 rpm) can be complicated. At first analysis, SACI has been developed for
1200 rpm 52% load working point; then feasibility of entirely flame propagation
combustion at full load is verified.

Figure 5.5: Revisited engine map, showing SACI points target

119
 As previously described both solvers have been used, indicating with ”Xi”
the one considering just flame propagation, while with ”SACI” the one consid-
ering both flame propagation and autoignition. In this analysis, the focus is
pointed on gross indicated work variation between the two combustion modes
and the compression ratio of 15 has been chosen for this comparison.

Figure 5.6: Pressure curves obtained by both solvers in the same initial condi-
tions

As illustrated in figure 5.6, spontaneous ignition occurs in correspondence

of maximum pressure value, evidently increasing SACI solver curve underlying
area. A spark timing of -16◦ has been imposed to both the cases, as well as
SNAM North Europe derived natural gas composition (the same that could not
autoignite in section 4) has been used. Because of spontaneous ignition, pressure
maximum value increases considerably, over than 20% in relative terms. Espe-
cially regarding cumulative heat release curve of figure 5.7, one can appreciate
how fast compression ignition combustion is in homogeneous mixture condi-
tions: in the case where it is included in the solver, combustion ends between
13-14◦ crank angle, almost 20◦ before entirely flame propagation combustion is
completed. This of course leads to a very severe and dangerous pressure rise,
that overcomes enormously the already high assumed limit.

Figure 5.7: Cumulative heat release curves obtained by both solvers in the same
initial conditions

120
Figure 5.8: Pressure rise curves obtained by both solvers in the same initial
conditions

As shown in figure 5.8, pressure rise is close to 30 bar/deg, more than 12

bar/deg over the engine limit but, since this analysis is focused on efficiency
variation, this is not the main analysis target at the moment. If cumulative
gross indicated works are compared, result obtained seems in contradiction with
pressure curves previously illustrated. As shown in figure 5.9, after spontaneous
ignition occurs curves intertwine two different times: at the very beginning
where SACI one takes benefits of the faster combustion, but in the end of
the expansion stroke gross indicated work curve computed by ”Xi” solver gets
higher, resulting in the end to have a better thermal efficiency.

Figure 5.9: Gross indicated work curves obtained by both solvers in the same
initial conditions

It is hard to detect differences in curves behaviors of figure 5.9 because they

are almost coincident, and it is also hard to recognize by sight which one is
presenting higher or lower values. To have a better knowledge of work during
the cycle, instantaneous gross indicated work is computed as illustrated in the
equation below:
pi +pi+1
p · ∆V 2 · (Vi+1 − Vi )
GIWinstantaneous = = (5.1)
∆θ θi+1 − θi

121
In equation 5.1, ”i” stands for time step considered, ”p” for cylinder pressures,
”V” for combustion chamber volume and, in the end, ”∆θ for the time step
used computed in crank angles. In this way it is much easier to follow gross
indicated work variation during the cycle, as in figure 5.10.

Figure 5.10: Instantaneous gross indicated work curves obtained by both solvers
in the same initial conditions

Before top dead centre, instantaneous gross indicated work is negative be-
cause piston motion is compressing air fuel mixture. In the expansion stroke,
work gets higher because of combustion presence, and cumulative gross indi-
cated work as figure 5.9 is considering both work spent in compression stroke
and useful one derived from expansion stroke. Returning to figure 5.10, the two
curves are coincident up to spontaneous combustion presence, which increases
blue curve pressure values hence instantaneous work made by burnt gases. Even
if the detach is clear, in less than 20◦ the curves crosses, and expansion com-
puted by ”Xi” solver is clearly more efficient with respect to the one computed
by the ”SACI” one whenever combustion process is completed.
As predicted by the theory, knock is characterized by the presence of huge pres-
sure waves increasing the heat transferred to the walls. As figure 5.11 confirms,
in knock presence minimum wall heat transfer values become huge, affecting
cylinder temperature hence useful work in the expansion stroke.

Figure 5.11: Wall heat transfer curves obtained by both solvers in the same
initial conditions

122
Figure 5.12: Temperature curves obtained by both solvers in the same initial
conditions

Wall heat transfer minimum value in knock presence is more than four times
lower than the one computed by solver ”betaFlameletXiEngineDyMFoam” re-
ported in red. Effects on temperature curves are enormous, since immediately
after compression ignition weighted average maximum values are higher of 300
- 400 K, but then burnt mixture cools down in a very fast way and blue curve
negative slope after compression ignition is considerable. Burnt mixture has
hence less energy to be exploited for almost all the second part of the expansion
stroke, and cumulative gross indicated work gets lower at the end of the cycle.
This leads to a sort of paradox, because with spark assisted compression ignition
a more ideal combustion mode has been reached, but losses are so important
to considerably reduce power cycle work and efficiency. As it can be deduced
by heat release curve, autoignited mixture fraction is in the order of magni-
tude of 40%, which is very high in homogeneous mixture condition. Figure 5.14
displayed below shows spontaneous ignition principle timing related to fraction
burnt because of flame propagation.

Figure 5.13: Autoignition propagation in CR 15 engine; 13 deg after TDC

leftwards, 14 deg after TDC rightwards

123
 Figure 5.14: Autoignition principle in CR 15 engine

Adopting such a high compression ratio, temperatures reached inside the

combustion chamber make spontaneous ignition propagation even faster, result-
ing in very heavy knock. Values reported in table 5.1 summarize this comparison
between different solver results. As evident, pressure and weighted average tem-
perature maximum values increase, but especially cumulative wall heat transfer
rise is impressive, of more than 46%. This leads to the apparent contradiction
that a more ideal Otto cycle such as the one computed by SACI solver has a
lower thermal efficiency with respect to a less ideal one. In figure 5.15 it is
evident how spark assisted combustion is faster and closer to a constant volume
one, however resulting efficiency is lower because of too huge losses.

Figure 5.15: Pressure-volume curves obtained by both solvers in the same initial
conditions

η GIW cum Heat losses Pmax Tmax

SACI 40.78% 2759 J -1359 J/deg 113.2 bar 2633 K
Xi 41.89% 2827 J -927 J/deg 95.2 bar 2276 K

Table 5.1: SACI and Xi solvers comparison

124
5.2 Wall temperature, spark timing and load
sweeps
Spark assisted compression ignition combustion in previous subsection turned to
be more ideal but also less efficient because of huge wall heat transfer presence.
In order to reduce heat losses, temperatures imposed to the walls have been
increased by 80◦ C (per wall), in order to verify if a thermal efficiency increase
is associated.

Head Piston Liner Ad. Piston Ad. Liner

SACI 551 K 499 K 413 K 499 K 413 K
SACI high T 631 K 579 K 493 K 579 K 493 K

Table 5.2: Wall temperature values

To distinguish this configuration with the previous one, the name ”SACI
high T” has been chosen to indicate the configuration with high wall temper-
atures imposed; with the label ”SACI” instead the previously analysed one is
reported. Since cylinder walls are hotter in this case, heat losses should decrease
because of lower temperature difference between air-fuel mixture and cylinder
walls and, consequently, thermal efficiency is predicted to increase. This is what
happens when combustion is totally driven by flame propagation, while if also
spontaneous ignition is considered this can turn into an enhanced probability
for knock to occur. As it is possible to see in figure 5.16, autoignition is an-
ticipated in ”SACI high T” configuration, presenting also an increased pressure
maximum value.

Figure 5.16: Autoignition anticipation when engine walls are at higher temper-
atures

In cumulative wall heat transfer curves of figure 5.17, the high wall temper-
ature curve is clearly above the ”SACI” one before spontaneous ignition occurs,
and heat losses are hence contained; but once knock presents, red and light
blue curves intertwine, and at the end of the power cycle the configuration that
should theoretically contain wall heat transfer is worse behaving because of the
heavier knock.

125
Figure 5.17: Heat losses curves when engine walls are at higher temperatures

From this very brief investigation, increasing walls temperature is not a good
strategy to take advantage from homogeneous compression ignition combustion.
As it is possible to see in instantaneous gross indicated work curves in figure
5.18, during the expansion stroke the light blue one is always above the dark red
one, hence the lower the amount of mixture autoignited the higher the work done
by burnt gases. This is also summarized in table 5.3 where thermal efficiency of
the engine decreases of about 1% instead of growing up, also introducing higher
loads to the mechanical structure of the machine. By a first sight analysis, an
increase of autoignited fraction leads to a thermal efficiency decrease.

Figure 5.18: Instantaneous gross indicated work curves when engine walls are
at higher temperatures

η GIW cumulative Heat losses Pmax Tmax

SACI 40.78% 2759 J -1359 J 113.3 bar 2633 K
SACI high T 39.92% 2700 J -1470 J 118.3 bar 2648 K

Table 5.3: Standard and high wall temperature performance comparison

To verify this behavior and to consider even more extreme conditions, a

spark advance sweep has been performed, considering a delayed mixture ignition
(corresponding to -14◦ ) and an anticipated one (at -19◦ ). During this analysis,
also autoignited mixture fractions are compared between different cases.

126
Figure 5.19: Cumulative heat release in high compression ratio spark advance
sweep

From heat release curves of figure 5.19, it is possible to see how combustion is
further speeded up when spark ignition is anticipated. The more severe pressure
and temperature conditions contribute to a more ideal combustion, and it gets
even more similar to a constant volume one. While pressure curve underlying
area varies evidently, temperature curve behavior is far way more similar: both
of them are weighted average values taking into account all the cells in the
CFD domain. This means that in presence of compression ignition temperature
weighted average maximum values change because of the higher number of cells
involved in combustion process, but maximum cell burnt temperatures do not
vary considerably.

Figure 5.20: Pressure and temperature curves in high compression ratio spark
advance sweep

127
Figure 5.21: Instantaneous gross indicated work in high compression ratio spark
advance sweep

As in all the other cases, instantaneous gross indicated work provides very
important indications of points of strength and weakness of this combustion
mode during the cycle, showing how most delayed spark timing configuration
is the best one in this sweep. Once engine performances are considered, a
higher underlying pressure - θ curve area is not directly corresponding to higher
performances, because energy released by the fuel contributes to increase more
heat losses and fluid kinetic energy than useful work done by burnt gases. If a
deeper analysis is performed, pressure and turbulent kinetic energy values must
be regarded cell by cell during all the different combustion phases. To make
this analysis even clearer, the most knock severe case is investigated, hence the
one at anticipated spark timing with increased cylinder wall temperatures.

Figure 5.22: Flame front and cylinder pressure in an entire flame propagation
combustion

Figure 5.22 shows pressure increasing inside the cylinder in an entire flame
propagation environment, before spontaneous ignition occurs. For these repre-
sentations, it has been chosen the regress variable ”b” to indicate flame front
and surface positions, and the normalized progress variable ”cf resh ” to refer to
compression ignition combustion. In flame propagation combustion pressure rise
is homogeneous inside the cylinder, with very low almost numerical gradients in

128
regions where mesh orthogonal quality is particularly poor. Temperature values
are not as uniform as pressure ones, such that in burnt gas region 2700 - 2800
K are reached, while fresh mixture temperatures are stable between 1000 K and
1100 K. This provokes some density differences between different areas, hence
fresh mixture mass concentrates in most external cylinder regions.

Figure 5.23: Flame front and cylinder temperature in an entire flame propaga-
tion combustion

In the very first phases of compression ignition development, because of huge

amount of heat released high pressure values appear in the most external region
of the cylinder, introducing significant pressure gradients inside the combustion
chamber. As a consequence, the working fluid is moving from high density re-
gions towards the more internal one, at very high velocity. Figure 5.24 displayed
below captures one of the very first autoignition phases, with the presence of
an always higher pressure gradient originated in the most external part of the
cylinder. Since heat release is huge and squish transversal area is constant, that
high pressure fluid finds a lot of difficulties in expanding, originating the wave
and hence derived pressure gradient.

Figure 5.24: Pressure gradient at compression ignition origin, 8 deg after TDC

It has previously discussed in this thesis work how low time step imposed

129
and considered 2D geometry are a limit of this analysis introduced to opti-
mized computational resources available; but also using these solver settings,
pressure gradient after some crank angle degrees is moving from the external
part of the cylinder to the internal one, increasing of two orders of magnitude
fluid turbulent kinetic energy. This behavior is strongly connected to pressure
waves presence, which move from the most external cylinder region to the in-
ner of the combustion chamber to then rebound in correspondence of cylinder
walls. Turbulence k value that have been plotted so far are derived performing
a weighted averaged considering all mesh cells, hence maximum κ values are far
way higher when the whole CFD domain is displayed. This kind of behavior is
actually typical of all knock met so far, and it provides a further explanation on
performance decrease encountered. Some crank angle degrees after spontaneous
ignition is completed, pressure gradient inside the combustion chamber becomes
maximum, with pressure differences of over than 30% in most severe knock con-
dition. All of it is contributing into increasing kinetic energy of the working
fluid, and as a consequence heat transfer coefficient between the fluid and the
wall, explaining also why heat transfer curve and turbulent kinetic energy ones
have such a similar behavior.

Figure 5.25: Pressure gradient immediately after compression ignition is com-

pleted, 9 deg after TDC

Figure 5.26: Pressure gradient after wave has propagated inside the combustion
chamber, 11 deg after TDC

Between figure 5.24, 5.25 and 5.26 there are just three crank angle degrees

130
of difference, but cell pressure values change completely. All of them are of
the same case, but these figures in sequence provides an idea on how wave
propagates inside the whole combustion chamber and how difficult is for the
fluid to expand in squish region. If in the CFD domain turbulent kinetic energy
during flame propagation is compared to the one after compression ignition
occurring, behavior described is further verified. Therefore, in figure 5.27 it
is possible to see two turbulent kinetic energy time step values of the same
case: leftwards, during flame propagation but before autoignition appearance;
rightwards once spontaneous ignition is almost completed and huge heat has
been released. As evident, maximum κ values differs of more than two orders of
magnitude. In an entire spark ignition context, turbulent kinetic energy reaches
its maximum values very close to squish area, which is designed to increase
fluid turbulence and hence flame velocity; after autoignition occurs, κ reaches
its maximum values as soon as the fluid is exiting from the squish area to go
towards inner parts of the combustion chamber, as soon as there is more space
for the pressure wave to propagate.

Figure 5.27: Turbulence intensity comparison between before (left) and after
(right) autoignition

As a consequence, the higher the quantity of spontaneously ignited mixture

fraction the higher the efficiency loss. This is clearly visible by figures 5.28
and 5.29 displayed below, where the best efficiency case corresponds to lower
autoignition fraction one.

Figure 5.28: Efficiency curve with spark timing in a heavy knock engine

131
Figure 5.29: Autoignited mixture fraction in spark timing sweep. AI indicator
is the ratio between autoignited mixture mass and the whole mixture mass

In chapter 3 where engine has been optimized, it has been discussed how
efficiency curve in an entire flame propagation context is stable, varying of
some tenths of percentage for each spark timing degree of change. Figure 5.28
illustrated above clearly shows that this is not the case anymore: there is an
absolute efficiency variation of more than 4% between most spark delayed case
and most spark anticipated one. As summarized by table 5.4 below reported,
minimum of wall heat transfer value changes of more than four times between
these two cases, conditioning cumulative curves in a huge way. From first to the
last case (always referred to table 5.4), heat losses becomes almost the double,
with a consequent reduction of gross indicated work. Pressure and temperature
maximum values reach very high numbers for a half load condition.
To summarize what has been derived so far:

- Autoignition is originated in squish area, to then propagate into inner region

of the combustion chamber

- In squish area, once autoignition occurs the compressed burnt mixture has no
space to expand: this is enhancing pressure wave presence and turbulent
kinetic energy, hence heat transfer between the working fluid and the walls

- About heat losses, in squish area distance between the piston and the head
of the cylinder is very small, hence heat transfer area between the wall
and the working fluid is very big. Volume/surface ratio in that region
when spontaneous ignition occurs is very low, and the just burnt mixture
at very high temperature dissipate almost all of heat released, with very
poor thermal inertia.

θSpark η Heat transfermin Heat losses GIW Pmax Tmax

-14◦ 41.72% -52.4 J/deg -1100 J 2818 J 104.7 bar 2618 K
-16◦ 39.32% -128.1 J/deg -1470 J 2700 J 118.3 bar 2648 K
-19◦ 36.96 % -219.2 J/deg -1946 J 2506 J 131.7 bar 2702 K

Table 5.4: Spark timing sweep performance table

132
In these pages, very heavy knocks have been obtained more than a controlled
spontaneous ignition. Before modifying combustion chamber, some other con-
siderations are needed, hence a load sweep has been performed to check what
happens if cylinder works at lower pressures. When engine is running at de-
creased loads, there is a lower quantity of mixture mass inside the cylinder,
with a consequent density reduction. Therefore, in one hand, the lower thermal
inertia should increase heat losses if related to mixture mass, as it happened in
an entire flame propagation context; but in the other one, spontaneous ignition
should delay, leading to a lighter knock and, for what has been obtained so far,
an increase of useful work from power cycle. Also, full load condition has been
taken into account, to check maximum pressure values reached and how knock
occurs even at very severe cylinder temperatures.

Figure 5.30: Pressure curve in load sweep, in knock presence

As illustrated by pressure curves of figure 5.30, the higher the load, the
heavier the knock. Temperature curves are easier to be compared, since engine
is working in stoichiometric condition hence adiabatic flame temperature does
not change within the sweep. Full load maximum value in figure 5.31 is both
the highest one and the most anticipated, but it soon crosses with lower loads
temperature curve leading to the presence of more important heat losses.

Figure 5.31: Temperature curve in load sweep, in knock presence

Wall heat transfer as shown in figure 5.32 presents very low values, even if

133
compared with spark timing sweep previously performed. At full load condition,
knock is so heavy that wall heat transfer is lower than -300 J/deg.

Figure 5.32: Wall heat transfer in load sweep, in knock presence

When a load sweep is performed, it is difficult to compare gross indicated

work because of the different fuel amount injected. Instantaneous gross indi-
cated work has been displayed in figure 5.33, presenting curves almost moving
on y-axis direction with the different load imposed. The only thing that needs
to be remarked is the vibration increase once knock occurs, which can be also
detected by wall heat transfer figure above, and it is strongly connected with
knock intensity.

Figure 5.33: Instantaneous gross indicated work in load sweep, in knock presence

Passing to efficiency curve, once again violent autoignition leads to no gross

indicated work benefit. At the contrary of what has been obtained in section 3,
low load working points present a better efficiency with respect to high load ones,
just because spontaneous ignition is weaker. In any case, efficiency curve has
not a constant shape or behavior, but it varies between different load conditions:
by the first three point, engine efficiency is decreasing in always faster way, even
with η difference of 1.5% between half load and 68% load conditions. Between
68% and full load working points, slope of the curve is much higher, hence an
efficiency decrease of just 0.35% is present. It has been shown how full load
autoignition strength is far way the highest computed in this work, and to
provide explanation of this behavior, compression ignition progress variable is
displayed in CFD domain, as represented in figure 5.35.

134
 Figure 5.34: Efficiency curve in load sweep, in knock presence

Figure 5.35: Autoignition principles in full load condition

As it is possible to see in figure 5.35, mixture is spontaneously ignited in two

different points: one in squish region as in all other simulations, another one in
the combustion chamber in correspondence of the piston curvature. Both are
very violent causing huge pressure waves, but in squish area burnt gases are
transferring heat almost to the walls, while a big portion of the whole second
autoignition core surface is facing the mixture already burnt by flame propaga-
tion. Therefore, instead of losing that fraction of heat release, energy transfer
takes place between both kinds of burnt gases, not wasting fuel energy. Engine
is taking some advantages from the faster combustion, and a lower percentage
of spontaneous ignition heat release is dissipated through the cylinder walls.
Hence, homogeneous mixture compression ignition can be an opportunity to in-
crease efficiency of the engine, but it has to be correctly exploited to not waste
that kind of heat release enhancing losses.
This analysis concerning full load condition has been performed just to have
a better knowledge of how compression ignition can influence engine perfor-
mances. As showed by figure 5.5 at the very beginning of this section, full load
working point is not of interests of spark assisted compression ignition, because
of too high pressure values reached.

135
 Figure 5.36: Pressure maximum values in load sweep, in knock condition

As shown in figure 5.36, computed maximum pressure in full load condition

is 225.2 bar, more than 45 bar over the assumed limit. Such high stresses are
unbearable for the engine, hence in spark assisted compression ignition design,
high loads and vibrations must be considered.
To conclude this subsection, turbulent kinetic energy in CFD domain is shown in
figure 5.37. It is possible to see how second core of heat release is not increasing
working fluid turbulent kinetic energy as much as squish region one. Value scale
has been edited in order to show that fluid kinetic energy close to the piston
is in the range of 100-200 m2 /s2 , while at the inlet of squish region computed
values can overcome even 10000 m2 /s2 .

Figure 5.37: Turbulent kinetic energy values in full load condition

In these two first subsections knock phenomenon has been further analysed
and an almost complete knowledge of its behavior is provided. Combustion
always results into a more ideal one, visible especially in pressure volume plane.
To take advantage of it, two different strategies are investigated: piston shape
changes and leaner air fuel ratio. The first one can be explained by these last
considerations, showing that pressure waves are enhance by squish region; the
second aims to have both a far way less violent compression ignition and to have
more thermal inertia of gases inside the cylinder, in order to reduce heat losses.

136
5.3 Piston geometry influence
It has been demonstrated how squish region is important for flame properties but
also detrimental for a good exploitation of compression ignition. Heat released
from homogeneous charge autoignition is far way faster than the one coming
from flame propagation; hence combustion chamber has been redesigned in the
external region with the aim of taking benefits of this combustion mode, keeping
the same compression ratio. Most important geometrical parameters such as
stroke, bore or cylinder head shape have not been changed to not impose huge
modifications to the engine: once again piston shape is investigated, focusing
on most external regions.

Figure 5.38: Piston geometry comparison at top dead centre with respective
names

In figure 5.38 geometry changes are illustrated, with both images taken in
correspondence of top dead centre piston position. Some more space has been
added in correspondence of squish area, in such a way to expand the pressure
wave originated by compression ignition. The extension of this region forces the
designer to reduce the distance between cylinder head and piston in the central
part of the combustion chamber (in correspondence of the axis of symmetry),
and for flame propagation this is a huge disadvantage because flame front sur-
face will be lower and combustion is predicted slower than before. Taking in
consideration these assumptions, also a third piston geometry has been simu-
lated, called ”Turb”. Just this last combustion chamber design has a slightly
lower compression ratio (14.8), and it has been edited to find a good exploitation
of flame surface properties.

Figure 5.39: ”Turb” piston geometry at top dead centre

137
 In both of these two last edited geometries, importance and influence of
squish area have been remarkably sacrificed to exploit homogeneous charge com-
pression ignition, which in CNG-heavy duty geometry was too violent. In figure
5.40 displayed below, an ”Eng” geometry render has been showed, evidencing
how squish area influence has been considerably decreased.

Figure 5.40: 3D render of ”Eng” combustion chamber

Therefore, a first set of simulations has been performed, keeping the same
working point parameters such as initial pressure, temperatures, stoichiometric
air-fuel mixture ratio and turbulence, with a spark timing of -16◦ imposed for
all the different geometries.

Figure 5.41: Pressure curves associated to different piston geometries

From figure 5.41 displayed above, it is evident how combustion is far way
slower in ”Eng” and ”Turb” geometries, which have almost coincident pressure
curves. Spark ignition combustion in CNG-heavy duty engine is considerably
faster but pressure rises due to autoignition of the three cases are comparable.
As predicted, combustion velocity is reduced because of the lower flame front
surface but also to the lower turbulence intensity presence, as shown in the
figure below.

138
Figure 5.42: Flame front visualization in CFD domain associated to different
piston geometries

Figure 5.43: Turbulence intensity associated to different piston geometries

Figure 5.42 reported above illustrates in a good way behavior previously

described: CNG-heavy duty geometry presents a flame front with a higher sur-
face, leading to considerable advantages for combustion. If the two flame front
distances along piston head are compared, it is evident how CNG-heavy duty
flame is also faster, leading to a better spark ignition combustion under every
aspect. These CFD domains shown are in correspondence of 4◦ after top dead
centre.
Even if combustion development has been slowed down, power cycle efficiency
values between the three different cases are comparable:

solver η GIW Heat Losses

CNG-heavy duty SACI 40.78% 2759 J - 1359 J
Eng SACI 41.69% 2831 J - 1169 J
Turb SACI 42.01% 2869 J - 1080 J
CNG-heavy duty Xi 41.89% 2827 J - 927 J

Table 5.5: Efficiencies comparison of different piston geometries, with the same
initial conditions and spark timing

139
 In table 5.5, efficiency values are reported in such a way they can be com-
pared. When ”SACI” solver is chosen compression ignition always occurs, but
both ”Eng” and ”Turb” efficiency values are higher with respect to CNG-heavy
duty one. In the previous subsection, a comparison between ”SACI” and ”Xi”
solver results has been performed, to verify efficiency gain just due to new com-
bustion mode; by the very same table, Eng and Turb geometry efficiency values
are very close to the target of 41.89%, even overtaken of one tens by Turb piston
geometry. This results have been obtained in a non-optimized condition as the
one displayed in figure 5.42, demonstrating that in these two edited geometries
combustion can take advantages from a partial compression ignition even when
air-fuel mixture is stoichiometric. Therefore, a spark time sweep has been per-
formed, not just to find an optimized value for these two engines, but also to
understand how compression ignition behaves and if there are differences be-
tween these geometries and CNG-heavy duty one.
The spark advance sweep has been performed for both the geometries, but just
”Turb” one is illustrated since they lead to very similar considerations.

Figure 5.44: ”Turb” pressure curves in spark timing sweep

Figure 5.45: ”Turb” heat release curves in spark timing sweep

140
 Pressure curves behavior (shown in figure 5.44) is consistent with the one
previously analysed. Compression ignition is occurring in every cycle, with an
increasing intensity as spark timing is anticipated. By cumulative heat release
curves of figure 5.45 it is possible to see how autoignition strength is important
in all cases. When spark timing is very delayed for that geometry (16◦ before
top dead centre), autoignition occurs in correspondence of 18◦ , which is very
delayed if compared to previous cases.
It has been discussed how flame propagation has been sacrificed to favor com-
pression ignition. Especially, geometry aims to slow spontaneous ignition down
of even some tens of degree, in order to have a smoother pressure rise hence
lower pressure wave intensity. In figure 5.46 it is shown autoignition develop-
ment from its origin (11◦ ) to its ending (14◦ ). Changing in squish area is also
important to give a shape or compression ignition development, as happens with
flames. When squish area was adopted, as soon as autoignited mixture has more
space, it developed in a messy way, originating pressure waves. With this geom-
etry, cf resh has a more defined shape very similar to the one of a flame coming
from most external regions, hence wave generated is far way lower leading to
heat losses reduction. In figure 5.47 also ”Eng” geometry compression ignition
is reported, to verify that this smoother kind of compression ignition is due just
to the redesign of the most external cylinder region, not properly a squish area
anymore.

Figure 5.46: Compression ignition evolution between 11 deg and 14 deg in

”Turb” geometry

141
Figure 5.47: Compression ignition evolution between 11 deg and 14 deg in ”Eng”
geometry

Spark timing influences spontaneous ignition strength, as cylinder fresh mix-

ture can reach different temperature values. As previously described, weighted
average turbulent kinetic energy can provide an indication of pressure wave in-
tensity, but maximum mesh values can be very different from weighted averaged
ones. Figure 5.48 presents a sort of hyperbolic relation between turbulent ki-
netic energy peaks, leading to absolutely not stable cumulative heat transfer
curves.

Figure 5.48: ”Turb” turbulent kinetic energy curves in spark timing sweep

In previous section relation between heat transfer and kinetic energy has
been described, emphasizing how fluid motion originated by spontaneous igni-
tion is increasing heat transfer coefficient. This simulation set provides another
confirm of that behavior, with contained heat transfer minimum values when
high intensity pressure waves are not present, to then fall down as soon as
compression ignition gets stronger.

142
 Figure 5.49: ”Turb” wall heat transfer curves in spark timing sweep

In figure 5.49, wall heat transfer curves are shown. Minimum values of
”Spark Advance -16” and ”Spark Advance -20” curves are contained in the
interval between -20 and -40 J/deg, far way higher with respect to the one
seen in CNG-heavy duty when autoignition was occurring. With such a behav-
ior, engine can take advantage of a faster spark assisted compression ignition
combustion even in stoichiometric homogeneous charge, since heat losses are
considerably contained. When cumulative curves are compared, this behavior
is even more evident: as illustrated in figure 5.50 reported below, end stroke
cumulative heat transfer values decrease more than linearly with spark timing,
leading to a heat losses increase and a huge performance reduction in the second
part of the expansion stroke, once combustion is completed.

Figure 5.50: ”Turb” cumulative wall heat transfer curves in spark timing sweep

But this effect is not as detrimental as for CNG-heavy duty geometry, where
efficiency was always higher with the decreasing of autoignition strength, hence
spark timing delaying. When both ”Eng” and ”Turb” geometries are adopted,
efficiency reaches its maximum value when a spark timing of -20◦ is imposed.

143
 Figure 5.51: ”Turb” efficiency curve in spark timing sweep

From figure 5.51, several very important considerations can be derived: first
of all, efficiency curve is not monotonous with autoignition strength as happened
for CNG-heavy duty engine; second, efficiency target of 41.8%, imposed by ”Xi”
solver simulation of very high compression ratio spark ignition engine, is defi-
nitely overtaken. These geometries edited can work with very high compression
ratio engines in a spark assisted context without penalizing their performances,
but taking advantage of a faster combustion completing phase. All the flame
propagation phases are hugely penalized by lower turbulence presence in the
engine, as it leads to a slower flame and to a more anticipated spark timing
optimum value. As declared at the beginning of this section, not in the whole
engine map it is possible to perform a spark assisted combustion because of the
high loads transmitted to the engine. Working point analysed up to this sub-
section is 52% load working point, which can be a realistic threshold dividing
the two combustion modes. By looking at curves in figure 5.52, pressure rise in
the optimum condition is slightly above the already high assumed limit of 18
bar/deg. Therefore, it is difficult to develop entirely spark assisted compression
ignition engines at the state of the art, but they must be designed to be able to
work also in normal spark ignition mode.

Figure 5.52: ”Turb” pressure rise curves in spark timing sweep

In tables 5.6 and 5.7 results of the two spark timing sweep are displayed to be

144
compared. Both efficiency increases are appreciable, with the ”Turb” geometry
better performing with respect to ”Eng” one not just in efficiency terms, but
also in maximum pressure and pressure rise ones. In the whole sweep, dp/dθ
changes massively: in 2◦ of spark time difference, it rises of almost 9 bar/deg
leading to much higher loads on the engine. It is far way more controlled with
respect to the one previously obtained, but spark control must be performed
in a precise way. At contrary of what has been expected, best efficiency point
is stable for very low spark timing variation, to have then a massive decrease
if it is further anticipated; a delayed spark timing leads to not huge efficiency
differences. As pressure rise, also pressure maximum value is strongly influenced
by autoignition strength, reaching 114.5 bar at half load.

θspark η GIW Heat losses Pmax Tmax dp/dθmax

-16◦ 42.01% 2839 J -938.5 J 92.6 bar 2622 K 12.04 bar/deg
-20◦ 42.42% 2868 J -1080 J 114.5 bar 2695 K 20.4 bar/deg
-24◦ 41.68% 2821 J -1285 J 132.2 bar 2754 K 36.4 bar/deg
-28◦ 39.25% 2662 J -1669 J 144.8 bar 2798 K 57.2 bar/deg

Table 5.6: Simulation summary for ”Turb” spark timing sweep

θspark η GIW Heat losses Pmax Tmax dp/dθmax

-16◦ 41.69% 2815 J -999 J 95.4 bar 2622 K 13.76 bar/deg
-20◦ 41.89% 2831 J -1169 J 117.6 bar 2683 K 24.4 bar/deg
-22◦ 41.87% 2831 J -1229 J 126.9 bar 2719 K 33.2 bar/deg

Table 5.7: Simulation summary for ”Eng” spark timing sweep

These two tables previously illustrated confirm that a very precise spark time
control is needed, and that both pressure values must be reduced to preserve the
mechanical integrability of the engine. But a change of the piston geometry is
needed to exploit compression ignition, and to have a further analysis the three
are compared together. For CNG-heavy duty one, a spark timing of -16◦ has
been chosen since it is the one taken as reference so far; for ”Eng” and ”Turb”
ones, spark timing chosen is -20◦ as it is the best efficiency one.

Figure 5.53: Geometry comparison pressure curves

145
 Figure 5.53 illustrates the three pressure curves, showing that they are al-
most coincident and hence the comparison between the three geometries can
be considered fair. While CNG-heavy duty engine efficiency is penalized by
compression ignition because it takes place in the squish area, in the two edited
geometries there is not a proper squish region, and also turbulent kinetic energy
is influenced. As it can be seen in figure 5.54, turbulence maximum value for
CNG-heavy duty engine is almost four times ”Eng” and ”Turb” geometries one.
Piston shape hence plays a very important role in compression ignition devel-
opment and leaving space to the most external cylinder regions can be a good
criterion to have very fast combustion exploitation.

Figure 5.54: Geometry comparison turbulent kinetic energy curves

This leads also to a sort of contradiction in engine design phase, since flame
propagation combustion needs to be speeded up, while spontaneous ignition
combustion propagation needs to be slowed down. As shown in figure 5.55,
CNG-heavy duty flame velocity gain is considerable, leading to lower compres-
sion work and a better engine behaving where spontaneous ignition is unwanted.
Developing a SACI engine means hence finding the best compromise between
a fast flame propagation and a good compression ignition, but it can lead to
appreciable efficiency advantages: In section 3, for 1200 rpm half load working
point, efficiency obtained was 40.09% while, in this one, best value is 2.33%
higher with lower natural gas injected in the combustion chamber because of
the volume reduction.

146
 Figure 5.55: Geometry comparison cumulative heat release curves

Another important thing focusing on autoignition analysis is to distinguish

knock from a well exploited compression ignition. Knock is defined by theory
as an abnormal combustion in a spark ignition Otto cycle engine that can be
recognized by the presence of metallic noise coming from the engine. It origins
pressure waves which, propagating in the combustion chamber and rebounding
in correspondence of cylinder walls, reduce power cycle efficiency and hence
gross indicated work, even if the whole combustion process needs less time to
be completed. In this thesis work, knock caused performance loss is associated
to a wall heat transfer increase due to turbulence generated by the wave itself.
As a consequence, to decrease that performance loss, piston geometry can be
changed to have a smoother and organized spontaneous ignition flame front.
Heat losses, detrimental for engine performances, are hence strongly reduced as
figure 5.56 shows, keeping very similar pressure-crank angle curves.

Figure 5.56: Geometry comparison wall heat transfer curves

This wall heat transfer difference between the geometries remain consis-
tent even considering wall per wall values, with just a small discrepancy in the
liner minimum value of the ”Eng” combustion chamber. In figure 5.56, ”Eng”
and ”CNG-heavy duty” have such a similar minimum value because ”adiabatic
walls” in ”Eng” geometry are bigger with respect to other cases, leading to a
huge instantaneous loss. But this loss does not have a considerable overlying
area, meaning that in cumulative terms it is not as important as reference ge-
ometry one. Another phenomenon easier to be detected in curves of figure 5.57
is the difference in vibration intensity between the geometries. These vibration

147
propagation of almost all specific quantities considered by the solver are the
one caused by pressure waves, typical of knock. For all these reasons merged
together, autoignition occurring in ”CNG-heavy duty” geometry is considered
knock, while the one presented by ”Eng” and ”Turb” ones can be considered a
controlled spark assisted compression ignition.

Figure 5.57: Wall per wall heat transfer geometry comparison

In addition, differences in gross indicated work due to this phenomenon are

considerable, since between best performing piston configuration and worst one
GIW variation is about 100 J, corresponding to 50 Nm of torque. This result has
been obtained with no variation in amount of fuel injected, meaning that this
efficiency variation can be addressed in a decreased amount of fuel consumed
by the engine.

148
Figure 5.58: Gross indicated work geometry comparison, with corresponding
thermal efficiency values reported in percentage

In this subsection piston influence has been investigated, and very good
efficiency have been reached in half load condition. While pressure maximum
constrain has been fulfilled, pressure rise one has not, since a value of 20 bar/deg
has been reported for the best efficiency condition. To further slow compression
ignition down, in the next subsection lean air-fuel ratios are considered, in order
to compare obtained results with stoichiometric mixture ones.

5.4 Lean SACI combustion

To chose a lean air-fuel ratio instead of a stoichiometric one can lead to con-
siderable advantages, such as a reduction of cylinder maximum temperature or
having higher gas thermal inertia. Best efficiency point of Otto cycle engine is
in slightly lean condition, in order not to compromise too much flame velocity
but also to have a better exploitation of the fuel injected in the cylinder. In
previous subsection, it has been discussed how geometrical criteria for a good
spark assisted compression ignition and the ones for a fast flame propagation are
somehow opposite. Therefore, to make autoignition useful even for the ”CNG-
heavy duty” geometry can lead to several advantages, especially in working
points where SACI combustion is unwanted. Equivalence ratio, as introduced
in previous sections, represents the ratio of mass of air in stoichiometric con-
dition and mass of air present in the engine. Therefore, the lower the φ, the
higher the air excess imposed, while rich mixtures are not considered because
even theoretically efficiency predicted decreases.
Equivalence ratio imposed in this subsection is 0.8, so the engine is evidently
lean, but no flame extinguish risk is present. As working points initial condi-
tions are well defined, there are two ways to obtained such a lean φ: to keep
constant initial cylinder pressure hence reducing quantity of natural gas used,
or to keep constant amount of fuel injected and increasing cylinder pressure at
the intake valve closing.

149
 Figure 5.59: Pressure curves when engine is running in lean conditions

In the legend of figure 5.59, ”Φ air ” indicates that inlet cylinder pressure
has been kept fixed, while ”Φ fuel” denotes that amount of fuel has been kept
constant. In both lean cases, it is evident even from pressure curves how com-
bustion is far way slower than before. Piston geometry considered in the figure
above is the ”CNG-heavy duty” one, and the same spark timing of -16◦ has
been imposed for all the three cases. Therefore, a spark timing sweep is needed
since lean pressure curves maximum value is too low, and one of the two must
be chosen. ”Φ fuel” is the selected one, because engine efficiency variation with
the same amount of fuel injected is the objective of this investigation. As fuel
quantity is kept fixed, combustion chamber needs to be fed with more air, and
derived intake valve closing pressure rises from 1.32 bar to 1.46 bar.

Figure 5.60: Pressure curves spark timing sweep in lean conditions

In figure 5.60 a very wide spark timing range needs to be considered: flame
propagation is slower in all its phases and especially ignition delay is affected.
By anticipating spark timing up to -24◦ , a similar pressure curve with the stoi-
chiometric mixture one can be obtained, hence also different computed results
can be compared.

150
 Figure 5.61: Temperature curves spark timing sweep in lean conditions

By both figures 5.60 and 5.61, it is possible to note that compression ignition
is occurring in all the simulations performed. Since geometry used presents an
important squish area, flame velocity is not that much compromised, and just a
8◦ of spark timing further anticipation is needed to obtain comparable pressure
curves. Temperature ones of figure 5.61 present a lower maximum value, as
the ones obtained in all the previous simulations were over 2600 K. This is a
result of the lower adiabatic flame temperature due to air excess, leading to
an important contribution in heat losses reduction. Slopes of the temperature
curves once combustion is completed is not as low as before, both due to lower
temperatures in the combustion chamber but also to the higher thermal inertia
since there is more air mass inside the cylinder. From a pollutants point of
view, an oxygen excess can enhance NOx production in the combustion process,
even if weighted averaged cylinder temperatures reached are lower. In any
case, a subsection regarding combustion pollutants formation has been reported,
considering CO and unburnt fuel hydrocarbons.

Figure 5.62: Cumulative heat release curves spark timing sweep in lean condi-
tions

In figure 5.62 cumulative heat transfer curves are illustrated. Even from this
chart, it is possible to see how combustion is slower especially in first phases
and it takes more time to develop, with autoignition occurring in every curve

151
(even if slightly visible in the most ignition delayed one). Also, autoignition
appears to be slower as some slopes even in cumulative heat release curves can
be seen. Therefore, the pressure wave originated inside the combustion chamber
is weaker and, consequently, heat losses during the power cycle are far way lower.

Figure 5.63: Cumulative wall heat transfer curves spark timing sweep in lean
conditions

In the reference -16◦ spark timing stoichiometric mixture case, wall heat
transfer assumed a cumulative value of -1470 J while, in figure 5.63 displayed
above, all the illustrated curves do not cross the horizontal line corresponding to
-1200 J. Containing heat losses can lead to several advantages in the expansion
stroke, as instantaneous gross indicated work shown in figure 5.64 shows.

Figure 5.64: Instantaneous gross indicated work curves spark timing sweep in
lean conditions

When engine was running stoichiometric, there was a huge difference between
high and low autoignition intensity curves, especially once compression ignition
occurred and combustion is considered completed. In figure 5.64 instead, after
40◦ all the four curves are almost coincident, not wasting the whole heat released
by compression ignition. Cumulative gross indicated work hence benefits of
spontaneous ignition even if it occurs in squish area, and the engine is not
knocking.

152
Figure 5.65: Gross indicated work curves spark timing sweep in lean conditions

Even if geometry is presenting the squish area, fresh mixture autoignition is

not something to be avoided anymore, as far as gross indicated work is maxi-
mized in correspondence of -24◦ of spark timing. This behavior is reflected also
once pressure values are displayed in CFD domain, as in figure 5.66: pressure
gradient is very smooth, and difference between maximum and minimum local
values is just the 5% of the maximum one. When heavy stoichiometric mixture
knock occurred (as in figure 5.26), this difference could even overcome the 30%
in extreme wall temperature condition. Therefore, an important result has been
obtained, since making the engine work in lean condition could be an oppor-
tunity especially for low load working point, where the reduced thermal inertia
compromises the cumulative gross indicated work.

Figure 5.66: Pressure gradient in the CFD domain after lean compression igni-
tion occurred

153
Figure 5.67: Pressure rise comparison between stoichiometric and lean mixtures
conditions

In the whole lean spark timing sweep, engine reaches the highest efficiency
when ignition starts at 24◦ before top dead centre, as illustrated in figure 5.65.
If a comparison between stoichiometric and optimized lean charges is performed,
it is evident how autoignition is smoother when cylinder works with some air
excess, leading to an opportunity to reduce pressure rise without compromis-
ing useful work. As in figure 5.67, one can see how maximum pressure time
derivative has been considerably reduced, decreasing also mechanical stresses
induced to the engine. In all the previous simulations (even when piston geom-
etry has been changed), all the pressure rise maximum values were higher than
the assumed limit. In a spark assisted compression ignition engine, a lot of not
predictable variables can condition fuel-air chemical reaction velocity, such as
turbulence variations due to cycle variability or even a change in natural gas
composition. Therefore, adopting a lean air-fuel ratio can be a wise choice to
have a slowed down homogeneous charge compression ignition, easy applicable
to all the possible combustion chamber geometries. Rate of heat release curves
(displayed in figure 5.68) provides another confirm of this behavior.

Figure 5.68: Rate of heat release comparison between stoichiometric and lean
mixtures conditions

154
 Since fuel quantity has been kept fixed, cumulative heat release values are
the same for both the cases. The two shown in figure above have similar amount
of autoignited mixture fraction, but combustion behavior is very different: in
the first phases, the flame velocity reduction due to the air excess is evident,
since lean flame propagation rate of heat released maximum value is definitely
lower than stoichiometric one, even if its ignition is anticipated. Comparing
the two autoignition regions, it is possible to recognize that in lean conditions
compression ignition is slightly anticipated, even if cylinder temperature values
are lower if compared to the stoichiometric charge case. But second and more
important thing, for almost the same underlying area, compression ignition in
lean engine needs more time to be completed, leading to lower rate of heat re-
leased and hence being a less violent combustion mode. Duration differences
are important, since lean engine autoignition last almost two times the stoichio-
metric one, and from this behavior the weaker the pressure wave displayed in
figure 5.66, the lower the heat losses, represented in figure 5.63.
Therefore, it has been proved that lean charge can take several benefits to en-
gine efficiency, even leading to a good exploitation of spontaneous ignition at
high compression ratio. If efficiency curve of figure 5.69 is observed, one can
note that the 42.42% efficiency value of ”Turb” piston shape has been not just
reached but also overtaken. The low difference between -19◦ and -24◦ spark
timing efficiency values presumes a good curve stability, which is essential to
have a flexible load control with spark timing.

Figure 5.69: Engine efficiency of lean CNG-heavy duty piston shape in spark
timing sweep

Since lean air-fuel mixture can improve significantly SACI combustion ef-
ficiency and control, also the other piston shapes have been tested. In these
cases, both air excess and combustion chamber shape slow the flame down,
hence a wide spark timing range must be considered. In this brief spark sweep
range, ignition has been anticipated up to 30◦ before top dead centre. Load is
constant at 52%, and no other change has been imposed. For the same reason
of subsection 5.3, just ”Turb” geometry related curves are displayed, because
behavior of the two edited combustion chamber is very similar. In any case, a
spark advance sweep has been performed even for ”Eng” one, and it is reported
at the end of this analysis.

155
Figure 5.70: Pressure curves of lean ”Turb” piston shape in spark timing sweep

In pressure curves of figure 5.70, just the most advanced one is comparable
with the previous cases. Flame propagation is very slow especially in the first
phases, and even when compression ignition occurs a pressure curve slope can be
recognized at sight. Dealing with efficiencies, very high values can be reached:
as figure 5.71 reports, maximum value is about 43.34%, a very high one for this
kind of thermal engines.

Figure 5.71: Efficiency curves of lean ”Turb” piston shape in spark timing sweep

This is the result of adopting high compression ratio and air excess inside the
combustion chamber. As previously specified, cumulative heat released does not
vary within this analysis, and the whole efficiency gain is due to an increase of
the gross indicated work. The strength of these kinds of configurations are defi-
nitely the heat losses reduction: from the -1400 J of the stoichiometric mixture
CNG-heavy duty geometry to the over -1000 J obtained in this spark advance
sweep, even with a very anticipated mixture ignition. In figure 5.72 heat losses
of the whole sweep are reported: more than the low values reached, what is
even more important is the very low distance between the -28◦ and -30◦ igni-
tion timing curves, meaning that pressure wave generated by this compression
ignition combustion is almost negligible.

156
Figure 5.72: Heat Losses curves of lean ”Turb” piston shape in spark timing
sweep

Figure 5.73: Turbulent kinetic energy of lean ”Turb” piston shape in CFD
domain

As predictable, even turbulent kinetic energy in the whole CFD domain

assumes very low values, as an indicator of how compression ignition has been
slowed down. Maximum fluid kinetic energy region is always in correspondence
of spontaneous combustion reactions, but values reported in figure 5.73 are not
comparable with all the ones previously obtained in this thesis work.
In the tables 5.8 and 5.9, both spark timing sweeps main results are illustrated.

θspark η GIW Heat losses Pmax Tmax dp/dθmax

-19◦ 41.05% 2735 J -748 J 68.3 bar 2298 K 1.82 bar/deg
-28◦ 43.24% 2882 J -915 J 106.7 bar 2474 K 8 bar/deg
-30◦ 43.34% 2889 J -953 J 114.2 bar 2500 K 10.4 bar/deg

Table 5.8: ”Turb” piston shape lean spark timing sweep

157
 θspark η GIW Heat losses Pmax Tmax dp/dθmax
-19◦ 41.09% 2736 J -746 J 71.4 bar 2338 K 3.44 bar/deg
-28◦ 42.90% 2858 J -1005 J 112.7 bar 2481 K 11.2 bar/deg

Table 5.9: ”Eng” piston shape lean spark timing sweep

Efficiency values in both cases are very high: 42.9% for ”Eng” geometry and
43.34% obtained by ”Turb” one are an appreciable increasing with respect to
the starting target of 41.8% obtained by the solver ”betaFlameletXiEngineDyM-
Foam”, hence not considering spontaneous ignition. In addition, these best ef-
ficiency values are obtained reducing stresses induced to the engine at the same
time. In figure 5.74 reported below, a comparison of maximum rate of heat
released between the most important simulations have been reported. Since
in all the illustrated cases compression ignition is occurring, heat release rate
maximum value is going to be the one caused by spontaneous ignition as this
combustion mode is faster than flame propagation one. This can be consid-
ered a measure of autoignition strength: as it is possible to notice, between
”CNG-heavy duty” geometry and ”Turb” lean one, heat release rate has been
more than halved, with a consequent reduction of autoignition strength. This
decrease does not involve amount of mixture autoignited (since in all cases it is
pretty similar) but the velocity of compression ignition, leading to a definitely
better exploitation of it. This behavior influences almost all specific quantities
considered in the solver, including pressure gradient, turbulent kinetic energy
and, because of these two, heat transfer.

Figure 5.74: Main piston shape and equivalence ratio comparison in heat release
maximum values

And figure 5.75 is a direct consequence of this. Pressure rise intensity and
pressure waves generated by compression ignition are strongly connected, as it
has been discussed. Previously, ”CNG-heavy duty” lean engine was good be-
having even with spark assisted combustion mode, but pressure rise obtained
was higher the assumed engine limit. Adopting both ”Eng” and ”Turb” geome-
tries, an increase of gross indicated work and a reduction of stresses induced to
the engine are obtained at the same time: pressure rise maximum value in best
efficiency condition is 10.4 bar/deg, but if it is needed to be further reduced, by
delaying spark timing it is possible to get 8 bar/deg losing just a tenth of the

158
whole thermal efficiency.

Figure 5.75: Main piston shape and equivalence ratio comparison in pressure
rise maximum values

52% load working point has been chosen as the design reference for spark
assisted compression ignition engines, but also all the other points of the whole
engine map needs to be considered; in particular, 1000 and 1200 rpm full load
once are tricky, since for the same or lower engine speed both kinds of com-
bustion mode should be possible to be performed. Piston shapes analysed in
this section so far have a compression ratio of 15, too high for an entire spark
ignition engine: at full load, 1200 rpm spark timing must be delayed in order
not to have any knock. If for medium-low loads working points a huge efficiency
gain is predicted, for high load ones (especially at low regimes) ignition timing
needs to be such delayed that thermal efficiency is not going to increase that
much with respect to optimized values shown in section 3.
Fixing 1200 rpm full load working point as a reference for entirely flame prop-
agation combustion, adopting a compression ratio of 15 means having knock
even at -8◦ of spark timing when ”CNG-heavy duty” geometry is used, and
that would lead to very low thermal efficiency values; when ”Eng” or ”Turb”
geometry are used, spark timing must be delayed up to -10◦ , but flame propaga-
tion is slower and thermal efficiency is getting even worse. To develop an entire
engine map, ”CNG-heavy duty” geometry with a slightly reduced compression
ratio of 14.5 has been chosen: the presence of squish area should increase flame
velocity when combustion is performed entirely through spark ignition, while
autoignition strength is controlled via air excess resulting in a more ideal Otto
cycle.
In this section, an equivalence ratio of 0.8 is considered, since it is enough lean
to take several benefits but not that much to have flame extinguish. When an
equivalence ratio of 0.7 is considered, flame can propagate through combustion
chamber, but as illustrated in figure 5.76 the regress variable b has a minimum
value of 0.038, hence the fuel is not predicted to release all of its energy in flame
propagation combustion process. Adopting a different spark light able to release
higher quantities of energy can be a solution of this problem for this equivalence
ratio, but it is not an object of study of this thesis work as no huge thermal
efficiency gains are predicted.

159
Figure 5.76: Flame propagation when an equivalence ratio of 0.7 is adopted,
showing that the minimum b value can not be considered 0

When engine is running even leaner, flame extinction occurs. As illustrated

in figure 5.77, near the spark plug b assumes low values, meaning a small volume
of fuel-air mixture is igniting. Heat released by these small volumes is not
enough to generate a real flame front and then a good spark ignition combustion,
and after some time the flame extinguish and both kind of combustions are not
possible anymore. Three CFD domains of the figure below represent (from left
to right) the regress variable at -25◦ , -20◦ and -15◦ , and hence combustion can
not take place.

Figure 5.77: Flame extinction when an equivalence ratio of 0.6 is adopted

Therefore, to conclude this lean engine analysis, adopting an equivalence

ratio of 0.8 can be an interesting opportunity to reduce autoignition velocity
hence to take advantage of spontaneous ignition combustion. Leaner fuel-air
mixtures can not ensure a reliable control on flame propagation development,
since not the whole heat can be released by the mixture itself. This characteristic
can also depend on the kind of natural gas used, since in SNAM North Europe
derived composition one almost 7% of inerts are present in the fuel itself, further
reducing flame propagation velocity.

5.5 SACI engine map

As explained in previous pages, for the development of a spark assisted compres-
sion ignition combustion engine map has been chosen the ”CNG-heavy duty”

160
piston geometry with a compression ratio of 14.5. In high load or high regime
working points, an entire flame propagation combustion is wanted, while in low
load working points a spark assisted compression ignition combustion has been
performed. Each working point illustrated in the figure 5.5 has been optimized,
considering the same criteria of section 3.

Figure 5.78: Efficiency SACI engine map

In figure 5.78, in green thermal efficiencies of SACI combustion are reported,

while corresponding to red written number no spontaneous ignition occurs.
Maximum efficiency points has been reached at 1600 rpm in half load condi-
tion, with the very high value of 43.24%. The higher the regime, the lower
the heat transfer and it can be considered important when compression igni-
tion occurs, since the increased turbulence in combustion chamber is enhancing
engine heat losses. If this efficiency map is compared to the one derived in
section 3, one can note that in figure 5.78 maximum efficiency points are low
load ones, while in the other one the higher the load, the higher the thermal
efficiency. This can be an important consideration when dealing with real life
operation, since engine is not predicted to run in all working points with the
same probability, and a well chosen design reference can further decrease fuel
consumption. In fact, if thermal efficiencies are plotted within a spark timing
sweep, it is possible to note how values are decreased once the engine passes
from a SACI combustion to a flame propagation one.

Figure 5.79: Efficiency curve for a SACI engine in load sweep

161
 Load sweep of figure 5.79 is performed at 1200 rpm, as the presence of six
different working points suggests. Spark timing and its controlling are essential
to perform a good combustion mode: at high load, engine risks to knock in the
whole map, leading to a problem especially when engine rotational speeds are
low. For this reason, squish area is important for flame propagation develop-
ment, since in these point a very fast flame is needed. Adopting a ”Turb” or
an ”Eng” geometry is a more risky choice, since they would present higher effi-
ciency values in low load SACI working points, but a definitely lower in entire
flame propagation ones.

Figure 5.80: Spark timing SACI engine map

In figure 5.80, spark advance chosen in the whole engine map are reported.
Both in 1200 and in 1600 rpm sweep, spark advance is imposed by knock pres-
ence, which does not vary that much between considered working points. While,
in low load ignition must be very advance, and even a -34◦ spark timing is set at
1600 low load engines. Mixture fraction autoignited is very important within the
load sweep range, since it aims to be controlled through ignition timing. For
a brief remind, green SACI points works with lean air-fuel ratio, while flame
propagation ones with stoichiometric φ. Transition between the two different
conditions can be an object of study for future works. As figure 5.81 high-
lights, as soon as transition in entire flame propagation is completed, optimum
spark timing is constant with the load. Between minimum and full loads, huge
optimum ignition timing variations are present.

Figure 5.81: Spark timing in load sweep at 1200 rpm

162
 In points displayed in greed, a spark assisted compression ignition combus-
tion of a lean mixture is performed, adopting an equivalence ratio of 0.8. Due to
the high compression ratio, the entirely flame propagation combustion is com-
promised, in order to avoid knock presence. As anticipated in the first section,
load (defined by fuel mass injected at the intake valve closing) has been kept
fixed to develop this combustion mode; in table 5.10, gross indicated works are
reported, always keeping 1200 rpm as fixed engine speed.

entirely Flame propagation SACI mode ∆η

CR 11.7 14.5
17% load 1202 J 1245 J + 2.71%
35% load 2054 J 2116 J + 2.86%
52% load 2894 J 2953 J + 2.68%
68% load 3800 J 3567 J + 0.11%
85% load 4500 J 4216 J + 0.06%
100% load 5522 J 5164 J + 0.02%

Table 5.10: Differences in gross indicated work between starting flame propa-
gation combustion and obtained spark assisted one, with differences of thermal
efficiencies reported rightwards

∆η = ηSACI − ηentF lameP ropagation (5.2)

Entirely flame propagation combustion in table 5.10 has been obtained using
methane as a fuel (hence higher heat released with respect to SNAM North
Europe one due to no inerts presence) and since a higher compression ratio is
adopted, for the same initial pressure fuel mass injected is slightly higher, as
illustrated in figure 5.3 at the beginning of this chapter.
In conclusion of this brief subsection, focus is pointed on pressure rise. It has
been discussed how it is important to have a limited value of it in order to
not induce too high mechanical stresses to the engine; as figure 5.82 shows,
there is just 1200 rpm half load condition when pressure time derivative reaches
dangerous values. In all others working points, because of a load reduction or
a faster engine speeds, values assumed by pressure rise are considerably lower,
not too much severe for a heavy duty engine.

Figure 5.82: Pressure rise SACI engine map

163
 The importance of having a good control of spontaneous ignition needs once
more to be highlighted, since a different natural gas composition can move
optimized spark timing or vary maximum pressure rise. Since at low load com-
pression ignition is something wanted, a more reacting fuel should lead to an
efficiency increase in low load points, but managing high load ones could be
an issue. In any case, an almost versatile natural gas composition has been
taken as a reference. In Italy, a lot of sources reports a natural gas reached in
methane and poorer in heavy hydrocarbons, as illustrated in section 1. This
kind of change in practice should not lead to huge thermal efficiency variations,
but maximum pressure rise values are predicted to decrease because of the lower
chemical reactivity of the fuel itself.

5.6 Pollutants analysis

Running the engine lean could be a problem for the three-phase catalyst since,
as explained in the introduction section, conversion efficiency is optimized in
stoichiometric air-fuel mixture condition, where it can overtake even 95% and
consequently decrease pollutant emissions of the whole engine. Since in this
CFD solver after-treatment systems are not considered, focus is pointed first
on the carbon-monoxide formation during the combustion process and then on
the amount of unburnt hydrocarbons at the end of the power cycle, in such a
way to have some values to be compared with possible future works. However,
NOx transport equation has not been implemented in the solver, hence it is
possible just to give general considerations but not to quantify nitrogen-oxides
production in the combustion process.
Working with some air excess in a homogeneous mixture flame propagation
engine implies that adiabatic flame temperatures are going to be lower and
some oxygen, able to react with reaction products, is remaining even after the
combustion is completed. As previously described, air excess is present just
in SACI combustion mode working points, hence as point of interest the half
load one at 1200 rpm has been chosen. Piston shape chosen is the CNG-heavy
duty one with a compression ratio of 14.5, and its pollutants emission have been
compared with the optimized one in section 3 (the entirely flame propagation
engine adopting 11.7 as compression ratio).

Figure 5.83: CO production in 1200-half load working point comparison. φ =

0.8 SACI reported in green, while stoichiometric flame propagation in red

164
 First pollutants to be analysed is the carbon monoxide. As evident by the
figure 5.83, air excess enhances carbon oxidation both during and after the
combustion process. Reducing the equivalence ratio from 1 to 0.8, less than
the 25% of the CO is globally formed at the exhaust valve opening, and this
quantity is even predicted to decrease once exhaust gases cross the catalyst.
Carbon oxidation proceeds with an appreciable velocity until to the end of the
expansion stroke (with a slowdown at around 90◦ crank angle because of gases
temperature reduction) and the pollutants is homogeneously distributed within
the whole combustion chamber.

Figure 5.84: CO2 production in 1200-half load working point comparison. φ =

0.8 SACI reported in green, while stoichiometric flame propagation in red

Passing to CO2 as shown in figure 5.84, it is considered a complete reaction

product connected to the amount of fuel used, detached by quality of combustion
process. Therefore, the only way to decrease carbon dioxide emission of an
internal combustion engine is to increase its thermal efficiency, in such a way to
use a lower fuel mass to satisfy every demanded torque.

Figure 5.85: unburnt hydrocarbons after the whole combustion process in 1200-
half load working point comparison. φ = 0.8 SACI reported in green, while
stoichiometric flame propagation in red

165
 However, not all the fuel particles can burn and then release heat: as shown
by figure 5.85, even if at higher compression ratios a lower fuel quantity is in-
jected in the engine (for the same initial pressure), unburnt natural gas quantity
increases, probably due to a wider flame extinguish region near to cylinder walls.
Always in the same figure, respective percentages representing the ratio of un-
burnt hydrocarbons with respect to mass of fuel injected are reported. In the
lean SACI engine, the 0.7% of the fuel is not taking part to the combustion
process, and to decrease this amount can be a target for future combustion
chamber designs.
For what regards nitrogen-oxides emission, however the solver can not predict
NOx formation. For sure, the oxygen excess is enhancing their production, but
adopting a leaner mixture permits to have a lower adiabatic flame temperature
hence decrease NOx production.

Figure 5.86: Cell temperatures in 1200-half load working point comparison, at

the same engine time. φ = 0.8 SACI reported leftwards, while stoichiometric
flame propagation rightwards

Figure 5.86 previously reported compares cell by cell temperatures reached in

a SACI at high compression ratio with the ones of an entirely flame propagation
combustion. Temperature weighted average values are higher when spontaneous
combustion takes place, but this increase is given by the presence of two dif-
ferent combustion modes, which speed up fuel oxidation reactions. Since cell
temperature maximum values in lean conditions are almost 200 K lower with
respect to the stoichiometric case, these effects combined should limit the over-
all NOx production during the power cycle.
As already written, these calculations were performed ignoring the presence of
any after-treatment system at the turbine outlet. In any case, three-phase cata-
lyst conversion efficiency is function of NOx, CO and hydrocarbons inlet mass,
and this can be an interesting study case for future works. Figures displayed
so far permits to compare lean SACI pollutants emission with a stoichiometric
flame propagation reference case. These considerations are consistent even if
higher regimes are analysed: as reported by figures 5.87, 5.88 and 5.89, at 1600
rpm pollutants formation behavior is consistent with what has been discussed
so far.

166
Figure 5.87: CO production in 1600-half load working point comparison. φ =
0.8 SACI reported in green, while stoichiometric flame propagation in red

Figure 5.88: CO2 production in 1600-half load working point comparison. φ =

0.8 SACI reported in green, while stoichiometric flame propagation in red

Figure 5.89: Unburnt hydrocarbons after the whole combustion process in 1600-
half load working point comparison. φ = 0.8 SACI reported in green, while
stoichiometric flame propagation in red

167
 In this kind of engines designed to exploit homogeneous charge compression
ignition, even high load working points where autoignition is unwanted must be
taken into account. As a consequence, downwards a flame propagation engine
running with a compression ratio of 11.7 (adopting -23◦ as spark timing) and the
one derived so far at 14.5 as compression ratio (which, as presented in subsection
before, has a very delayed spark timing corresponding to -10◦ ) are compared.

Figure 5.90: CO production in 1200-full load working point comparison. Both

power cycles performed through entire flame propagation

The ongoing of these charts is very similar: as shown in figure 5.90, the
two curves have a much more rounded peak and a lower decreasing afterwards,
because of a lower oxygen partial pressure and then CO2 formation results more
complex and slower. The lower fuel amount used and the lower temperatures
reached (due to more delayed spark timing) allow to have a carbon monoxide
production far way lower such that, even if both cases run in stoichiometric
conditions, at high compression ratio CO outlet mass is almost halved.

Figure 5.91: CO2 production in 1200-full load working point comparison. Both
power cycles performed through entire flame propagation

As previously written, CO2 is a complete combustion product and its mass

is directly related to the amount of natural gas used. For sure, to have a fuel
composed by heavier hydrocarbons enhances carbon dioxide production, but

168
when different commercial natural gas compositions are compared, differences
are not so evident. Therefore, the whole CO2 reduction is totally connected to
the lower amount of fuel used.

Figure 5.92: Unburnt hydrocarbons after the whole combustion process in 1200-
full load working point comparison. Both power cycles performed through entire
flame propagation

For what regards unburnt hydrocarbons mass, as illustrated in figure 5.92

stoichiometric air-fuel mixture ratio helps to have lower unburnt fuel released
after the exhaust valve opening: while previously mass of unburnt natural gas
were over the 0.7% of the whole fuel injected, in this case this percentage de-
creases up to 0.5%. Since at full load mixture results stoichiometric, three-way
catalyst is predicted to work at maximum conversion efficiency to drastically
reduce the whole engine pollutants emissions.

169
170
Conclusions

Purpose of this thesis was the prediction of knock and combustion under the
so-called spark-assisted mode in a heavy duty engine operating with natural
gas. Onset of knock limits the spark-advance and compression ratio and its
correct estimation makes possible to identify the maximum engine performance
and define suitable modifications to the combustion chamber layout to avoid it.
Engine operation under the spark-assisted mode is similar to knock, but with
a reduced pressure rise rate. Understanding the difference between knock and
spark-assisted combustion is one of the main objectives of this thesis work: both
of them can considerably speed up the combustion process, but in case of knock
pressure gradients are so extreme to dissipate a huge part of energy released in
heat losses. Therefore, if pressure gradient inside the combustion chamber is
low and controlled, engine performances can take advantage of this faster com-
bustion mode. Two different strategies to reduce pressure gradients have been
analysed: the design based one (which acts on piston shape in correspondence
of the squish area) and the mixture based one (which instead changes air-fuel
stoichiometric ratio into a leaner one). In any case, not the whole engine map
is suitable for a spark assisted compression ignition combustion: it induces very
high stresses to the mechanical structure of the engine. For this reason, the
whole map has been divided in two regions:
- from zero to half load where spark assisted compression ignition combustion
is used to improve engine performances;
- form half to full load where spontaneous ignition of any nature is undesired
to not have too high pressure values in the combustion chamber.
To further preserve the mechanical integrability of the engine, two more lim-
its have been added: a maximum pressure of 180 bar and a pressure-rise not
to overtake the value of 18 bar/deg. If piston design is investigated, region of
interest it the furthest one from spark plug, where spontaneous ignition usu-
ally starts. In this region, squish area is located in such a way to have higher
turbulence intensity for a faster flame propagation; but when autoignition takes
place, a non negligible part of the heat released is lost to the walls. To avoid this
effect which compromises the efficiency, extreme cylinder regions must have a
wider compression ignition flame front surface in order to exchange more energy
with flame propagation burnt gases and at the same time reduce the pressure
wave intensity. Adopting this strategy, efficiency can overcome the value of 42%
at half load exploiting spark assisted compression ignition combustion but tur-
bulences generated inside the combustion chamber are far way lower and flame
propagation results slower. In SACI combustion working points, this undesired

171
effect can be counterbalanced anticipating the spark timing of almost 4◦ to
have similar pressure curve maximum values; but in high-load working points
where spark assisted compression ignition combustion is not wanted, the lower
flame velocity reduces the engine efficiency, even if compression ratio adopted is
higher. Then the second strategy has been investigated: the possible presence
of air excess in the engine can be easily regulated by the throttle valve installed
before the intake port, in such a way to have a lean mixture when spontaneous
ignition is wanted and a stoichiometric mixture in entirely flame propagation
working points at high loads. Equivalence ratio chosen for lean working points
is 0.8 to avoid the possibility of the flame to be extinguished. Thermal effi-
ciency improvements obtained are significant, with values even overcoming 43%
at 1600 rpm half-load working point. However, no performance improvement is
predicted at high load: compression ratio adopted (14.5) is very high for such
a big engine (having 2.15 liters as displacement) and spark timing must be de-
layed to avoid knock. Therefore, maximum efficiency point (which in common
Otto cycle engine is the full load one) moves to the half load condition because
of the different combustion mode used and fuel consumption of the overall map
has been appreciably reduced. The main drawback of adopting leaner engine is
represented by the three-way catalyst, as its conversion efficiency is maximum
when air-fuel mixture used is stoichiometric. But if pollutants formations just
by combustion are analysed comparing the derived SACI combustion with the
entirely flame propagation one, when engine is running lean carbon monoxide
emissions are reduced up to the 75% and maximum temperature reached are
more than 150 ◦ C lower, which should theoretically discourage NOx formation.
However, no nitrogen oxides transport equation is implemented in the solver,
hence no quantification has been performed.

172
Bibliography

[1] H. Versteeg and W. Malalasekra, An Introduction to Computational Fluid

Dynamics, 2, Pearson India Education Services (2009)

[2] Giancarlo Ferrari, Motori a combustione interna, 1, Socierà editrice Escula-

pio (2016)

[3] Nicolò Bachschmid, Stefano Bruni, Andrea Collina, Bruno Pizzigoni, Fer-
ruccio Resta, Alberto Zasso, Fondamenti di meccanica teorica e applicata,
Graw Hill, 2015

[4] D.S. Malik, Introduction to C++ Programming, 1, Apogeo Education (2009)

[5] Tommaso Lucchini, Angelo Onorati, Gianluca D’Errico, Alessandro Stagni

and Alessio Frassoldati, Modeling non-premixed combustion using tabulated
kinetics and different flame structure assumptions, Dipartimento di Energia,
Politecnico di Milano (2017)

[6] H.G. Weller, The Development of a New Flame Area Combustion Model
Using Conditional Averaging, Imperial College Mechanical Engineering De-
partment (1993)

[7] H.G. Weller, S. Uslu, A.D. Gosman, R.R. Maly, R. Herweg, B. Heel, Predic-
tion of Combustion in Homogeneous-Charge Spark-Ignition Engines, Imperial
College Mechanical Engineering Department (1994)

[8] Maurizio Mastropasqua, Modellazione del processo di combustione in motori

Diesel mediante modelli di combustione basati su cinetica chimica tabulata,
Dipartimento di Energia, Politecnico di Milano (2016)

[9] Alberto Comolli, CFD Modeling of Diesel Combustion with Tabulated Ki-
netics Based on Homogeneous Reactor Assumption, Dipartimento di Energia,
Politecnico di Milano (2018)

[10] Jiri Vavra, Michal Takats, Vojtech Klir and Marcel Skarohlid, Influence of
Natural Gas Composition on Turbocharged Stoichiometric SI Engine Perfor-
mances, Czech Technical Univ. (2012)

[11] Akira Kikusato, Hiroyuki Fukasawa, Kazutoshi Nomura, Jin Kusaka and
Yasuhiro Daisho, A Study on the Characteristics of Natural Gas Combustion
at a High Compression Ratio by Using a Rapid Compression and Expansion
Machine, Waseda Univ. (2012)

173
[12] Ahmed Abdul Moiz, Zainal Abidin, Robert Mitchell, and Michael Koc-
sis, Development of a Natural Gas Engine with Diesel Engine-like Efficiency
Using Computational Fluid Dynamics, Southwest Research Institute (2019)

[13] Yalan Liu, Xuexiang Zhang, Junxia Ding, Chemical effect of NO on CH4
oxidation during combustion in O2 /NO environments, University of Chinese
Academy of Sciences (2019)

[14] Hrvoje Jasak, Numerical Solution Algorithms for Compressible Flows, Uni-
versity of Zagreb, Croatia (2006)

[15] Jinlong Liu and Cosmin Dumitrescu, CFD Simulation of Metal and Optical
Configuration of a Heavy-Duty CI Engine Converted to SI Natural Gas. Part
2: In-Cylinder Flow and Emissions, West Virginia University (2019)

[16] Fubai Li, Changpeng Liu, Heping Song, and Zhi Wang, Improving Com-
bustion and Emission Characteristics in Heavy-Duty Natural-Gas Engine by
Using Pistons Enhancing Turbulence, Tsinghua University (2019)

[17] William P. Attard, Elisa Toulson, Harry Watson and Ferenc Hamori, Ab-
normal Combustion including Mega Knock in a 60% Downsized Highly Tur-
bocharged PFI Engine, The University of Melbourne, Australia (2010)

[18] G. Brecq, A. Ramesh, M. Tazerout and O. Le Corre, An Experimental

Study of Knock in a Natural Gas Fuelled Spark Ignition Engine, Ecole Des
Mines De Nantes,(2001)

[19] Junseok Chang, Orgun Güralp, Zoran Filipi, and Dennis Assanis, New
Heat Transfer Correlation for an HCCI Engine Derived from Measurements
of Instantaneous Surface Heat Flux, University of Michigan (2004)

[20] Jeremie Dernotte, John Dec, and Chunsheng Ji, Investigation of the Sources
of Combustion Noise in HCCI Engines, Sandia National Labs (2014)

[21] Magnus Sjöberg and John E. Dec, Nicholas P. Cernansky, Potential of

Thermal Stratification and Combustion Retard for Reducing Pressure-Rise
Rates in HCCI Engines, Based on Multi-Zone Modeling and Experiments,
Sandia National Labs and Mechanical Engineering Department, Drexel Uni-
versity (2005)

[22] Patrick Pertl, Alexander Trattner, Andrea Abis , Stephan Schmidt and
Roland Kirchberger, Takaaki Sato Expansion to Higher Efficiency - Investi-
gations of the Atkinson Cycle in Small Combustion Engines, Graz University
of Technology and DENSO Automotive Deutschland GmbH (2012)

[23] Laura Manofsky, Jiri Vavra, Dennis Assanis and Aristotelis Babajimopou-
los Bridging the Gap between HCCI and SI: Spark-Assisted Compression Ig-
nition, Univ. of Michigan and Czech Technical Univ (2011)

[24] William P. Attard and Hugh Blaxill, Eric K. Anderson, Paul Litke, Knock
Limit Extension with a Gasoline Fueled Pre-Chamber Jet Igniter in a Modern
Vehicle Powertrain, MAHLE Powertrain, National Research Council and US
Air Force Research Laboratory (2012)

174
[25] Anne Prieur and Richard Tilagone, A Detailed Well to Wheel Analysis of
CNG Compared to Diesel Oil and Gasoline for the French and the European
Markets, IFP (2007)

[26] Chang, J., Kim, M. And Min, K., Detection of misfire and knock in spark
ignition engines by wavelet transform of engine block vibration signals, Mea-
surement Science and Technology, 2002.
[27] Ohtubo, H., Yamane, K., Kawasaki, K., Nakazono, T. and Shirouzu, T.,
PCCI Combustion for Multi Cylinder Natural Gas Engine (Second Report) -
Leading Auto-ignition Reduction of Cylinder-to cylinder Variations by using
Spark Ignition JSAE Proceeding Paper, 2007
[28] Roberts, C.E., Snyder, J.C., Stovell, C., Dodge, L.G. et al., The Heavy-Duty
Gasoline Engine-An Alternative to Meet Emissions Standards of Tomorrow
SAE Technical Paper, 2004
[29] Liu, J. and Dumitrescu, C.E., Combustion Visualization in a Single-
Cylinder Heavy-Duty CI Engine Converted to Natural Gas SI Operation, State
College USA, 2018
[30] Yu, X., Liu, Z., Wang, Z., and Dou, H., Optimize Combustion of Com-
pressed Natural Gas Engine by Improving In-Cylinder Flows International
Journal of Automotive Technology, 2013
[31] Wang, Z., Wang, J., Shuai, S., Tian, G., An, X., and Ma, Q., Study of the
effect of spark ignition on gasoline HCCI combustion, Journal of Automobile
Engineering, 2006.

175