PROGRAMA DE DOCTORADO EN INGENIERÍAS

DOCTORAL THESIS:

CONTRIBUTION TO NON-INTRUSIVE COMBUSTION

PRESSURE DIAGNOSTICS BY MEANS OF IONIZATION

CURRENT CHARACTERIZATION, APPLIED TO SPARK

IGNITION COMBUSTION ENGINES

AUTHOR:

MAURICIO MONROY JARAMILLO

PEREIRA-2025

APORTACIÓN AL DIAGNÓSTICO NO INTRUSIVO DE LA

PRESIÓN DE COMBUSTIÓN A PARTIR DE LA

CARACTERIZACIÓN DE LA CORRIENTE DE IONIZACIÓN

APLICADO A MOTORES DE COMBUSTIÓN INTERNA DE

ENCENDIDO POR CHISPA

MAURICIO MONROY JARAMILLO

Proyecto de grado presentado como requisito para optar por el título de Doctor en Ingeniería

Director: Ph.D Carlos Alberto Romero Piedrahita

UNIVERSIDAD TECNOLÓGICA DE PEREIRA

PROGRAMA DE DOCTORADO EN INGENIERÍAS

PEREIRA

2025

Año 2025

NOTA DE ACEPTACIÓN

Sobresaliente

PhD. Vicente Bermudez. CMT Universidad Politécnica de Valencia PhD. Oscar de la Garza Universidad Autónoma de Nuevo León PhD. Mauricio Holguín Universidad Tecnológica de Pereira Jurado

Jurado

Pereira 11 de Noviembre de 2025

DEDICATION

This work is dedicated to my parents, to my brother and my sister.

Mauricio Monroy.

ACKNOWLEDGEMENTS

This has been a long journey, that began with a conversation, which I should desestimate to save me a lot of troubles...anyway, sails were deployed. At midpoint, there was a time when the vessel was burned far away from homeland…and I didn’t listen to the mutiny or my burns. From then, moments of rejouissance passed, but also valleys of death of work progression and then again mountains in eruption. Certainly, I’ve found true supporters, but also peaky stones below my feet. All of them, somehow, contributed to the development of this work.

Huge Thanks to my thesis director, PhD. Carlos Alberto Romero, for his openness and endless share of knowledge on combustion engines and their nuances. Additionally, for encouraging such a bitter gypsy to write and report.

Thanks to PhD. Gabriel Calle and PhD. Juan Carlos Burbano, for recommending me on the beginning of this journey.

Thanks to Ph.D. Edison Henao and PhD. Héctor Quintero, for sharing their views on crucial moments when blockages arrived.

Thanks to PhD. Ciro Iglesias Coronel and PhD. Alexander Diaz, for indicating me the nonwritten age limit -hell, I know it now!- to pursue this.

Thanks to the “engine witch-doctor”, soon-to-be PhD., Juan David Ramirez for his support on the implementation of all the engine setups, publications, his nobility…and his humour.

Thanks to soon-to-be PhDs. Wilson Pérez and Juan Camilo Mejía for their magnificent support on the programming of that NI-demon acquisition system.

Thanks to dean of faculty, PhD. Valentina Kallewaard, Consejo de Facultad and Dirección de programa, as well as Vicerrectoría Académica, in particular, Beatriz Tangarife. Additionally, Marisol Agudelo and Paula Andrea Pulido from the Doctorate Program, for all their prompt and invaluable deeds on this process.

Thanks to PhD. Ivan Dario Bedoya and PhD. Yamid Alberto Carranza for their votes of confidence on this work.

Thanks to my parents, for their moral support on the darkest days.

Mauricio Monroy.

Resumen Este trabajo aborda la implementación de un sistema no intrusivo de estimación de características de la presión en cámara de combustión, basado en un esquema de detección de corriente de ionización tipo Low-side, aplicado a un motor monocilíndrico instrumentado de encendido por chispa. La sección 2 del trabajo, relativa a la revisión de literatura, comprende tópicos de diagnóstico de motores de combustión, sistemas de encendido y su modelado eléctrico, conceptos básicos de formación del núcleo de llama, química de la combustión, así como los principales modelos de generación de iones, esquemas eléctricos de detección de corriente de ionizacion, contornos principales de señales de corriente de iones y sus capacidades de diagnóstico asociadas. La sección 3 muestra el instrumental electrónico propietario implementado, caracterizado y validado, para medir corriente de ionización, incluyendo una fuente promotora de ionización galvánicamente aislada y un microamperímetro de derivación de resistencia seleccionable, en conjunción con un limitador de señal. Para el sistema de ignición se implementaron sondas de alta tensión de tipo resistivo e inductivo. La tensión y corriente de devanado primario son medidas con amplificadores analógicos aislados especialmente diseñados. Para cada variable de ignición se realizan caracterizaciones estáticas y de respuesta en frecuencia. Se elaboraron sensores de carrera de apertura de válvulas de admisión y escape, basados en transformadores de núcleo móvil con coeficiente de acoplamiento variable. Variables adicionales instrumentadas incluyen par motor, presión en cámara de combustión, posición angular de cigüeñal, relación aire-combustible y temperaturas de motor. La sección 4 describe la definición de las condiciones operacionales y campañas de pruebas del motor, así como hallazgos preliminares en arrastre y combustión, con y sin precámara. Las señales de corriente de ionización, tensiones y corrientes de bobina de ignición y otras variables se validan en las condiciones de arrastre y combustión.

La sección 5 profundiza en los hallazgos de la sección 4, enfocándose en la estimación de variación cíclica de la presión en cilindro, correlacionando con los accidentes de la señal de ionización, bajo diversas condiciones de par de carga, frecuencia rotacional y ajuste de parámetros. Se propone un criterio de selección del resistor de derivación del microamperímetro Rion, basado en la duración de la oscilación residual y la amplitud de la señal de ionización. También se presenta un novedoso análisis de frecuencia natural amortiguada, previo al pico de quimo-ionización, para estimar la capacitancia de huelgo de bujía, para correlacionarla con la magnitud y posición angular del pico de presión, y se compara con el área bajo la curva de corriente de ionización, encontrando buen ajuste en ambos casos en condiciones de carga parcial. Adicionalmente, las combinaciones frecuencia amortiguada-área bajo señal iónica y frecuencia amortiguada-centroide de área de curva iónica mejoran la bondad de ajuste para magnitud y posición angular de pico de presión respectivamente. Resultados adicionales del análisis de señal de corriente de ionización incluyen detección de preignición, identificación de chispa perdida, ausencia de encendido y eventos de combustión.

Abstract

This work tackles the implementation of a non-intrusive in-cylinder pressure features estimation system, based on a low-side scheme of ionization current signal detection, applied to an instrumented single-cylinder spark-ignition engine.

Section 2 of literature review comprises topics of engine diagnostics, ignition systems and their electrical modeling, basics of flame kernel development, combustion chemistry, and main ion generation models, electrical schemes of ionization current detection, main contours of ionization current signals, and their associated diagnostics capabilities. Section 3 addresses the proprietary electronic instrumentals implemented, characterized and validated, to measure ionization current, including a galvanically isolated promoter power source and a selectable shunt microammeter in conjunction with a signal limiter. For the ignition system, resistive and inductive secondary voltage probes are implemented, and the primary side voltage and current are measured with especially designed analog isolated amplifiers. For each ignition variable, static and frequency response characterizations were performed. Intake and exhaust valve opening sensors were built, based on movable-nucleus variable coupling coefficient transformers. Additional variables instrumented include torque, in-cylinder pressure, crank angle, AFR, and engine temperatures.

Section 4 comprises the definition of engine operational conditions and test campaigns, and the preliminary findings during motoring and combustion with and without prechamber. The ion current signals, ignition voltages and currents, valve opening, and other engine variables were validated in motoring and combustion conditions.

Section 5 advances further in the findings of section 4, focusing on cyclic variation estimation of in-cylinder pressure, correlating with ion signal features, under diverse conditions of load torque, rotational frequency and parameter settings. A criterion of selection of the microammeter shunt resistor Rion is proposed, based on the duration of residual oscillation and ion signal amplitude. It is also presented a novel ignition damped frequency analysis, prior to chemi-ionization peak, that estimates the spark gap capacitance, to correlate it with the magnitude and crank angle of peak pressure, and it is compared with the ion current signal area, finding good fittings for both, in partial load conditions. In addition, the combinations damped frequency-ion signal area, and damped frequency-centroid of ion area improved the fitting goodnesses for magnitude and angle of peak pressure respectively. Additional outcomes from ion signal analysis include detection of preignition, identification of wasted spark, misfire, or combustion events.

I

Contents Abstract ................................................................................................................................. 7 Contents ................................................................................................................................ 1 List of Figures ................................................................................................................... 4 List of Tables ........................................................................................................................ 9 Glossary ............................................................................................................................... 10 List of symbols and acronyms ........................................................................................... 11

1

Introduction ................................................................................................................. 13

1.1

Background and motivation .............................................................................. 13

1.2

Objectives and methodology ............................................................................. 18

1.3

Thesis outline ...................................................................................................... 20

2

Literature review ........................................................................................................ 21

2.1

Introduction ........................................................................................................ 21

2.2

Chapter overview ............................................................................................... 24

2.3

On methods of fault detection and diagnosis ................................................... 24

2.4

Engine ignition systems using spark ................................................................. 27

2.5

Ion current detection.......................................................................................... 48

2.5.1

Main schematic circuits for ion current detection and microammeters ... 49

2.5.2

The independent probe circuit. ..................................................................... 50

2.5.3

The high-side circuit ...................................................................................... 56

2.5.4

The low-side circuit ........................................................................................ 62

2.5.5

Comments about the polarity of the power supply voltage U .................... 69

2.5.6

Comments about using AC for power supply voltage U ............................ 70

2.6

Correlations between combustion chamber pressure and ion current features ............................................................................................................................ 71

2.7

Conclusions from literature review .................................................................. 86

3

Engine setup and instrumentation for ignition system and ion current measurement ....................................................................................................................... 88

3.1

Ignition system measurement setup ................................................................. 92

3.1.1

The ignition coil .............................................................................................. 94

3.1.2

The resistive high-tension probe ................................................................... 97

3.1.3

The inductive voltage probe ........................................................................ 105

3.1.4

The power source U ..................................................................................... 107

3.1.5

The microammeter and limiter for ion current measurement ................ 111

II

3.1.6

The isolation amplifiers ............................................................................... 116

3.2

Engine Crank and distribution ....................................................................... 121

3.2.1

Crank angle sensing ..................................................................................... 121

3.2.2

Torque sensing .............................................................................................. 123

3.2.3

Valve displacement sensing ......................................................................... 129

3.3

Acquisition system ............................................................................................ 135

3.4

Conclusions from the implementations .......................................................... 138

4

Test procedure and preliminary findings ............................................................... 141

4.1

Preliminary considerations for the tests ........................................................ 142

4.1.1

Limiting the acquisition campaigns ............................................................ 147 Table 15. (continued) ..................................................................................................... 151

4.2

Description of the software .............................................................................. 153

4.2.1

Conditioning and referencing of the pressure signal data ....................... 156

4.2.2

Preprocessing of the ion current signal. ..................................................... 158

4.3

Preliminary findings on motoring tests .......................................................... 159

4.3.1

Motoring tests without spark ...................................................................... 161

4.3.2

Motoring tests with ignition system on....................................................... 163

4.4

Preliminary findings on combustion tests ...................................................... 173

4.4.1

Findings with prechamber .......................................................................... 173

4.4.2

Findings without prechamber ..................................................................... 179

4.5

Conclusions ....................................................................................................... 184

5

Analysis and correlations between ion current and pressure ............................... 186

5.1

The ion current signal before chemi-ionization and its relationship with peak of combustion chamber pressure. ............................................................................... 187

5.2

The area of the ionization current curve during combustion and its relationship with pressure. .......................................................................................... 198

5.3

Correlations between free-run damped frequency, area of ion current and pressure ......................................................................................................................... 201

5.3.1

Cyclic variation: pressure peak magnitude ............................................... 201

5.3.2

Cyclic variation: determination of pressure peak position ...................... 204

5.3.3

Effect of engine variables on the ionization current based estimation of pressure peak and its crank angular position ........................................................ 209

5.4

Ignition delay estimation ................................................................................. 223

5.5

Mass fraction burned and heat release rate estimations .............................. 224

5.6

Conclusions of the chapter .............................................................................. 227

6

Final conclusions and future work .......................................................................... 228

6.1

Introduction ...................................................................................................... 228

III

6.2

Main conclusions .............................................................................................. 228

6.3

Future work ...................................................................................................... 233

6.4

Contributions .................................................................................................... 235 A Appendix 1 .................................................................................................................. 237 B Appendix 2 .................................................................................................................. 238 C Appendix 3 ................................................................................................................. 239

7

Bibliography ............................................................................................................... 241

IV

List of Figures Figure 1. Schematic of the methodology ............................................................................. 19 Figure 2. Schematic of the state-of-the-art review of ionization current related issues in reciprocating combustion engines ........................................................................................ 22 Figure 3. Overall scheme of fault detection and diagnosis, based on Isermann. [16] ......... 25 Figure 4. Effects of pressure, spark gap and electrode temperature on breakdown voltage, based on Pashley et al. [37] .................................................................................................. 28 Figure 5. Classic inductive ignition systems, based on Thurman. [39] ............................... 29 Figure 6. Schematic diagram of contact-breaker ignition system, from [40] ...................... 30 Figure 7. Schematic diagrams of contact-breaker ignition system in charging phase: a) complete, b) simplified, from [40] ....................................................................................... 31 Figure 8. Schematic diagrams of contact-breaker ignition system in release phase: a) complete, b) simplified, from [40] ....................................................................................... 32 Figure 9. Schematic diagrams of transistor-controlled ignition a) single, b) double, with wasted spark, based on Sforza et al. [50]and Pischinger et al. [51] ..................................... 36 Figure 10. Schematic diagrams for coil-on-plug ignition system a) single cylinder, b) multicylinder integrated with ECU, based on Shimasaki et al. [54] .................................... 37 Figure 11. Schematic diagram for CDI ignition system. Based on Shimasaki et al. [58]and Thurman [39] ....................................................................................................................... 38 Figure 12. Model of the thermodynamic system of flame kernel. Based on Herweg and Maly [66], Song and Sunwoo [67], Arcoumanis and Kamimoto [68], Shen et al. [70] ...... 42 Figure 13. Magnitudes of voltage and current of spark plug electrodes during combustion. Based on Pashley et al. [37] and Maly [72]. Phases: PB, pre-breakdown; BD, breakdown; T1, transition; AR, arc; T2, transition; and GL, glow. ......................................................... 44 Figure 14. Divider network circuits to measure secondary ignition voltage. Based on Shimasaki et al. [54] ............................................................................................................. 46 Figure 15. a) Ideal circuit for ion current sensing, b) shunt microammeter, c) feedback microammeter ...................................................................................................................... 49 Figure 16. Independent probe circuits: a) shunt resistor, b) shunt resistor plus amplifier and

c) with voltage divider (after Gazis et al. [21]) .................................................................... 50

Figure 17. Independent probe circuits. a) classsical, with ammeter connected to negative and b) with feedback ammeter and isolation amplifier (after Hu et al. [96])....................... 54 Figure 18. High-side circuit for ion current detection. Based on Eriksson. [25] ................. 56 Figure 19. High-side circuit for ion current detection with power sources from ignition coil. Based on Eriksson. [25], Wang et al. [42] and [40]. ............................................................ 57 Figure 20. High-side circuit for ion current detection. Based on Anderson [92]................. 58 Figure 21. High-side circuit for ion current detection. Based on Yoshiyama-Tomita. [111].

.............................................................................................................................................. 59

Figure 22. High-side circuit for ion current detection. Based on Laganá et al. [113]. ......... 60 Figure 23. High-side circuit for ion current detection. Based on Hunicz et al. [74]. ........... 61 Figure 24. Low-side circuit for ion current detection. Based on Shimasaki-Maki et al. [58] and Wang et al. [42] ............................................................................................................. 62 Figure 25. Transient power sources of ignition in Low-side scheme for ion current detection. Based on [40] ....................................................................................................... 63 Figure 26. Low-side circuit for ion current detection. Based on Förster et al. [84] ............. 66 Figure 27. Normalized ion generation model predictions of (1) Saitzkoff et al., (2) Naoumov et al., and (3) Measured ion current. Based on Naoumov et al. [13]. .................. 76

V

Figure 28. Typical ion current waveform features in SI engines. Based on Andersson [49] and Fiedkiewicz [131] .......................................................................................................... 77 Figure 29. Correlation between cylinder pressure(1), ion current (2), MFB (3) and gas temperature (4). The features of ion current are: (2.1)Primary charge; (2.2)Spark release; (2.3) Chemoionization; (2.4) Thermoionization. Based on Gürbüz [18] ............................. 78 Figure 30. Schematic of the chapter ..................................................................................... 89 Figure 31. Photo of engine setup.......................................................................................... 89 Figure 32. Schematic of the engine setup ............................................................................ 90 Figure 33. Ignition and ion current measurement system .................................................... 92 Figure 34. Test circuits for ignition coils, a) autotransformer ignition coil (based on [40]),

b) transformer ignition coil, c) schematic of the amplifier (2), and d) a photo of the setup 95

Figure 35. Transformation ratio as function of the frequency for the ignition coil ............. 96 Figure 36. Schematic diagram of the resistive high-tension probe ...................................... 98 Figure 37. Schematic diagram of the setup for transformation ratio of elevator transformer

.............................................................................................................................................. 99

Figure 38. Setup for static calibration of resistive probe (preliminary) ............................. 100 Figure 39. Setup for HVDC calibration of resistive probe, a) schematic, b)photo. ........... 101 Figure 40. Setup for frequency response of the high voltage probe, a) Schematic diagram, and b) photo of the mockup. .............................................................................................. 103 Figure 41. Normalized frequency response of the resistive high voltage probe ................ 105 Figure 42. Schematic diagram of the setup for frequency response of the inductive voltage probe ................................................................................................................................... 106 Figure 43. Frequency response of the resistive high voltage probe ................................... 107 Figure 44. Explanation of spurious contribution of power source to microammeter current

............................................................................................................................................ 108

Figure 45. Schematic diagram of the power supply for ion current measurement. ........... 110 Figure 46. Schematic of the shunt resistor and zener limiter implemented ....................... 112 Figure 47. Schematic of the test circuit for frequency response of the zener limiter. ....... 113 Figure 48. Schematic diagram of the microammeter, limiter and amplifier ...................... 114 Figure 49. Photo of the complete ion current measurement setup. From left to right:microammeter, power source and VARIAC ............................................................. 115 Figure 50. Schematic diagram of the isolator amplifier. Stage 1 shown for primary winding voltage measurement. ......................................................................................................... 117 Figure 51. Transient response of the isolation amplifer. .................................................... 119 Figure 52. Frequency response at 10kHz ........................................................................... 119 Figure 53. Frequency response at 100kHz ......................................................................... 120 Figure 54. Frequency response at 230kHz ......................................................................... 120 Figure 55. Schematic diagram of the signal conditioner for the encoder .......................... 122 Figure 56. (A+Z) signal indicating 0,1° motion and TDC position. .................................. 122 Figure 57. Dynamometer for load torque ........................................................................... 124 Figure 58. Free body diagrams of the dynamometer and engine, for a) engine stopped and

b) engine in motion ............................................................................................................ 124

Figure 59. Schematic diagram of the loadcell signal conditioning circuit. ........................ 126 Figure 60.Calibration of loadcell using INST P3............................................................... 127 Figure 61. Test points for Usig,, Uexc, Uamp y Uabs ............................................................... 127 Figure 62. Calibration of Usig : a)compression, b)traction ................................................ 128 Figure 63. Calibration of UAmp : a)compression, b)traction .............................................. 128 Figure 64. Calibration of a)AD620 amplifier gain, b) absolute value circuit gain ............ 129

VI

Figure 65. Simplified schematic diagram of the valve opening sensing system ............... 130 Figure 66. Detailed schematic diagram of the valve opening sensing system ................... 131 Figure 67. Photos of the transformers installed on the valve cover and circuit board. ...... 132 Figure 68. Tests of frequency for transformer of intake valve .......................................... 133 Figure 69. Photo of the setup to measure the displacement of the nucleus ....................... 133 Figure 70. Characterization of the intake valve stroke measurement ................................ 134 Figure 71. Characterization of the exhaust valve stroke measurement .............................. 134 Figure 72. Signal acquired for the exhuast valve (blue) and digitally treated opening stroke (red) .................................................................................................................................... 135 Figure 73. Setup for delay between channels, a) complete. b) detail of connections ........ 136 Figure 74. Schematic diagram of the multiple acqusition system. .................................... 137 Figure 75. Superimposed channels 1 to 8, at f=1kHz. Horizontal axis in “samples”. ....... 137 Figure 76. Details of the superimposed channels 1 to 8, at f=1kHz. a) full rising slope, b) detail of rise. Horizontal axis is in “samples”. .................................................................. 138 Figure 77. Schematic view of the chapter .......................................................................... 141 Figure 78. Test campaign schematic .................................................................................. 148 Figure 79. Matlab® user interface ..................................................................................... 153 Figure 80. Sample of gathered signals. .............................................................................. 154 Figure 81. Sample of a scaled primary current .................................................................. 154 Figure 82. Scheme of the conversion of a sampled signal to cyclic signal arrays ............. 155 Figure 83. Log-log plots of a) transformed pressure signal, in red. Transformed pressure with removed components, in blue. b) transformed pressure with removed components, cleaned. Original pressure signal obtained in motored condition at 500min-1. .................. 157 Figure 84. a) Plots of motored pressure at 500min-1. Original signal in red, referenced signal in blue. And b), detail of unfiltered motored pressure signal in black and reconstructed (filtered) signal from inverse FFT in cyan................................................... 157 Figure 85. Scheme of the code for pressure signal correction. .......................................... 158 Figure 86. Scheme of the coding for correction of ion current signal and area calculation

............................................................................................................................................ 158

Figure 87. Initial setup for motoring with AC induction motor, a) Initial mounting of belt drive and b) Additional instrumentation. ........................................................................... 159 Figure 88. Setup for motoring with DC compound motor ................................................. 160 Figure 89. Schematic of the power supply for the DC compound motor .......................... 161 Figure 90. Sample of ion current signal as a function of time during motoring without spark. .................................................................................................................................. 162 Figure 91. Sample of cyclic ion current during motoring without spark ........................... 162 Figure 92. Sample of mean ion current, pressure and valves displacement as function of crank angle during motoring without spark. ...................................................................... 163 Figure 93. Primary winding signals related to ion current (secondary current) and pressure

............................................................................................................................................ 164

Figure 94. Details of B1 (compression) and B2 (wasted spark) ........................................ 165 Figure 95. Motoring waveforms for pressure, secondary current and secondary voltage measured with both probes, being a) test with prechamber and b) test without prechamber.

............................................................................................................................................ 167

Figure 96. Motoring waveforms for pressure, secondary current and secondary voltage measured with both probes, without prechamber .............................................................. 168 Figure 97. Comparisons between ion factors a) 104 and b) 105 for the microammeter. .... 169 Figure 98. Comparisons between combinations of Rion and G for the microammeter. .... 171

VII

Figure 99. Motoring waveforms for pressure, secondary current and secondary voltage measured with both probes. U = -100V ............................................................................. 172 Figure 100. Modification to the ignition module ............................................................... 173 Figure 101. Cyclic plots of a) ion current, and b) pressure. ............................................... 174 Figure 102. a) Exclusion of atypical values of pressure, and b) average of 60 cycles of pressure, ion current and secondary voltage (both probes). ............................................... 175 Figure 103. Cyclic plots of a) ion current, b) pressure. c) average of ion current, pressure and secondary voltage (both probes). U = -300V. ............................................................. 176 Figure 104. Cyclic plots of a) pressure, b) average of ion current, pressure and secondary voltage (both probes). U = 100V. ...................................................................................... 177 Figure 105. Combustion waveforms for pressure, secondary current and secondary voltage measured with both probes, with prechamber.................................................................... 178 Figure 106. Cyclic plots for combustion without prechamber, a) Pressure, b) box plot of pressure deviation, c) ion current, d) secondary voltage (clamp) ...................................... 180 Figure 107. Average of cycles for combustion without prechamber, showing pressure, ion current, valve displacements and secondary voltage (both probes) ................................... 181 Figure 108. Cyclic plots for combustion without prechamber, showing abnormal pressures and ion currents .................................................................................................................. 182 Figure 109. a) Average of cycles for combustion without prechamber, b) Atypical cycle. Waves: pressure, ion current, valve displacements and secondary voltage (both probes). 183 Figure 110. Schematic diagram of the electrical nets of the ignition core module. Ignition core module comprises the spark plug, the secondary winding of the ignition coil, the resistance of the high tension wire (and internal spark plug resistor), the promoter power source and the microammmeter. The model of the primary and secondary electrical loops, includes the inductive effects of both coils. ....................................................................... 187 Figure 111. Results of the simulations of free-run: a) with primary coupling , b) without coupling. ............................................................................................................................. 192 Figure 112. Free run frequencies observed on ion current signal as function of time. ...... 193 Figure 113. Details of the ionization current waveform: a) and c) Oscillations of transistor engaging, b) free-running oscillation in wasted spark and d) free-running oscillation preceding the ionization of the air-fuel mixture. ................................................................ 194 Figure 114. Results of the simulation of free-run without primary coupling for diverse values of Rion. ..................................................................................................................... 196 Figure 115. Comparison of average ionization current signals for varying Rion. Common engine settings: throttle opening = 100% ; spark advance = 12° BTDC. Ion amplifier gain G = 1, except for Rion = 100 . .............................................................................................. 197 Figure 116. Sample waveforms of a) full ionization current signal, b) in-cylinder pressure,

c) windowed ionization current, d) ionization current integral. ......................................... 200

Figure 117. a) magnitude of maximum pressure compared with ionization current area under the curve, and b) regression model of the variables. ................................................ 201 Figure 118. Magnitude of maximum pressure compared with free-run damped frequency and regression model of the variables. ............................................................................... 202 Figure 119. Magnitude of maximum pressure compared with magnitude of thermoionization peak of ionization current signal and regression model of the variables. ......... 202 Figure 120. Three-dimensional surface fit modeling normalized peak pressure as a function of normalized ion signal area and damped natural frequency............................................ 203 Figure 121. Peak pressure position compared with the centroid of ion area. .................... 204

VIII

Figure 122. Percentage of maximum ionization current integral signal corresponding to the crank angle position of peak pressure. ............................................................................... 206 Figure 123. a) Peak pressure position compared with angular positon of thermo-ionization peak, and b) correlation between them. ............................................................................. 206 Figure 124. Correlation between normalized pressure peak and its normalized angular position ............................................................................................................................... 207 Figure 125. Comparison of three-dimensional surface fitting results for predicting crank angle position of pressure peak: a) normalized damped natural frequency and centroid of ion current area; b) normalized ion current area and centroid of ion current area. The z-axis represents the crank angle position of pressure peak ......................................................... 208 Figure 126. Correlation between normalized magnitude of pressure and normalized frequency, and superposition of pressure cycles for test 1................................................. 218 Figure 127. Correlation between normalized magnitude of pressure and normalized frequency, and superposition of pressure cycles for test 1................................................. 218 Figure 128. Three-dimensional surface fitting result for predicting crank angle position of pressure peak from normalized damped natural frequency and centroid of ion current area (Test 4). The z-axis represents the crank angle position of pressure peak. ........................ 222 Figure 129. Ignition delay and angular position of pressure peak. .................................... 223 Figure 130. Plots of normalized in-cylinder pressure, ionization current, MFB and HRR as functions of crank angle for test 8. ..................................................................................... 225

IX

List of Tables Table 1. Energy balance for plasma discharge for flame kernel stages, under ideal conditions and small electrodes. Based on Maly ............................................................45 Table 2. Summary of contributions of authors in ion current…………………………...82 Table 3. Transformation ratio of transformer T2………………………………………..99 Table 4. Calibration results for resistive high voltage probe with AC 60 Hz....................100 Table 5. HVDC Calibration results for resistive high voltage probe................................102 Table 6. Results of frequency response for the resistive high voltage probe...................104 Table 7. Results of frequency response for the inductive high voltage probe...................106

Table 8. Values of Rion and fuels used for different authors…………………………….111 Table 9. Measured resistance of the shunt resistor with the zener limiter…………..…..113 Table 10. Analog isolation amplifiers……………………………………………………116 Table 11. Combinations of Rion and G for the tests………………………………………143 Table 12. Values of U, schemes and fuels used for different authors…………………...143 Table 13. list of variables and assignment of channels and racks in the acquisition system……………………………………………………………………………………146 Table 14. Motoring tests. (*) indicates no prechamber......................................................149 Table 15. Combustion tests with prechamber…………………………………………….150 Table 16. List of tests under combustion conditions without prechamber………………………………………………………………………………..152 Table 17. Accurate and maximum ion current indications for each factor………………170 Table 18. Summary of correlation results between ion current features and pressure peak characteristics across engine…………………………………………………215 Table 19. Correlation coefficients for the tests performed (normalized damped frequency and normalized maximum pressure.217

Table 20. Comparisons of correlations with normalized peak pressure between normalized damped frequency, normalized ionization current area and the combination of both………………………………………..…219 Table 21. Results of the angular difference ppp - cent………………………………...220 Table 22. Results of the percentage of maximum ion current integral related to PPP...221 Table 23. Comparison of correlations: centroid-PPP, combined centroid-damped frequency- PPP and combined centroid-ion signal area-PPP……………………………………...221 Table 24. Ignition delay and PPP estimation from ignition delay, for tests 1 to 9........223 Table 25. Crank angular positions of ion current……………………………………..225 Table 26. Angular position estimations………………………………………………..226

Glossary  Astable multivibrator: Oscillator that produces square wave.  Burden voltage: The voltage drop of a shunt resistor used to measure current.  Crosstalk: electrical interference between two channels of an acquisition system.  Dwell time: magnetic field charging time of the ignition coil.  Equivalence ratio (): fuel – air relation.

 Flyback: voltage elevator.

 Free run, ringing: oscillation of a system after removing excitation.  Monostable multivibrator: Pulse generator triggered by edge.  Variac or VARIAC: Autotransformer with accessible center tap to have variable amplitude AC output.

 Victim: On signal integrity, the device affected by strong signal from offender.  Offender: On signal integrity, the device that affects another (the victim) with a strong signal.

List of symbols and acronyms

Acronyms AC Alternating Current

AFR

Air to Fuel Ratio

AMSC

Amplitude Modulation with Supressed Carrier

ANN

Artificial Neural Network

ATDC

After Top Dead Center A/D Analog to Digital CAN Bus Controller Area Network Bus

CDI

Capacitor Discharge Ignition

CFD

Computer Fluid Dynamics CI Compression Ignition

CMOS

Complementary Metal Oxide Semiconductor

CMTI

Common Mode Transient Immunity CV Constant Volume (chamber) DC Direct Current D/A Digital to Analog

ECU

Electronic Control Unit

EGR

Exhaust Gas Recirculation e.m.f.

Electromotive force

EOC

End Of Combustion GM General Motors

HCCI

Homogeneous Charge Compression Ignition HT High Tension (wire) HV High Voltage

HVDC

High Voltage Direct Current

ICE

Internal Combustion Engine

IGBT

Isolated Gate Bipolar Transistor

ISO

International Standards Organization kS/s Kilosamples per second

LCD

Liquid Crystal Display

LED

Light Emitting Diode

LVDT

Linear Variable Differential Transformer

MAF

Mass Air Flow

MAP

Manifold Absolute Pressure

MBT

Maximum Brake Torque

MFB

Mass Fraction Burned

MLP

MultiLayer Perceptron

MSD

Multiple Spark Discharge

MSI

Multiple Spark Ignition Mux Multiplexer

OBD

On-Board Diagnostics

OHV

Overhead Valve

PC Personal Computer

PPP

Peak Pressure Point ppr Pulses per revolution

PVC

Polyvynil Chloride RF Radiofrequency rms Root Mean Square (value)

RON

Research Octane Number

SCR

Silicon Controlled Rectifier

SDL

Spark Duration Limit SI Spark Ignition

SNR

Signal to Noise Ratio

SOC

Start Of Combustion s&h Sample and Hold

TCI

Transistor-Controlled Ignition

TDC

Top Dead Center

THD

Total Harmonic Distortion

UEGO

Universal Exhaust Gas Oxygen (sensor)

ULSD

Ultra Low Sulfur Diesel (fuel) UV Ultraviolet (radiation)

VFD

Variable Frequency Drive VW Volkswagen

1 Introduction

The present thesis comprises the review of a significant stacking of published works related to the ionization current signal as a non-intrusive inference method of in-cylinder pressure features, in conjunction with ignition systems variables, aimed to combustion diagnostics. This work describes the especially built electronic instrumentals, along with their corresponding validations for ion current measurement, ignition variables, valve opening, rotational frequency, torque, in-cylinder pressure. The instrumentation was integrated to a modified single-cylinder port-injected spark ignition engine. The motoring and combustion acquisition campaigns were defined and performed, presenting the signals processing and their sensitivities’ checkings to diverse conditions. A model of the natural damped frequency at the end of ignition pulse is presented, and its corresponding experimental analyses are combined with existent ion signal features methods to assess the cycle-to-cycle variations of pressure.

1.1 Background and motivation

Why performing research on combustion engines today?

In the work of Lešnik et al. [1], it is written: “The majority of on-road vehicles today are powered by internal combustion engines, which are, in most cases, burning petroleumderived liquid fuels mixed with bio-components. The power to weight ratio of internal combustion engines combined with the high energy content of conventional fuels, which can be refilled easily in matter of minutes, makes them ideal for all kinds of road transportation... In recent years, a lot of effort was also put into promotion of electric vehicles as zero emissions vehicles. This statement should be reconsidered, since the greenhouse impact of electrical vehicles is not negligible. Conversely, in some cases, an electrical vehicle can have an even higher emission impact than modern vehicles with sophisticated internal combustion engines. This is characteristic for countries where the majority of the electricity is produced in coal power plants. With the decrease of greenhouse gas emissions in the Electricity Production sector, and with the increase of battery capacity, the role of electric vehicles in the Transport sector will probably increase. Despite significant research and financial investments in electric vehicles development, the transport sector in near future will be mostly powered by internal combustion engines and petroleum-derived liquid fuels”. In the same work, relative to the electrical energy needed to power all the potential surrogate electric vehicles that would be applied to the Transport sector, it is stated that the current amount of electricity is not enough to do so.

In a similar line, Kalghatgi [2] states that the elimination of global greenhouse gas emissions will not happen by 2050, let alone 2030 because of the huge scale of the problem. In addition, the transportation sector is very difficult to decarbonize, as well as steel, cement, gas industries.

Sinigaglia et al. [3] indicate that the technological maturity rate of internal combustion engines was 81,77% by 2018, with subtechnologies with a current diffusion speed of 32,14%

for friction loss subjects, 21,05% for water injection, and 20,88% for hybrid technology. In addition, they forecast 27 years to encounter patent saturation in the subject of ICE.

The local case (Colombia) exhibits similar trends to the global one, where prime means of transportation, both for cargo and passengers, are strongly dependent on vehicles powered by combustion machines [4]. Backup electrical power is also dependent on engine-generator modules. Recent local government efforts aim for policies related to the decarbonization of the economy, as well as the transition to clean energies.

Thus, even considering the explained drawbacks, there is still room for improvement of the combustion machine, and the present work points to that research line.

Combustion diagnostics and pressure measurement

Combustion analysis is used in the industrial as well as the academic communities as a convenient method to quantify the effects of modifications on engine design and calibration, and the effect in velocity and wholeness of combustion.

Engine diagnostics tasks include:

 Assessing of fluid and thermodynamic processes that condition the development of the chemical energy release from fuels in SI engines.

 Explaining the causes of operational anomalies in SI engines.  Determining the actual technical condition of the combustion machine.  Determining the necessity of adjustments, calibrations, maintenance, inspection, substitution of engine components during their operation.

 Evaluate the quality of tuning, maintenance and repair tasks performed.

To investigate engine combustion processses, there are two main methods:  Optical technologies, such as Laser induced fluorescence, Schlieren images, Emission spectroscopy, Laser Doppler velocimetry.

 Pressure-based methods, such as piezoelectric, piezorresistive, fiber optics.

Vollberg et al. [5] indicate that the accurate in-cylinder pressure sensing was tackled decades ago. However, that technology is not widespread in today’s mass-produced passenger cars and other ICEs. Vollberg et al. explain that in-cylinder pressure sensors applied to combustion engines have challenges like cost-effectiveness and production viability, a reliable operation for up to 109 cycles, easiness of installation and packaging, and must withstand the harsh environmental conditions of the combustion chamber.

Similarly, it is pointed that for test bench operations, that is, engine research, the requirements are different, and piezoelectric sensors are extensively used for those purposes. In [5], it is specified a lower limit of lifetime expectancy for piezoelectric sensors of 12000 hours.

Kurtz et al. [6] point out that piezorresistive sensors can read both static and dynamic components of pressure, whereas piezoelectric are better intended to dynamic pressure sensing. In addition, piezoelectric sensors require an external charge amplifier, opposed to the Wheatstone bridge strain gauge primary sensor and instrumentation amplifier sections of the piezorresistive one, that can be integrated onto the same package.

Indirect pressure measurement approaches had taken place, using piezoelectric washershaped sensors on the washer of the spark plug [7], [8], or an engine head screw [9]. Nevertheless, those sensors rely their sensitivity on the mounting position and the mechanical stiffness of the surrounding construction [5].

Fiber optics sensors offer insensitivity to EMI interference, high bandwidth, long life expectancy, high accuracy and stability of signal. However, they are significantly difficult to manufacture [10].

Ion current detection leverages the existent spark plug electrodes (SI engines) or a modified glow plug [11], to sense the current through the gap between electrodes, promoted by an electrical potential difference across the electrodes, due to the ionized substances present at the gap. This ionization takes place during the breakdown of the air-fuel mixture that later results in an in-cylinder pressure rise caused by combustion. Besides the existent electrodes, no additional components are required inside the combustion chamber for pressure estimation from ionization current signal. The cost of the spark plug as primary sensor is lower than the previous alternatives, and is a very robust component.

Although the shortcomings of the ion signal, in comparison with the in-cylinder pressure, during intake and compression strokes, publications that will be referred in detail in chapter 2, for example Gazis et al., Rivara et al., Gao et al., indicate the use of the ion current signal as an in-cylinder pressure estimator, “training” artificial neural networks to predict pressure peaks. Moreover, the ion signal can be used to on-line monitoring of the combustion process, revealing the ignition onset occurrence, flame propagation, in particular its speed.

Local research context.

Henao [12] states that the combustion engine research efforts point mostly, with the exception of research groups affiliated to Universidad de Antioquia, Universidad Nacional, Universidad del Norte, Universidad de los Andes, to exploitation properties (external characteristics), not necessarily from engineering perspective, due to the chronicly dominant country’s import policy.

The material results of Henao’s work, affiliated to the research group “Grupo de Manufactura y Diseño de Máquinas” of “Facultad de Mecánica Aplicada” in “Universidad Tecnológica de Pereira”, and implemented in the Engine Research Laboratory, constituted a base point of constructive and technological appropriation, where a dynamometric test bench has been implemented and used to quantify load torque and engine speed characteristics, instantaneous in-cylinder pressure, several engine temperatures, mean fuel consumption. With these measurements, energy characteristics, as well as emissions are related, under steady operation points. Additionally, engine performance is evaluated using conventional and alternative fuels. Moreover, monitoring and performance diagnostics, including technical and mechanical condition, from external characteristics, indicator diagrams assessment, data correlation and analysis methods, have been performed.

Nevertheless, the aforementioned research group realizes that the comprehension of in-depth chemical and physical phenomena, that take place inside the engine subsystems, provide the

high-level scientific culture to participate in design, modeling, simulation, and diagnosis of combustion engines at their different levels and study activities. Thus, the author of the present thesis looks for an unveiling of the mechanisms, phenomenology and underlying laws within the combustion processes of SI engines, emerging the challenge of building proprietary electronic equipment, and tackling methodologies related to the dynamics of ignition systems and their electrical variables and parameters, ignition and combustion phenomena, in conjunction with the implementations of complementary ionization current measurement placed in series into the secondary winding loop of the ignition coil. Another motivation comes from the absence of non-intrusive instrumentation systems within the Engine Research Laboratory of the University, based on ionization current, that allow combustion dynamics diagnostics.

This work constitutes an answer to local fundamental research demand on thermal systems and mechanical power, energy characteristics of fuels, in particular alternative ones, and it is a part of the commitment that the Research group has assumed in the subject of internal combustion engines.

When an air-fuel mixture burns, the substances become ionized, resulting in an electrical conductivity. This was leveraged by Schnauffer in 1934 to assess the speed and form of the flame front inside the engine’s combustion chamber. There are works that date from the fifties of the past century related to the ionization that takes place on rocket exhausts, following the path to the study of the progression of engine combustion. One example is Arrigoni et al., who tackled the effect of fuels, engine speed, load torque, ignition timing, AFR on the burning velocity, using ion sensing. Other authors like Witze, Hu et al., Nicholson, Meyer et al., explored the use of a modified head gasket as ion sensor, to assess the flame location and arrival time, swirl, tumble and squish characterization. Shimasaki et al. explains the challenges of the three ion current detection schemes, performing both ion signal and secondary voltage measurements. Auzins et al. explores the misfire and knock detection using a low-side ion current detection scheme. Eriksson proposed three methods to correlate the ion signal current with the in-cylinder pressure, based on the area under the ion curve, its centroid and a sum of gaussian functions. Leveraging that gaussian summation, Nielsen, Hellring, Andersson propose variations to achieve closed-loop spark timing control, PPP estimation and virtual pressure sensing. The modeling of ion formation starts from Calcote, following with Saitzkoff et al. that considered only the thermal peak, and then Naoumov et al. improves the prediction of chemi-ionization peak event. Daniels explored the estimation of MFB and HRR, from the in-cylinder pressure and ion current signal derivatives respect to the crank angle, comparing the Wiebe functions built from each one. Yoshiyama et al. tackled the difficulties of correlation of ion current and pressure in low-load conditions. Also, they explored the correlation between ion signal and

AFR.

Artificial Neural Networks have been used in conjunction with ion current: Huang et al. explored the detection of AFR in cold start. Gazis et al., Panousakis et al., Rivara et al., Gao et al. to reconstruct the in-cylinder pressure from ion signal features.The ionization current integral has been used as a misfire diagnosis on the works of Yamada et al., Shimasaki et al., Laganá et al. Budko et al. used the ion signal integral to estimate the PPP. Gürbüz performs simultaneous correlations between ionization current, in-cylinder pressure, combustion chamber temperature, MFB.

The ionization signal sensing is strongly explored in spark-ignition engines, most of them fueled by gasoline or natural gas. Nevertheless, Glavmo pioneered the ion current signal analysis on compression ignition engines. Then, additional research efforts in the matter arise from the works of Henein et al., Estefanous, George, Assaad. Mehresh et al. also tackled the research of ion current signals in HCCI mode of combustion, finding the subsequent works of Panousakis et al., Chen et al., Phan et al., Dong et al., Hunicz et al. in the matter.

However, to the best of the author’s knowledge, only two works highlight into the colombian research groups in the subject, aboarding the knock detection and partial burn.

The ion current signal, by itself, contains combustion events information within the interval between the chemi-ionization and thermo-ionization. Nevertheless, when the ignition system is electrically coupled into the same loop with the ion current detection system, secondary winding oscillations that come from the magnetic charge and high-voltage release of the ignition coil are detected by the ionization microammeter. Therefore, compared to the incylinder pressure waveform, the ionization current, adding the secondary winding contributions into the same current signal, the richness of information of the total current is greater, indicating:

 The onset of primary charging  The ignition strike release.

 The final contribution of ignition: the free-running oscillation.  The chemi-ionization peak, taken in literature as the combustion onset.  The thermo-ionization peak, close indicator of maximum temperature, 90% of MFB and proximity to the pressure peak, according to Gürbüz’s work.

Thus, the development of a combined gathering of ignition variables and ionization current detection to enhance combustion features recognition is justified. Moreover, the in-cylinder pressure in cooperative mode, as indicated by Zhu et al. can provide additional outcomes of combustion assessment.

The results of the present work can be applied in:

 Engine combustion process diagnosis in laboratory and operational conditions.  Calibration of engine control systems and engine components design.  Engine testing, as a cooperative control system tool, used in combination with the ECU, including for example in-cylinder pressure or knock sensors, and combustion data registry.

 Didactic tool for thermodynamics courses, ICE’s, to demonstrate the different variables data acquisition.

This thesis aims to answer the following research question:

How to detect and leverage the secondary winding breakdown voltage and ionization current that takes place on spark plug electrodes to diagnose the quality of combustion on SI engines under different load torque, rotational frequency and parameter setting conditions?

This work bases its feasibility on the hypothesis that the ionization current signal, promoted by the ignition and combustion processes, as well as the ignition system variables can be reliably measured, allowing to assess the combustion process in a similar and competitive manner, compared with the intrusive piezoelectric in-cylinder pressure sensors.

1.2 Objectives and methodology

The main objective of this work is the development of a non-intrusive measurement system, based on ionization current detection inside the combustion chamber, to determine the incylinder pressure during the combustion process, from experimental and numerical approaches on a SI engine, under defined load torque and rotational frequency conditions.

From this main goal, specific objectives emerge as follows:

 Define experimental methodologies for the development and fine-tuning of measurement instrumentals and digital signal processing of voltages and currents present on both primary and secondary sides of the ignition coil, as well as ionization current between the spark gap electrodes of a combustion chamber for combustion studies. The aforementioned variables are subjected to noise superposition. Thus, substantial work is needed to carry out on the implementations of reliable experimental and mathematical filtering tools to ensure reliable combustion diagnosis.

 Develop and implement algorithms and methods for the characterization of experimentally acquired ionization currents and their correlation with the course of combustion processes in test engines.

 To compare and assess the quality and limitations of combustion diagnosis based on ionization current with combustion diagnosis based on in-cylinder gas pressure measurement.

 To develop a methodology for assessing the quality of the combustion process in engines, relating the ionization current signal profile to certain combustion process parameters, such as the start of combustion, the instant of maximum pressure, the end of combustion, the ignition delay time, the pressure gradient, and the heat release gradient.

The methodology proposed in the present work branches into four groups of activities destined to tackle the aforementioned objectives, as well as the main outcome related to nonintrusive engine diagnosis and pressure features inference from ion current signal. The schematic view of the methodology is presented in Figure 1.

First, there is a gathering and aprehension of information referent to tools, methods and algorithms of engine performance diagnostics, tools and methods of combustion diagnostics, ion formation and transport phenomena applied to combustion processes within open and closed media and piston engines. In addition, methods, circuital schemes and algorithms to measure and process ignition coil variables and ion current are explored.

A second portion points to an experimental study of currents and voltages of the ignition coil, implementing an ignition system test bench. Then, it is developped a proprietary ionization

current signal detection system, including the currents and voltages of both primary and secondary windings of ignition coil, attaching the ion current sensor to the secondary loop.

Third part comprises tests in open air environment, using the implemented ignition mockup to characterize their corresponding electric arc and ion formation phenomena. A subsequent task involves the instrumentation of a single-cylinder engine that encompass a variable load torque provided by a dynamometer setup. The instrumental of the engine test bench adds the distribution system (intake and exhaust), ignition system, crank angle, in-cylinder pressure, temperatures and ionization current measurands.

Figure 1. Schematic of the methodology Fourth segment refers to the offline signals processing, collecting in-cylinder pressure, voltages and currents of both primary and secondary side of ignition transformer, ionization current, distribution system quantities, performing comparisons and correlations between them, and assessing their linked sensitivities, along the different engine operating conditions

such as engine temperature and speed, load torque, air-fuel mixture, spark advance, for combustion diagnostics purposes. Once the corresponding correlations between in-cylinder pressure, voltages and currents are met, a set of criteria are defined and proposed to recommend the use of the experimental tool developped for engine combustion diagnosis.

1.3 Thesis outline

For the information gathering, chapter 2 addresses the literature review, beginning with the basics of engine diagnostics, focusing on methods of fault detection and diagnosis. Subsequently, electrical schemes of ignition systems are explored. Next subsection comprises the fundamentals of flame kernel development and corresponding electrical relations. Next subsection describes the three general electrical schemes of ionization current detection and their variants. Then, the basics of combustion chemistry and main models of ion generation are presented. Another part comprises the main features of ionization current signals, and their diagnostics capabilities.

To cover the experimental study and instrumentals, chapter 3 describes the implementation, static and dynamic characterization of proprietary electronic devices applied to ionization current measurement, voltages and currents of both primary and secondary sides of the ignition coil. In addition, the intake and exhaust valves opening sensing subsystem, combined crank angle and TDC indication, load torque, air-fuel ratio (AFR), in-cylinder pressure and engine temperatures.

For the instrumented engine, chapter 4 defines the ranges of variables and parameters, as well as acquisition campaigns. Next, the referencing and filtering of the in-cylinder pressure and other software implementations are described. Last subsections present preliminary findings on ionization current, ignition and valve opening signals during motoring and combustion tests, with and without prechamber.

Additional signal processing is covered in chapter 5, focusing on the inference of cyclic variations of peak pressure from the ionization current signal features, as well as a novel analysis of damped natural frequency of the remaining energy of the ignition strike, and further combinations of these methods to highlight the corresponding correlations and their fitting goodness.

Chapter 6 comprises the final conclusions of the present work, highlighting the main conclusions, future work and contributions, related to the static and dynamic characterization of ignition voltmeters and ammeters, the ionization current measurement components, the distribution system valve opening transducers, the ignition system modeling conjoined to a low-side scheme of ion current detection, damped natural frequency analysis, ion current features analysis, estimation of misfire, preignition and cyclic variation.

Literature review

2.1 Introduction

Diagnostics (diagnosability) of combustion engines is an attribute demanded by the user and by the international regulations [1], both for engines under development and for engines put into service in transportation systems. Indeed, according to both the European and American legal regulations, motor vehicles are required to monitor and detect misfires as well as to locate the cylinders where combustion is not complete; it is enforced that a 100% misfire detection covers the activity of the engine within all load and speed ranges (misfire detection constitutes an essential ingredient of On Board Diagnostics, OBD). On the other hand, precise monitoring of combustion and feedback control becomes paramount for safe and efficient engine operation.

The diagnosability of the engines is considered during their design, modeling, simulation, development and tuning process, at the engineering level, as well as in their exploitation cycle - during their operation, monitoring, service and repair. To this end, research work is currently being carried out to design electronic control systems for vehicle engines by embedding diagnostic functions that facilitate condition-based service in workshops, pursuing mechanical and thermal performance targets for vehicle engines. The diagnostic functions are based on experimentally validated modeling of the stationary and transient dynamics of engine mechanisms and systems, and of the entire engine unit, under different foreseeable operating conditions.

Electrical conductivity of reactants and products of combustion is today pondered as a property of great potential to study the combustion behavior of different fuels and additives, to delve into the intricacies of chemical kinetics and flame chemistry underpinning the understanding of reaction mechanisms, and in the overall combustion process in combustion engines.

Combustion diagnostics is a subject of great interest to the community involved in the study of internal combustion engines (ICE) because of its direct correlation with the performance and thermodynamic efficiency of the ICE energy conversion process. In addition to this, tracking the kinetics of the chemical reactions that modulate the dynamic response of torque, emissions formation, and ignition irregularities (auto-ignition and knock) based on the measurement of the conductivity of the reaction products, is convenient in multiple ways for purposes of approaching the dynamics of heat release and flame front velocities, within the space and over time inside the combustion chamber. The discussion about the possible control modes of forward and feedback control of the combustion process of vehicle engines is not recent; in spark ignition engines, the ion current can be the key parameter to optimize the engine performance (torque, fuel consumption, misfiring, knocking, air-fuel ratio) and to minimize the exhaust emissions (HC, NOx, CO). All this is a motive to review the state of the art of combustion diagnostics based on ionization currents.

The study of the state of the art of ionization current based ICE diagnosis in this thesis has been approached primarily with the objective of exploring the current models and experimental works used in the diagnosis and prediction of combustion inside the engine, as

well as delving into the coupled electrical phenomena of the engine-ignition-system and the electrical properties of the combustion gases, including, of course, the intricacies of the ignition systems´ dynamics. The review has been carried out according to the scheme shown in Figure 2.

---

Figure 2. Schematic of the state-of-the-art review of ionization current related issues in reciprocating combustion engines

To give a background to the ionization current and voltage-current performance of the ignition systems in relation to combustion process diagnostics of combustion engines, a literature study is performed and presented in this chapter. Linked to this thesis, primary references about battery-coil ignition systems date back to the period from 1926 to 1946; research on the ionization in flames appears in the mid-1950s concerning the optimization of magnetohydrodynamic generators and the study of ionized particle formation in combustion products of propulsion systems [13]. Following from the conducted review, a first systematic summary of the hitherto understanding of the mechanisms of ion formation in flames was achieved by H.F. Calcote in 1957 [14]. Ion current based engine diagnostics and combustion quality monitoring has been explored theoretically and experimentally by many investigators since then: Collings et al. (1991), Shimasaki et al. (1993), Witze (1993), Auzins et al. (1995), Lee and Pyko (1995), Saitzkoff et al. (1997), Nielsen and Eriksson (1998), Asano et al. (1998), Balles et al. (1998), Daniels (1998), Reinmann et al. (1998), Wilstermann (1999),

Förster et al. (1999), Hellring et al. (1999), Andersson and Eriksson (2000), Naumov (2001),

Franke (2002), to mention a few of them. The physics and effects of spark ignition on IC engines have been broadly overviewed by Maly (1984). It can be said that the experimental approaches have not changed much over the last decade, and what has been making a difference is the advances in the works based on computational methods for the ion current signal analysis, information content, and correlations with combustion process´s parameters. More work remains to be done on detailed models of the ionization current-voltage parameters of the ignition system during the spark break-up and the mathematical algorithms for processing these signals.

The chapter opens with an overview of the engine diagnosis methods, followed by the modeling fundamentals of the most used electrical ignition systems, i.e. battery conventional, transistor-controlled, capacitor discharge (CDI) ignition systems, among others, altogether with the conservation-expressions-based models related to flame kernel formation and its development phases, along with the electrical relationships underpinning the engine diagnostics suitability of secondary voltage measurements. The central section of the chapter delves into the main schematic circuits for ion current detection and microammeters, such as the independent probe, and high and low side circuits, highlighting the relevance of the power supply voltage polarity, and offering comments on the use of AC power supply. The final section of the chapter addresses the basics of combustion chemistry behind the primary ion generation processes, represented by chemi-ionization and thermo-ionization. It tackles the distinctive features of typical ion current waveforms in SI engines emphasizing the diagnostic possibilities that arise from ion current detection.

2.2 Chapter overview

This chapter is pursued to give an appropriate context to this thesis. To this end, the review begins with a broad glance at the methods of fault detection and diagnosis, focusing on the ion current measurement as a method that can be applied to detect and diagnose anomalies of operation in SI engines. Afterwards, this review presents relevant subjects on ignition systems and flame kernel modeling, required to gather information on the energy provided by ignition systems in relation to combustion chamber pressure, the ranges of voltages and currents on the components of ignition systems, linked to ion current measurement (in particular, to low-side ion current detection circuits), and with that having proper instruments to quantify those variables. The third part of the review describes the most common ion current detection circuits, including different variations present in literature. The fourth part is dedicated to explaining the phenomena and the models of ion current generation along with the ion current signal interpretation. The last part deals with the possibilities of engine fault detection and diagnosis based on ion current measurement, and also the extension and limitations of the technique.

2.3 On methods of fault detection and diagnosis

As it has been reported in the work of Mejía et al. [15], many diagnostics advances in the last three decades have been possible through the introduction of mechatronic components in the combustion engine; until the advent of the on-board diagnostics facilities, the engine diagnostics mostly consisted of limit checking of important variables such as the cylinder compression, compression leaks, intake manifold vacuum, oil pressure, coolant temperature, spark advance, current voltage traces in the oscilloscope. After that, digital engine analysis became the standard, particularly with On Board Diagnostics OBD, plugging to ISO 15765-

4 normalized connection, that is, CAN Bus; it followed then the wireless with internet

connection remote diagnosis.

Following Isermann [16], the internal combustion engine is a mechatronic machine, a combination of mechanic and electronic hardware with increasing informatic capabilities. Now, not only sensor information reading, and engine feedback control is implied, but also supervision and fault diagnosis. With that comes fault classification (performance, emissions, drivability, safety) and corrective actions of the engine itself (disconnect, factory default parameter setting, limp mode). Isermann describes the overall supervision, detection and fault diagnosis tasks for open-loop and closed-loop engine management systems, using either detectors (measurement instruments) or observation of variables by skilled technicians, see Figure 3.

For the case of the measurements, there is an analytical approach: the measured variables feed a data processing stage, preceding a feature extraction and a change detection stage. The analytical knowledge database has the model of the process, estimation and filtering, and a set of normal features, to generate analytical symptoms if a fault occurs. For the case of observation of variables, there is a heuristic approach. The observed variables feed another feature extraction, after which process history and statistics are accomplished. The heuristic approach uses an experience-based model, records of normal behavior, typical causes of faults and their usual contributions. With this, heuristic symptoms are generated upon the presence of a fault. Both analytical and heuristic symptoms can be integrated to classify

fault(s) on the system, having pattern recognition or inference rules, and finally fault diagnosis, if present.

Figure 3. Overall scheme of fault detection and diagnosis, based on Isermann. [16] For fault detection, Isermann presents a survey of analytical fault-detection methods, using a single variable or a plural number of them. For the case of a single variable, checking of limits and trends are applied, having possibilities of fault detection by fixed or adaptive threshold, or change-detection methods [17]. For multiple signals, the checking can be based on the signal model, process model or a multi-variant data analysis. Among tools for fault detection using the signal model, there are correlation [18], spectrum analysis [19] or wavelet analysis [20]. When using the process model, possible tools are parameter estimation, artificial neural network [21], [22] , [23] , [24], state observers, state estimators or parity equations.

For multi-variant analysis, Isermann indicates principal component analysis as fault detection tool.

For fault diagnosis, Isermann shows a summary of fault-diagnosis methods, branching in two paths: classification methods and inference methods. For classification purposes there are tools like pattern recognition (using decision tables), statistical classification (Bayesian [25], [26] or decision-tree [27]), approximation (polynomial classifier), probability density (geometrical classifier [25]), and artificial intelligence methods (fuzzy or ANN-based classifiers [21], [22], [23], [24], [26]). For the case of inference methods, Isermann presents the binary (predicate logic) and approximate reasoning (using fuzzy logic or ANN) as tools for fault diagnosis. For fault diagnosis methods applied to internal combustion engines, there are two approaches: external to the ICE and internal to the combustion chamber of the ICE.

Possible external measurements methods include the following:

- Starting current. To detect individual compression leaks for multicylinder engines, a

Bourdon manometer can be put into the spark plug hole of each cylinder. As an alternative, in Hanson [28], [29] it is described a procedure using an oscilloscope and a low-ohm shunt resistor to capture the cranking current in steady-state, disabling ignition. There are current consumption peaks during compression stroke of each cylinder. The cylinder with compression leaks will show a lesser current peak.

- Output torque variation. On alternative ICE, there are inherent periodic fluctuations in

output torque, given by the periodic opening and closing of the intake and exhaust valves and the reciprocating motion of the pistons, added to the variability of the combustion process itself (combustion phase and traces of the previous cycle and the next cycle of the engine are constantly changing, resulting in different peak combustion pressures in each cycle). The irregularities under steady operating conditions are the aggregate of the contributions of each cylinder of the ICE, the so-called torque signature. If on any particular a misfire occurs, the torque signature signal gets altered, having less mean torque and a reduction of engine rotational speed. That diagnosis method is described in the works of Mahieu et al. [30], Förster et al. [31], Wu and Lee [24]. Mahieu et al. [30] comments the two components of torque waveform: combustion of each cylinder and dynamics of the mechanisms of the ICE. The dynamics component tends to mask the reduction of mean torque given by a misfire event for high regime or low load, whereas in conditions of low speed and high load, the effect of absence of ignition in torque and speed is more evident.

- Variation of crankshaft rotational frequency. Förster et al. [31] indicate that, for low regime

and high load, a misfire event can cause a reduction of mean engine speed down to 8% whereas for high speed and low load, mean engine speed decreases 0,2%. In the work of Wu and Lee [24], ANN are “trained” to interpret the crankshaft rotational speed, and with it deduce and indicate a misfire event. Wu and Lee refer to a detection capability of 98,5% of misfires for 100000 cycles.

- Exhaust gas pressure measurement. The work of Willimowski et al. [32] indicates that the

exhaust gas pressure signal can be used to detect misfire. The frequency spectrum of gas pressure signal shows normal components with certain amplitudes for harmonics 6, 12 and

18. If a misfire occurs, the frequency spectrum shows additional harmonics 1, 2, 3, 4, 5, due

to the missing pulsation of the failing cylinder. A similar outcome is shown in Nybäck [33], testing for a diesel engine in normal operation and misfire, including the effect of the Jacobs brake actuated and the waste-gate valve.

- Vibration firm. A rapid pressure change in a cylinder during combustion traduces in engine

structural vibrations. In Chang et al. [20], accelerometers coupled to engine block and wavelets are used to determine misfire and knock [34].

- Engine sound. The work of Deptula et al. [27] addresses the acoustic signals obtained from

outside of the engine enable diagnosing faults on turbocharger, camshafts, and injectors. The work of Cavina et al. [35] illustrates diagnostics possibilities of knock, misfire, abnormalities in rotational speed of turbocharger axle, leaks on intake manifold using acoustic records on intake, exhaust and other locations of engine.

On internal measurements, the most relevant variable is the combustion chamber pressure. It is the closest indicator of torque production, a reference signal that helps to obtain features like mass fraction burned, heat release rate, thermal efficiency, or to diagnose knock, and misfire, among other features.

Given the difficult installation, long-term reliability issues, and cost limitations of typical piezoelectric pressure sensors, the detection of voltages and currents of ignition system has attracted the attention of many researchers, as shown in Martychenko et al. [36]. In particular, the spark plug is responsible for the ignition of fuel, but for a short interval, it has the potential to become a combustion sensor for the rest of the cycle. On the other hand, with dedicated electrodes or with the spark plug terminals, the ion current that appears at the onset and development stages of the combustion process allows detecting certain features of pressure signal or combustion conditions [19], which is to be treated on the upcoming lines.

2.4 Engine ignition systems using spark

In spark ignition engines, a brief high-voltage pulse is applied on the terminals of a spark plug to promote the ignition of an air-fuel mixture, with the aim of releasing the energy of fuel in the shortest lapse possible. The required magnitude of such high tension, known as breakdown voltage, is in the vicinity of 10 kV, and is mainly a function of the distance between electrodes, pressure and temperature of combustion chamber gases, properties of fuel, air-fuel ratio, and with less specific weight, the form and material of spark plug electrodes, their temperatures, humidity content in air-fuel mixture, ion concentration [36].

Figure 4 shows the effects of chamber pressure p, spark gap d, and temperature of spark plug electrodes T on breakdown voltage. Breakdown voltage is directly proportional to pressure, whereas electrodes’ temperature and breakdown voltage are inversely proportional between each other.

Figure 4. Effects of pressure, spark gap and electrode temperature on breakdown voltage, based on Pashley et al. [37]

Next lines will show the circuit schemes of ignition systems used to cause the required transient breakdown voltage, their features and principle of operation.

Classic ignition systems Ignition systems are composed of circuits known in the field of electronics as “flyback” and “step-up” [38]. Their function is to elevate the voltage of the power source by means of a transformer or autotransformer. Figure 5 shows two typical schemes, a) for a single-cylinder engine and b) for multicylinder engine, particularly with 4 cylinders in the figure.

Figure 5. Classic inductive ignition systems, based on Thurman. [39] The ignition coil M, is an elevator transformer whose primary winding (p+ and p- terminals) is energized by the power source vs (battery, alternator, magneto). A switch S0 enables the ignition system, a ballast resistor Rb limits the primary current, and the breaker contact S1 - that has a condenser in parallel Cs - is responsible to charge the primary winding (connect) and release it (disconnect). The breaker contact is actuated by a cam that lifts and drops the contact at the pace of once every axle’s full rotation. The terminal (s+) of the secondary winding of ignition coil is internally tied to the primary terminal (p+), making accessible three terminals of the ignition coil in an autotransformer configuration. SP1 corresponds to the spark plug, which is connected to (s-) terminal of secondary winding.

In Figure 5.b) the only differences are that the distributor Dis switches the secondary winding terminal (s-) to one spark plug at a time -for the example of the figure, SP1, SP3, SP4, SP2 driven by the rotation of the axle, and that the cam that actuates the contact breaker S1 lifts n times every full rotation of the axle, where n is the number of cylinders.

Considering contact S1 as ideal switch, there are two general conditions of operation: charging (connect) and releasing (disconnect). The models of the ignition system components in charging and releasing stages are presented and analyzed in [40], based on the works of Stevenson et al. [41], Wang et al. [42], Rohwein et al. [43], Khan et al. [44], Rozowicz et al. [45].

Figure 6. Schematic diagram of contact-breaker ignition system, from [40] Figure 6 shows the components of the ignition system with the internal parameters: Batt is the power source, an ideal equivalent independent voltage source Vs in series with its internal resistance r0; the switch S0 enables the ignition system; the ballast resistor Rh may be used to limit the primary winding current and with it avoid overheating.

The ignition coil has a primary winding with resistance R1, self-inductance L1 and equivalent capacitance of primary winding C1 due to its turns. For the secondary winding, there is resistance R2, self-inductance L2 and equivalent capacitance C2. Between windings, there is mutual inductance M from secondary to primary and viceversa and equivalent capacitance between windings C12. The secondary winding has a times more turns than primary winding, being usual a value of a from 100 to 200 [39].

The high tension (HT) wire has its resistance R3 and capacitance respect to negative of power source C3; the spark plug has its capacitances C4 and C5, and resistance R4. The impedance of the gap between the electrodes Zg is dependent on the contents of air-fuel mixture and its properties.

Figure 7. Schematic diagrams of contact-breaker ignition system in charging phase: a) complete, b) simplified, from [40] Figure 7.a) illustrates the charging phase with all transient power sources (initial conditions) due to each capacitor and winding of the ignition system: e1 for L1, e2 for L2, the induced from primary to secondary voltage e12 due to M, the induced from secondary to primary voltage e21 due to M, ec1 for C1, ec2 for C2, ec12 for C12; for the high tension wire e3 of C3, for the spark plug e4 of C4 and e5 of C5, and es1 represents the initial condition of Cs.

Figure 7.b) simplifies the schematic putting convenient short-circuits or open-circuits for initial condition sources: es1 is zero since S1 puts Cs in short-circuit. Assuming that windings were not previously energized, e12=e21= e1= e2 = 0; capacitances are discharged, and as C1, C2, C12, C3, C4, C5 are in the order of picofarads [42], [41], they are neglected for the analysis. From Figure 7.b), a loop Vs–r0-Rh-R1-L1-S1 appears, having a resistor-inductor circuit with a differential equation:

𝑣௦= (𝑅ଵ+ 𝑅௛+ 𝑟଴) ∙𝑖ଵ+ 𝐿ଵ 𝑑 𝑑𝑡𝑖ଵ Eq. 1

Solving Eq. 1 for the primary current i1, with a step-function forced-excitation vs = Vs u(t): 𝑖ଵ= ௏ೞ (ோభାோ೓ା௥బ) ቆ1 −𝑒ି (ೃభశೃ೓శೝబ) ಽభ ௧ቇ𝑢(𝑡)

Eq. 2

The primary current will rise to its maximum within approximately: 𝑡௖௛= ସ௅భ ோభାோ೓ା௥బ [𝑠] Eq. 3 That value tch is related to the dwell time, needed to charge the primary winding.

Figure 8. Schematic diagrams of contact-breaker ignition system in release phase: a) complete, b) simplified, from [40] With the primary charged with a magnetic field density, the contact S1 opens (release phase), causing in the secondary winding a high-tension spike explained by Faraday’s law. Both coils have free-running oscillations due to the releasing of S1 and initial conditions from e1, e2, e12, ec1, ec2, ec12, e3, e4, e5, see Figure 8.a).

Figure 8.b) shows a simplified version of the system, ignoring certain contributions of ignition coil. Sources e1 at L1, e2 at L2, e21 and e12 appear from the change in magnetic field density. Initially es1 of capacitor Cs is zero, since S1 was a short-circuit when charging.

A primary loop Vs-r0-Rh-R1-L1-e1-e21-Cs-es1, arises with:

𝑣௦+ 𝑒ଶଵ= (𝑅ଵ+ 𝑅௛+ 𝑟଴) ∙𝑖ଵ+ 𝐿ଵ 𝑑 𝑑𝑡𝑖ଵ+ 𝑢஼௦ Eq. 4 𝑖ଵ= 𝐶௦ 𝑑 𝑑𝑡𝑢஼௦ Eq. 5

𝑒ଶଵ= 𝑀𝑑 𝑑𝑡𝑖ଶ Eq. 6 For secondary loop Vs-r0-Rh-R2-L2-e2-e12-R3-(C3║C4║C5║Zg), the expressions are: 𝑣௦−𝑒ଵଶ= (𝑅ଶ+ 𝑅௛+ 𝑟଴+ 𝑅ଷ) ∙𝑖ଶ−(𝑅௛+ 𝑟଴)𝑖ଵ+ 𝐿ଶ 𝑑 𝑑𝑡𝑖ଶ+ 𝑢௚ Eq. 7 𝑖ଶ= (𝐶ଷ+ 𝐶ସ+ 𝐶ହ) 𝑑 𝑑𝑡𝑢௚+ 1 𝑍௚ 𝑢௚ Eq. 8 𝑒ଵଶ= 𝑀𝑑 𝑑𝑡𝑖ଵ Eq. 9 Using the Laplace transform with Zg as open circuit [40], the behavior of the primary and secondary currents will be given by the expressions: 𝐼ଵ(𝑠) = 𝑁1(𝑠) 𝐷(𝑠) 𝑉௦(𝑠) + 𝑁2(𝑠) 𝐷(𝑠) ቆ𝐿ଶ𝑖ଶ(𝑡ଶି) + 𝑀𝑖ଵ(𝑡ଶି) −1 𝑠𝑢௚(𝑡ଶି)ቇ+ 𝑁3(𝑠) 𝐷(𝑠) (𝐿ଵ𝑖ଵ(𝑡ଶି) + 𝑀𝑖ଶ(𝑡ଶି)) Eq. 10 𝐼ଶ(𝑠) = 𝑁4(𝑠) 𝐷(𝑠) 𝑉௦(𝑠) + 𝑁5(𝑠) 𝐷(𝑠) ൫𝐿ଵ𝑖ଵ(𝑡ଶି) −𝑀𝑖ଶ(𝑡ଶି)൯+ 𝑁6(𝑠) 𝐷(𝑠) (𝐿ଶ𝑖ଶ(𝑡ଶି) + 𝑀𝑖ଵ(𝑡ଶି) −1 𝑠𝑢௚(𝑡ଶି)) Eq. 11

Where:

𝜔௡ଵ= ට ଵ ௅భ஼ೞ Eq. 12 is the uncoupled natural frequency of the primary winding alone. 𝜁ଵ= (ோభାோ೓ା௥బ) ଶ ට ஼ೞ ௅భ Eq. 13 is the damping factor of primary winding.

The effect of C1 can be neglected and not added to Eq. 12 and Eq. 13, since for contactbreaker ignition systems Cs is several orders of magnitude higher than the primary equivalent capacitance C1 [40].

𝜔௡ଶ= ඨ

1

𝐿ଶ((𝐶ଶ+ 𝐶ଷ+ 𝐶ସ+ 𝐶ହ) Eq. 14 is the uncoupled natural frequency of the secondary winding, adding the effect of C2. 𝜁ଶ= (𝑅ଶ+ 𝑅௛+ 𝑟଴+ 𝑅ଷ)

2

ඨ(𝐶ଶ+ 𝐶ଷ+ 𝐶ସ+ 𝐶ହ) 𝐿ଶ

Eq. 15 is the damping factor of secondary winding, adding the effect of C2. 𝑀= 𝑘௖ඥ𝐿ଵ𝐿ଶ Eq. 16 is the mutual inductance [46], [47] , [48], where kc is the coupling coefficient between coils, usually 0,95 as presented in Stevenson et al. [41].

The coefficients of transformed currents I1(s) and I2(s) of Eq. 10 and Eq. 11 respectively, stand for:

𝐷(𝑠) = (𝑀ଶ𝐶௦൫(𝐶ଷ+ 𝐶ସ+ 𝐶ହ)𝜔௡ଵ ଶ𝜔௡ଶ ଶ+ 1൯𝑠ସ+ 2(𝜁ଵ𝜔௡ଵ+ 𝜁ଶ𝜔௡ଶ)𝑠ଷ+ +((𝑀(𝑅௛+ 𝑟௢)𝐶௦+ 1) 𝜔௡ଵ ଶ+4𝜁ଵ𝜁ଶ𝜔௡ଵ𝜔௡ଶ+ 𝜔௡ଶ ଶ)𝑠ଶ+ +2(𝜁ଵ𝜔௡ଵ𝜔௡ଶ ଶ+ 𝜁ଶ𝜔௡ଵ ଶ𝜔௡ଶ)𝑠+ 𝜔௡ଵ ଶ𝜔௡ଶ ଶ Eq. 17 𝑁1(𝑠) = 𝐶௦𝜔௡ଵ ଶ𝑠((𝑀(𝐶ଷ+ 𝐶ସ+ 𝐶ହ)𝜔௡ଶ ଶ+ 1)𝑠ଶ+ 2𝜁ଶ𝜔௡ଶ𝑠+ 𝜔௡ଶ ଶ) Eq. 18 𝑁2(𝑠) = 𝑀𝐶௦(𝐶ଷ+ 𝐶ସ+ 𝐶ହ)𝜔௡ଵ ଶ𝜔௡ଶ ଶ𝑠ଷ Eq. 19 𝑁3(𝑠) = 𝐶௦𝜔௡ଵ ଶ𝑠(𝑠ଶ+ 2𝜁ଶ𝜔௡ଶ𝑠+ 𝜔௡ଶ ଶ) Eq. 20 𝑁4(𝑠) = (𝐶ଷ+ 𝐶ସ+ 𝐶ହ)𝜔௡ଶ ଶ𝑠൫(1 −𝐶௦𝑀𝜔௡ଵ ଶ)𝑠ଶ+ (2𝜁ଵ𝜔௡ଵ+ (𝑅௛+ 𝑟௢)𝐶௦𝜔௡ଵ ଶ)𝑠 + 𝜔௡ଵ ଶ൯ Eq. 21 𝑁5(𝑠) = (𝐶ଷ+ 𝐶ସ+ 𝐶ହ)𝜔௡ଶ ଶ𝐶௦𝜔௡ଵ ଶ(𝑅௛+ 𝑟௢−𝑀𝑠)𝑠ଶ Eq. 22 𝑁6(𝑠) = (𝐶ଷ+ 𝐶ସ+ 𝐶ହ)𝜔௡ଶ ଶ𝑠(𝑠ଶ+ 2𝜁ଵ𝜔௡ଵ𝑠+ 𝜔௡ଵ ଶ) Eq. 23 The transformed primary and secondary voltages are: 𝑈ଵ(𝑠) = (𝑅ଵ+ 𝑅௛+ 𝑟଴) ∙𝐼ଵ(𝑠) + 𝐿ଵ𝑠𝐼ଵ(𝑠) −𝐿ଵ𝑖ଵ(𝑡ଶି) −𝑀 𝑠 𝐼ଶ(𝑠) + 𝑀 𝑖ଶ(𝑡ଶି) Eq. 24

𝑈ଶ(𝑠) = (𝑅ଶ+ 𝑅௛+ 𝑟଴+ 𝑅ଷ) ∙𝐼ଶ(𝑠) + 𝐿ଶ 𝑠 𝐼ଶ(𝑠) −(𝑅௛+ 𝑟଴)𝐼ଵ(𝑠) + 𝑀 𝑠 𝐼ଵ(𝑠) − 𝑀 𝑖ଵ(𝑡ଶି) −𝐿ଶ𝑖ଶ(𝑡ଶି) Eq. 25

According to Eq. 11, the secondary winding current depends on the initial condition term of primary current 𝐿ଵ𝑖ଵ(𝑡ଶି) and mutual coupling 𝑀𝑖ଵ(𝑡ଶି),.generating the transient e.m.f. between secondary winding terminals (s+) and (s-) after the fall of i1. This causes the highvoltage pulse that appears in the gap Zg to initiate the electrical spark.

The term ௗ ௗ௧𝑖ଵ becomes a large number when primary current falls from its maximum down to zero, requiring a certain dwell time to promote enough gap voltage Ug for the spark.

During the charging phase, the ignition coil is energized, meaning forced excitation. However, in the discharge phase, the windings are energized themselves by the initial condition sources and oscillate in a free-run manner.

The roots of the denominator polynomial of Eq. 17 (poles) manifest in the currents and voltages at primary and secondary windings (Eq. 10, Eq. 11, Eq. 24, Eq. 25) establishing oscillations, known as ringing, at a frequency that is a combination of the uncoupled primary natural frequency ωn1, and the uncoupled secondary natural frequency ωn2, due to the magnetic coupling M between windings [40].

The transient terms L1i1(t2-), L2i2(t2-), M i1(t2-), M i2(t2-) can be expressed as Dirac impulses, thus the primary and secondary currents will fall to zero until a new charging phase, and the voltage U1 will fall to zero as well asU2.

As explained in [40], opening-closing actions of the mechanical contact-breaker are far from ideal step actions. Indeed, there are initial imperfect disconnect phases, followed by short forced-excitation and then steady excitation during the charging, due to the mechanical bouncing of the points of the contact breaker. For the same reason, during the releasing, there are also imperfect short forced-excitations followed by short free-run until steady open contact. This in turn affects the spark generation, makes the contact-breaker ignition prone to generate less voltage at high engine speed, and requires the capacitor Cs to guarantee a good spark.

The previous analysis does not take into account the variance of the spark gap impedance Zg, which during combustion exerts an important load on the ignition system. Wang et al. [42] considers during the releasing process, the phases of pre-breakdown, breakdown, arc, and glow of the flame kernel formation, having four sets of initial conditions for the spark gap voltage and impedance and the other transient power sources mentioned. However, their sets of equations for charging and releasing correspond to a more modern transformer-type ignition coil.

The instantaneous polarity of the spark voltage generated is such that the electrical connections and mutual coupling of ignition coil force terminals (p+) y (p-) to be positive and negative respectively, and with that the accessible secondary terminal (s-) will be biased negatively respect to the negative terminal of power source Vs.

Thus, during spark strike, the central electrode of the spark plug will be negative respect to chassis ground. This is justified in Thurman [39] explaining that the central electrode of spark plug is hotter than the ground one. Electrons are negative charges lighter than positive ions [49] and can be excited and forced to conduction band by means of electric field as well as thermal energy, this last known as thermoionic emission. Thus, the gap can be ionized with 20% to 40% less electrical voltage than using the opposite polarity.

Transistor-controlled ignition systems The maintenance requirements, limitations of contact-breaker ignition systems, and technological advances in electronics enabled the introduction of the semiconductor transistor to perform the switching of the contact S1. Initially, the contact breaker was used to provide or cut the base current to the transistor to perform the switching. Then, other features to control circuit of transistor were added and the contact-breaker was no longer in vogue. Instead, inductive or infrared sensors combined with phonic wheels or slotted discs provide the trigger signal to connect and disconnect the transistor.

Figure 9. Schematic diagrams of transistor-controlled ignition a) single, b) double, with wasted spark, based on Sforza et al. [50]and Pischinger et al. [51] Figure 9.a) shows the substitution of the contact-breaker for a transistor Q1 to provide spark to an individual cylinder. The sensor X1 detects one specific position of the driving axle and sends a signal to the transistor driver circuit CT commanding to open or close the contact. Figure 9.b) shows a variation known as wasted spark, in which the ignition coil is a true transformer [52], [53] and both terminals of secondary winding are used to send a spark strike to two cylinders simultaneously. The engine timing ensures that only one cylinder is in compression stroke ready to ignition at a time, and the other one in exhaust stroke. Some designs of wasted-spark ignition system, have an internal connection between primary and secondary through a central tap in secondary connected to (p-) of primary [52], [53].

The transistor-controlled ignition can be applied to engines with distributor, distributorless wasted-spark having one coil per 2 cylinders, or even have one ignition coil per cylinder. The high-tension wires can be used to send ignition pulses to spark plugs. Another possibility

is to use a compact integrated control-transistor-coil unit known as coil-on-plug, in which the distance between coil and spark plug is much reduced.

Figure 10.a) illustrates the components of an integrated coil-on-plug assembly. The circuit CD is used to detect secondary current, for misfire diagnosis. Figure 10.b) shows the scheme used for multiple cylinders. The ECU controls the spark strike of every cylinder independently, promoting individual ignition timing and multiple sparks sending, according to engine parameters and conditions.

Figure 10. Schematic diagrams for coil-on-plug ignition system a) single cylinder, b) multicylinder integrated with ECU, based on Shimasaki et al. [54] As explained in Yamada et al. [17], the coil-on-plug scheme has advantages like low energy loss, wide range for spark advance angle. In addition, it is possible to have multiple spark strikes to improve the flammability of the air-fuel mixture. The ignition timing and number of pulses can be varied according to engine speed and load.

Poggiani et al. [55] used a Multiple Spark Discharge module set to fire 1, 4 and 9 strikes and a fixed volume chamber with high-speed optical detection. The multiple pulses increase the brightness of the spark, its energy, and tend to maintain the average position of the spark channel. In Forte et al. [56], it is shown a setup for a Ducati engine with two spark plugs. With the aim of reducing the cyclic variation, the generation of two flame fronts contributed to improve combustion stability in part and full load conditions. However, at full load, the tendency of knock increased. In Jung et al. [57], up to ten coils were used feeding pulses to one spark plug, showing that in very lean conditions, around  = 1,94, a thermal efficiency of 47% can be achieved. However, in conditions of  = 1,67, the higher spark discharge energy leads to an early initial combustion phase, promoting knock.

Capacitor Discharge Ignition (CDI)

Figure 11. Schematic diagram for CDI ignition system. Based on Shimasaki et al. [58]and Thurman [39] The CDI scheme uses a capacitor CC as power source for the primary of ignition coil, as shown in Figure 11. This capacitor is charged at around vc = 400 V [58] by means of a boost DC-DC converter, or similar device, that is fed with the low voltage of Vs. Instead of using a transistor, a thyristor TH1 is triggered to send a pulse to primary winding from capacitor CC. Thyristors withstand better the voltage and current transients generated by the coil due to their specifications against typical transistors. The triggering of TH1 is performed by the CT circuit.

According to Goulart et al. [59], for CDI systems, the secondary winding voltages can be up to 45 kV with a spark discharge energy of 496 mJ and a duration of 200 μs. In comparison, a TCI setup can deliver a smaller energy between 30 mJ to 80 mJ. The CDI is preferred for high-speed regime engines, like the racing ones or two-stroke ones, since the CDI strikes can be discharged 10 times faster than TCI [58]. Yamada et al. [17] comment that CDI has as weaknesses of short spark timing and poor flammability.

Although, by definition, a high tension pulse is -from the electronics and communications disciplines perspective- a vile and mean offender, that is, a huge contributor of electromagnetic interference, a CDI system produces more interference in comparison with a TCI scheme.

Other ignition systems to mention Not only inductive (TCI) single or multi-spark, or capacitive (CDI) are fitted to ignite the airfuel mixture in a controlled manner. As the engine developments tend to lean mixtures, bordering misfire, other ignition systems have been tested. Dale et al. [60] reviewed the high energy ignition systems, mentioning additional DC bias of 3 kV to spark plug voltage in conjunction with the secondary winding pulse, laser ignition, corona discharge spark plugs, plasma jets, railplug electrode. In their work, high energy systems are compared in terms of deposited energy (fuel jets are the most energetic, followed by plasma jets and then lasers), duration (fuel jets take the lead, then multi-spark, whereas plasma jets and laser provide the shortest pulses), efficiency, possibility of flame enhancement (plasma jets favour early flame growth), burn rate (multi-spark and fuel jets have better burn rate), durability (traditional TCI

and CDI last longer and cost less), cost and security, as well as production status. Massive production systems are TCI and CDI, leaving the rest for experimental and research purposes.

Gao et al. [61] show in their work the design of a railplug electrode applied to an engine with large bore, using natural gas in very lean mixture conditions, low crankshaft rotational speed, high load. The discharge process was studied using high-speed photography, a railplug electronic circuit with a conventional inductive ignition system in conjunction with a followon module to provide high current and accelerate the plasma down the rails. Railplugs increase the burning rate and stability limit for lean mixtures (=1.88) in comparison to a conventional spark plug.

Lang et al. [62] show the potential of using guided jets of fuel with direct piezoelectric injectors set at 140 bar in conjunction with turbocharging, rendering reductions in specific fuel consumption of 8%, operation with alternative fuels (E85), better cold-start and warmup performances. Compilation works of Günther et al. [52], [53] and Yu et al. [63] gather information about high-energy TCI ignition systems, with instantaneous discharge power of

4 MW (for comparison, usual TCI systems release around 1 MW) and discharge-stage

energies of 10 mJ, compared to 2 mJ for traditional TCI. For total energy, TCI systems can gather 200 mJ, while high-energy CDI, multicharge and double-coil can deposit up to 5 J.

Balmelli et al. [64], carried out an experimental study of breakdown voltage as function of spark gap, pulse duration, pulse waveform and rise rate of pulse for a nanosecond pulse discharge ignition. They concluded that high breakdown voltages create a more favorable ignition area, promoting fast and reproducible transition to self-sustained chemical reactions of combustion. The ideal pulse duration should be restricted to 20 ns to avoid transition to arc, which erodes the electrodes.

Yu et al. [63] found that transient plasma systems provide instantaneous power up to 6,85 MW and hundreds of millijoules for ignition energy. Radiofrequency ignition systems, having similar outcomes to conventional TCI, have promising advantages such as high ignition volume, continuous energy delivery and even combustion diagnostics. RF Corona ignition systems deliver up to 4 J for up to 4 ms, although requiring a fine-tuning of the resonant circuit. Microwave plasma ignition, which use quarter-wavelength antennas, have similar energy outcomes to TCI, but with targets to achieve lean-burn limit beyond  = 2. Laser induced plasma ignition deliver instantaneous power up to 10 MW, energy up to 10 mJ, with 10 ns duration, having very fast energy deposition, desirable to ignite very lean mixtures.

In the next part, a brief summary of the stages of formation of the flame kernel will be presented, as well as their associated values of voltage across the spark gap and currents through it.

Flame kernel formation From the onset of ICE, requirements for better performance, as well as reducing harmful exhaust emissions, led to electronic closed-loop fuel injection and spark timing control, fault detection and diagnosis. However, other research interests pointed to the chemical and physical phenomena and, with it, to the ICE’s mechanical and electrical components involved. Thus, models of the subsystems of the ICE for improvement of performance, as well as the quest for understanding the phenomena involved, were developed. A combustion model is a description, based on physics, of the combustion process inside the combustion chamber of an ICE, used to predict the mass burned and heat release rates of the mixture, among other parameters, as functions of the engine design and operation variables [65]. Herweg and Maly [66] described the importance of the turbulence intensity, load, chemical effects of air-fuel ratio and fuel type, pressure, and temperature at the beginning of spark discharge. Referring to the work of the latter authors, Song and Sunwoo [67] described their approach to the turbulent flame speed that includes strain, turbulence intensity and scales, along with residual gases, assuming constant turbulence intensity and integral length scale. The model of Pischinger and Heywood [51] includes the electrical characteristics of spark discharge, heat transfer effect, spark plug geometry among other factors, though they assumed laminar the flame speed and used a simplified heat transfer coefficient. Recent developments of optical and other experimental techniques, coupled to the advances in numerical resources and computational tools have advanced the comprehensive understanding of combustion processes in combustion engines, refining the predictive and diagnostic combustion models, including those for the formation, development, and propagation of flame kernel. Though used today, the considerations of [66] have been revised and also updated by researchers like Arcoumanis and Kamimoto [68]. As the time passes, new complexities and new challenges of combustion modeling, including those related to the flame kernel appear. For instance, Zadeh et al. [69] comment that, although numerous investigations studied the effect of spark plug geometry and size on spark features, there are limited data for strong crossed flows. Identification of the evolution of the flame kernel and flame front is essential to characterize the ignition and overall combustion processes.

Not being addressed the present thesis to the combustion process; some comments and considerations related to it are given in the following with the aim of having a light context associated with the ignition process, flame kernel, and ionization current (electrical conductance of the combustion products during the chemical reactions).

Though a comprehensive view of all the previously mentioned works and the ones of other authors will be impractical, some bases can be grasped from [65], [66], [67], [68], [70] for the modeling of flame kernel development. In brief, modeling considerations can be stated the following [66], [67], and [70]:

-Flame front is an open thermodynamic system in which the flame front is a thin reaction leaf, which means that the initial growth of flame kernel is laminar, at least for low and middle regime.

-Spark electrical energy released and heat losses toward spark plug lead to an early nonadiabatic flame kernel development.

-Mixture is burned immediately after entering flame region. Burned gas inside the flame is in chemical equilibrium, that is, once burned, it doesn’t react anymore. -Uniform pressure for burned and unburned zones is assumed.

-General law of gases is assumed applicable.

-The initial volume of flame kernel is much lower than the combustion chamber volume. -The increment of pressure inside combustion chamber during flame kernel formation is negligible.

-There is transport of the flame kernel away from the spark plug once the flame becomes turbulent. The reaction leaflet wrinkles and get distorted when the inflamed region grows up.

Conservation expressions to model the flame kernel development For the combustion process modeling, the used approaches are: -Zero-dimensional models: they are based on thermodynamics principles, without involving details of flow field and are used to analyze energy conversion aspects for the ICE. -Multidimensional models: they are based on solving Navier-Stokes balance equations for energy, mass and momentum, and turbulence. They relate directly processes within the cylinder with its flow details.

-Quasidimensional models: They include a spatial dependency on combustion and heat transfer processes explicitly and phenomenologically. relate the outcomes of the model to the geometry of combustion chamber, adding flow field parameters.

The expressions for the modeling of flame kernel initiation are: -Mass conservation (continuity) -Energy conservation (first law of thermodynamics) -Conservation of chemical species -Momentum conservation -Turbulent kinetic energy conservation -Conservation of the rate of dissipation of turbulent kinetic energy.

Figure 12 shows a general model of the thermodynamic system of the flame kernel that will be created within the gap of spark plug. The high tension pulse across electrodes forms a plasma channel that heats the nearby air-fuel mixture, leading to the development of a spherical volume known as flame kernel. That kernel expands and ignites the rest of the airfuel mixture on combustion chamber.

Shen et al. [70], Herweg and Maly [66], Song and Sunwoo [67] used the first two conservation equations to obtain expressions for the mass burning rate (Eq. 26), enthalpy variation of burned mass (Eq. 27), and volumetric expansion of flame kernel (Eq. 28); assuming spherical shape, they obtained the expansion radius of flame kernel (Eq. 29) and temperature variation of kernel (Eq. 30).

𝑑𝑚௞ 𝑑𝑡= 𝜌௨𝐴௞൫𝑆௧+ 𝑆௣௟௔௦൯+ 𝐶𝜌௨𝑠௜௡ Eq. 26

𝑑ℎ௞ 𝑑𝑡= 𝜌௨(ℎ௨−ℎ௞)𝐴௞ 𝜌௞𝑉௞ (𝑆௧+ 𝑆௣௟௔௦) +

1

𝜌௞𝑉௞ (𝑑𝐸௦௣ 𝑑𝑡−𝑑𝑄௛௟ 𝑑𝑡) −1 𝜌௞ 𝑑𝑝 𝑑𝑡 Eq. 27 𝑑𝑉௞ 𝑑𝑡= 𝜌௨ 𝜌௞ 𝐴௞൫𝑆௧+ 𝑆௣௟௔௦൯+ 𝑉௞( 1 𝑇௞ 𝑑𝑇௞ 𝑑𝑡−1 𝑝 𝑑𝑝 𝑑𝑡) Eq. 28 𝑑𝑟௞ 𝑑𝑡= 𝜌௨ 𝜌௞ ൫𝑆௧+ 𝑆௣௟௔௦൯+ 𝑉௞ 𝐴௞

( 1

𝑇௞ 𝑑𝑇௞ 𝑑𝑡−1 𝑝 𝑑𝑝 𝑑𝑡) Eq. 29 𝑑𝑇௞ 𝑑𝑡= (ℎ௨−ℎ௞) 𝑐௣𝜌௞𝑉௞ 𝜌௨𝐴௞൫𝑆௧+ 𝑆௣௟௔௦൯+

1

𝑐௣𝜌௞𝑉௞ ൬𝑑𝐸௦௣ 𝑑𝑡−𝑑𝑄௛௟ 𝑑𝑡൰+ −

1

𝑐௣𝜌௞ 𝑑𝑝 𝑑𝑡 Eq. 30 In the former expressions: mk is the mass of flame kernel, hk its enthalpy per unit mass, Vk its volume, Ak the surface area of kernel, rk the radial expansion of kernel, Tk its temperature, ρk its density, ρu the density of unburned mixture, hu the enthalpy per unit mass of unburned mixture, the value C = 0,8∙10-7 [67], sin the incoming speed of mixture to kernel, St the turbulent combustion speed, Splas the expansion rate of plasma channel, cp the specific heat, Esp the electrical energy delivered by the spark plug, Qhl the heat lost in spark plug electrodes, and p the chamber pressure.

Song and Sunwoo [67] include on Eq. 26 the term C ρu sin of the incoming mass, not taken into account by [66], [70].

Figure 12. Model of the thermodynamic system of flame kernel. Based on Herweg and Maly [66], Song and Sunwoo [67], Arcoumanis and Kamimoto [68], Shen et al. [70]

Herweg and Maly state that, within the plasma channel, the energy Eplas comes from a fraction of the electrical energy of the spark Esp which corresponds to the electrical breakdown energy Eb, electric arc energy Ea and its fraction a, the glow energy Eg and its fraction g, and the chemical energy of substances present in the plasma channel Ech, as it is condensed in Eq. 31.

𝐸௣௟௔௦= 𝜂𝐸௦௣+ 𝐸௖௛= 𝐸௕+ 𝜂௔𝐸௔+ 𝜂௚𝐸௚+ 𝐸௖௛ Eq. 31 The aforementioned energy transfer efficiencies will be treated with more detail shortly.

The work of Lucchini et al. [71] presents a lagrangian ignition model used to predict the initial stages of combustion; the arcing and restrike effects of plasma channel caused by the air-fuel mixture are modeled. The calculations for temperature and particle diameter are based on Song and Sunwoo [67]. The arcing and restrike of plasma channel is present also in the work of Zadeh et al. [69], except that the phenomenon is observed directly by means of optical techniques.

Flame kernel formation phases and electrical relations Figure 13 shows approximate log-log plots of the waveforms of the electrode voltage uSP and electrode current iSP as functions of time, based on Maly [72]. Those plots summarize the 4 phases of generation of flame kernel:

Pre-Breakdown: Ferguson and Kirkpatrick [73] state that in pre-breakdown phase, energy is added to gas molecules to ionize them, the secondary winding voltage rises to the point that an electrical current starts to flow through the spark gap. Herweg and Maly mention a rising slope of 10 kV/ms for transistor-controlled ignition and 100 kV/ms for capacitive discharge ignition. Pashley et al. [37] state that uSP at the end of pre-breakdown is in the order of 10kV; also in Arcoumanis and Kamimoto [68], the electric field magnitude is between 50 kV/cm and 100 kV/cm. This intense electric field accelerates the electrons, ionizes gas molecules, and at the same time produces UV radiation, which promotes an ionizing chain reaction that accelerates more and more electrons.

Breakdown: In this phase, the spark gap is not a dielectric, its impedance drops quickly and current iSP rises to values in the order of 100 A [73], approximately in 10 ns [66], [70], [73]. A low-impedance plasma channel is established, and considering it cylindrical in shape, as shown previously in Figure 12, its radius is around 40 μm, having dissociated particles at temperatures near 60000 K [66], [68], [70]. Herweg and Maly refer an energy transfer efficiency close to 80%. Also an instantaneous pressure rising of 300 bar at the flame kernel, causing an initial expansion velocity greater than Mach 1. At the end of this stage, a flame kernel, assumed spherical with a radius of 1 mm should appear [70].

Figure 13. Magnitudes of voltage and current of spark plug electrodes during combustion. Based on Pashley et al. [37] and Maly [72]. Phases: PB, pre-breakdown; BD, breakdown; T1, transition; AR, arc; T2, transition; and GL, glow.

Arc: the third phase can be distinguished by a low uSP in the order of 50 V to 100 V and current iSP goes down to 1 ampere [73]. This happens when the plasma “streamers” generated by the spark are present across electrodes and gap impedance is again reduced [74]. Flame kernel temperature lowers to ranges of 4000 K to 10000 K, being dominant the diffusive processes of heat conduction and radiation. The energy transfer efficiency drops to 50% due to flame kernel expansion. Additionally, expansion velocity is lower, having a greater duration for this phase, around 1 μs [70].

Glow: The fourth and last phase lasts longer, having an estimate time lapse around 1 ms [70]. Spark gap voltage rises up to 500 V, and current iSP goes down to 100 mA [73]. Kernel temperatures are estimated near 3000 K and energy transfer efficiency is less than 30% [66]. Around 20 ms to 100 ms [73] after ionization onset, if chemical energy released exceeds conduction heat transfer to unburned mixture, the combustion reactions become selfsustained and the flame propagates radially away of spark plug.

Table 1 summarizes the energy losses for each one of the phases of flame kernel formation.

Table 1. Energy balance for plasma discharge for flame kernel stages, under ideal conditions and small electrodes. Based on Maly [72].

Martychenko et al. [36] state that spark plug electrodes are by definition a condenser with a capacitance Csp and its electrical energy stored is calculated as follows: 𝐸௦௣= 𝐶௦௣ 𝑈௕ ଶ

2

Eq. 32 Being Ub the breakdown voltage. Having Csp = 5 pF (including wires) and Ub=7 kV, Esp is just 0,12 mJ. The requirement to strike with more than 30 mJ from the ignition system to the spark plug obeys to the aforementioned energy losses on flame kernel formation. On this subject, Rohwein [43] tested an increment of Csp up to 90 pF, improving transfer efficiency of the ignition system up to 50%.

Engine diagnostics based on secondary voltage measurements

The external measurements mentioned at the beginning of this chapter offer indirect inferences of combustion process based on the features of their signals. However, it is desirable to obtain, by means of internal measurements, a better approximation to the happenings of combustion chamber. Pressure is a very direct indicator, besides being a thermodynamic indicator, but costs and handling challenges encourage the measurement of other variables inside combustion chamber.

In particular, the spark plug sends ignition strikes for a small time interval, and can be used for the rest of the cycle to detect variations inside combustion chamber. Knowing that breakdown voltage varies with pressure and temperature of combustion chamber gases, a possible candidate variable for diagnosis is the spark plug voltage or secondary winding voltage.

To measure the secondary voltage, there are inductive pick-up sensors that take advantage of the current transformer principle and capture a fraction of the spark plug voltage during the

Breakdown Arc Glow Radiation losses <1% 5% <1% Heat conduction through electrodes 5% 45%

70%

Total losses 6% 50% 70% Total energy on plasma 94% 50% 30%

ignition pulse. This sensor has the disadvantage of capturing only the AC components of secondary voltage, with bandwidths between 30 Hz and 5 kHz, or up to 20 MHz for Rogowski coils [37], [40].

Another way to measure such secondary signal is to use voltage dividers. One case corresponds to capacitive voltage dividers, which are shown in Figure 14.a) and Figure 14.b) where the capacitors C1 and C2 act as impedances. For the scheme of Figure 14.a), the capacitors must be specified for dielectric strength above 20 kV or more.

The relationship between secondary voltage usp and input to acquisition system U2 is [75]: 𝑈ଶ= 𝐶ଵ 𝐶ଵ+ 𝐶ଶ 𝑢௦௣ Eq. 33 The scheme of Figure 14.b) is similar, leveraging the capacitor C1 formed with the ignition wire and its cover with a metal clamp that is part of the probe and forms the other capacitor plate.

Schemes a) and b) of Figure 14 have the same disadvantage of detecting only AC components of ignition voltage. To detect both DC and AC components of secondary ignition voltage, a resistive divider network can be applied, as shown in Figure 14.c). For resistive divider network, the relationship between secondary voltage usp and input to acquisition system U2 is [76]:

𝑈ଶ= 𝑅ଶ 𝑅ଵ+ 𝑅ଶ 𝑢௦௣ Eq. 34 The value of R1 must be considerable, between 100 M and 1000 M, and must be specified to have dielectric strength above 20 kV.

Figure 14. Divider network circuits to measure secondary ignition voltage. Based on Shimasaki et al. [54] As for the engine diagnosis possibilities of ignition voltage, the work of Shimasaki et al. [54] tackles the use of capacitive and resistive dividers to asses misfire detection from usp, finding that without combustion, usp is higher than for the case of good combustion. In addition, the duration of discharge is longer with combustion. The capacitive divider shown in Figure

14.b) is preferred by Shimasaki et al., due to the isolation difficulties of resistive dividers. Additionally, in Shimasaki et al. the analysis is complemented with ion current detection. That ion current analysis will be explained in further lines of the present chapter.

Yamada et al. [17] compares secondary voltage, chamber pressure and ion current for a classic TCI ignition system and a multi-spark ignition, with the purpose of detecting misfire from usp and ion current. For the case of regular TCI, minor usp waveform differences between combustion and misfire can be appreciated, but for multi-spark systems, the voltage detection cannot provide a significant criterion of differentiation. For their results, it is better to use ion current technique to detect misfire and also knock in MSI systems.

Martychenko et al. [36] made comparisons between combustion chamber pressure and secondary voltage as function of crank angle. The work used a resistive divider network and shows some similarities in the waveforms of breakdown voltage and pressure along a 4stroke cycle, detecting the opening of intake and exhaust valves. However, the bias voltage used to pressure reconstruction from voltage is high, around 25 kV, above the spark gap breakdown threshold.

Pashley et al. [37] used a resistive divider to find the aforementioned experimental correlations between pressure, temperature and breakdown voltage. On durability, Soldera et al. [77] describes the spark discharge process that erodes the spark plug electrodes; secondary voltages are measured with high tension probes connected to oscilloscope, images of the sparks and electrodes’ surfaces are presented. The work of Khan et al. [44] analyzes the secondary voltages waveforms, finding differences in them for rich mixture, lean mixture, spark plug in short-circuit, excessive gap distance. Rozowicz [78] presents the variations in secondary voltages and currents, as well as electrode erosion, when varying the fuel type. A mathematical model of the ignition system is presented, using space-state.

The work of Sebok et al. [79] includes, besides secondary voltage, primary winding waveforms, and a thermographic analysis of the ignition coil. For faulty spark plugs, the secondary voltage is lower. Zadeh et al. [69] used a resistive Tektronix P6015A high-voltage probe to capture the secondary waveforms and assess the effects of gas mixture at high speed crossflows, the ground electrode orientation on discharge energy, as well as duration and flame propagation speed.

2.5 Ion current detection

The magnitude and development of working fluid pressure throughout the combustion process is an essential factor in describing the combustion process in internal combustion engines. Piezoelectric and piezoresistive transducers are used to measure combustion pressure directly. However, the transducer's parameters must be adequate due to the harsh working conditions inside the combustion chamber and the transient nature of the pressure itself, which drives up the cost of those sensors.

The detection of ion current to characterize the combustion process has been an appealing technique since a long time ago. The work of Amman [80] shows that the detection of ionic currents was studied since 1920, particularly referring to the work of MacKenzie and Honaman [81]. That technique was used to measure the flame propagation speeds along different points of a combustion chamber. The document of Amman refers also a setup made by Schnauffer [82] in 1936, with 24 probes distributed on a cylinder, to visualize the form and speed of the flame front.

Related to ICE, the ion current technique becomes an alternative to pressure sensors. It relies on an electrical potential difference originated after the ignition of the fuel mixture, that attracts and repels electrons and ions to the correspondent electrodes, producing then an electrical intensity of current. That current depends, among other factors, on the air-fuel ratio, the composition of fuel, the temperature, and pressure inside the flame kernel and combustion chamber. There has been a quest to correlate the ion current signal given after the spark ignition -or after injection of fuel- and the pressure on the combustion chamber, for combustion analysis. In several works, as can be seen in Andersson [49], Saitzkoff, Reinmann et al. [83], Förster et al. [84], Shimasaki et al. [54], it is described a first rise in ion current known as chemi-ionization, just after the spark, where the molecules of the mixture split and become ionized, forming radicals (branching) that recombine several times in several stages, and then, given that most of the chemical reactions are exothermal, there is an elevation of temperature and pressure in the chamber, and with that another rise in ionic current, known as thermo-ionization peak.

In the next pages, an overview with a brief description of the circuits, ion generation processes and possibilities of diagnosis of combustion features will be presented.

2.5.1 Main schematic circuits for ion current detection and microammeters

Figure 15. a) Ideal circuit for ion current sensing, b) shunt microammeter, c) feedback microammeter Figure 15.a) shows the main components of an ideal ion current sensing system: the microammeter sensor S, the power source U and the electrodes SP1 present in the chamber. The term ideal comes from considering the power source U as capable of providing U volts, irrespective of the current consumption, that is, zero ohms of output impedance and perfect voltage regulation. For the microammeter, ideal means the meter S is a short-circuit between terminals and zero burden voltage, also infinitely immune to noise when measuring.

In reality, to measure low currents, in the order of microamperes, there are two methods: shunt resistor and feedback ammeter. Figure 15.b) shows the microammeter subsystem based on a shunt resistor Rion and an ideal voltmeter V (ideal means infinitely high input impedance, so all of the current i goes through Rion). The current i is converted to a burden voltage Uion based on ohms’ law:

𝑈௜௢௡= 𝑅௜௢௡ 𝑖 Eq. 35 In Figure 15.c), an alternative is presented, called feedback microammeter, composed of an operational amplifier G, and a feedback resistor Rf [85]. As the operational amplifier G operates in closed loop, the voltages of the non-inverting terminal <e+> and inverting terminal <e-> become very close as <e-> ≈ <e+>, and voltage Uion can be obtained by using:

𝑈௜௢௡= −𝑅௙ 𝑖 Eq. 36 Advantages of feedback ammeters are a very low voltage burden and low input impedance [85]. However, no matter the microammeter sensor, a resistor is implied in the measurement.

Although the reviewed literature presents several circuits to gather the ionic currents, there are commonalities in them, thus basic circuits used to detect ion current can be classified as: Independent probe, low-side of ignition coil and high-side of ignition coil.

Historically, the independent probe circuit was the first development. Then, as early ignition systems used distributor with the accessible secondary terminal going to spark plug (highside), the high-side scheme appeared second for ion current detection and diagnosis. Once the ignition systems adopted an individual coil per cylinder, having available the other secondary terminal (low side), the third scheme appeared. In the following, those three circuit configurations will be presented, along with their schematics, features, variations, as well as some contributions to ion current detection and diagnosis offered by the authors of those setups.

2.5.2 The independent probe circuit.

Figure 16. Independent probe circuits: a) shunt resistor, b) shunt resistor plus amplifier and c) with voltage divider (after Gazis et al. [21]) Figure 16 shows the components of an independent probe circuit for ion current detection: the power source U, the meter S and the electrodes SP1 forming a single electrical loop.

In Figure 16.a) the voltage Uion manifested on the resistor Rion due to the ion current, goes directly to the acquisition system Acq, in general, an analog-to-digital converter that gives as result a digital word proportional to Uion. For that circuit, it is assumed that the input impedance of the acquisition system Acq is high enough and thus would minimally affect the measurement.

Figure 16.b) shows the case when the input to the analog-to-digital converter is the signal from resistor Rion, but rescaled with the amplifier G. For that case, expression (2.3) applies: 𝑈௜௢௡= 𝑅௜௢௡ 𝐴ீ 𝑖 Eq. 37 Where AG is the amplification factor of the amplifier G.

For Figure 16.b circuit, the sensitivity is AG times that of the Figure 16.a circuit.

Figure 16.c) shows a different approach, using the combination of the resistor Rion and a voltage divider R2-R3 to get a limited voltage input to the analog-to-digital converter. Capacitor C1 is connected in parallel with R3 to act as a lowpass filter. That approach can be seen in the work of Gazis et al. [21]. The expression to obtain the ion current is given by: 𝑈௜௢௡= 𝑅ଷ 𝑅ଷ+ 𝑅ଶ+ 𝑅௜௢௡ (𝑈−𝑅௜௢௡ 𝑖) Eq. 38 For SI engines the electrodes are not necessarily the same as the terminals of the spark plug, that is, the electrical loop of ignition system is independent of the ion current sensing loop. The independent circuit has the advantage of not requiring protection components that withstand the high-voltage pulses of the ignition, as explained in Eriksson [86]1.

The polarity of power source U is such that the internal electrode of spark plug is positively charged respect to the cylinder head. That positive biasing is seen very frequently for ion current studies, as shown in Yoshiyama et al. [87], Pfeffer et al. [88], Gazis et al. [21], Şahin [22]. The work of Abdel-Rehim et al. [89]1 refers the use of the opposite polarity and the effects on the signal when changing to negative bias. That detail of biasing will be discussed later.

Although it is advantageous to have independent circuits for ignition and ion sensing, there is the need to have a hole for each one in the combustion chamber, so this independent probe setup is mostly used for engine research, as it will be discussed below.

For the case of fixed-volume chambers, besides MacKenzie and Honaman [81] and Schnauffer [82], in Yoshiyama et al. [87], an independent probe circuit and an imaging system were used to characterize the flame front formation and relate it with the ion current signal. In Labuda et al. [90], a similar fixed volume chamber setup is used, albeit it features a dedicated spark plug for ignition combined with two spark plugs becoming ion sensors placed inline at different distances from the igniter electrode, with the possibility to have a third spark plug as ion sensor, or a pressure sensor, at the farthest position of the chamber. These instruments were used to correlate pressure and ion current during knocking phenomena.

For in-engine tests, the work of Arrigoni et al. [91], dated 1973, used a 4-cylinder engine with a twin ignition system, left one spark plug for ignition and the second hole was used for pressure measurement or ion probing. Fuels used were gasoline with tetraethyl lead, isooctane, benzene, methanol, and nitro-ethane. Instead of atmospheric air supply, they used oxygen with argon or helium. The power supply used was a 70V battery coupled to an unspecified value of Rion, and lowpass filters. The work shows plots for each fuel, varying rotational speeds, load torque, spark timing, air-fuel ratio, to assess the stages of combustion pressure development aside from compression, and correlate them with the turbulent flame velocity, turbulence intensity, flame speed, burning velocity, ion current signal and its derivative. The flame front characteristics were obtained from the theories of their time 1 NOTE: In these works the independent circuit is not used.

(Sokolik, Ricardo, DeSoete, Longwell, Summerfield) the chemical and thermal ionization peaks were observed, mentioning the main chemical ion formation process.

In Anderson's work [92], dated 1986, an ion probe in the cylinder head, near the border, is used as the main sensor. A special spark plug design enclosing one ignition electrode and an additional one as the second ion sensor is also used. He used the independent circuit and high-side circuits to detect ion current. The primary objective of the study was to evaluate how well the ion probe signal can estimate the timing of peak pressure while varying ignition timing and air-fuel ratio. The indicated work and the mass fraction burned at 90% (calculated using the Rassweiler-Withrow approach) were correlated with the ion current. The results indicated a strong linear relationship between the peak pressure angle and the ion current angle when using a stoichiometric air-fuel ratio. However, for lean mixtures, the ion current showed inconsistency.

Shimasaki et al. [54] opted for a special spark plug design with ignition electrode and an adjacent electrode for ion sensor. In fact, they used the independent circuit, low-side and high-side circuits to detect ion current. Secondary voltage detection was performed using the high-side circuit. The tests were also performed in distributor ignition as well as individual ignition coil per cylinder. The power source U was 300 V, using typical household AC supply and a full-wave rectifier. They investigated the relation between equivalence ratio and ion current, and the challenges of every ion current detection circuit.

Low-side and High-side circuit configurations for ion current detection will be discussed later.

In a two-cylinder 4-valve engine, Pfeffer et al. [88] allocated twelve ion sensing electrodes in each cylinder atop of the cylinder head, distributed in two concentric circles, having 4 electrodes in the inner circle and 8 in the outer one. At high-speed regimes of 17000 min-1, common in Formula One engines at that time, the work investigated the flame front propagation, flame speed, combustion stability and effect of low and high tumble motion. Pfeffer asserts the use of individual amplifiers for each ion sensing electrode and shows a further signal conditioner circuit with a C-R first-order high-pass filter tuned to 250 Hz, followed by a unity gain buffer and an adjustable signal inverter amplifier made with operational amplifiers, ending with another first-order C-R high-pass filter.

The setup in the single-cylinder 4-stroke engine of Gazis et al. [21] features a hole for pressure sensor, and holes for three electrodes atop of cylinder head, one electrode used for ignition and the other two for ion sensing. The electrical circuit used was shown previously in Figure 16.c), with the addition of two switches to select ion electrode 1 or ion electrode 2. The purpose of the work is to emulate and reconstruct the combustion chamber pressure signal from the ion current signal features, “training” an artificial neural network (ANN) to do so.

Şahin [22] used a single-cylinder engine with one spark plug for ignition and a second one for ion current sensing. Power supply used was U=75 V and an unspecified value of Rion, connected to a high-input impedance instrumentation amplifier. Similarly to Gazis et al. [21],

Şahin developed and “trained” an ANN to correlate ion current signal with equivalence ratio, for a set of engine speed, load, and spark advance cases, obtaining a very good agreement.

Presenting test results under both motoring and firing conditions, Gürbüz [18] applied a similar scheme in a single-cylinder OHV engine, and used one spark plug for ignition with the separated ion current sensor electrode compounded into a special well with a fast response C-type thermocouple. For U=64 V and R=150 k, Gürbüz obtained a comprehensive graph with the correlations between in-cylinder pressure and ion current, chamber temperature, mass fraction burned and stages of ionization, researching the effects of ignition timing, equivalence ratio, engine speed and throttle opening.

Works where the head gasket is used as place for ion current sensors are also found in the literature reviewed. The work of Witze [93] shows three geometrical configurations (symmetrical, clustered and symmetrical wide electrode) to allocate up to eight ion sensors, having 8 independent circuits with U = 300 V, Rion = 300 k, amplifiers and filters, connected to pulse generators (that is, 1-bit A/D converters based on a threshold voltage). Its purpose was to characterize the flame propagation, the shape of the burned volume, and the swirl motion, and relate it to cyclical variations. Another work of Witze et al. [94], with a similar setup, shows the measurement of swirl motion, its direction and speed of flame propagation, agreeing with laser Doppler accounting.

Hu et al. [95] used 4 ion probes 5 mm above the gasket for a pentroof-chamber spark-ignition engine. Comparisons were presented between ion currents, pressures, and heat release; laser doppler velocimetry was used to assess correlations between combustion duration and flame speeds for turbulent mixture motion in several conditions of throttle opening, load, and airfuel ratio.

Another work of Hu et al. [96] presents the design of an ionization probe with a coaxial configuration of electrodes and an amplification circuit with feedback ammeter G1 and analog galvanic isolation G2 for the ion signal, as shown in Figure 17.b). The isolation amplifier G2 allows to have an input that floats over the value of the bias power supply U, and an output close to the negative terminal of the supply. That is, a U volts difference between input and output. For comparison purposes, the common layout for independent circuit is shown in Figure 17.a) where the ammeter S is connected near the ground reference. The bandwidth for the scheme of Hu et al. was 15 kHz; records of pressure, and ion current were presented, having the same 4 ion probe design of Hu et al. [95], and showing the ion current magnitude for several axial distances from the electrodes and bias voltages.

The work of Nicholson and Witze [97] uses the same 8 ion sensors located into the head gasket with two different widths to account for flame location, travel and geometry. Meyer et al. [98] added to the Witze setup an optical spark plug, to obtain correlations between flame arrival times, pressure and ion currents, speed and load, swirl and tumble. In Meyer et al., no correlations were found between early flame kernel shape and correspondent flame shape for low or high tumble swirl. Flame arrival times correlate acceptably with 0% to 90 % heat release duration and ion current.

Figure 17. Independent probe circuits. a) classsical, with ammeter connected to negative and b) with feedback ammeter and isolation amplifier (after Hu et al. [96])

Russ et al. [99] uses the symmetrical configuration (electrodes placed separated by 45°) shown in Witze [93] for a head gasket, with 8 ion sensors as well, but with a special design of head gasket that eventually failed due to high temperatures and cooling issues. The head gasket design was made to reduce the crosstalk between ion sensors, specially, the adjacent ones, which could give false readings. The purpose of the work was to account for 2%, 10% and 50% of mass fraction burned in terms of flame radius, cycle-to-cycle variation in conditions of part throttle and 100% opening.

Park et al. [100] use the head-gasket symmetrical configuration with the 8-ion-sensor configuration of Witze [93], added to simultaneous pressure measurement and high-speed photography to visualize the flame propagation process. They followed the ion current to find possible points of onset of knock, in the pentroof cylinder-head with 3 and 4 valves per cylinder.

Yoshiyama et al. [101] uses a head-gasket ion sensor with one ring as electrode in a 2 valveper-cylinder engine and two spark plugs for ignition, one centered and the other offset. The authors present a relation of equivalence ratio and ion current, as well as the effect of the position of the spark plug in ion current signal, maximum combustion pressure, and relation of pressure peak with ion current.

The independent circuit has been applied also for ion current detection in compression ignition engines.

An example is the pioneer work of Glavmo et al. [102], where the ion sensor is mounted into the glow plug hole of a diesel engine. The ion current signals obtained can detect the start of combustion for both pilot and main fuel injection, misfire, and relation between EGR signal and ion current. A problem mentioned in this work is the additional current that appears in the ion current detection circuit due to soot formation, since the soot has a noticeable conductivity. The glow plug, aside from its regular warming use, is used as cleaning element and ion sensor.

In Henein et al. [103], the independent circuit is applied for a single-cylinder gasoline engine and two single-cylinder diesel engines, the last to operate in HCCI mode, to gather a comprehensive comparison of all the expected features of the ion current signal in each mode of combustion, particularly its peaks and chemical phenomena associated to them. One remark is the multiple peaks that appear on the ion current in a diesel engine, while spark ignition engines typically show the chemical and thermal peaks, giving a glance of the complexity of combustion process in compression ignition.

Estefanous [104] uses a setup, where the glow plug is also ion sensor. As Estefanous described, the tip of the ion sensor must be outside the path of the jets from fuel injector to avoid false readings. The values of power supply U tested were 25 V, 50 V, 75 V, 100 V, with a Rion = 100  and a 5B40 amplifier with Ag = 50. The work of Estefanous contributes a correlation between load, injection pressure, NOx species formation and ion current, along with cycle-to-cycle variation, effects of equivalence ratio on chemical and thermal ionization, and prediction of soot formation.

The circuit described in the works of Henein [103] and Estefanous [104] is also used in the studies conducted by George [105] and Assaad [106]. In these adaptations, a switch has been added for the glow plug and ion sensor selection. Assaad utilized a 7.2-liter, 6-cylinder engine with HEUI injectors, while George employed a 2.0-liter, 4-cylinder diesel engine equipped with piezo injectors, using ULSD and jet fuels. Their aim was to analyze the ion current signal under a set of load/speed conditions, and with various fuel types, injector pressures, and injection profiles.

The independent circuit is also used for ion current detection in research concerning the HCCI mode of combustion. Apart from Henein et al. [103], the works of Mehresh et al. [107], Phan et al. [108], Panousakis et al. [23], Dong et al. [109], mention its use.

In the study conducted by Mehresh et al. [107], a 4-cylinder, 1.9 l VW diesel engine was modified to operate on propane. The circuit values used were U = 27 V and Rion = 500 k, with an additional 1 M resistor connected to the input of the acquisition system for equipment protection. The research focuses on the chemical kinetics involved in ion generation during HCCI operation, through a single-zone numerical model, comparing the results with experimental measurements of ion currents and pressures. The work shows very good agreement between the rise of positive-slope pressure waveforms and the corresponding increase in positive-slope ion currents that occur at the onset of combustion. There is also satisfactory agreement in detecting the cumulative 50% mass fraction burned, as derived from both pressure and ion signals. In this study, the electrodes used for ion sensing were modeled as a capacitor, which, in series with Rion, creates an inherent R-C lowpass filter. Rion was replaced by a 100 kΩ resistor, to demonstrate the ion signal delay caused by this R-C network. This adjustment led to improved agreement of the ion signal with the pressure signal using the lower value of Rion.

The research of Panousakis et al. [23] focuses on isolating one cylinder of a 4-cylinder sparkignition GM engine to operate in HCCI mode. The circuit values used were U = 200 V and Rion = 200 k. This work utilizes a voltage divider network approach, like the one previously illustrated in Figure 14.c, also referenced in Gazis et al. [21]. Artificial Neural Networks and

wavelet techniques were compared to correlate the ion current signal with chamber pressure, revealing neural networks better in prediction of peak pressure crank angle, with a delay between 1,5 and 3 degrees of crank angle.

The engine utilized in the study conducted by Phan et al. [108] was a 4-cylinder, 1.9 l diesel VW adapted to operate on gasoline. The circuit parameters used were U = 237 V and Rion = 241 k. The goal of the work was to compare three signal conditioning circuits: a unity gain, 60 Hz notch filter, and a custom-built integrator, to determine which produced the strongest ion signal. The integrator circuit produced the strongest signal when the fuel-air equivalence ratio was 0,16 or higher. However, this custom integrator circuit introduced a 2-degree crank angle delay between the pressure signal and the ion signal response.

Dong et al. [109] used a 2-cylinder gasoline direct-injection variable valve timing engine with the circuit parameters U = 200 V and Rion unspecified. This work shows an extensive CFD modeling, including primary reference fuel oxidation and hydrocarbon flame ionization mechanisms for prediction of ion current signal and autoignition time delay. The research reveals a good correspondence between the peak of ion concentration and ion current amplitudes for different equivalence ratios.

2.5.3 The high-side circuit

Figure 18. High-side circuit for ion current detection. Based on Eriksson. [25] Figure 18 illustrates the basic components of a high-side circuit for ion current detection. The circuit includes the power source U, the meter S and the electrodes SP1, like those found in an independent probe. However, this configuration features a “T” joint to connect the secondary winding of the ignition coil (terminal s-), creating two electrical loops that share the electrodes SP1: one loop dedicated to the ignition system, while the other is linked to the power source and meter. A high-voltage diode D1 is meant to prevent the current pass from the ignition coil to the meter S when ignition is triggered, while allowing the current from U to pass when ignition ceases. The voltage Uion manifested on the resistor Rion due to the current, goes to the acquisition system Acq, rescaled with the amplifier G.

Things not explicit in the literature about ion current detection when using the high-side circuit are the combined contributions from the primary and secondary windings during spark generation, as well as the effects of the high-tension wire and spark plug parameters. These contributions are illustrated in Figure 19.

Figure 19. High-side circuit for ion current detection with power sources from ignition coil. Based on Eriksson. [25], Wang et al. [42] and [40].

The secondary winding includes its resistance R2 and self-inductance L2. The primary winding has its resistance R1 and self-inductance L1. M stands for the mutual inductance between coils. There is a high-tension wire HT with a resistance Rw between the ignition coil and the spark plug electrodes. The internal resistance Rp in spark plugs is aimed at reducing electromagnetic interference. There are also capacitances Cw for the high-tension wire and Cp1 and Cp2 for the spark plug. Those components are usually ignored in the literature concerning ion current detection. In Wang et al. [42] the impedance of the electrodes SP1 is considered as Zg, a value dependent on the conditions of the combustion chamber. To ensure that the ignition pulse reaches the spark plug with only one polarity, an optional impedance Zop, such as another high-voltage diode, can be utilized.

The e.m.f. e12 is the transient ignition voltage that is generated by the collapse (and also the charging) of the primary winding current of ignition coil, required to produce the highvoltage spike, responsible of igniting the air-fuel mixture inside the combustion chamber. The term e2 is the transient e.m.f. associated to the self-inductance L2.

An electrical loop e12-e2-L2-R2-Zop-Rw-Rp-Zg can be seen in Figure 19, where the current i comes from two sources: e12 and e2. The terms e12 and e2 are present during the ignition

process, causing a current i that raises in nanoseconds to tens of amperes, drop to milliamperes (Pashley et al. [37] and Maly [72]) and finishes nil.

Since e12 and e2 are temporary, once their voltages are lower than the power source tension U, diode D1 is forward-biased and the ion current appears in the loop U-D1-Rw-Rp-Zg-Rion, as the air-fuel mixture starts burning between electrodes SP1, making the expression (2.3) again valid for ion current detection after ignition pulse, that is, e12 and e2 are zero. A warning comes considering Eriksson [25], who comments that the HV diode has an inherent capacitance that is charged by the ignition strike, therefore the diode doesn’t get reversebiased immediately. Then, until that capacitance gets discharged, certain parts of ion current can be missed.

Taking into account the previous comment, it´s a good practice to use of an electronic limiter circuit Lim, in parallel with Rion (as shown in Figure 19) that becomes a less-resistance path for the current if the ignition contribution manages to get to Rion.

This circuit has the advantage of using the spark plug simultaneously as fuel igniter and ion sensor, but the disadvantage of masking the ion current during the ignition pulse and ringing of the coils, that is, the ignition current is up to 7 orders of magnitude higher than ion current, which is in the range of microamperes (Saitzkoff, Reinmann et al. [83], Förster et al. [84], Eriksson [25]) not reliably detectable when ignition is just triggered.

As the high-side terminal of ignition coil is available to connection to the spark plug, there is no need to have independent secondary winding terminals.

There are several works in which the high-side circuit is used for ion current detection and diagnosis:

The outcomes of the work of Anderson [92], dated 1986, were already explained in the independent circuit. Yet, he used also the high-side circuit; instead of forming a “T” with two wires and a diode, the circuit of Anderson used a spark gap Zop in series with the high-tension wire and a 20 M resistor instead of the diode D1 (see Figure 20). The power supply used was U = 300 V and Rion = 100 k coupled in parallel with a 3,6 V Zener in the Lim box, connected to an inverter amplifier G.

Figure 20. High-side circuit for ion current detection. Based on Anderson [92].

The results of the work of Shimasaki et al. [54], dated 1993, were already explained in the independent circuit subchapter. However, for the high-side circuit (see the scheme of Figure 18, where Zop = 0 and D1 is a HV diode) Shimasaki et al. commented about the issues associated with the failure of the HV diode D1, where the worst-case-scenario is the diode fault in short-circuit condition.

In the work of Daniels [110] the scheme shown in Anderson [92] is used (see Figure 20), having: U = 150 V, Rion = 100 k, and RD1 = 5 M. In this work, ion current and pressure signals are compared to obtain mass fraction burned characteristics. A Wiebe function with its first, second and third derivatives are declared, while from the inflection points of second derivative (maximum heat release point) and third derivative (maximum acceleration point), coefficients a and m of Wiebe function are stated. Using ion current first derivative and its inflection points (maximum heat release point, maximum acceleration point and end of combustion), a Wiebe function leading to the mass fraction burned is found and then compared against the Rassweiler-Withrow method that uses chamber pressure as input. The work displays very similar outcomes of mass fraction burned from pressure and ion current signals.

The scheme of Yoshiyama and Tomita [111] uses two diodes with the cathodes connected to the spark plug, as seen in Figure 21. Yoshiyama and Tomita disagree with the consideration that the ion current is only a local phenomenon near the spark plug. Their tests under partial load and idle conditions allowed them to conclude that for partial loads, the correlation between ion current and pressure features is weak, at the time that for the idle condition that correlation improves. Also, they consider that the ion current is dominated by the area touched by the flame, which includes the cylinder walls and even the exhaust port. This consideration is pertinent when the internal electrode of the spark plug is positively biased with respect to the cylinder walls.

Figure 21. High-side circuit for ion current detection. Based on Yoshiyama-Tomita. [111].

The work of Gao et al. [112] uses the scheme shown in Figure 18. The test engine was a 4cylinder diesel converted to spark ignition with stratified charge and direct injection, using methanol as fuel. The components of the ion current detection circuit were U = 400 V; Rion = 100 k in parallel with a 22 uF capacitance for prefiltering. D1 consisted of a silicon stack diode specified to 180 kV, which caused the absence of the ignition pulse seen usually in ion current signals; however, the chemical and thermal peaks were present. The work reports the ion current signals in conditions of normal operation and knock, under low and high engine loads, and different ignition timings. In conditions of abnormal combustion, the work reveals peaks of ion current for backfire, knock and post-combustion, while the pressure waveforms barely present high frequency oscillation after peak pressure point (knock). As the knocking level increases, so does the amplitude of oscillations in ion current (after thermo-ionization and peak pressure angle). There were two filtering approaches, lowpass below 7 kHz, and 7 kHz to 10 kHz bandpass, showing for both the possibility of knock detection.

The work of Laganá et al. [113] uses a scheme shown in Figure 22, in which the ignition system sends sparks for 2 cylinders (wasted spark). The tests were performed on a 2,0 liter, 4-cylinder VW engine. The components for the ion current detection were: Rion = 10 k in parallel with a 1N5347 zener limiting diode. D1 consisted of a group of 3 ESJA53-16A silicon diodes. For the high-tension wire, Rw = 5 k. The authors performed tests with new and used spark plugs, big and small electrodes, power supply U values of 50 V, 150 V and 300 V, and gasoline with ethanol addition of 0%, 15% and 25%.

Figure 22. High-side circuit for ion current detection. Based on Laganá et al. [113]. Laganá et al. used a digital bandpass Hanning window to filter and detect knock conditions. Additionally, the work compares the knock and ion current signals to find correlations between them, finding a best correlation using the Bohman window technique. The area under the curve of ion current signal is used as a method for detection of misfire, having more area in combustion than in misfire. Bigger spark plug electrodes provide more ion current magnitudes.

Another work of Gao et al. [114], uses the scheme shown in Figure 18. The tested machine was a 4-cylinder spark-ignition engine with port injection, using gasoline as fuel. Components of ion current detection were U = 12 V and Rion = 100 k, in parallel with a 15 uF capacitance for pre-filtering. D1 consisted of a silicon stack diode specified to 150 kV. The ion currents obtained for different combinations of rotational speed, load, spark advance, and equivalence ratio were used to train an artificial neural network, and to perform a deep-

learning model of autoencoder with semantic representation technique, destined to estimate

and predict pressure change and peak angle. The estimations of pressure were 7,84%, better than the work of Rivara et al. [115]. Nevertheless, while the prediction for peak pressure was good, the pressure change prediction was not as good.

Chen et al. [116] conducted a study on Homogeneous Charge Compression Ignition (HCCI) using a modified 4-cylinder, 1,9 l diesel engine. In their experiment, they focused on one cylinder and primarily used ethanol as the fuel, adding CsOAc to enhance the stability of ion current detection. Components used for ion current detection were U = 237 V and Rion = 241 k (same scheme as shown in Figure 18) in parallel with a non-specified filter. The diode D1 used in the study was of a non-specified silicon stack diode in parallel with a resistor. Although an ignition coil was part of the configuration, it was not utilized to ignite the fuel. The work examined the cycle-by-cycle variations by integrating a well-mixed reactor model with a heat loss model based on the Woschni´ equation. The predictions of cyclic variations derived from maximum ion current outperformed those based on maximum pressure, owing to the more deterministic structure of the former one.

On the other hand, and also for HCCI, in the work of Hunicz et al. [74], a scheme with two diodes is presented, however with different orientation of the diodes and power supply of Figure 21, as can be seen in Figure 23.

Figure 23. High-side circuit for ion current detection. Based on Hunicz et al. [74]. The work of Hunicz et al. used U = 200 V and Rion = 33 k. D1 and D2 are diodes specified to 30 kV. The ignition coil is off during the tests, the fuel used was gasoline with 95 RON. There were performed analyses of heat release rate and mean effective pressure, correlated with ion current traces. The ion current integral was calculated to estimate the energy released during combustion. Then, the ion current integral was compared with the cumulative heat

release, obtaining good correlation. Also, there was found a good correlation between ion current integral and calculated chamber temperature. The angle of peak ion current is well correlated with the angle of peak of heat release, except for lean mixture, where ion current was behind heat release.

Dittrich et al. [117] shows a test scheme like that of Hunicz et al. [74], as shown in Figure 23, changing D1 for three BY16 diodes in series. U = 300 V; Rion is not specified. The test was conducted on a three-cylinder spark-ignition engine. Although the schematic circuit of the work of Dittrich et al. precedes the one of Hunicz et al., the former presents just a basic theory of ion current generation, typical ion signals seen in literature and the test circuit is a preliminar development for future work.

2.5.4 The low-side circuit

Figure 24. Low-side circuit for ion current detection. Based on Shimasaki-Maki et al. [58] and Wang et al. [42] Figure 24 shows the main components of a low-side circuit for ion current detection: the power source U, the meter S, and the electrodes SP1, arranged similarly as in the independent probe and high-side circuits, but adding in series the secondary winding of the ignition coil (terminals s+ and s-), forming a single electrical loop. The secondary winding includes its resistance R2 and self-inductance L2. The primary winding has its resistance R1 and selfinductance L1. Between coils, there is the mutual inductance M.

Also, there is a high-tension wire HT with a resistance Rw between the ignition coil and the spark plug electrodes. The built-in spark plug internal resistance Rp is intended to reduce electromagnetic interference. There are also capacitances Cw for the high-tension wire and Cp1 and Cp2 for the spark plug. Those components are usually ignored in literature related to ion current detection when using low-side circuits. The voltage Uion manifested on the resistor Rion due to the current, goes to the acquisition system Acq, rescaled with the amplifier G.

Again, the addition of the contributions of the primary and secondary windings in condition of spark generation is not explicit in the literature related to ion current detection when using the low-side circuit.

Figure 25. Transient power sources of ignition in Low-side scheme for ion current detection. Based on [40] Figure 25 shows those contributions for the low-side case. The e.m.f. e12 is the transient ignition voltage that is generated by the collapse (and the charging) of the primary winding current of ignition coil, while the term e2 is the transient e.m.f. associated with the selfinductance L2. Ignoring the contributions of the capacitors Cw, Cp1 and Cp2, an electrical loop

U-e12-e2-L2-R2-Rw-Rp-Zg-Rion is seen in Figure 25, where the current i comes from three sources connected in series: U, e12 and e2.

The terms e12 and e2 are present during the ignition process, causing a current i that, as described before, raises in nanoseconds to tens of amperes, drops to milliamperes and finishes zero. Though, in the high-side configuration, it is good practice to include an electronic limiter, for the low-side circuit it is essential, as there is only one electrical loop. In Figure 25, the electronic limiter circuit Lim, which operates in parallel with Rion, becomes a lowerresistance path for the current during the ignition process.

Since e12 and e2 are temporary, it is the power source U that finally contributes to the ion current, when the air-fuel mixture burns between electrodes SP1, and the expression (2.3) remains valid for detecting the ion current after the ignition pulse, at which point, e12 and e2 become zero.

This circuit offers the advantage of using the spark plug both as fuel igniter and ion sensor, but the disadvantage of masking the ion current during the ignition pulse and the ringing of the coils. Indeed, the ignition current can be up to seven orders of magnitude higher than the ion current, which is in the range of microamperes (Saitzkoff, Reinmann et al. [83], Förster et al. [84] Eriksson [25]). Consequently, it is difficult to reliably detect the ion current right when ignition is triggered. Another important consideration is that the terminals of the secondary winding must be independent of those of the primary winding in the ignition coil, hence certain ignition systems in which the ignition coil is electrically configured as autotransformer, with a common terminal between primary and secondary (see Figure 5) would require a modification to their connections.

The patent for low-side circuit for a CDI ignition system was granted in 1989 to Gillbrand et al. [118]. This low-side configuration is commonly referenced in literature related to combustion engines. Notably, Auzins et al. [119] discuss its application in systems with individual ignition coils for each cylinder, as this technology allows for multiple ignition strikes and configurable spark durations.

The work of Shimasaki et al. [54], dated 1993, previously mentioned for the independent circuit, presents the use of the low-side circuit, exposing the challenges of the ion current detection circuit along with the relation between equivalence ratio and ion current. It is needed to consider that Shimasaki et al. mention an early reference of Terada and Suzuki [120], around 1978, also mentioned in the work of Dittrich et al. [117], although in Dittrich, the circuit used is a high-side one. Asano et al. [121] refer the correspondence between ion current energy and its frequency with the combustion chamber pressure, in knocking condition found in the work of Terada and Suzuki. Miyata et al. [122] also quote this latter work which points the ion density as a function of the distance to the flame front, with the ion density maximum near the bright flame region. In Miyata et al. [122], the low-side and high-side circuits are used, and the secondary voltage and ion current are registered to indirectly detect the flame ion density, by means of the decay time of spark plug voltage.

In the study conducted by Auzins et al. [119], nine different engines were tested for misfire detection and eight for knock detection using ion current. To confirm the results, a pressure sensor and a knock sensor were also employed. The misfire detection proved to be 100%

successful and was more effective than the speed variation method. Additionally, the detection of knock using ion current outperformed the knock sensor.

For misfire detection, two thresholds are used: one for the amplitude of the ion current and another for the crank angle duration of the ion current, measured from the point of chemiionization to its decay. These thresholds vary depending on the engine load and rotational speed. In the case of knock, a high-frequency component above 5 kHz can be observed in the ion current trace, which is confirmed by oscillations in combustion pressure from its peak. By applying a bandpass filter, this component can be assessed, enabling the identification of such combustion abnormality. Auzins et al. also discuss additional detection possibilities using the low-frequency components of the ion current. These include ignition timing feedback control, cam phase sensing, exhaust gas temperature monitoring, air-fuel ratio prediction, spark voltage control, misfire detection, and pre-ignition assessment.

The work of Lee and Pyko [123] shows an arrangement for a V6 engine featuring six coilon-plug ignition packs connected to a set of six independent power sources -each operating at 100 V- and six integrators designed to process ion current signals. These integrators are disabled during the ignition pulses. The power sources consist of capacitors that are charged by the ignition pulses themselves. The integration of the signals allows for independence from the location of ion current peaks, flow turbulence, and variations in the air-fuel ratio. This setup enables the detection of misfires and short circuits in the spark plugs by analyzing the integrated ion current. Synchronization is achieved through signals from the camshaft and crankshaft positions. The analog signals generated are distinctly identifiable between misfires and normal combustion, demonstrating robustness under conditions such as changes in exhaust gas recirculation (EGR) and crankshaft speed, and other hard conditions like idle and low load. The criterion for assessing a misfire was a minimum signal/noise ratio of 7.

Eriksson [25] mentions the use of the Mecel principle for his experiments on ion current measurement, pointing to the low-side scheme [119], [118]. Eriksson suggests the use of an analog lowpass filter set to 15 kHz for ion signal and develops three algorithms to correlate pressure peak and ion current traits. The first algorithm looks for the local maximum of ion current at the post-flame phase (thermo-ionization), which seems to correlate with peak pressure angle. The second algorithm calculates the centroid of the area under the curve of ion current signal for the same purpose. The area in question follows the ignition pulse and ends when ion current is nil. The third algorithm approximates the ion signal to a gaussian distribution. The first algorithm is not very effective at accurately identifying the peak pressure angle. The second algorithm performs well only when the post-flame stage is clear; however, it fails when the flame front is large, leading to an early and incorrect prediction of the pressure peak. The third algorithm is more promising, though computationally more demanding and challenging to implement.

In Nielsen and Eriksson [75], the previous setup was used, and the ion signal was reduced to gaussian distribution, essentially a superposition of gaussian functions, to use it as feedback input for spark ignition control purposes.

Lee and Pyko [124] were granted the patent for a low-side scheme in 1996. In the scheme, the ammeter is feedback type (see Figure 15.c) with the addition of antiparallel diodes to limit the current in spark strike. The circuit includes hardware integration of ion signal.

In his thesis, Eriksson [86] tackles the closed-loop ignition timing control using the ion current signal as input. Eriksson´s setup used the low-side scheme, where the ringing phase of the ignition coil -after spark- was windowed away, letting flame front and post-flame phases of ion signal pass, and used as input for a gaussian interpretation algorithm. The work shows the connections between pressure and ion signal features needed to detect maximum brake torque. In particular, with the parameterized model and ion current, its interpretation led to estimate peak pressure angle. Eriksson found that the cyclic variation is reduced with a proper ignition timing, and that the peak pressure point can be detected and set from ion current signal features. Eriksson obtained an increase in the torque of the tested engine of up to 3% by combining an active injection of water and a closed-loop ignition timing control. Eriksson demonstrates the potential for fine-tuning the peak pressure angle and identifying heat release parameters from ion signals. Heat release calculations were made using the approaches of Gatowski et al. [125]. The best predicted fractions were 45% and 50% MFB, with better prediction results for the 45% MFB fraction.

Förster et al. [84] used a primary-side variation of the low-side scheme as is illustrated in Figure 26. Transistor Q1 controls the ignition strike and an additional transistor Q2 puts the primary coil in short circuit after the ignition pulse to reduce the ringing of the coils and with it the spark duration. The ECU manages the switching of Q1 and sends the spark duration limit (SDL) signal to Q2. D1 is implemented for spark suppression during the switch-on phase. The specifications for the ammeter, U = 150 V, and the resistance Rion is not provided. The shorting of primary winding allows an earlier detection of ion current and an extended bandwidth of the ion current measurement circuit from 7 kHz (without shorting) up to 16 kHz, sufficient for knock detection. Misfire detection is performed by filtering the ion signal using a fourth-order Bessel low-pass trap circuit.

Figure 26. Low-side circuit for ion current detection. Based on Förster et al. [84]

In this study, tests were conducted on a 4-cylinder direct injection gasoline engine at idle, 3000 min-1 and 5000 min-1, all under no load conditions. Low load conditions, as discussed by Förster et al., are critical due to the considerable variation in ion current observed. The addition of MMT and MTBE additives increased the ion current magnitude. Exhaust gas recirculation (EGR) was found to reduce ion current levels. It was also noted that a larger spark plug electrode increased the ion signal; a similar increase in the ion signal was observed when the gap distance was widened from 0,7 mm to 1,1 mm. In the research conducted by Förster, pressure and ion signal frequency spectrums were compared, revealing identical peaks under knock conditions at full load, specifically at 6 kHz, 11 kHz, and 15 kHz. The 6 kHz peak appears on normal signals, which means that it is not sufficient to infer knock from it. Therefore, a high-pass filter set to frequencies greater than 6 kHz was implemented.

Förster et al. conducted additional tests using ion signals for closed-loop knock control on an 8-cylinder engine. They found a sufficient degree of correlation for detecting knock. Yamada et al. [17] applied a low-side scheme for ion current detection in a multi-spark ignition system. In this setup, the power supply U is a capacitor charged by the ignition pulses. The work shows that, when used with ignition coils that have less secondary winding inductance, multiple sparks provide improved detection of misfires and knock compared to conventional single-strike ignition systems. For knock detection, Yamada et al. utilized a bandpass filter tuned between 3 kHz and 20 kHz, along with an integrator to extract the knock signal. Their work also indicated that ion current signals can detect changes in the insulation resistance of the spark plug, exhibiting increased current flow with lower insulation resistance.

The thesis of Andersson [49] uses data from a low-side scheme, similar to Eriksson [25], [75], [86]; the ion current is used to validate a model for combustion chamber pressure, temperature and NO species formation, deriving a virtual pressure sensor from ion signal detection system. As Andersson states, thermal ionization explains the second peak of ion current after ignition (post flame phase). Additional information about the modeling of ion current generation and typical waveforms will be presented later.

Shimasaki et al. [58] used a low-side ion measurement system in a Formula One engine with CDI type ignition system. The elevated regime of Formula One engines and the trend to operate them in lean mixture condition to reduce the fuel consumption impairs the stability of combustion, making the diagnosis task particularly challenging. The power source employed for ion current measurement was a capacitor in parallel with a 300 V zener diode, having that value of U. The value of Rion = 18 k was specified to establish a relation between the area of the spark-plug electrode and the ion current. The voltage output from the ion signal is directed to a differential amplifier, which then feeds into a resettable active integrator based on an operational amplifier. This integrator processes the ion current integral to identify misfire conditions. If the final voltage from the integrator is high, it indicates that combustion is normal. Conversely, if the final integrated voltage is low, it signifies a misfire.

With the ion integral signal, faltering could also be detected, as mixture becomes leaner when throttle is released. Both signals were used to identify faltering from a threshold in ion integral amplitude and correct automatically for leaner or richer fuel injection. Knock could also be detected using the ion current signal before the integrator, passing it through a bandpass filter set between 5 kHz and 12 kHz. Air-fuel ratio measurements were made, comparing

UEGO sensors and ion currents, detecting air-fuel ratios among each of the 10 cylinders of the engine.

In Huang et al. [126], a constant volume chamber and a SAAB engine were used for their tests. The ion current measurement system appears to be the same Mecel type seen in Auzins et al. [119] and the ignition system was coil-on-plug type. For the constant volume combustion chamber, they had a high-speed camera, a pressure sensor and a UEGO sensor for air-fuel ratio. The features of the ion signal respect to time: -Initial ionization peak (flame front), Time of second peak of ionization from spark command, Amplitude of second peak of ionization, Area under the curve of ionization signal- were compared against photos of combustion chamber at certain instants to understand ionization phases. With that features, varying air-fuel ratio, an artificial neural network was trained to predict the air-fuel ratio, obtaining good results.

For the engine tests, varying rotational frequency, load, coolant temperature, and using the outcomes of constant volume chamber, air-fuel ratio was obtained by “training” another neural network. Then, ion current was used to account for air-fuel ratio during cold cranking conditions. With that, the air-fuel ratio was predicted before lambda sensor warmed up. The authors quote that this outcome can be used to avoid overfueling at cranking and warmup. They also indicate that the post-flame peak of ion current moves toward the spark event when load increases.

The work of Abdel-Rehim et al. [89] used the low-side scheme, but with the opposite polarity for power supply U. A more detailed discussion of the use of that polarization will be mentioned later.

In the work of Zhu et al. [127], a 3-liter V6 engine with coil-on-plug ignitors was used. Although not explicitly stated in the work, it is possible to infer that the scheme is low-side, as it is associated with coil-on-plug ignition, looking at the previous works shown. They combined three control strategies combined for closed-loop ignition timing: Maximum Brake Torque (MBT) timing control, borderline knock limit control, and retard limit control. All three strategies gather input from ion signal features: First inflection point (from chemiionization peak to valley of thermal ionization), Second inflection point (from valley to peak of thermo-ionization), Second peak (thermo-ionization), that is, second and third derivatives respect to the crank angle, and the integral of ion signal. Then, those points and values can help to construct the MFB curve and determine the 50% MFB (as explained in Daniels [110]), obtain the peak pressure angle, and the maximum torque angle. With the MBT timing control, the goal is to advance the ignition timing as closer as possible to the point of maximum torque, limited by knock, having as additional input engine speed and load. That also reduces cyclic combustion variations.

The borderline knock limit control strategy sets a moving maximum ignition advance to avoid knock. It uses a stochastic limit control algorithm to have as less conservative advance timing for knock as possible, still avoiding it. For the retard limit control, the stochastic algorithm is the same, except for its purpose of retarding ignition timing, to ensure combustion stability. With all the three strategies combined, engine tests were made, obtaining good results in avoiding knock, getting the maximum torque and combustion stability. Additionally, ion signal fed the control strategy to account for a faster catalyst heating, improving from 20 s to 12 s.

The work of Rivara et al. [115] appears to use the setup of low-side scheme, applied on a 1,6-liter 4-cylinder Ford Zetec engine, using only one cylinder for measurements and a “trained” neural network, to predict the Peak Pressure Point (PPP) from ion signal (virtual pressure sensor). In this work, besides ion current input, the signals from air bleed valve and manifold absolute pressure (MAP) were input to the neural network. From their engine tests, the algorithm was found to be robust for finding and setting a PPP under step load and speed disturbances.

The works of Molina et al. [128], [129] present the application of a low-side scheme to a two-cylinder engine that has been converted to spark ignition, utilizing natural gas as fuel. They refer to the circuit design illustrated in the Butler-Delphi patent [130], noting that modifications were necessary due to the unavailability of certain components. According to the specifications of that patent, the voltage U is set at 100V, and the ion sensing resistance Rion is 53 kΩ. The patent [130] uses two Silicon Controlled Rectifiers (SCRs) in addition to the low-side ion sensing components. One SCR is designed to prevent the discharge of the capacitor that serves as the power source U during the ignition strike, while the other SCR protects the ion sensing circuit operating within the same phase. Molina et al. were able to detect partial burning of fuel from ion current, finding a higher thermo-ionization peak value for normal combustion and a lower one for partial burn. Oscillations of knock abnormality were also possible to detect after thermo-ionization peak of ion signal corroborated with pressure signal.

Fiedkiewicz and Pielecha [131] explored the use of ion current to analyze the combustion process in a natural gas-fueled, single-cylinder engine that was converted to spark ignition. This engine utilized an ignition system adapted from a Mazda Skyactiv G setup, which already includes a low-side scheme for measuring ion current. By taking the first derivative of the measured ion current with respect to the crank angle and in-cylinder pressure, they determined the heat release rate and established a correlation. They found that the peak pressure point showed a strong correlation with the thermal peak of ionization, although this correlation was limited to a specific range of ignition spark advance. For heat release diagnostics, there was a notable correlation between the maximum of the first derivative at the thermal peak and the maximum heat release.

2.5.5 Comments about the polarity of the power supply voltage U

According to Thurman's research [39], the inner electrode of a spark plug typically reaches a higher temperature than the ground electrode. When the inner electrode is negatively biased in relation to the ground electrode, it enhances a thermoionic effect, allowing the inner electrode to emit electrons more easily than if the polarity were reversed. However, this holds only during spark strikes.

Arrigoni et al. [91] pioneered the discussion of the impact of power supply polarity on ion current detection when using a DC power source. Conducting tests between -600 V and 600 V, they found that positive polarity generates a greater ion current compared to negative polarity.

Wilstermann et al. [132] assumed that the spark plug gap functions as a diode in series with a variable impedance attributable to the mixture content between the electrodes, allowing current to flow in only one direction.

Yoshiyama et al. [87] analyzed the ion current waveforms during combustion when utilizing a U = 350 V voltage source. The ion current trace presented two distinct peaks, the second of which located close to the pressure peak. In contrast, with U = -350 V, the second peak associated with the pressure peak was not noticeable.

The interest in correlating ion current with in-cylinder pressure prompted the experiments of Arrigoni et al. [91], Wilstermann et al. [132], and Yoshiyama et al. [87], all of whom preferred using positive polarity. Since then, many authors opted for positive polarity, with the exception made by Abdel-Rehim et al. [89], who chose a negative bias power supply. They argued that the positive polarity for U promotes a first peak (chemi-oionic or flame front for some authors) several orders of magnitude higher than what is observed with negative U. This occurs because electrons, which are lighter and smaller than positive ions, are collected by the inner electrode (anode), while the positive ions are collected by the ground electrode (cathode), represented by the cylinder head and cylinder walls. In other words, the cathode has a significantly larger surface area compared to the anode, an explanation also supported by Shimasaki et al. [58].

Now, with a negative voltage U, the anode becomes significantly larger in comparison with the cathode. The peak of chemi-ionization is weak, but a subsequent thermal ionization peak appears when the flame reaches the spark plug. According to Abdel-Rehim, with negative polarity, the second peak is bigger than the chemical one. However, the magnitude of U used in the work of Abdel-Rehim et al. is 380 V.

The diode consideration of Wilstermann and the huge magnitude of U for Abdel-Rehim, arouse questions: What about the breakdown condition of a diode when it is reverse-biased? From figure 3 in the work of Martychenko et al. [36], a sufficiently high tension will promote an attraction of charges, just because of its electric field strength, and with this an avalanche of current in the spark plug, that is, a negative current.

2.5.6 Comments about using AC for power supply voltage U

The use of AC voltage supply for ion current measurement was first introduced in the work conducted by Wilstermann et al. [132], where, as explained earlier, the diode ensures robust sensing. According to Wilstermann, the AC supply allows to separate the shunting impedance of the spark plug from the ionization impedance.

Figure 25 already illustrated the temporary power source e12, derived from the primarysecondary coupling. That power source normally appears in ignition transients when the transistor is switched-off. However, in the Wilstermann work that source appears again after ignition by switching the transistor with a lower duty cycle, generating a sinusoidal power

source e12 which becomes U. The ion signal appears as a sinusoidal variation in Rion. The amplitude of these sinusoidal changes depends on both chemical and thermal ionization, eliminating the initial spark strike component of ion signal. Subsequently, this signal is demodulated in amplitude and filtered using a low-pass trap to extract the ion feature.

Nilsson [133] uses a scheme like the one used by Wilstermann. His work found it possible to detect ion signals using an AC voltage supply. However, Nilsson faced problems in differentiating misfires from normal combustion, extracting information from the carrier wave, and selecting the frequency of the power source. Perhaps, the need to demodulate the AC ion signal and the election of frequency discouraged other authors from implementing AC ion current measurement systems.

2.6 Correlations between combustion chamber pressure and ion current

features

As stated by Naumov et al. [13], the spark ignition of the air-fuel mixture in the engine cylinder produces many charged particles such as free electrons, plus or minus ions and free radicals, generating gas conductivity. In fact, during combustion phases, the gas in the vicinity of the spark plug gap in a particular time domain represents a plasma state produced by the combined effects of high-pressure ionization, chemical ionization, thermal ionization and other ionization sources. When the spark plug terminals are connected to a DC bias voltage, an electric field is generated within the spark plug gap. Under the influence of the applied electric field, these charged particles move in a specific direction: positive ions migrate toward the cathode, while electrons and negative ions move toward the anode, with an intensity dependent on the gas temperature, strongly connected to the pressure of the cylinder. This movement establishes an ion current in the spark plug, carrying valuable information about engine combustion and operation. By acquiring and processing this signal, the extracted characteristic parameters can be used to monitor and control the combustion process of the engine, ultimately leading to improved performance.

Although pressure and ionization occur throughout the whole combustion chamber, the ion current detection is limited to a region where the electrodes are present, as commented by Henein et al. [103]. For a SI combustion engine, the electrical spark interacts with the airfuel mixture and triggers the combustion process that promotes both a rise in pressure from the value due to compression and ion formation, giving the chance to correlate pressure and ion current.

In the following, this part of the chapter is devoted to presenting a few comments about ion generation, chemical kinetics models, ion signal features and their interpretation, with the aim of revealing the diagnosis capabilities that ion current can offer.

Basics of chemistry of combustion

Combustion is an exothermal irreversible chemical reaction which, according to Khovakh [134], occurs when fuel is in vapor phase and the oxidation reactions take place at sufficiently high rates. The rise of combustion, its development and climax are defined by particular events and rates of such reactions, heat and mass transfer conditions on flame kernel and to the chamber walls. Combustion is the sequential disassembly of a (hydrocarbon) fuel molecule, atom by atom, eventually producing stable products (CO2 and H2O), through intermediate chemical species, occurring through elementary reactions prompted by chain carriers or free radicals H, OH, O, HO2, CH3, among others.

In fact, combustion is the result of a sequence of reactions encompassed in what is called the reaction mechanism of a fuel with oxygen. It is not a process that yields results after a single reaction, but rather a given number of intermediate reactions depending on the molecular complexity of the fuel, the environment in which it takes place, and the thermodynamic conditions in which it takes place. The same fuel can burn through different routes or reaction mechanisms producing in the process even final products, intermediate products, ions and energy in the form of heat. In the cylinder conditions of a combustion engine, constructive, technological and operational factors are involved in the reaction process, demanding the narrowing of the reaction mechanism for possible configurations of these factors and at the same time, turning these factors into possibilities of observation and control, eventually. As the reaction proceeds, as noted, ions are released (active radicals) that make the reacting mass itself and the intermediate products conductive, especially those present in the vicinity of the combustion nucleus and the flame front. As an example, a typical combustion reaction for propane is:

C3H8 + 5O2  3CO2 + 4H2O + 

This reaction is considered a typical reaction in an internal combustion engine cylinder. In it, the hydrocarbon molecule is combined with oxygen to produce carbon dioxide, water and thermal energy. However, this expression includes just initial reactants and final products. According to [134], [135], [136] during combustion, chemical chain reactions occur forming intermediate species. Some of them form stable molecules, others build active radicals. This is known as degenerate branching.

From Semenov [135], the oxidation process of hydrocarbons is presented, beginning with an initial chain, then chain propagations, degenerate branching and the end of chain reactions. The combustion mechanism determines the favored chemical reactions, intermediate species and formation, as well as the ion concentration for that stage. During the ongoing reactions, electrically charged molecules (ionized) and free electrons appear.

From Taylor [136], for certain chemical reactions, it happens that not necessarily all the initial reactants are consumed. Instead, once reached the equilibrium, some quantities of initial reactants and intermediate species remain. Additionally, some reactions are exothermal, but others don’t. In particular, exothermal reactions are less complete at high temperatures than at low ones. Most of the combustion processes involve a change in the final number of molecules, even in complete reactions. This numeric change depends on fuel

composition, air-fuel ratio and completeness of reaction. So, for instance, in [135], to evaluate the chemical composition of the final products in spark ignition engines, an example based on the C8H18-air mixture is presented, for which twelve chemical species are considered in equilibrium upon the end of the final reaction: fuel, O2, N2, CO2, H2O, H, H2, O, N, OH, CO, and NO. Thermal NO formation can be considered using the Zeldovich kinetic model.

Though the previous equilibrium consideration is the most common reduced approach to model the mixture final composition, D’Errico and Lucchini [137] state that it is not the only one, particularly for natural gas combustion. They developed a reduced kinetical model to evaluate NO and CO species for premixed flames, using a spark ignition natural gas fueled engine. They also took into consideration the formation of NO, in which the dominant reaction mechanism is the thermal one, the extended Zeldovich mechanism:

O + N2  NO + N

As well as the reactions:

N + O2  NO + O N + OH  NO + H

These reactions are present only at high temperatures because of the required high activation energy. D’Errico and Lucchini object that the classical approach assumes steady-state approximation for nitrogen radical, while the rest of species are assumed in equilibrium. Since in regions near the flame front, the free-radical concentration and NO formation are higher than the outcome predicted with equilibrium concentration, D’Errico and Lucchini attempt a reduced chemical kinetics model with 3 transport equations for NO, CO, H species, as well as a partial equilibrium for OH, O, and H2 species. For the NO species, they use the Fenimore -“prompt”- model, also used by Ramajo and Nigro [138].

Zheng et al. [139] uses a reduced chemical kinetics model for HCCI, stating that the chemical kinetics of combustion is extremely complicated, especially for autoignition processes. From the degenerate branching mechanisms and reduced chemical kinetics, more detailed chemical kinetics models have been developped, having hundreds of species and equations.

Due to the stochastic character and complexity, the study of transition of chemical species requires powerful software and computers to do so. As an example, Kravchik et al. [140] shows a list of 97 intermediate reactions and their coefficients for Arrhenius expressions, using Chemkin. Forigua and Mantilla [141] use KIVA, adding the Cantera toolbox.

Revealing the importance and complexity of hydrocarbon-air chemical kinetics, Curran et al. [142] mention the number or species with their associated rate constants and reactions required at present to describe the combustion reaction of some fuels. For instance, the oxidation of hydrogen can be described using the global reaction H2 + ½O2 = H2O and that for methane by CH4 + 2O2 = CO2 + 2H2O. However, these reactions do not occur as written. Hydrogen requires eight species and approximately 30 elementary reactions to describe its oxidation over a wide range of pressure and temperature [143]. For methane the mechanism is even more complex requiring approximately 30 species and 200 elementary reactions. The

complete reaction mechanism for n-heptane oxidation included 2450 elementary reactions among 550 chemical species [142].

Atef et al. [143] presented a comprehensive iso-octane combustion model with improved thermochemistry and chemical kinetics. Atef cites the comprehensive chemical kinetic model for iso-octane presented by Curran et al. [142].The model was developed based on the lowand high-temperature reaction pathways and rate rules proposed earlier for the Lawrence Livermore National Laboratory (LLNL) n-heptane model [144]. An updated comprehensive chemical kinetic model for iso-octane whs been developed by jointly researchers at KAUST, LLNL, NUIG, and UCONN. The thermochemical data for all species in the iso-octane submechanism have been updated using new group values and detailed analysis of intramolecular interactions. The chemical kinetic reaction mechanism was also updated with new reaction classes and rate rules [145].

All the above was presented just to get a brief glance about related chemistry issues involved in the combustion process, although it is not a subject in the present document.

The main ion generation processes

The above-mentioned reactions don’t consider the releasing of positive ions and free electrons. Thus, by assessing them during the reactions, it is possible to correlate combustion processes with ion currents, those ones favoured by a potential difference across electrodes. As summarized in Shimasaki et al. [58], Henein et al. [103], Glavmo [102], Nielsen and Eriksson [75], and Mohanesen et al. [146], the main processes are chemi-ionization and thermo-ionization. Chemi-ionization rises on spark strike, while thermo-ionization derives from temperature increase. In the following, a summary of these processes will be presented.

The chemi-ionization

In the combustion chamber of the spark ignition engine, shortly after the spark discharge, molecules such as OH, CH, C2, CO appear, indicating that the combustion reactions are taking place [135]. Now, from Shimasaki et al. [54] as mentioned by Eriksson [25], some of the elemental hydrocarbon reactions that cause ionization are:

CH+O→CHO++e-

CHO++H2O→H3O++CO

CH+C2H2→C3H3++ e-

On the other hand, Vressner [147] for his HCCI analysis, cites the reaction: H3O++e-→ H2O+H

Included also by Franke [148]. Mehresh et al. [107] explored a chemical kinetics model in a propane-fueled HCCI engine, based on the Warnatz mechanism C–H–N–NO (C1–C4 chemistry) augmented with a skeletal ion formation and consumption mechanism. Their ionization model contains 34 reactions involving 9 ionic species (NO+, N+, N2+, O2+, OH+, O+, H3O+, HCO+, and electron).

In Dong et al. [109], another chemical kinetics model for HCCI with 189 reactions and 62 species is developed, having details of intermediate reactions, leaps in HC species formation and their concentrations, emphasizing the H3O+ species as ionization contributor. For the case of compression ignition, Estefanous [104], George [105], and Assaad [106], considered detailed hydrocarbon reactions and ion formation processes.

The thermoionization

For the thermal-based ion formation, Andersson [49] quotes the approaches of Saitzkoff- Reinmann, Calcote and Yoshiyama-Tomita, using for his work the outcomes from Saitzkoff et al. [83], who establish the NO species as the main source of free charges inside the combustion chamber. The Saitzkoff-Reinmann model is based on the NO formed by the Zeldovich mechanism, refered also in [135] as dynamic NO formation as follows:

O+N2  NO +N N+O2NO+O N+OHNO+H

In Andersson [49], the concentration of NO ions rises during compression from approximately 20° BTDC to its maximum until 18° ATDC. Then, the NO ion concentration decreases approximately at 60°ATDC. Comparing the species H, H2, O, O2, OH, H2O, CO, CO2, NO, NO2, the NO species has the lesser ionization energy requirement (9,2 eV), meaning that it is easier to snatch electrons from that species, which explains its dominance on the onset of ion current [49]. The works of Andersson [49], Eriksson [25], Glavmo et al. [102], Vressner [147], Franke [148], are based upon the model of Saitzkoff et al. [83], which assumes full equilibrium of species.

Naoumov et al. [13] tackles the formation of HC and NO species from non-equilibrium chemi-ionization and thermo-ionization in equilibrium. Naoumov et al. [13] investigated the mechanism of reactions of ionization that are responsible for ionization process during the combustion in gasoline and natural gas engines. The analog of gasoline presented by with the conventional formula C7.653 H14.884. corresponded to the mass fraction: C - 86%, H - 14%. The mechanism consists of 65 species and 247 chemical reactions, including the following, which produce ionized substances:

Chemi-ionozation:

CH + O ⇔ CHO+ + e- Mixture transfer reactions:

CHO+ + H2O ⇔ H3O+ + CO

CHO+ + NO⇔ NO+ + HCO

Mixture recombination:

H3O+ + e- ⇔ 2H + OH

NO+ + e- ⇔ N + O Collision and thermo-ionization:

NO + M ⇔ NO+ + e- + M

Figure 27. Normalized ion generation model predictions of (1) Saitzkoff et al., (2) Naoumov et al., and (3) Measured ion current. Based on Naoumov et al. [13]. Figure 27 shows a comparison between ion generation models and actual ion current. The traditional Saitzkoff model (1) predicts better the thermoionization peak, failing to show the chemical ion peak. On the other hand, the non-equilibrium model of Naoumov (2) has a better ion prediction for chemi-ionization peak, but overpredicts thermo-ionic current peak.

Estefanous [104] uses the Naoumov model in his work on CI engines, although he says that it doesn’t predict completely the ion current waveform. According to Estefanous, most works related to ion current are applicated to SI engines, where the air-fuel ratio is kept within a very limited range and data are easily collected.

Figure 27 illustrates the typical ion current waveform for SI engines, having two peaks, one for chemi-ionization, known also as flame-front phase, and the other for thermo-ionization, known also as post-flame phase [49]. A more detailed discussion comes next.

Main features of typical ion current waveforms in SI engines From Andersson [49] and Fiedkiewicz [131], the main features of ion current can be distinguished, as follows (see Figure 28):

a) Charging of primary winding of ignition coil

b) Releasing of the primary current, having the ignition pulse on secondary winding

c) Free-run ringing of the windings.

d) Chemi-ionization phase.

e) Thermo-ionization phase.

The resolution of the ion current waveform features is dependent on the ion detection circuit used. The independent circuit -unless there is crosstalk- should not show the contributions of ignition system pulse, that is, the charging a), releasing b) and ringing c) of ignition.

The high-side circuit, with its HV diode, is expected to have no contribution from the ignition system. However, Eriksson [25] mentions that the capacitance of the diode may allow some ignition signal to reach the detection circuit.

The low-side circuit will have all features from a) to e) in combustion conditions, making it possible the detection of misfire if d) and e) are not present.

Figure 28. Typical ion current waveform features in SI engines. Based on Andersson [49] and Fiedkiewicz [131] Figure 28 is not representative of the exact values of crank angle nor the amplitude of current, since the beginning of the charging a) depends on the ignition timing, its end depends on the dwell time, the releasing b) and ringing c) depend on the dynamics of the ignition coil, d) and e), as well as the current magnitude depend on fuel type, air-fuel ratio, load, crank rotational frequency, position of electrodes within combustion chamber, temperature, magnitude, and polarity of power source U, as stated in the literature.

A graph that gathers the traces shapes and relative positions of pressure, temperature, ion current, and mass fraction burned is the one presented in Gürbüz [18], reproduced in Figure

29. In this figure, the point (2.3) corresponds to the chemi-ionization peak that often matches

the rising-off in-cylinder pressure from compression value and the activation of burning, for this case 14% of MFB. The point (2.4) corresponds to thermo-ionization peak that usually matches the angle of peak pressure and 90% of MFB.

Figure 29. Correlation between cylinder pressure(1), ion current (2), MFB (3) and gas temperature (4). The features of ion current are: (2.1)Primary charge; (2.2)Spark release; (2.3) Chemoionization; (2.4) Thermoionization. Based on Gürbüz [18]

Relevant features in the studies of ion current are the areas under its curve. Particularly those of the regions d) and e), A1 and A2 respectively, see Figure 28.

Shimasaki et al. [58] used the area under the curve for closed-loop control of combustion, detecting misfire, hesitation, detonation, and lean burn. Their integration was implemented using operational amplifiers alongside integral-resetting electronics. Similarly, Yamada et al. [17] adopted the same approach, but applied it to multi-spark ignition systems.

Lee et al. [123] used a hardware integrator similar to the one employed by Shimasaki et al. to detect misfires under “rough road” automotive conditions (crankshaft velocity fluctuations). They found that ion current integral remains unaffected by these conditions. The robustness of their detection system is based on its reliance on a minimum signal-tonoise ratio criterion.

Phan et al. [108] also employ a hardware integrator, a band supression filter and a unity gain amplifier as signal conditioning elements for ion signal. They found that these three setups could accurately predict combustion events. A good correlation was observed between the ion current and the pressure for equivalence ratio from 0,16 on. Below this value, both signalto-noise ratio and ion formation are poor (in HCCI mode). Among the setups, the integrator provided the strongest signal, although it introduced a delay of 2 degrees of crank angle for peak pressure estimation.

The work of Budko et al. [149] shows the estimation of chamber pressure by means of the definite integral of ion current, digitally calculated, not by hardware integration. With ion integral, a maximum of current Imax is obtained. When the ion integral curve reaches the 80%

of Imax, combustion ends and the pressure peak is achieved. With their method, there is low dependency on the size of electrode and air-fuel ratio.

In the study conducted by Laganá et al. [113], the ion current integral is used to assess misfire events, through digital data integration. When comparing the accuracy of the knock detection between the ion current sensor and the knock sensor, the ion current signal demonstrates a faster detection capability. To identify misfires, a Hanning bandpass filter is applied.

In the work of Hunicz et al. [74], the ion current integral is obtained digitally to calculate energy released during the combustion process, as well as the temperature. Hunicz et al. show correlations between ion current, pressure and heat release rate for HCCI mode of combustion.

In another approach, Eriksson [25] examines three algorithms designed to correlate chamber pressure and ion signal. The first algorithm identifies the local maximum of ion current during thermo-ionization (see Figure 28.e)), However, it encounters challenges in aligning the crank angle of peak pressure with the ion peak. The second algorithm calculates the centroid of the area under the ion signal curve (A1 and A2 are considered) to determine the crank angle of peak pressure based on the horizontal coordinate of the centroid. This algorithm is effective when the thermo-ionization phase is clearly distinguishable. If, however, the thermo-ionization component of the ion signal is larger than average, the centroid's horizontal coordinate shifts toward A2, leading to an early and incorrect estimation of peak pressure. The third algorithm attempts to fit the chemi-ionization and thermoionization peaks as gaussian functions. Accurately positioning the gaussian curves along the horizontal axis to align with the actual ion signal proved to be quite demanding, making its implementation difficult.

In the study conducted by Nielsen et al. [75], a gaussian algorithm was employed to reconstruct the ion signal. This algorithm utilized a sum of two parameterized gaussian functions: one function represented chemi-ionization, while the other represented thermoionization. This approach is grounded in the theoretical framework established by Saitzkoff et al. [83], which provides an analytical expression for ion current formation as a function of pressure (p):

𝐼 𝐼௠ =

1

( 𝑝 𝑝௠) ଴,ହି଴.଻ହఊିଵ ఊ 𝑒ି ா೔ ଶ௞்೘[ቀ௣ ௣೘ቁ షംషభ ംି ଵ] Eq. 39 Where Im is the maximum of ion current, estimated between 10 μA and 40 μA for SI engines [123], [25], pm is the maximum chamber pressure, γ is the ratio of specific heats, Ei is the ionization energy, k is the Boltzmann’s constant, and Tm is the maximum temperature.

The gaussian function approximation appears to be a good fit, as indicated by the expression in Eq. 39, which includes an exponential component, Additionally, Figure 27 shows that the Saitzkoff approach takes on a gaussian function form. Eriksson [86] addresses closed-loop ignition timing control by utilizing the ion signal, optimizing for maximum brake torque, and also employs the gaussian algorithm in this context.

Another significant feature to analyze in ion current is its variation respect to crank angle.

Daniels's work [110] employs the first, second, and third derivatives with respect to crank angle, both from ion current and pressure to get a curve fitting of a Wiebe function and the Rassweiler-Withrow method to find the mass fraction burned curve, along with key points of it, and the heat release rate. The study also compares the results obtained using ion current as input against those using pressure as input for the MFB, revealing favorable results. In the study of Fiedkiewicz et al. [131] the first derivative of ion current with respect to the crank angle is used to assess: the pass through the flame front phase, the maximum rate of mass fraction burning (herald of the ending of ion formation around spark plug electrodes), the maximum heat release point, and the end of combustion. The measurement of ion signal, obtaining its derivative, is used to attain the position of maximum heat release and from it the crank angle of peak of pressure. The results are evaluated for natural gas, having differences in the angles where the peaks of pressure and ionization occur, attributable to the differences in fuel used.

There are other approaches like chemical kinetics and CFD to calculate and model the ion generation. Particularly, the work of Mehresh et al. [107] uses that approach in HCCI to obtain pressure and the 50% of MFB. The work of Dong et al. [109] describes a CFD model to predict the combustion process and the ion current, also for HCCI. The work of Chen et al. [116] involves a mixed reactor model and the Woschni equation to study the heat losses and the pressure cyclic variations.

Artificial neural networks (ANN) are “trained” to obtain empirical correlations between ion current, pressure, heat release, and air-fuel ratio. For peak pressure angle, the work of Hellring et al. [26] compares three algorithms of peak detection, multilayer perceptron MLP (a neural network in its basic concept), gaussian fit and peak finder. For low loads, peak finder and MLP perform acceptably. For loads under 30% of the maximum, gaussian fit arouses poor results. Peak finder is, according to Hellring et al., more promising. Another work with the goal of finding the peak pressure angle is the one of Rivara et al. [115], who used an ANN to locate the angle of peak pressure from ion signal and to use it in a closed loop ignition timing control.

The works [21], [23] approach a reconstruction of the pressure signal from ion current. Panousakis et al. [23] compared wavelets and ANN used to reconstruct chamber pressure, finding better results with ANN. Gao [114] looked for pressure reconstruction also, although using a deep-learning ANN, varying A/F ratio, engine speed, load, and ignition timing. To estimate the air-fuel ratio from the ion signal, Huang et al. [126] employed an ANN to infer the air-fuel ratio from ion current values. This method measures the ion current to determine the air-fuel ratio during the engine starting process. They compared their findings with those from a wideband lambda sensor and discovered that the ion signal allowed for faster detection of the air-fuel ratio. Similarly, Sahin [22] also utilized an ANN to establish

a correlation between ion current and air-fuel ratio, while varying factors such as engine speed, load, and ignition timing.

Possibilities of diagnosis from ion current detection

In summary, the ion current signal can provide:

-Correlations with the in-cylinder pressure: pressure rise from compression value and peak pressure angle. However, it is not feasible to reproduce the whole pressure signal for the entire cycle from the ion current signal.

-Correlations with air-fuel ratio.

-Air-fuel ratio in cold start condition, before lambda sensor heats up [126]. -Knock detection.

-Misfire detection.

-If used in low-side scheme, charging of primary winding of ignition coil and ignition pulse. -Mass fraction burned and its related specific points.

-Estimation of cyclic variation of combustion pressure.

-Input signal for closed-loop ignition timing control.

-Phase angle for camshaft [86]

Table 2 summarizes the contributions of several authors to the subject of ion current. The column Stp stands for special setups. for example, CV=constant volume chamber and Gsk=Cylinder Head Gasket mounting of ion sensor. By default, an engine was used. The column Ign corresponds to the mode of combustion studied by the author; SI=Spark ignition, CI=Compression ignition, and HCCI=Homogeneous charge compression ignition. The column SCh corresponds to the scheme used, being 1=Independent probe, 2=High-side, 3=Low-side. Some authors used more than one scheme. The column Elc corresponds to the number of electrodes used for the combustion chamber.

Table 2. Summary of contributions of authors in ion current Year Author Stp Ign Sch Elc Contribution Observ

1920

MacKenzie et al. [81] CV SI

1

Flame speed

1934

Schnauffer [82] CV SI

1

24

Flame speed Form of flame front

1973

Arrigoni et al. [91]

SI

1

1

Effect of fuels, speed, load, timing, A/F Flame velocity, burning velocity…

1986

Anderson [92]

SI

1,2

2

Peak pressure,

90% MFB

Rassweiler- Withrow

1989

Witze [93] Gsk SI

1

8

Flame propag, swirl Burned volume shape

1989

Gillbrand et al. [118]

SI

3

1

Patent ion lowside

CDI

1991

Witze et al. [94] Gsk SI

1

8

Swirl, flame speed

1992

Hu et al. [95] Gsk SI

1

4

Pressure, Heat release Combustion duration, flame spd.

1993

Hu et al. [96]

SI

1

4

Ion signal magnitude as function of distance and U Feedback ammeter, isolation amplifier

1993

Nicholson et al. [97] Gsk SI

1

8

Flame location Travel, geometry

1993

Meyer et al. [98] Gsk SI

1

8

Flame arrival time, pressure, speed, load, swirl, tumble Heat release, optical spark plug

1993

Shimasaki et al. [54]

SI

1,2,3

1

Challenges of ion schemes, A/F Secondary voltage and ion current measurements

1993

Miyata et al. [122]

SI

2,3

1

Ion density as a function of distance to flame front Decay of spark plug voltage

1995

Auzins et al. [119]

SI

3

1

Coil-on-plug for ion signal detection Misfire, comparison ion current-knock sensor for knock.

1995

Lee et al. [123]

SI

3

1

Misfire,“rough road” tests for immunity of ion current to crankshaft speed variations Shorted spark plug from ion signal integral.

Discrimination by minimum S/N ratio.

1995

Eriksson [25]

SI

3

1

Methods to correlate ion signal and pressure Local maximum,centroid of area, gaussian function approximation

1996

Saitzkoff et al. [83]

SI

3

1

First ion current model Only thermoionization

1996

Nielsen et al. [75]

SI

3

1

Spark timing control Gaussian algorithm

1997

Russ et al. [99] Gsk SI

1

8

2%, 10%, 50%

MFB, flame readius Cyclic variation

1997

Park [100] Gsk SI

1

8

Flame propagation Knock onset, highspeed photo

Table 2. (continued)

Year Author Stp Ign Sch Elc Contribution Observ

1998

Asano et al. [121]

SI

2

1

Correlation between ion current energy and pressure Knock

1998

Daniels [110]

SI

2

1

Comparison of MFB from pressure and ion signal First,second and third derivatives of pressure an ion signal to fit to Wiebe equation.

Rassweiler- Withrow for MFB.

1999

Eriksson [86]

SI

3

1

Closed-loop ignition timing control with ion current Maximum torque, pressure and heat release. Gaussian algorithm

1999

Förster et al. [84]

SI

3

1

Extended bandwidth for ion current Knock, misfire.

Ringing supression

1999

Yamada et al. [17]

SI

3

1

Ion current for Multispark ign.

Misfire, knock from ion integral.

1999

Glavmo et al. [102]

CI

1

1

First Ion measurement in diesel Start of combustion,misfire, EGR. Sensor in glow plug.

2000

Yoshiyama et al. [87] CV SI

1

1

Correlation flame front formation-ion signal

2000

Wilstermann et al. [132]

SI

3

1

First AC ion current AM demodulation

2001

Butler [130]

SI

3

1

Patent for ion current, low side TCI ignition

2001

Hellring et al. [26]

SI

3

1

Peak pressure angle Comparison MLP (ANN), Gaussian fit and Peakfinder

2002

Pfeffer et al. [88]

SI

1

12

Flame front, flame speed, combustion stability Tumble motion effect. High engine speed.

2002

Yoshiyama et al. [111]

SI

2

1

Bad correlation ion current-low load Ion current dominated by area touched by flame

2002

Andersson [49]

SI

3

1

Virtual pressure sensor from ion signal Pressure, Temperature and NO formation models

2002

Naoumov et al. [13]

SI

Ion formation model Chemical and thermal

2002

Franke [148]

SI

Ion formation model Hydrocarbon species

2003

Yoshiyama et al. [101] Gsk SI

1

1

Correlation A/Fion current Pressure peak.

Effect of electrode position

Table 2. (continued)

Year Author Stp Ign Sch Elc Contribution Observ

2003

Suzuki et al. [150]

SI

1

2

Correlation ion current in knock and light emission Knock

2004

Huang et al. [126]

SI

3

1

A/F from ion current.

ANN.

Detect A/F in jumpstart, compared with lambda sensor.

2005

Mehresh et al. [107]

HC CI

1

1

Chemical kinetics for ion formation, pressure Single zone model.

Lower Rion for less delay in peak pressure estimation

2006

Gazis et al. [21]

SI

1

1

Pressure from ion current Neural network

2006

Panousakis et al. [23]

HC CI

1

1

Pressure from ion current Comparison Wavelets and ANN to reconstruct p.

2007

Abdel-Rehim et al. [89]

SI

3

1

Ion current with negative polarity More thermoionization peak.

2007

Zhu et al. [127]

SI

3

1

Ion signal derivatives 2,3 and integral MFB, borderline knock operation for closed-loop of ign.

Timing

2008

Nilsson [133]

SI

3

1

AC ion current Possibility of ion current detection, lots of problems

2009

Abhijit et al. [151]

SI

3

1

Correlation A/Fion current Cyclic variation, effects of engine speed, load, fuel

2009

Rivara et al. [115]

SI

3

1

Peak pressure angle from ion current

ANN.

Ignition timing closed-loop.

2010

Henein et al. [103]

CI

1

1

Ion features SI,

CI, HCCI

2011

Estefanous [104]

CI

1

1

Correlations ion current-load, injection pressure NO formation, A/F, cyclic variation

2011

Labuda et al. [90] CV SI

1

2

Correlation ion current-pressure Knock

2014

George [105]

CI

1

1

Correlations ion current-load, injection pressure Injection profile

2014

Gao et al. [112]

SI

2

1

Ion current for knock detection Absence of ignition pulse

2015

Sahin [22]

SI

1

1

Correlation A/Fion current

ANN

2015

Dittrich et al. [117]

SI

2

1

Setup for ion current measurement

2015

Molina et al. [128]

SI

3

1

Partial burn of fuel Knock detection from ion current.

Table 2. (continued)

Year Author Stp Ign Sch Elc Contribution Observ

2015

Molina et al. [129]

SI

3

1

Partial burn of fuel Knock detection from ion current.

2016

Chen et al. [116]

HC CI

2

1

Mixed reactor model Woschni equations.

Heat loss, cyclic variation

2017

Phan et al. [108]

HC CI

1

1

Ion signal to detect combustion events Hardware integrator, band rejection filter, unity gain amplifier

2017

Budko et al. [149]

SI

3

1

Ion current integral to find pressure peak Software integration

2017

Dong et al. [109]

HC CI

1

1

CFD modeling, prediction ion signal Air-fuel ratio

2017

Gürbüz [18]

SI

1

1

Correlation ion current, pressure, MFB, temperature C type thermocouple

2017

Assaad [106]

CI

1

1

Correlations ion current-load, injection pressure Piezoelectric injector

2018

Fiedkiewicz et al. [131]

SI

3

1

Heat release and pressure Ion current first derivative respect to crank angle.

2018

Laganá et al. [113]

SI

2

1

Misfire detection by ion signal integral Comparison ion current-knock sensor. Hanning bandpass filter

2018

Hunicz et al. [74]

HC CI

2

1

Correlation ion current-pressure and HRR Ion integral for fuel energy release and temperature

2020

Gao et al. [114]

SI

2

1

Pressure from ion current ANN deep-learning

2020

Zhu et al. [152]

SI

2

1

SI and HCCI,

ANN

Cooperative ion current and pressure signals for combustion control

2.7 Conclusions from literature review

This review chapter condensed a substantial amount of published work in key areas involved in the ion current phenomena during combustion process inside the combustion chambers of internal combustion engines, to reveal the relationships between spark ignition current/voltage parameters, ion current formation features, in-cylinder combustion pressure, and combustion diagnostics. To this end, they have been approached the engine diagnosis methods, the modeling of most used inductive ignition systems, the basics of flame kernel formation and development, the design of ion current detection circuits, and the basics of combustion chemistry behind the ion generation processes. The review aimed at gathering knowledge of ignition system design and performance modeling, circuit designs for measurement of electrical variables of the ignition systems, together with the ion current, and also the techniques and algorithms to correlate the ion current features with the combustion process development.

Conclusions drawn from the literature review, to meet the objectives of the present thesis, can be summarized as follows:

 The ion current signal has been present in literature of combustion phenomena for decades, and there is still room to explore its features as a non-intrusive means of combustion engine diagnostics.

 The ion current signal by itself is rich in information from primary charging up to the end of the thermo-ionization. It has more peaks and valleys than the pressure signal along the crank angle interval. However, outside the crank angle interval, there are no considerations for ion current in relation to cylinder pressure.

 There has been, since a long time ago, an interest in implementing the ion current sensor as a virtual pressure sensor, but there stands the requirement of the current signal interpretation and the correlation with other engine performance variables. Some efforts have ended in the implementation of empirical approaches that use gaussian fitting or ANN to convolve a correlation between cylinder pressure and ion signal.

 Ion current is fast in detecting some combustion abnormalities such as knock, before the knock sensor gets excited by the pinging and vibration of engine block. The airfuel ratio can be estimated with the ion current signal at engine cold-start before the lambda sensor heats up. However, there is little information about the frequency response of the ion sensors.

 The preferred voltage supply for ion current is a DC one with positive polarity, since it is more difficult to obtain ion current information from an AC power source.

 The low-side and independent schemes have been studied more extensively in literature than the high-side scheme, primarily because of the difficulties associated with measuring on the high-voltage side of the ignition coil. The ion signal obtained

from the independent scheme is not interfered by the ignition system. However, implementing the independent circuit necessitates either an additional hole in the cylinder head or a specialized setup in the cylinder head gasket for detection purposes.

 The ion signal from low-side has the ignition contribution, using that feature for additional diagnosis. Current commercial ignition systems prefer the coil-on-plug design due to the multi-sparking feature, to its independent ignition timing control capabilities, and to the already integrated ion current sensing.

 In comparison, secondary voltage is not a common technique for engine diagnosis, due to the inconveniences of measuring high voltage.

 This literature review has been of valuable help for a better understanding of the electrical phenomena related to combustion charge properties, electronic instrumentation issues related to combustion process and ignition systems, and combustion diagnostics based on the ion current; it is also valuable for the analysis of combustion phenomena in internal combustion engines.

 The voids found in the literature reviewed serve to justify the research objectives of the present thesis intended to contribute to the deep comprehension of the ignition system architecture and its modeling, the design intricacies of the circuits for the ion current detection, as well as the identification of the ion current features representative of the combustion process development, as a means for engine diagnosis.

Along these lines, some questions arose and stand to be answered:

 What are the criteria for choosing the ohmic value of the shunt resistor required for ion current sensing?

 Is it possible to detect anything else outside the usually studied crank angle interval of ion current?

Can the secondary or even primary winding voltage provide additional diagnostics information to complement the outcomes from ion current signal?

Engine setup and instrumentation for ignition system and ion current measurement The literature review focused on addressing key questions related to ignition system architecture, modeling, ion current detection methods and its associated with combustion process development features, all with the goal of developing an engine diagnosis system. In line with this objective, the present chapter describes the test facility designed and the instrumentation process of a single-cylinder engine, highlighting the specially designed subsystems for this purpose. Thus, electronics design considerations, preliminary static calibrations, verifications, and frequency response tests for the various components of the instrumentated system are presented. The development of this chapter is illustrated in the diagram of Figure 30.

Beginning with a general overview of the setup, the second section addresses the instrumentation of the ignition subsystem components, where the voltages and currents of both windings of the ignition coil are measured using proprietary isolation amplifiers for the primary side and two types of high-voltage probes for the secondary side. A further subsection engages the ion current detection subsystems that are invariably conjoined to the secondary of the ignition system. The ion current detection subsystem includes the power source and the microammeter, both with special designs to change their parameters on the go, as the measurement tasks demand it.

The third section presents the measurement of crank angle. A specially designed circuit multiplexes the information of an incremental encoder in one single signal, solving the limitation in the number of available fast channels of the acquisition system.

The fourth section outlines the dynamometer installation. The measurement of engine torque is accomplished by means of a load cell, coupled through an amplifier to the acquisition system, independently providing the absolute-value visualization of the loading torque. Threafter, it is depicted a self-developped subsystem for the detection of intake and exhaust valves events, based on transformers with movable nucleuses, which leverage the variation of coupling coefficient to provide a rectified signal proportional to the stroke. In the last section, the acquisition system is tested independently, to assess the delays between channels and to perform static calibration of each channel. All these tasks are completed to ensure the reliability of the measurements.

General arrangement of the test stand. To study the ionization current, a setup was arranged that comprised a stationary engine, a dynamometer, and an instrumentation system. Figure 31 shows the instrumented single-cylinder 418 cm3 Changfa 186F engine, naturally aspirated and originally diesel, converted to port-injected spark ignition, with its correspondent electronic control unit, a TCI ignition system, with a manually adjustable ignition advance, a manually adjustable throttle, an Klam-CFK90 Eddy brake dynamometer and the cabinet that receives the conditioned signals from the engine.

Figure 30. Schematic of the chapter

Figure 31. Photo of engine setup

For motoring tests, a Siemens 7,5 kW AC three-phase induction motor was used, and also the DC starter motor integrated to the Changfa Engine. A more detailed explanation of the setup goes next.

Figure 32. Schematic of the engine setup Figure 32 illustrates a schematic diagram of the experimental setup of the instrumented engine. The engine (1) through a belt-driven transmission, drives the eddy-current brake dynamometer (2). The force exerted on the load cell (3), due to the brake torque is converted to a voltage signal that is sent to the amplifier (4), whose higher magnitude voltage signal in the output, proportional to the brake torque, is received by the acquisition system (5). The braking torque of the dynamometer is adjusted manually through its power source (6). In the hole originally used for the diesel injector in the cylinder head, is placed a liquid-cooled 7061B Kistler piezoelectric transducer P to measure the in-cylinder gas pressure. The electric charge output of the pressure transducer is input to one of the channels of a 5065A 4-channel Kistler amplifier, the conditioned voltage signal of which is gathered by the acquisition system. A second hole in the cylinder head has been drilled to place a spark plug connected to an ignition coil. This ignition coil is instrumented to measure its primary winding current and primary winding voltage with proprietary signal-conditioning devices. Analog isolated amplifiers are used to prevent possible damages to the acquisition system in the event of free-run transient voltage spikes of the primary winding after contact release.

The secondary winding voltage of the ignition coil is captured by two probes, one being a proprietary amplified 1000-megohm resistive divider Sec.V1, and the second an inductive pickup Sec. V2. The resistive probe Sec.V1 has a bandwidth between DC to 3 kHz, whereas the inductive probe Sec. V2 captures AC components from 600 Hz up to 50 kHz.

The secondary winding current is captured by means of a Low-side scheme, implemented to overcome the difficulty to integrate an independent ion-current measurement scheme. This would require another hole in the cylinder head, and to leverage the single electrical loop characteristic of Low-side circuit to detect primary winding charging -from secondary side-, secondary voltage release and, certainly, the ion current signal. A more detailed explanation is presented in chapter 3.1.

The crank angle position is sensed by an Autonics E50S8-3600-3-T-24 incremental encoder, with a 3600 ppr resolution, and totem-pole outputs A,B and Z. There is also a proprietary signal conditioning circuit to send a unique combined signal to the acquisition system.

Self developed electrical transformers, with variable coupling coefficients and movable nucleus, similar in principle of operation to a LVDT, have been installed to follow the displacement of the intake and exhaust valves, especially the opening and closing instants of the valves, and to reference the in-cylinder events during the engine cycle process. The transformers are coupled with special signal conditioning circuits to send voltage signals IntakeV and Exhaust to the acquisition system. The systems for measuring valve opening, crankshaft angle and torque signal conditioning are explained in section 3.2.

A wideband UEGO-LSU4.9 transducer senses the amount of oxygen present in the exhaust, sending an equivalent voltage signal (related to air fuel ratio of the working gas, AFR) to the acquisition system. The sensor provides an analog output between 0V and 5V, digitally converted to the AFR ratio between 10 and 18. From the specifications sheet of the lambda sensor, the AFR ratio can be deduced from the output voltage Vox as: 𝐴 𝐹= 2𝑉௢௫+ 10 Eq. 40 The reference values of ambient air pressure are measured by the MAP sensor, whose output, in the form of proportional voltage signal, is sent to the data acquisition system. Although the engine is instrumented with more sensors, such as intake, exhaust, oil and engine block thermocouples, only the above-mentioned signals are relevant for this work. The details of each of the proprietary signal conditioning circuits will be explained below.

3.1 Ignition system measurement setup

Figure 33. Ignition and ion current measurement system As it was shown in the literature review, the low-side scheme is a typical configuration found on coil-on-plug ignition systems. Commercial coil-on-plug units integrate the commutator transistor, ignition coil, power source, ion current measurement, dwell-time circuit, trigger circuit and more in one device. However, they are sold as sealed units, making quite difficult

to extract electrical signals from them. Thus, an entire low-side scheme with all components and terminals accessible was built. Figure 33 illustrates the schematic diagram of the measurement components for the ignition system and ion current.

The ignition coil M is a transformer originally intended for using it as a wasted-spark coil for two cylinders, but for this setup the terminal (s+) is connected to one terminal of the power source U, that is used to promote the ion current, whereas the terminal (s-) goes to the spark plug of the Changfa engine. Section 3.1.1. shows additional details on the ignition coil.

The terminal (s-) of the ignition coil is connected to one side of the high-tension wire W that in turn is connected on its other side to a “T” connector that branches to the spark plug SP1 and to the resistive high-voltage probe Sec V1, that sends the conditioned voltage signal U2LF to the acquisition system. The resistive high-voltage probe Sec V1 has 1000 M of input impedance and it is capable of measuring up to 40 kV. It consists of an encapsulated resistive divider Rd-Rc that sends a scaled voltage, proportional to secondary voltage, to an instrumentation amplifier with unitary gain. On the other hand, the inductive clamp sensor Sec V2 is a Pearson-type coil that embraces the high-tension wire W, to capture the AC high frequency components of the secondary voltage and send them to the acquisition system as the voltage signal U2HF.

The power source U is variable at will, from 0 to 300 V DC and with the possibility to put its terminals in open-circuit, short-circuit or normal constant-voltage mode of operation, and separately to invert its polarity if needed. The other terminal of the power source U is connected to the microammeter system S, required to measure ion current.

The microammeter S has the shunt resistor Rion that can be changed with a switch to the values: 10  ,100 , 1 k, 10 k and 100 k. The amplifier connected to the shunt resistor has three possible gains: G = 1, G = 10 and G = 100, selectable with another switch. In parallel with the shunt resistor, there is a semiconductor-based limiter Lim that absorbs the huge transient of secondary winding current during the spark, whereas the much lower ion current is captured by the shunt resistor after the spark. The output of the microammeter S is the voltage Uion, which represents the ion current i2 by the expression: 𝑈௜௢௡= 𝑅௜௢௡𝐺 𝑖ଶ Eq. 41 The primary winding voltage UP1 goes to a resistive divider that reduces the magnitude of the primary voltage to ±100 mV, that in turn goes to an analog isolation amplifier ISO1 with a gain of 10, that sends the amplified voltage signal U1 to the acqusition system. The isolation amplifier ISO1 is required because the primary voltage UP1 is a differential measurement with respect to ground: the terminal (p+) of primary winding is connected to the ignition power source Vs (a 12V battery) having a typical potential of 12 V respect to ground. On the other hand, the terminal (p-) of primary winding can have the same static 12 V value respect to ground, but additionally a transient potential above 300 V, and negative respect to the ground, when the spark is released [40].

The primary winding current i1 is detected by the shunt resistor Ra and converted to the voltage UP2 by applying Ohms law as follows:

𝑈௉ଶ= 𝑅௔ 𝑖ଵ Eq. 42 The voltage UP2 is another differential quantity, with typical potentials above ground in both terminals of Ra, requiring the isolation amplifier ISO2, as explained earlier. The gain of ISO2 is also 10 and its output Ui1 is connected to the acquisition system, representing the primary current signal i1.

The ignition module is a classsic GM 4-pin HEI, TCI type, with no timing retard for total control of ignition timing. It has two terminals {G} and {W} for the inductive pickup sensor ign_puls that sends the ignition trigger signal, based on the crankshaft angle. The ignition timing can be manually adjusted with a proprietary mechanical setup. The {C} terminal of the ignition module is connected to the primary winding terminal (p-) while the terminal {B} is connected to the positive of battery Vs. The manufacturer of the GM 4-pin HEI module states the module should not operate for more than 5 minutes with the engine not running.

The switch S0 enables the current to the ignition coil to provide the spark for the operation of the engine.

Next lines provide additional detail of the considerations for each subsystems of the ignition measurement setup.

3.1.1 The ignition coil

The ion current measurement scheme plays a prime role in the selection of the ignition coil. In [40] it is shown a classic cartridge ignition coil, in which the windings are internally connected, forming an autotransformer. The Low-side scheme presented in the present work requires an ignition coil with transformer-type configuration, with all four terminals of the coil available.

To get a proper parameterization of the ignition coil to be used, primary and secondary resistance measurements are performed.

Additionally, the frequency response characterization for the classic ignition coil shown in [40] can be applied on the transformertype ignition coil to be used, with a slight variation in connections.

Figure 34 illustrates the setup components for the ignition coil frequency response characterization. Figure 34.a) shows the schematic when the ignition coil has autotransformer configuration. On the other hand, the schematic of Figure 34.b) shows the variation in connections when the ignition coil is configured as transformer. Figure 34.d) displays a photo of the test bench implemented with its components: (1) sinewave signal generator G, (2) audio amplifier required to drive the primary winding of ignition coil, (3) oscilloscope that captures the primary and secondary waveforms, connecting U1 to the first channel and U2 to the second channel, and (4) the transformer-type ignition coil used in the present work.

Figure 34. Test circuits for ignition coils, a) autotransformer ignition coil (based on [40]),

b) transformer ignition coil, c) schematic of the amplifier (2), and d) a photo of the setup

The schematic diagram of the TDA20230A audio amplifier [153] used to drive the primary winding is displayed in Figure 34.c). As explained in [40], the amplitude of the sinewave applied to the primary U1 is fixed to 100 mVp-p and the scale of secondary wave U2 to be observed on the oscilloscope should be adjusted to 100 Vp-p and lowered accordingly during tests, due to the transformation ratio.

The transformer turns ratio a = N2/N1 is obtained by dividing the secondary turns N2 by primary turns N1. However, that constant ratio applies for low-frequency transformers, typically 50 Hz or 60 Hz. To obtain a more accurate transformation ratio for the ignition coil, a sweep of frequencies is performed, increasing gradually the frequency of the sinewave that the generator G provides. The transformation ratio as function of the frequency, that is, the amplitude ratio U2/U1, for the ignition coil used in the present work is presented in Figure

35. From the transformation ratios found between 100 Hz and 1000 Hz, the turns ratio can

be deduced as aproximately a = 66 for the coil. Additionally, Figure 35 illustrates a resonant peak with a transformation ratio of 203,8 times at 8000 Hz. The adjustment of the amplitude of U1 that the primary winding receives must be limited to 100 mVp-p, checked with the oscilloscope, avoiding variations in amplitude due to possible frequency response variations of the audio amplifier (2). The TDA20230A audio amplifier used results adequate to drive the primary winding, not so for a previous amplifier design based on the LM386 chip [154] that failed.

Figure 35. Transformation ratio as function of the frequency for the ignition coil

0

50

100

150

200

250

10

100

1000

10000

U2/U1 Frequency (Hz) Transformation ratio as a function of frequency

3.1.2 The resistive high-tension probe

The literature review referred that the resistive dividers can detect both DC and AC components of secondary voltage, while other types of sensors can detect only its AC components. Now, the work of Shimasaki et al. [54] referred the isolation difficulties of resistive dividers. Despite that difficulty, low frequency components of secondary voltage can provide relevant information of the combustion chamber phenomena, using the spark plug as sensor. For this reason, the present work leverages the use of a resistive divider, complementing the secondary voltage measurements with an AC Pearson clamp.

As it was said before, the probe Sec V1 has an input impedance of 1000 M and can measure up to 40 kV. It is a passive probe and is designed to connect with multimeters or oscilloscopes characterized by an input impedance of 10 M. The input impedance of the correspondent channel of the acquisition system is 100 k, therefore, it is needed an active circuit that couples the impedances of the passive probe and the acquisition system.

Figure 36 illustrates the schematic diagram of the HV probe. The secondary winding voltage reaches the passive divider network Rc-Rd that reduces the voltage 1000 times. The divider is then connected to a 9,1M resistor and a 1M variable resistor R1 that matches the 10 M input impedance for the correct operation of the passive probe. Then, the reduced input signal is sent to an instrumentation amplifier based on the AD620 chip [155] that sends the signal to the acquisition system with a total atenuation of 10000 from input to output. The gain adjustment of the instrumentation amplifier is performed by R2 - and R1 - and the output offset is adjusted with R3. The offset adjustment uses two precision 2,5V regulators with low temperature-coefficient configuration [156] to get ±200 mV of bias. The amplifier is powered by a dual stage regulator [157], [158] that provides ±12 V from a non-regulated dual power supply that in turn is connected to a common household supply of 120V AC.

The HV probe has a PVC tubular chassis that integrates the passive probe and the instrumentation amplifier. The power supply is mounted on a metal box one meter away from the probe chassis to avoid induction of 60 Hz components to the secondary voltage measurement. The tubular chassis is covered inside with aluminium tape to avoid electromagnetic interferences from any source nearby.

Initially, a BK Precision BK-PR28A passive probe was used for the experiments, but the lack of information of bandwidth and an earlier malfunction of that probe urged to substitute for a PINTech HVP-40 probe with a specification of 20 MHz of bandwidth. Nevertheless, the warnings of the work of Shimasaki et al. [54] were considered and two types of tests were performed, static calibration and frequency response, to ensure proper measurements.

A high-voltage generator that comprises a variable transformer (Variac), an elevator transformer, and a half-wave Cockroft-Walton multiplier network was implemented to ensure isolation and to perform a DC calibration of the high voltage probe.

The first step made for the calibration was to ensure the transformation ratio of the elevator transformer powered by the Variac. Figure 37 illustrates the schematic of the setup.

Figure 36. Schematic diagram of the resistive high-tension probe

Figure 37. Schematic diagram of the setup for transformation ratio of elevator transformer The Variac T1 provides an AC supply from 0 V to 120 V to the elevator transformer T2. The measurement of primary voltage is U1 and U2 corresponds to secondary voltage. The Fluke 289 voltmeter, with av range of measurement of 1000 V [159], is used to measure U1 and U2, one at a time.

Table 3 Illustrates the results for the transformation ratio of elevator transformer T2. The rows two to five indicate that the voltage ratio of T2 is U2 / U1 = 13,13. The values of U2 from rows 5 and on are calculated and provide a reference for the expected voltage applied to the Cockroft-Walton multiplier, based on the voltage ratio obtained.

Table 3. Transformation ratio of transformer T2 Previous to the application of a high DC voltage from the multiplier to the probe, a preliminary AC test was performed. Figure 38 shows the schematic diagram of the setup implemented. The Variac T1 provides a variable AC-60 Hz voltage supply U1 to the elevator transformer T2, that in turn feeds its secondary voltage U2 into the high voltage probe SecV1. The AC voltage of the output of the high voltage probe U3 is proportional to U1. The Fluke 289 voltmeter is used to measure U1, U2 and U3, one at a time, all measurements in AC volts rms.

Table 4 shows the results of the preliminary calibration. The values of U2 above 702 V are calculated with the voltage ratio of the elevator transformer.

U1 (V rms) U2 (V rms) U2 / U1

0,00

0,0

12,39

162,0

13,07

26,02

341,8

13,13

39,60

520,0

13,13

53,50

703,0

13,14

65,10

75,90

88,50

854,7

996,5

1162

.

.

.

99,30

1304

.

111,6

1465

.

124,7

1637

.

Figure 38. Setup for static calibration of resistive probe (preliminary)

Table 4. Calibration results for resistive high voltage probe with AC 60 Hz. From Table 4, the average ratio of the resistive probe is 0,09929 mV/V, with a R2 = 0,9999. Figure 39.a) illustrates the schematic diagram of the setup implemented to perform DC calibration of the resistive high-tension probe. The Variac T1 provides a variable AC voltage supply U1 to the elevator transformer T2. This transformer feeds a Cockroft-Walton multiplier network, which converts the secondary voltage from T2 into a high DC voltage. The probe SecV1 reduces this output, resulting in a DC voltage U2 that is proportional to the input voltage U1. It is important to note that the DC output from the Cockroft-Walton multiplier is negative in relation to ground, due to its operating principle [160], [161]. For testing, a Fluke 289 voltmeter is used to measure U1 and U2, one at a time, with U1 measured in AC volts rms and U2 measured in DC volts.

Figure 39.b) shows a photo of the mockup implemented, with its components: (1) resistive high voltage probe, (2) power source, (3) Variac, (4) elevator transformer, (5) Cockroft- Walton multiplier, and (6) voltmeter.

U1 (V rms) U2 (V rms) U3 (mV rms)

0,00

0,0

0,0

12,39

162

16,1

26,02

341,8

34

39,6

520

51,8

53,5

703

70

65,1

854,763

85

75,9

996,567

98,1

88,5

1162,005

115,6

99,3

1303,809

129,7

111,6

124,7

1465,308

1637,311

145,3

Figure 39. Setup for HVDC calibration of resistive probe, a) schematic, b)photo. Table 5 condenses the results of the calibration of the resistive probe using the multiplier. The first column corresponds to the angular setting of the knob of the VARIAC, U1 is the AC voltage applied to primary winding of the elevator transformer T2, UT2 is the AC voltage of the secondary winding of T2, measured up to 961 V and the other values presented in the column are obtained based on the 13,13 voltage ratio of T2. The fourth column corresponds to the DC reading of the output of the resistive probe and the fifth one to the AC reading of

such output. That AC value corresponds to the ripple of the Cockroft-Walton multiplier given by the pulsed charging of its capacitors.

Table 5. HVDC Calibration results for resistive high voltage probe. The last row of Table 5 indicates, using the aforementioned 0,09929 mV/V ratio of the resistive probe, that the maximum excitation applied to the probe was -11314 V. The Cockroft-Walton multiplier should provide a no-load output of 2√2 𝑛 𝑈்ଶ(஺஼). Having in the present work a multiplier with n = 5 stages and a maximum UT2(AC) = 1845,3 V the maximum excitation applied should be -26096 V [160], [161]. However, the load resistance IL, capacitance C, forward bias voltage of diodes Vf and excitation frequency f determine the DC output of the multiplier, as follows [162] :

|𝑉௢௨௧| = 2√2 𝑛 𝑈்ଶ(஺஼) − 𝐼௅ 2𝜋𝑓𝐶(4𝑛ଷ+ 3𝑛ଶ−𝑛) −2𝑛𝑉௙ Eq. 43 Solving Eq. 43 for f=60 Hz; C=2,2 nF; Vf=10 V; |Vout| =11314 V, it is found that the actual input impedance of the resistive probe is 580 M.

VARIAC

indication U1 rms (V) UT2 rms (V) U2 DC (V)

U2 AC (Vrms)

0

0

0

0

0

5

6,88

90,09

-0,0659

0,0132

10

12,55

164,72

-0,1258

0,009

15

17,98

236,39

-0,183

0,014

20

23,3

306,41

-0,2391

0,0135

25

30

35

40

28,52

34,1

40,58

45,4

374,31

448,13

532,9

603,1

-0,2927

-0,3512

-0,4134

-0,464

0,017

0,0212

0,021

0,022

45

51,9

684,3

-0,5211

0,025

50

55

57,2

63,5

751,5

836,2

-0,5692

-0,6246

0,0276

0,029

60

65

70

75

80

85

90

95

100

105

110

115

120

125

130

68,2

73,2

78,4

83,4

89,3

93,6

99,3

104,7

109,8

115,8

121,1

126,1

130,4

135,9

140,3

897,6

961,4

1031,2

1096,9

1174,5

1231,1

1306,1

1377,1

1444,2

1523,1

1592,8

1658,9

1715,1

1787,4

1845,3

-0,6616

-0,7003

-0,7413

-0,7783

-0,821

-0,8527

-0,888

-0,9231

-0,9548

-0,9943

-1,0224

-1,0482

-1,072

-1,0985

-1,1235

0,031

0,034

0,035

0,038

0,041

0,044

0,047

0,049

0,052

0,056

0,059

0,062

0,065

0,067

0,069

The DC and low frequency (60 Hz) tests demonstrated the linearity and isolation of the high voltage probe. The next step is a broader frequency response analysis to verify the gain of the probe with respect of the frequency.

Figure 40. Setup for frequency response of the high voltage probe, a) Schematic diagram, and b) photo of the mockup.

Figure 40.a) illustrates the schematic diagram of the mockup implemented, and Figure 40.b) displays a photo with its components. The signal generator (1) is configured to produce a sinusoidal output. This output is sent to the audio amplifier (2), which then feeds the primary winding of the ignition coil (5). The secondary winding of the ignition coil is connected to the resistive probe (3), which is powered by its own power source (4). The oscilloscope (6) is set up to receive the primary voltage U1 on the first channel and the output from the resistive probe U2 on the second channel.

Table 6 shows the results of the frequency response obtained with the implemented setup. The primary voltage U1 was kept as large as possible, except for frequencies below 100 Hz, to avoid overheating of the amplifier and coil. The voltage U2 is determined by the product of the primary voltage,the transformation ratio of the ignition coil, which varies with the

frequency as described in section 3.1.1 and illustrated in Figure 35, and the attenuation factor of the resistive probe. The column U2c divides U2 by the transformation ratio of the ignition coil, to separate the frequency effect of the ignition coil on the measurement. The last column shows the normalized gain of the high voltage probe. As shown in Table 6, at 14000 Hz, the probe withstood a 20 kVp-p input.

Table 6. Results of frequency response for the resistive high voltage probe.

Freq.

(Hz) U1 (Vp-p) U2 (Vp-p) U2c (Vp-p) U2 / U2c (gain)

10

20

30

40

50

60

70

80

90

100

200

300

400

500

600

700

800

900

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

11000

12000

13000

14000

20000

3,6

4,7

6,4

7,96

9,44

10,72

12,04

13,12

14,2

14,9

20,1

21,8

22,4

23

23,1

23,1

23,3

23,5

23,7

23,6

23,7

23,7

23,9

23,7

23,6

23,4

23,5

23,1

23

21,6

18,1

14,1

22,6

0,0128

0,025

0,0372

0,049

0,0596

0,0684

0,076

0,0852

0,0904

0,0976

0,1216

0,1184

0,1052

0,0912

0,0784

0,0668

0,0596

0,05

0,0444

0,062

0,1052

0,147

0,199

0,254

0,319

0,394

0,51

0,632

0,844

1,184

1,72

2,04

0,306

0,0097

0,0211

0,0332

0,0435

0,0560

0,0643

0,0741

0,0856

0,0979

0,0981

0,1334

0,1424

0,1483

0,1483

0,1499

0,1526

0,1512

0,1539

0,1573

0,1639

0,1772

0,2087

0,2546

0,3375

0,4776

0,4769

0,3040

0,1861

0,1347

0,0940

0,0603

0,0366

0,0244

1,3206

1,1873

1,1194

1,1252

1,0638

1,0634

1,0257

0,9958

0,9234

0,9953

0,9117

0,8315

0,7092

0,6150

0,5228

0,4376

0,3940

0,3248

0,2822

0,3781

0,5935

0,7042

0,7815

0,7524

0,6678

0,8261

1,6778

3,3951

6,2651

12,5957

28,5082

55,7216

12,5491

Figure 41 illustrates the logarithmic plot of the normalized gain of the resistive voltage probe as a function of frequency. An ideal resistive probe should have a normalized gain of 1 and an infinite bandwidth. The voltage probe tested manifested a lowpass behavior, especially between 100 Hz and 1000 Hz and a resonant peak at 14000 Hz. That implies the presence of additional capacitances in the probe, not associated with the instrumentation amplifier, which was tested previously, having a flat frequency response up to 100 kHz.

Figure 41. Normalized frequency response of the resistive high voltage probe

3.1.3 The inductive voltage probe

The inductive probe operates as a transformer in which, according to Faraday’s law of electromagnetic induction, the change in a magnetic field due to a change in secondary voltage of ignition coil becomes a proportional surge in the coil of the inductive probe. Such coil acts as the secondary winding of a transformer, and the high tension wire is the primary winding. The inductive probe reacts to variations in voltage, therefore, only AC components of voltage are detected. However, one question arises: how the inductive probe responds to different frequencies?

In the next lines, the characterization of the frequency response of the inductive probe is presented. The schematic diagram of the circuit implemented for the characterization is illustrated in Figure 42. It comprises the components: signal generator (1), audio amplifier (2), resistive probe (3), inductive probe (4), ignition coil (5), and the oscilloscope (6), similar components of the setup of Figure 40. Unlike the previous setup, the resistive probe is not powered, acting as a load for the ignition coil. The oscilloscope receives on the first channel the primary voltage U1, and on the second channel, the output of the inductive probe U2 or Usec, one at a time.

Table 7 summarizes the results of frequency response obtained with the implemented setup. The voltage Usec corresponds to the secondary winding output of the ignition coil. U2 /Usec is the gain factor of the inductive probe in [V/V].

0,1

1

10

100

10

100

1000

10000

normalized gain U2/U2c Frequency [Hz]

Figure 42. Schematic diagram of the setup for frequency response of the inductive voltage probe

Table 7. Results of frequency response for the inductive high voltage probe. Figure 43 illustrates the gain logarithmic plot of the inductive voltage probe as a function of frequency. The voltage probe tested presented a flatter response from 400 Hz, and the gain factor remained relatively flat up to the third decimal until the maximum test frequency of 20000 Hz.

Freq.

(Hz) U1 (Vp-p) Usec (Vp-p) U2 (Vp-p)

U2 / Usec (gain)

30

70

100

300

700

1000

3000

7000

10000

11000

12000

13000

14000

17000

20000

0,88

1,24

1,5

2,22

2,4

2,44

2,38

2,34

2,16

1,9

1,54

1,9

2,12

2,32

2,3

37,6

78,4

100

151,2

166

169

184

258

486

652

714

518

340

150

92

0,013

0,016

0,028

0,102

0,196

0,202

0,214

0,348

0,73

0,92

1,1

0,79

0,49

0,26

0,124

0,000345

0,000204

0,00028

0,000674

0,001180

0,001195

0,001163

0,001348

0,001502

0,001411

0,001540

0,001525

0,001441

0,001733

0,001347

Figure 43. Frequency response of the resistive high voltage probe

3.1.4 The power source U

As shown in the literature review, to promote the ion current, a power source is required. In the low-side scheme, see Figure 25, there is only one electrical loop U-e12-e2-L2-R2-Rw-Rp- Zg-Rion with three power sources, two of them e12, e2 appear temporarily from ignition and U is a sustained enabler of ion current. Common knowledge would indicate that selecting or designing a power source for an ion current measurement setup presents no hassle. However, the work of Shimasaki et al. [54] reveals the complexity of this challenge:

"...Also, the antenna effect fluctuates depending on the condition of the ion probes. It is unclear whether this is caused by the bias voltage being 200 - 300V or the configuration of the probes, but nevertheless it seems that there is a limit in the quantity of ions that can be detected...";"... How to generate DC bias voltage of several hundred volts. The problem remains on how to generate a DC bias voltage of several hundred volts. The general practise involves using the condenser on the side of the secondary...".

The pioneer work of Arrigoni et al. [91] presents the use of a battery as power source. Although such battery inherently provides an advantageous no-AC-component output, given its slow-varying electrochemical principle, it requires connecting multiple batteries in series to produce sufficient voltage U to promote ionization current. According to the literature reviewed, the required ionization voltage typically ranges from 70 V [91] to 300 V [58], [54] when using gasoline as fuel.

The work of Miyata et al. [122] adresses the use of a DC-DC step-up converter, that bootstraps from 12 V from a commercial battery to 100 V. One disadvantage of the converter is the oscillating output [163] caused by the switching of the transistor, that can induce highfrequency AC components on the microammeter resistor Rion. The works of Shimasaki et al. [54] and Dittrich et al. [117] approach the use of a classical non-regulated full-wave-rectified AC-DC power source (household AC input-Transformer-4 diodes-Capacitor) to provide the voltage output. Laganá et al. [113] added a regulator based on a transistor and several zener diodes in series (known as linear power supply) to have a constant voltage output.

0

0,0002

0,0004

0,0006

0,0008

0,001

0,0012

0,0014

0,0016

0,0018

0,002

30

300

3000

30000

Gain U2 / Usec Frequency [Hz]

Another approach consists on leveraging the ignition coil to charge a capacitor that sustains a potencial difference and several zener diodes to limit the voltage output [119], [123], [124], [130]. This approach works because the ignition coil charges the capacitor with the ignition strikes [53].

For the present work, a regulated household AC to DC linear power supply was initially implemented, like the one used in Laganá’s setup [113]. This power supply offered a variable output ranging from 30 V to 160 V. Nevertheless, an overload condition destroyed the regulator. Subsequently, a non-regulated approach with a fixed 170 V output was introduced. This setup also encountered issues, as the microammeter resistor became saturated. Another attempt included a variable transformer (VARIAC), before the transformer of the nonregulated supply. While this modification allowed for an output ranging from 0 V to 300 V, the issue of microammeter resistor saturation persisted once again.

Although the 60 Hz AC supply is, by definition, considered low frequency, it was found that he capacitance between windings of a transformer C12 (see Figure 44) leads to a small leakage current from primary winding of the transformer. This leakage current reaches the microammeter resistor, which results in the saturation mentioned earlier.

Figure 44. Explanation of spurious contribution of power source to microammeter current The current work adopted a more practical approach of charging a capacitor and maintaining its voltage, discarding the impractical battery method referenced in [91]. According to the literature, a zener diode -or a series of zener diodes- can be placed in parallel with the capacitor to limit or sustain the voltage. However, utilizing multiple zener diodes results in a limited number of discrete test voltages. Additionally, when a zener diode reaches its breakdown voltage, it becomes a low-resistance path. Below this breakdown voltage, it allows a leakage current [164]. This means that a zener diode can be modeled as a resistor in parallel with the capacitor, creating an undesirable path for the ion current, and potentially distorting the current readings (keeping in mind that the ion current is tipically in the microampere range). Consequently, an alternative approach was iimplemented.

Figure 45 illustrates the schematic of the circuit realized. The capacitor C3 is charged through capacitor C2, which, in turn, is charged by capacitor C1. Capacitor C1 receives power from the bridge rectifier, which is fed by transformer T2. To achieve variable voltage, a VARIAC T1 is placed before transformer T2. It is important to note that the capacitors are switched by dry contacts in series 13-14;23-24 and 33-34;43-44, which are actuated by two relays K1 and K2.

Additionally, there is an independent linear power supply powered by transformer T3. This supply energizes an astable multivibrator CD4047 [165] that provides three outputs. Those outputs serve as inputs to two AND gates of a CD4081 [166]. The outputs of the gates of the CD4081 provide base current to transistors Q1 and Q2, which in turn energize the coils of the relays K1 and K2 respectively.

The sequence of activation of the relays is as follows:

-Both relays are deenergized; capacitors C2 and C3 are discharged; capacitor C1 is charged by the group household AC supply-VARIAC T1-transformer T2- bridge rectifier. -Relay K1 is energized (and relay K2 is not); capacitor C1 is connected to capacitor C2 getting this last one charged.

-Both relays are deenergized, decoupling any AC 60 Hz component from bridge rectifier to capacitor C2.

- Relay K2 is energized (and relay K1 is not); capacitor C2 is connected to capacitor C3

getting it charged.

-Both relays are again deenergized and the cycle repeats itself.

Succesive cycles charge capacitor C3 up to the desired voltage. The charging of capacitor C3 due to ignition contribution is dissipated by R2 when relay K2 is energized. The switching speed is such that a cycle of charging capacitor C3 requires 1,2 s, below the 30 s charging time that comes from the ignition strikes.

The VARIAC T1 is put into a special metal box due to its size and to reduce 60 Hz induction. The capacitor C3 is put into the microammeter metal box to minimize 60 Hz induction even further.

With this scheme, the capacitor C3 constitutes the power source U needed to promote ion current. With the switching of relays K1 and K2, the capacitor C3 becomes a floating variable power supply, galvanically isolated -threefold- from AC supply.

The tests performed were satisfactory with a maximum peak-to-peak ripple of 40 mV with Rion = 100 k, G = 100 and up to 280 V, validating the use of the power source.

Figure 45. Schematic diagram of the power supply for ion current measurement.

3.1.5 The microammeter and limiter for ion current measurement

As mentioned in chapter 2, irrespective of feedback or shunt microammeter, a resistor is needed for measurements of low values of current. In multimeters, there is a very low-valued shunt resistor. For example, a Fluke 289 has 0,040  for the ampere range, 1,8  for the milliampere range and 102  for the microampere range [159]. On the other hand, Table 8 illustrates the values of Rion used by different authors in the subject of ion current detection, as a function of the mode of combustion and fuel. From Table 8 the shunts of hundreds of ohms are reserved for diesel engines, values beyond 100 k belong to the realm of HCCI engines, having then a typical range of shunts between 10 k and 100 k for SI engines. Nevertheless, there are no criteria for selecting the ohmic value of the shunt, except for the work of Chen et al. [116], in which the goal is to have a 5 V span of voltage reaching the acquisition system.

The present work proposes to use the lowest possible value of microammeter’ shunt resistor for SI engine test, based on both background noise and time-constant reduction. Therefore, an arrangement of several values of shunt resistor, in conjunction with three amplification factors is implemented.

Table 8. Values of Rion and fuels used for different authors The values selected for the shunt resistor of the microammeter are 10 , 100 , 1 k, 10 k, and 100 k. The amplification factors selected are 1, 10 and 100.

Author

Scheme Rion () Mode Fuel Estefanous [104] Indep.

100

CI Diesel George [105] Indep.

100

CI Diesel Assaad [106] Indep.

100

CI Diesel Laganá [113] Hi-side 10k SI Gasoline Shimasaki [58] Lo-side 18k SI Gasoline Hunicz [74] Hi-side 33k

HCCI

Gasoline Molina [128] Lo-side 53k SI Gas Anderson [92] Hi-side 100k SI Gasoline Daniels [110] Hi-side 100k SI Gasoline Gao [112] Hi-side 100k SI Methnol Gao [114] Hi-side 100k SI Gasoline Gürbüz [18] Indep.

150k SI Gas Panousakis [23] Indep.

200k

HCCI

Gasoline Phan [108] Indep.

241k

HCCI

Gasoline Chen [116] Hi-side 241k

HCCI

Gasoline Witze [93] Indep.

300k SI Gasoline Mehresh [107] 100k 500k

HCCI

Gasoline Gazis [21] Indep.

50k SI Gasoline

The resistor value is selected by turning the switch S1. A zener-based limiter network is put in parallel with the shunt resistor that protects the acquisition system from the ignition strike. As explained in [164], when a zener diode is reverse-biased, its PN junction has an inherent capacitance that can impair its usefulness. Thus, the limiter is implemented, according to the frequency response considerations design of [164], using 2 zener diodes, 2 high-speedswitching diodes and 2 capacitors, as shown in the scheme of Figure 46.

Figure 46. Schematic of the shunt resistor and zener limiter implemented Previous limiters with two antiparalleled 1N5819 schottky diodes and two 1N4007 diodes were tested, having lower measurements signals than expected. Further tests performed with a Fluke 289 ohmeter indicated that the limiters have an undesirable low-resistance path due to early clamping. That observation was backed up with ohms readings of 93  and 130  for the 100  and 1 k shunt resistors respectively, and 130  for upper values of the shunt resistor selected. The specifications of the diodes 1N5819 [167], and 1N4007 [168] indicate maximum forward-bias voltages of 0,55 V @ 1 A and 1,1 V @ 1 A respectively, but they allow enough current flow to lower the resistance measurement of the shunt resistor-diode limiter.

Table 9 summarizes the ohms measurements performed with the Fluke 289 ohmeter for the different shunt resistors, in conjunction with the implemented zener limiter. The range function of the ohmeter was set to manual to avoid that the ohmeter’ test voltage triggered the zener limiter clamping. The ohmeter sends a test current according to the range of measurements and its maximum open-circuit voltage is 5,2 V. Test currents, measured with another tester, were between 0,1 μA and 1 mA.

Checking the leakage current specifications of the zener [169] and 1N4148 diode [170], and as Table 9 indicates, the zener limiter has a higher impedance than the higher-valued shunt resistor Rion, as long as the test voltage is lower than 4,38 V. Therefore, for ion current measurements, the ion signal has the shunt resistor as the preferred path, and for ignition strike, the current has the zener limiter as preferred path, validating the suitability of the limiter and shunt resistors.

Table 9. Measured resistance of the shunt resistor with the zener limiter In addition, as shown in the schematic of Figure 47, using a signal generator (1), the zener limiter was tested applying a fixed sinusoidal voltage signal of 4 Vp-p, below the clamping threshold, and varying the frequency up to 100 kHz. The two channels of the oscilloscope

(3) were used to compare the input U1 against the output U2 of the limiter. The tests indicated

U1 = U2, therefore the limiter was not triggered, assuring its operation.

Figure 47. Schematic of the test circuit for frequency response of the zener limiter. The data acquisition system has a characteristic input impedance of approximately 100 k. Therefore, a coupling of impedances between the shunt resistor and acquisition system is needed. To achieve such a coupling, a non-inverting amplifier with 1, 10 and 100 gains was built. The schematic of the amplifier is shown in Figure 48. Switch S1 selects the value of the shunt resistor, switch S2 selects gain, switch S3 selects the power source U, and switch S4 selects its polarity.

Rion () Scale

500

5k 50k 500k

10

9,99

--

--

--

100

99,88

--

--

--

1k

--

985,4

--

--

10k

--

--

10129

10129

100k

--

--

--

100010

Figure 48. Schematic diagram of the microammeter, limiter and amplifier

The positions of the switch S1 to select Rion are as follows:

1: 10  2: 100  3: 1 k 4: 10 k 5: 100 k 6: custom

The positions of the switch S2 to select the gain of the amplifier are as follows: 1: G = 1 2: G = 10 3: G = 100 4: custom 5: custom 6: custom

The positions of the switch S3 to select the power supply U are as follows: 1: U = 0

2: Open-circuit

3: U from 0 V to 300 V The first position replaces the power source for a short-circuit. The second position replaces the power source for an open-circuit, and allows to use a different power supply if needed.

And the positions of the switch S4 to select the polarity of power supply U are as follows: 1: positive

2: negative

Capacitor C3 of the power supply U is placed inside the metal box of the microammeter to reduce external interference (see Figure 49). The power source of the amplifier is regulated to ±8 V, limiting amplifier output to approximately ±6,5 V, below the ±10 V of maximum input to the acquisition system.

Figure 49. Photo of the complete ion current measurement setup. From left to right:microammeter, power source and VARIAC

3.1.6 The isolation amplifiers

Figure 33 illustrates that the primary winding voltage and current signals are differential measurements with respect to ground. In [40] it is shown the use of 5 analog isolators for a classic ignition system. For the present work, only 2 analog isolators are required to measure the primary winding quantities.

Analog isolation amplifiers measure a voltage and convert it into a proportional signal for use in another circuit. They can maintain separate grounds for the input and output, allowing for isolation of up to hundreds of volts. That isolation technology is used in various applications, including electrocardiography, AC line monitoring and eliminating ground loops [171]. Analog isolators use capacitive [172], transformer [173] or optical coupling [174], [175] to transmit signals from the input side to the output side without maintaining electrical continuity between them.

Table 10. Analog isolation amplifiers Device Isolation technology

CMTI

(kV/μs) BW (kHz) Isol.

voltage (V) Nonlin.

(%)

Gain (V/V) AD215BY [173] Internal transformers for signal and power supplies.

Out. ripple:

2,5mVp-p.

Not specified

120

1500

0,005

1 to

100

ISO124

[172] Capacitive barrier, digital modulation and demodulation Out. ripple:

20mVp-p.

Not specified

50

1500

0,01

1

TLP7920

[174] Optocoupler.

Sigma-Delta modulation A/D-D/A converters.

20

230

5000

0,01

8,2

PS8551A L [175] Optocoupler.

Sigma-Delta modulation A/D-D/A converters.

28

100

5000

0,011

8

Si86IsoLi n [171]

CMOS

digital isolator.

35

100

2500

0,1

In the present work, the isolation amplifiers are used for differential voltage measurements. Table 10 presents the isolation amplifiers reviewed for the application with their specifications.

While the AD215 exhibits the best nonlinearity, it is outperformed by the TLP7920 in terms of bandwidth and isolation voltage features. Additionally, there is no data available regarding the common mode transient immunity (CMTI) of the AD215. This device is also susceptible to external electromagnetic induction [173] and comes at a higher cost. Conversely, the ISO124 provides a high input voltage range of ±10 V, comparable to the AD215, but exhibits poor bandwitdh and an unacceptable output ripple of 20 mVp-p. The Si86IsoLin boasts the best CMTI, but its nonlinearity is lacking, and the fourth-order filter in the output stage may introduce a phase lag. Ultimately, the TLP7920 device represents the best compromise for an analog isolation amplifier, making it the selected option.

Figure 50. Schematic diagram of the isolator amplifier. Stage 1 shown for primary winding voltage measurement.

Figure 50 illustrates the schematic diagram of the isolator amplifier, comprised of 7 stages. Stage 1 is itself the input of the amplifier. For the primary winding voltage U1 (recall Figure 33), it is a resistive voltage divider RA-RB that reduces the input voltage from ±400 V to ±100 mV. For the primary winding current i1 (Figure 33), RA = 0 and the shunt resistor is RB = 0,01 , allowing an input voltage of ±100 mV for an input current of ±10 A. To protect the inputs of the LF353 of Stage 2, a limiter is implemented with a 1k resistor and two opposed 3,3 V zener diodes D1 and D2. Stage 2 is an amplifier with unity gain and an input impedance of 1012 ohms. Additionally, there is an offset voltage of 1 V, provided in Stage 7 by the ZERO circuit, required by the isolator. The TLP7920 has an input impedance of 80 k [174], therefore Stage 2 is necessary to couple the impedances of Stage 1 and

TLP7920.

Stage 3 is a limiter that uses a 1 k resistor and two antiparalleled schottky diodes D3 and D4. The leakage current specification and the low forward-bias voltage of the 1N5819 [167] made it unfit for ion current signal limiting. However, for Stage 3, the low forward-bias voltage of the 1N5819 becomes useful, limiting the differential input of Stage 4 to ±200 mV, regarding that a differential input voltage above that range may put the TLP7920 into test mode or damage it.

Stage 4 is the TLP7920 device, which has a default gain of 8,2 V/V, differential input and differential output. It requires 5 V power supply in input and output sides.

Stage 5 is an instrumentation amplifier based on the AD620. Its gain can be adjusted with the variable resistor RG. The gain from Stage 2 to Stage 5 is set to 10, to have a ±1 V output. A 100 mV DC power source and a Fluke 289 were used to ensure such static calibration.

Stage 6 comprises the power supplies for the input side and output side. DC1 provides regulated outputs (+5V1, 0V1, -5V1) to Stage 7, Stage 2 and input side of Stage 3, whereas DC2 provides regulated outputs (+5V2, 0V2, -5V2) to Stage 5, and output side of Stage 3.

The power supplies DC1 and DC2 use the classical linear regulator scheme with an AC 120 V-60 Hz transformer, bridge rectifier, 2 capacitors, positive LM7805-based 5 V regulator and negative LM7905-based 5 V regulator, similar to the scheme shown in Figure 48, to avoid radiated high-frequency pulses emitted by switching power supplies which could cause interferences on stages 1 to 5.

The transformers of the power supplies and rectifiers are placed within the lower compartment of the metal box to avoid 60 Hz components on the input and output of the isolator amplifier. The metal box allocates three independent isolators with six independent power supplies, two for each isolator. However, for the present work, only two amplifiers are used, one for primary voltage, and the other for primary current. The TLP7920 is an optical isolation amplifier that uses both analog-to-digital (A/D) and digital-to-analog (D/A) conversion, featuring a digital optical link between its input and output. This setup reults in a delay between input and output signals. Figure 51 illustrates this delay in the isolation amplifier. In the test, a signal generator was set to produce a square wave for transient response analysis, while an oscilloscope was used to visualize the input (in blue) and the output (in green). The results of the test indicate a delay of 2 μs and a rise time of 2 μs.

Figure 51. Transient response of the isolation amplifer.

The ripple in the output results from Sigma-Delta demodulation of TLP7920, indicating the need for further digital processing of the signal once acquired. To test the frequency response of the isolation amplifier implemented, the signal generator is set to sinewave. Figure 52, Figure 53 and Figure 54 show the frequency responses at 10 kHz, 100 kHz and 230 kHz, respectively. The oscilloscope visualizes the input (blue) and output (green). The waveform shown in Figure 54 indicates a 230 kHz bandwidth with a signal attenuation of 70,7%.

Figure 52. Frequency response at 10kHz

Figure 53. Frequency response at 100kHz

Figure 54. Frequency response at 230kHz

3.2 Engine Crank and distribution

Under the scope of the present thesis, to obtain accurate correlations for the purposes of engine combustion diagnostics, it is essential to gather data from the ignition system variables and the processes within the combustion chamber, using crank angle as the basis for measurement. The following subsections will address the measuring and conditioning the engine's crank angle, torque, and valve position signals.

3.2.1 Crank angle sensing

An optical incremental encoder is used to measure the crank angle. For the present work, an Autonics encoder, with the model number E50S8-3600-3-T-24 was selected. The encoder is specified with a resolution of 3600 pulses per revolution, withstands a maximum rotational speed of 5000 min-1, and its maximum frequency output is 300 kHz. The encoder has three output signals A-B-Z, with a Totem-pole configuration, avoiding the use of external pull-up or pull-down resistors. The power supply for the encoder is 12 V, properly regulated with LM7812. With this regulation, all A-B-Z signals have an excursion up to 11 V. Both A and B signals generate 3600 pulses per full rotation. However, due to the unidirectionality of the engine rotation, signal B is not utilized.

The Z index signal indicates with one pulse when the encoder axle completed a full turn. For the present work, the encoder axle is aligned with the crankshaft in such a way that the pulse of the Z signal coincides with the piston geometrical top dead center. The procedure of alignment requires to remove the cylinder head and measure the movement of the piston. The acquisition system has a limited number of fast channels, and the crank angle input is limited to only one analog channel. This implies that it is not possible to send the two digital signals A and Z of the encoder to the acquisition system. Hence, a special multiplexer circuit was implemented.

Figure 55 illustrates the schematic diagram of the conditioning circuit for the encoder. One 12 V powered inverter chip CD4069 drives the LED indicators of the signals A, B and Z. Another CD4069, powered at 5 V, provides the 0 V-5 V A, B and Z buffered digital outputs. The diodes D1 and D2 create the special composite signal (A+Z) by combining the A and Z signals. When the encoder axle is not at the top dead center (TDC), the (A+Z) is identical to the digital 0V-5V signal A. However, when the encoder axle reaches TDC, Diode D1 becomes forward-biased. At this point, the Z signal from the encoder reaches 11 V. In turn, diode D1 receives 7V, which results in the (A+Z) signal reaching 6,8 V, indicating the TDC position. The (A+Z) signal is then sent to the acquisition system, which reads its amplitude. A reading 5 V - 0 V is one step, representing 0,1 degrees of crank angle displacement. On the other hand, a reading of 7 V indicates the TDC position. This is illustrated in Figure 56. The acquisition system has a maximum sampling rate of 500 kS/s and the encoder operates at a maximum fequency of 300 kHz. Thus, the maximum rotational speed of crankshaft is limited to 2600 min-1 to have recognizable encoder pulses.

Figure 55. Schematic diagram of the signal conditioner for the encoder

Figure 56. (A+Z) signal indicating 0,1° motion and TDC position.

The acquisition system presents a total of six racks of analog inputs with different sampling rates and signal formats. There are two (fast rack 1 and fast rack 2) that feature four analogvoltage channels with an acquisition rate of 500 kS/s. Additionally, another block (mid-speed rack) comprises three analog-voltage channels with an acquisition rate of 50 kS/s. An additional description of the data acquisition system will be given in section 3.3.

It was found that the data acquisition software requires a different acquisition session when the sampling rate is different. Hence, signals assigned to different sampling rates may become out of synchronization with respect to the crank angle, when triggering the data gathering. To overcome this, a monostable multivibrator CD4538 (Figure 55) is utilized to provide a copy of the index Z signal, with a longer duration, named ZZZ. Thus, the (A+Z) signal is sent to one channel of fast rack 1, becoming the crank angle position and TDC reference for the remaining seven inputs. In a similar manner, the ZZZ signal is sent to one channel of the mid-speed rack, to be the TDC reference for the remaining two inputs of it. With this arrangement, nine signals are referenced to a common crank angle position.

3.2.2 Torque sensing

The literature review indicates that ion current studies primarily focus on loaded engines. Therefore, measuring torque and crank angle is a must for evaluating and diagnosting of engine performance.

Figure 57 illustrates the schematic diagram of the setup used for torque measurement. Mechanical power is transmitted from the engine (1) to the dynamometer (2) through a belt system. The dynamometer is suspended from pivot point A, and its resistance torque causes lever A-C to rotate. Joint C is connected to the load cell (3), which measures force. The force measurement, when multiplied by the length LAC, is proportional to the engine torque value. The electrical signal generated by the load cell (3) is sent to the bridge amplifier conditioning system (4), which amplifies the signal before transmitting it to the acquisition system (5).

The dynamometer is energized by a power source (6) to provide an adjustable load torque. The dynamometer is a CFK-90 retarder, based on eddy currents, having electromagnets for the stator and a steel rotor with vanes for cooling. A DC current applied to the stator induces a magnetic field onto the rotor which gets magnetized. The slip betweeen rotor and stator generates currents inside the iron of the rotor that oppose to the stator’ magnetic field, having then a braking torque, directly proportional to the current of the stator’s electromagnets.

Figure 57. Dynamometer for load torque

Figure 58. Free body diagrams of the dynamometer and engine, for a) engine stopped and

b) engine in motion

Figure 58 presents the free body diagrams of dynamometer and engine in stopped and rotating condition of the last one. F1 is the force exerted by the taut strand of the belt, F2 is the force of its unladen strand, FA is the horizontal reaction component at the pivot A and FAy its vertical one. FAy and the weight of dynamometer W do not contribute to brake torque. F0 is the preload force of belts, obtained from FC when engine is not running. FC is the force to be measured by the load cell and M is the engine torque, obtained as:

 

0

1

*

1

1

2

2

AC C AB m d AB L F L F D R M D R R R L R R                        

Eq. 44 Where:

LAC: distance between pivot A and loadcell.

LAB: distance between pivot A and dinamometer axle.

Dd: diameter of dynamometer pulley.

Dm: diameter of engine pulley, with Dm < Dd.

R: ratio of forces between taut and unladen strands.

The ratio R can be obtained as [176]:

𝑅= 𝐹ଵ 𝐹ଶ = 𝑒௙ఏ Eq. 45 Being f the pulley-belt friction coefficient and θ the contact angle of the smaller pulley [rad].

The contact angle is given by [176]:

𝜃= 180 −57,3 𝐷ௗ−𝐷௠ 𝐿஻஽ [°] Eq. 46 With the dimensions: LAC = 520 mm; LAB =181 mm; Dd = 181 mm; Dm = 128 mm; LBD =650 mm; solving for Eq. 46, the contact angle is θ=175,3278°. From [177], the friction coefficient between aluminium and rubber approximates f = 0,5123. Solving for Eq. 45, R = 4,7812. The table in [176], confirms that R = 4,78 when θ = 175.

Figure 59 illustrates the schematic diagram of the signal conditioning circuit and visualization for the load cell, specified to 350 . The first stage is an instrumentation amplifier AD620 [155] with resistors R5 and R6 for gain and R7 for zero adjustment. The output signal of AD620, limited between -5 V and +5 V is connected to the data acquisition system. The positive excursion of voltage of AD620 output occurs when load cell operates at compression, whereas negative values of AD620 output correspond to traction force exerted onto the load cell. In addition, there is a derivation for another circuit with D1, D3, C1, three resistors of value R3 = 10 k, and the dual op-amp LF353 that conform a precision absolute-value circuit [178], intended to send a positive 0 V – 5 V voltage signal to the analog input of a microcontroller μC and an LCD screen, equal in magnitude to the output of the AD620.

Figure 59. Schematic diagram of the loadcell signal conditioning circuit.

The microcontroller μC converts the absolute voltage of LF353 output to a digital word, and this in turn to the force value Fc. With the parameters of Eq. 44 preloaded, the engine torque M is gathered and presented in the LCD screen. The value F0 is preloaded as a parameter into μC, once the belts are installed, being nil without the belts. Although the voltage signal sent to μC is always positive, it is likely to display negative torque with stationary engine, due to the accomodation of the belts into the pulleys. Therefore, for each test, it is necessary to consider the values of torque indicated with the engine not running and motored briefly by the starter motor, to use them as baseline.

The acquisition system receives the bipolar voltage signal from the AD620 and a similar process based on Eq. 44 is performed, with the advantage of possible negative excursions of torque signal.

The load cell has a full-scale sensitivity of 2,9995 mV/V, it’s excited with 10 V, and under a nominal compressive force of 4448 N, the output is 34,96 mV. The gain of the AD620

amplifier is adjusted to 143 with Rg = R5+R6 = 347,9  to achieve 4,9993 V output with 4448 N of load.

Load cell calibration. To calibrate the load cell amplifier module, digital voltmeters are connected to the circuit, as shown in Figure 61. Voltages Usig, Uexc, Uamp are measured with dedicated Fluke 289 multimeters, one for each signal, whereas Uabs is measured with a Fluke

116.

Initially, known loads are applied onto the loadcell, using an ELECAV Materials Testing Machine that can exert compressive and tractive load, and an INST-P3 load cell calibration equipment. The results of the calibration (compression) are shown in Figure 60.

Figure 60.Calibration of loadcell using INST P3

Figure 61. Test points for Usig,, Uexc, Uamp y Uabs y = 0,9953x - 2,984 R² = 1

-500

0

500

1000

1500

2000

2500

3000

0

500

1000

1500

2000

2500

INST P3 indication [N] Load of Testing machine [N]

Next, two load cells are put together inline, one connected to the INST-P3 equipment and the dynamometer load cell to the amplifier module. The load is exerted with the ELECAV Materials testing machine. The results of calibration of Usig are shown in Figure 62.

Figure 62. Calibration of Usig : a)compression, b)traction Another calibration tasks relate to the amplified output of the AD620 UAmp and absolute value circuit Uabs. Figure 63 indicates the output as function of load, whereas Figure 64 presents the calibration of the gain of the amplifier (144,74 V/V) and the gain of the absolute value circuit (131,19 V/V).

Figure 63. Calibration of UAmp : a)compression, b)traction y = 0,0067x - 0,1673 R² = 1

-5

0

5

10

15

20

0

500

1000

1500

2000

2500

U_sig [mV]

a) Load Fc, Inst P3 [N]

y = -0,0067x - 0,416 R² = 1

-25

-20

-15

-10

-5

0

0

500

1000

1500

2000

2500

U_sig [mV]

b) Load Fc, Inst P3 [N]

y = 0,001x - 0,1013 R² = 1

-0,5

0

0,5

1

1,5

2

2,5

0

500

1000

1500

2000

2500

U_Amp [V]

a) Inst P3 [N]

y = -0,001x - 0,1376 R² = 1

-3,5

-3

-2,5

-2

-1,5

-1

-0,5

0

-500

500

1500

2500

U_Amp [V]

b) Inst P3 [N]

Figure 64. Calibration of a)AD620 amplifier gain, b) absolute value circuit gain

3.2.3 Valve displacement sensing

A module that monitors the linear motion of intake and exhaust valves of the engine is developed to correlate the ion current signal traces with the position of the engine valves as functions of crank angle. In addition, further works for estimation of intake and exhaust gas mass flow rates, and related mathematical models can be fed from the displacements of the valves.

A candidate for measurement of linear displacement is the Linear Variable Differential Transformer, which is a broadly known type of sensor, due to its high accuracy and resolution, as well as its robustness [179]. A LVDT comprises a centered primary winding and two secondary windings at both sides, those connected in series with opposite instantaneous polarities, and a movable ferromagnetic nucleus attached to a nonferromagnetic rod. When the ferromagnetic nucleus is centered with respect to the three-coil arrangement, the total output of secondary windings is close to zero. That residual amplitude is caused by parasitic primary-secondary capacitances and non-symmetries between windings [180]. The relationship between the amplitude of secondary voltage vs and displacement l is given as [179]:

𝑣௦= 𝑘௧(1 −𝑘௡𝑙)𝑙

Eq. 47 Where kt and kn are the sensitivity and nonlinear coefficients, respectively. Songsuwankit et al. [179] state that the typical linear operating range of a commercial LVDT is around 10% to 20% of its stator case, having then a long structure in comparison with the stroke. In addition, to convert from motion to voltage signal, the usual approach is an Amplitude Modulation with Suppressed Carrier (AMSC) technique [179], [180].

For the present work, two variable transformers are used, with a configuration of one primary and one secondary winding for each one, maintaining the ferromagnetic nucleus and a nonferromagnetic rod of a typical LVDT, considering that the motion of the nucleus and the y = 144,74x - 0,0771 R² = 1

-0,5

0

0,5

1

1,5

2

2,5

-0,0005

0,0045

0,0095

0,0145

U_ Amp [V] U_Sig[V] y = 131,19x + 0,0616 R² = 0,9948

0

0,5

1

1,5

2

2,5

-0,0005

0,0045

0,0095

0,0145

U_ Abs [V] U_Sig[V]

secondary voltage will be proportional between each other. The non-ferromagnetic rods are connected to the valve spring retainers of intake and exhaust valves of the Changfa engine. The motion of the valve spring retainers causes a 1-to-1 motion of the nuclei of the transformers, and their secondary voltage outputs are proportional to the former ones.

With the 2-coil configuration, the length of the case of the variable transformer is 2/3 of the one of an equivalent LVDT, having a more compact structure.

Figure 65 illustrates a simplified version of the system with the transformers TR1 and TR2, having their nuclei attached to the intake valve and exhaust valve respectively. There is a signal generator uexc that provides a 6 Vp-p, 10 kHz sinusoidal wave to both primary windings. The outputs of TR1 and TR2 are rectified to convert them to DC voltages uintk and uexh as function of intake and exhaust strokes respectively.

Figure 65. Simplified schematic diagram of the valve opening sensing system

Figure 66. Detailed schematic diagram of the valve opening sensing system

A more detailed scheme of the electronic circuit for primary signal generation and secondary signal rectification is presented in Figure 66. The generation of the sinusoidal waveform is performed by means of an XR2206 chip, which has an adequate power supply sensitivity of 0,01% and a THD for sinewave of 0,5% [181]. The output of the XR2206 is sent to an LM386 amplifier, with a default gain of 20 V/V [154], that in turn provides excitation to both primary windings of the variable transformers.

The secondary windings of the transformers send inputs to half-wave precision rectifiers based on LF347 quadruple operational amplifier, 1N4148 diodes, 220 nF capacitors and 10

k resistors. The outputs of the rectifiers go to additional non-inverting gain stages, having the voltage outputs uintk and uexh for intake and exhaust stroke respectively.

Figure 67. Photos of the transformers installed on the valve cover and circuit board. The power supply for the XR2206 and LM386 chips is provided by a linear 12 V regulated power supply. The LF347 op-amp is powered on the positive side by the referred 12 volts, and for the negative side an LM7805 regulator feeds an ICL7660 that inverts to -5 V. Figure 67 shows photos of the circuit board implemented and the transformers’ setup attached to the valve cover. The strokes of both valves are 11mm, so all the 4 windings shown have a spool height of 11mm. The transformers’ structures have a total height of 25mm.

Although Yañez et al. [182] provides a selection of materials for nucleus, the same work indicates that ferrite has low tensile strength. The valve motion can exert quite strong linear acceleration on the nucleus. From the works of Guo et al. [183] and Zheng et al. [184] at a crank regime of 2400 min-1, a peak of acceleration of 1652 m/s2 is expected. Therefore, for the nuclei, two regular 8.8-grade partially-threaded M3 Allen screw rods were used. The nonferromagnetic joints to the valves are brass rods.

To validate the transfer characteristics of the nuclei, a set of frequencies were applied, using the frequency adjustment of the trimmer connected to the seventh pin of the XR2206. The amplitude was fixed to 300 mVp-p with the trimmer attached to the third pin of XR2206 (see Figure 66), having 6 Vp-p input for primary windings.

Figure 68 illustrates the test results of the peak of rectified output of the intake stroke measurement transformer as function of the displacement, for frequencies of 3 kHz, 6 kHz, 10 kHz and 20 kHz. Although the lower excitation frequencies offer a more linear response, the 10 kHz input exhibits less attenuation than the 20 kHz one. In addition, to get less ripple on the rectified output, a higher frequency is preferred. Thus, the 10 kHz excitation is selected, both for intake and exhaust stroke measurement transformers.

Figure 68. Tests of frequency for transformer of intake valve

Figure 69. Photo of the setup to measure the displacement of the nucleus Figure 69 shows the mockup used to characterize the rectified voltage output of the stroke measurement transformers, as function of the motion of the nuclei. For the displacement, a regular micrometer is used. For voltage measurement, a Fluke 289 is used (not shown). The results of characterization of the intake and exhaust valves measurement transformers are shown in Figure 70 and Figure 71 respectively.

u = -0,0216x2 + 0,618x + 2,1804 R² = 0,9957

0

1

2

3

4

5

6

7

8

0

2

4

6

8

10

12

14

Peak Rectified Out, u [V] Stroke, x [mm]

10 kHz

6 kHz

3 kHz

20 kHz

Figure 70. Characterization of the intake valve stroke measurement

Figure 71. Characterization of the exhaust valve stroke measurement Once installed in the engine the valve stroke measurement systems, a dynamic test is performed. As shown in Figure 72, the output of the circuit (blue, in volts) has a 10 kHz ripple due to the half-wave rectification. With a digital filtering, using the upper envelope of the blue waveform, that is, the peak voltage, and a conversion based on the aforementioned characterization performed, the stroke waveform (red, in mm) is obtained. The sample waveforms of Figure 72 correspond to the exhaust valve stroke, at a load of 14 N•m and a regime speed of 1900 min-1. Similar waveforms were found for the intake valve, validating the valves’ stroke measurement system.

3

3,5

4

4,5

5

5,5

6

6,5

7

0

2

4

6

8

10

12

Peak Rectified voltage [V] Stroke [mm]

2

2,5

3

3,5

4

4,5

5

5,5

6

6,5

7

0

2

4

6

8

10

12

Peak Rectified Voltage [V] Stroke [mm]

Figure 72. Signal acquired for the exhuast valve (blue) and digitally treated opening stroke (red)

3.3 Acquisition system

There are acquisition systems with analog low-level multiplexing, where an analog selector switch (mux) routes the signal from one sensor at a time to a chain of functional blocks: amplifier→lowpass filter→sample&hold→analog/digital converter, having a delay between channels proportional to their number. There are also high-level multiplexing systems, in which every channel has the chain: sensor→amplifier→lowpass filter, before the mux that sends one conditioned signal at a time to a single chain sample&hold→analog/digital converter. On the other hand, acquisition systems with digital multiplexing have for each channel a dedicated chain of blocks:

sensor→amplifier→lowpass filter→sample&hold→analog/digital converter. Every A/D in turn is connected to digital interfaces that do the digital throughput to a common bus that eventually sends the information of each channel to a tabulated register [185].

The acquisition system used in the present work is a National Instruments NI-cDAQ9178, which has the digital multiplexing architecture.

For verification purposes of the delay between channels, a multiple input test is performed.

Figure 73 displays photos of the multiple input setup that comprises a Snap-On signal generator (1), the National Instruments NI-cDAQ9178 acquisition system, that has 6 acquisition racks connected to its chassis (2), a SIGLENT SDS-1104 oscilloscope (3) to check the output waveform of (1), and a PC (not shown) connected to the acquisition system. The PC has preinstalled libraries to exchange information by means of the software MATLAB® with the acquisition system.

Figure 73. Setup for delay between channels, a) complete. b) detail of connections

The vibration signals rack (4), the slow signal rack (5) and the thermocouple input rack (6) are not connected to the signal generator, since vibration rack (4) is not used and the racks

(5) and (6) have a very slow acquisition rate. Nevertheless, they remain plugged to assess if

their presence affects the multiple-input acquisition campaign.

The mid-speed rack (7) has three channels with an acquisition rate of 50 kS/s and voltage input limits between -10 V and 10 V.

The fast rack 1 (8) and fast rack 2 (9) have four channels each, with an acquisition rate of 500 kS/s and voltage input limits between -10 V and 10 V.

A schematic connection diagram is shown in Figure 74. Conceptually, all channels are connected in parallel, receiving the same signal from the generator. However, Figure 73.b) shows a fence wiring, for each one of the signal generator terminals, to guarantee that the wires that reach each channel have approximately the same length and reduce any delay. The signal generator vs is set to 2 Vp-p square wave, zero offset, 50% duty cycle and test frequencies of 1 kHz, 10 kHz, 100 kHz.

A square waveform justified by its variations from -1 V to 1 V and vice versa constitute a close approximation to the slope of a step signal, a demanding input for any electronic system. Figure 75 and Figure 76 display a sample of the waveforms of the test results of the fast rack 1 (8), channels 1 to 4, and fast rack 2 (9), channels 5 to 8.

Figure 74. Schematic diagram of the multiple acqusition system.

Figure 75. Superimposed channels 1 to 8, at f=1kHz. Horizontal axis in “samples”.

Figure 76. Details of the superimposed channels 1 to 8, at f=1kHz. a) full rising slope, b) detail of rise. Horizontal axis is in “samples”.

The elbow of Figure 76.b) indicates that there is no delay between channels, as the rising slope begins on sample 698632 for all channels. Similar results were found at 10 kHz and 100 kHz, validating the use of the acquisition system.

3.4 Conclusions from the implementations

This chapter presented the proprietary setups and tests to validate the subsystems required for a trustable ion current sensing, as well as primary and secondary variables, and valve displacements. In addition to the validation of the usability of the subsystems, there are other outcomes that can be mentioned:

 A proper amplifier, with an adequate heat sink, must be selected to excite the primary winding of the ignition coil, specially when the characterization of the coil is done at low frequencies.

 It is important to check previously if the ignition coil is of transformer-type. An internal connection between primary and secondary terminal could lead to damage to the driver amplifier that excites the primary side or false readings.

 The characterization of the frequency response of the ignition coil allowed to test the resistive voltage probe, as well as the inductive probe, leveraging the resonant increase in transformation ratio above 8 kHz due to winding capacitances, helping it to obtain high voltages at high frequencies for the tests.

 Although the specification of the resistive voltage probe indicated a 20 MHz bandwidth, it was necessary to check it, since previous HV tests with a BK-Precision probe led to its failure, even with a thorough cleaning of any oily substance along its stem and flux residues on the solderings. Very few commercial HV probes offer specifications of the bandwidth, and this work offered a setup to check their real frequency response characteristics.

 The Cockroft-Walton multiplier implemented used HV capacitors with a low capacitance value in relation to the excitation frequency, 60 Hz, having less DC voltage than expected. It was not easy to find capacitors with higher values, specified for 2 kV or more. It is needed to consider the effect of the impedance of the load connected to the Cockroft-Walton multiplier, in this case the resistive probe.

 This work provided a setup to get a variable power supply for ion current measurement, applied to a Low-side scheme, in which both terminals of the power supply are galvanically isolated with respect to the ground. With the scheme of the power supply implemented, undesired contributions of leakage current that come from an AC outlet are avoided.

 It was found that, even at 60 Hz, a transformer can “push” displacement current from primary to secondary side, due to inherent inter-winding capacitance. As the impedance of a capacitor is 1 / 2π f C, a switching power supply, which has switching frequencies in the range of kilohertz, was not considered due of the lower impedance path of the interwindings.

 The tests of signal limiters must consider the leakage current of the semiconductor components, specially when attached to microammeters, given that the former can affect the ohmic value of the shunt resistor.

 The frequency response of the semiconductor limiter must be tested, as diodes do have reverse-bias capacitance, that can offer an undesired low-impedance path. The limiter must be designed to avoid alterations in frequency response of the component that is clamped by it.

 This work provided a microammeter setup to measure ion current and to assess whether the ohmic value of the shunt resistor influences the features of ion current.

 The isolation amplifiers implemented allowed for the input of a non-grounded signal to the acquisition system, to find possible correlations and outcomes from the lowerscaled primary voltage, reflex of the secondary side one, and to measure the primary current. Nevertheless, it is necessary to consider the demodulation ripple from the TLP7920 analog optocoupler, as well as its internal processing delay.

 There are analog isolation amplifiers that use diode-photodiode pairs, and no digital modulation-demodulation stages, with a promise of 20 MHz bandwidth. However, they are scarce and require complex feedback circuits, external to the optocouplers. In comparison, the TLP7920 is a one-chip solution.

 The filtering for the isolation amplifiers outputs must be digital, and has to take into account the delay of the TLP7920 optocoupler and the digital sampling delay of the filtering algorithm itself.

 The setup for torque measurement implemented in this work has elastic belts that can display a variation in the load cell signal due to whiplashes. Therefore, it is necessary to determine the rms value, maintaining constant load. In addition, the baseline value at standstill is required for the measurement.

 At cranking speeds, the valves open and close slowly, having with the variable transformer layout for valve stroke measurement closer values of crank angle positions of both valves, compared to those presented in the engine manual. For higher regimes, the dynamics of the valve train mechanism gets compromised.

Test procedure and preliminary findings The previous chapter presented the instrumentation of the engine, including ignition variables, ion current components, crank angle, torque and valve stroke measurement, providing a baseline of what can be expected from the different sensors installed, in terms of static measurements, frequency response, behavior of semiconductors for signal limitation, requirements of digital filtration to gather information from the signals.

In the present chapter, the procedures for testing are presented. In the first section, there they come the preliminary considerations for the tests: the maximum rotational speed to achieve considering the sampling limitation of the acquisition system, the spark advance angles, the microammeter settings, the ion promoter power source values chosen, the number of consecutive cycles for averaging, the variables that can be acquired simultaneously and the limitations of the test campaigns for engine motoring and combustion. The second section describes the acquisition routine implemented and the first two stages of offline postprocessing tasks. The third section presents the preliminary findings of motoring tests for the acquisition, first analyzing the effect of the promoter power source alone and then evaluating the additional contribution of ignition to the ion current signal. Then, the final part presents the tests for combustion conditions with the correspondent preliminary findings, when adding the effect of fuel breakdown on ion signal.

A birdseye view of this chapter is shown in the scheme of Figure 77.

Figure 77. Schematic view of the chapter

4.1 Preliminary considerations for the tests

The Changfa engine ignites the fuel by compression from factory, and includes a small generator specified to replenish the battery once it has delivered energy for cranking. As the engine was modified for spark ignition, the additional electronic components draw an additional current from the generator. The electronic battery charger used to replenish the battery comprises a switching power supply with current and voltage limiting, as well as additional pulsed charging to improve the battery charging. It was found that the combustion chamber pressure signal presented spikes when the battery charger was plugged to the AC outlet. Therefore, the battery charger must be connected to the battery only to recharge it after the tests.

The encoder used in this work to capture the crank angle signal can withstand a rotational speed of 5000 min-1, and generates 3600 pulses per revolution on channels A and B, having them a maximum electrical frequency of 300 kHz. The crank angle square wave is the fastest signal to be gathered. Now, the acquisition system has a maximum sample rate of 500 kS/s for certain channels. The remarks about Nyquist-Shannon sampling theorem (sampling frequency should be at least twice the maximum expected input frequency) presented in [186] state that the Nyquist criterion is not sufficient in certain cases, which could lead to consider expensive and memory-demanding oversampling that still can result impaired to gather important features of the signal. Additionally, it can take place an undersampling for choosing too low sampling frequency or inconvenient aliasing phenomena. Hence, all the engine tests are limited to a maximum rotational speed of 2600 min-1, to have recognizable crank angle steps. The square pulse signal acquired, shown previously in section 3.2.1, Figure 56, indicates that the maximum rotational speed criterion is good enough for encoder signal reconstruction.

The spark advance can be set manually, using a specially designed mechanical adjustment, shown in [12]. As the engine has been modified for spark ignition, it is necessary to set limits of spark advance, looking for a reasonably stable operation – for a single-cylinder engine -, avoiding knock whenever possible and providing the most mechanical power. Tests performed on the engine allow to establish a range of spark advances between 10° and 25°.

The sensitivity of the ammeter for ion current measurement has 5 shunt resistor values and 3 gain values. The product Rion • G represents the sensitivity in terms of voltage signal due to the current input. The present work opts for using Rion • G as the factor of the ammeter, expressed in V/A, looking for the use of lower Rion values, but with an amplification to get a factor closer to the values of Rion presented in literature, between 10 k and 100 k for spark ignition engines. Certain combinations of Rion and G will not fall into that range of factors. Thus, only certain resistors and gains are selected. The combinations Rion = 10  and gains 1, 10 and 100 represent a very low factor, and the microammeter shunt resistor present in multimeters starts from 100  [159]. Therefore, settings with Rion = 10  are discarded.

Table 11 presents the combinations of Rion and G selected for ion current tests. Values of Rion beyond 10k are less favourite, although tested.

Table 11. Combinations of Rion and G for the tests

The voltage of the ion current promoting power source U is also a parameter to vary on the tests. Table 12 shows the values used for different authors. As referred before, the power source voltage ranges between 70 V and 300 V for spark ignition engines.

Table 12. Values of U, schemes and fuels used for different authors

Rion () Gain Factor (V/A)

100

100

104

1k

1

1000

1k

10

104

1k

100

105

10k

1

104

10k

10

105

10k

100

106

100k

1

105

100k

10

106

Author

Scheme U [V] Mode Fuel Estefanous [104] Indep.

25,

50,

75,

100

CI Diesel Laganá [113] Hi-side

50,

150,

300

SI Gasoline Shimasaki [58] Lo-side

300

SI Gasoline Hunicz [74] Hi-side

200

HCCI

Gasoline Arrigoni [91] Indep.

70

SI Gasoline Anderson [92] Hi-side

300

SI Gasoline Daniels [110] Hi-side

150

SI Gasoline Gao [112] Hi-side

400

SI Methnol Gao [114] Hi-side

12

SI Gasoline Gürbüz [18] Indep.

64

SI Gas Panousakis [23] Indep.

200

HCCI

Gasoline Phan [108] Indep.

237

HCCI

Gasoline Chen [116] Hi-side

237

HCCI

Gasoline Witze [93] Indep.

300

SI Gasoline Abdel-Rehim [89] Indep

-380

SI Gasoline

Section 2.5.5 of this work commented on the magnitude and sign of the ion promoter source U. Arrigoni et al. [91] shows in appendix C-4 a directly proportional relation between ion current magnitude and ion promoter voltage, where the slope is steeper for positive polarity, meaning that there is more ion current if the inner electrode is positively biased. Similar outcome is found in the later work of Wilstermann et al. [132], resembling a similarity to a V-I semiconductor diode response curve [168]. Additionally, in [91], appendix C-4 indicates curves traced from a minimum value, for both polarities, around 50 V to 70V, regarding that requirement of voltage to promote ion current.

For this work, the test voltages are narrowed to the values: -300 V , -200 V, -100 V, 0 V, 100 V, 200 V and 300 V. The 0 V value provides a baseline for the ion current waveform in absence of biasing. The values 100 V, 200 V and 300 V are applied with both polarities to check for results with positive and negative bias. The minimum of 100V helps promotion of ion current, based on criterion of [91].

A mapping-programmable MicroSquirt® Electronic Control Unit (ECU) is used for port-fuel injection. The ECU receives data from the Throttle Position Sensor (TPS), the Manifold Absolute Pressure (MAP) sensor, and crank position (CKP) captured with the pickup coil. The software TunerStudio MS Lite is used to read and set the time-duration map for the injection electrical pulse. The ECU tuning is not shown in this work. More information on the tuning, and calibration of MAP sensor, is referred in [12].

Additional tuning of the ECU can be performed during the combustion tests, with the goal of maintaining the AFR as close as possible to the stoichiometric condition (14 to 15). A wideband UEGO oxygen sensor with AFR indicator display and voltage output, is put into the exhaust, to acquire the AFR simultaneously with the rest of variables. The presence of this sensor is justified by the known effect of concentration of fuel in mixture in the magnitude of ion current, as explained in [126], [151].

The throttle opening is adjusted manually and should be fixed during the tests, performing minor adjustments to compensate when the engine heats up.

The tests are performed aiming for fixed operation points of rotational speed, load, throttle opening, AFR, checking regularly for engine temperatures, and triggering the signal acquisition once thermal equilibrium is observed. The acquisition system gathers information of a certain number of consecutive cycles once it is triggered from the computer. The gathering of several cycles, as shown in Martin [187], allows getting initial average cycle waveform datasets to minimize the measurement uncertainty related to the cyclic dispersion. The averaging reduces the undesired varying amplitudes in an equivalent to the square root of the number of cycles, and also acts as a lowpass moving-average pre-filter [188]. Abbad [188] states that there is no consensus around the minimum number of combustion cycles required for averaging. For CI engines, 20 cycles [189] and 25 to 50 cycles [187] are mentioned, while for SI engines, 40 cycles [190] to 300 cycles [191] are gathered. Molina et al. [128] indicate 300 cycles for SI engine fueled by natural gas. The higher number of cycles for SI engines is justified on higher cyclic variation [188]. Abbad [188] indicates that the average cycle waveforms obtained require additional filtering to remove spurious peaks. For the present work, data processing after acquisition is

performed offline because, although the computer used to communicate with the NI-DAQ acquisition system presents a solid-state hard drive, faster than the mechanical ones, it is still unable to realize the post-filtering and visualization of the variables along with the acquisition.

The acquisition triggers a data capture limited to 5 seconds, obtaining: one matrix data with 2 500 000 rows and 8 columns for the eight fast channels, one data matrix with 256 000 rows and 3 columns for the three mid-speed channels and another data matrix with 15 rows and 4 columns for the four thermocouple channels. Once the acquisition ends, each test occupies typically 69MB of space on hard drive.

With this approach, the number of cycles is:

𝑁௖௬= 𝑇௔௖௤ 𝑛

120

Eq. 48 Therefore, with an example of rotational speed of n = 1900 min-1 and an acquisition time of Tacq = 5 s, an approximate of 79 consecutive 720-degree cycles are gathered on combustion tests. On motoring tests, with n = 500 min-1, an approximate of 20 cycles are piled up into the respective archive, having a compromise of acquisition time and memory stacking, and an acceptable averaging, in particular for combustion chamber pressure signal. The main variables to gather are: combustion chamber pressure, intake pressure, primary winding voltage, primary winding current, intake and exhaust valve displacement , secondary winding voltage with resistive probe, secondary winding voltage with inductive clamp, ion current, torque, AFR, crank angle.

Intake and exhaust valve displacement signals can not be gathered at the same time due to the limited number of fast channels. Therefore, two switches are used. The Switch 1 selects between primary winding voltage – intake valve stroke, whereas Switch 2 toggles between primary winding current – exhaust valve stroke. Tests for a determined operation point should be made twice, once for primary coil and then for valve displacement variables. Auxiliary variables are the temperatures of intake, exhaust gas, exhaust pipe, engine oil, captured by means of thermocouples and sent to the dedicated thermocouple rack of the acquisition system. These variables help determining steady conditions of the engine.

Table 13 shows the distribution of input variables connected to the acquisition system. The AnalogV type corresponds to signals limited to a range of -10 V to 10 V. K therm corresponds to K-type thermocouple inputs. Acc-mV and An-mV correspond to millivolt-level accelerometer and load cell type signals, not used in the present work. The TDC signal from ZZZ produces pulses only when the crankshaft is turning, that is, remains at 0 V in standstill, due to the edge triggering of the CD4538 monostable (recall Figure 55). Additionally, the ZZZ signal is used as input for an external tachometer.

The intake manifold sensor (MAP) measures an absolute pressure. SecV1 is the secondary voltage measured with the resistive probe, whereas SecV2 is the voltage measured with the inductive Pearson coil.

Table 13. list of variables and assignment of channels and racks in the acquisition system Before each test, the electrical connections are checked and two sample acquisitions with the engine at standstill are performed, having one set that includes primary coil variables and then the second set that comprises valve opening. The acquisition of variables for all the fast channels and the mid-speed channels provide vectors of values expressed in volts, whereas the thermocouple channels indicate the information directly in degrees Celsius. Then, for the vectors expressed in volts, the offline data processing must include a first stage of scaling to the correspondent units. More on the data processing is explained in section 0. Raw voltage from cylinder pressure signal should indicate 0 V, as well as signals SecV2, Ion current, Primary voltage, Primary current, ZZZ. The (A+Z) signal should indicate either 0 V, 5 V or 7 V; the MAP signal should indicate more than 3,3 V. SecV1 signal should indicate a DC scaled value in accordance with the sign and magnitude of the voltage of the ion current Variable Name Type Rack Channel Sampling rate (kS/s) Cylinder Pressure P AnalogV Fast 1

1

500

Crank angle+TDC (A+Z) AnalogV Fast 1

2

500

Intake pressure

MAP

AnalogV Fast 1

3

500

SecV2 (clamp) U2HF AnalogV Fast 1

4

500

SecV1 (resistive) U2LF AnalogV Fast 2

1

500

Ion current I2 AnalogV Fast 2

2

500

Primary voltage / Intake valve stroke U1/Intk AnalogV Fast 2

3

500

Primary current / Exh. valve stroke I1/Exh AnalogV Fast 2

4

500

TDC

ZZZ

AnalogV Mid-speed

1

51,2

AFR

AFR

AnalogV Mid-speed

2

51,2

Torque M AnalogV Mid-speed

3

51,2

Intake temperature T1 K therm.

Thermocouple

1

0,003

Oil temperature T2 K therm.

Thermocouple

2

0,003

Exh. Temperature1 T3 K therm.

Thermocouple

3

0,003

Exh. Temperature2 T4 K therm.

Thermocouple

4

0,003

Not used Acc-mV Vibration

1

51,2

Not used Acc-mV Vibration

2

51,2

Not used Acc-mV Vibration

3

51,2

Not used Acc-mV Vibration

4

51,2

Not used An-mV Load cell

1

0,003

Not used An-mV Load cell

2

0,003

Not used An-mV Load cell

3

0,003

Not used An-mV Load cell

4

0,003

promoter power source. For example, a setting of U = 200 V must lead to an indication around 0,02 V, given the 10000 times attenuation of the probe. The Torque signal is usually not zero, due to the pretensions of the belts in the pulleys. The values for intake and exhaust valve stroke signals must be between 1,8 V and 6 V. The values of Thermocouple channels should be referenced to the ambient temperature, unless there are previous warming cycles of the engine. Those preliminary acquisition tasks assure the baseline for the next measurements.

4.1.1 Limiting the acquisition campaigns

Given that there are fifteen engine variables, seven values of ion promoter power source, nine combinations of the Rion•G factor, several possible spark advances, loads, rotational speeds, it is impractical to consider a multidimensional array of tests.

This study proposes a set of test groups aimed at progressively acquiring operational knowledge of the implemented system based on motoring and combustion tests. Figure 78 presents the basic settings sheet, showing the engine the throttle opening, the spark advance limits, as well as the defined loads and operating regimes. For the ion current measurement system, it includes the Rion settings, gain, promoter source voltage U, and the acquisition of primary winding or valve displacement variables.

The motoring tests are conducted without spark contribution in order to evaluate the ion current detector’s sensitivity to non-combustion events. Then, that sparking contribution is added, to establish normal ignition waveforms present in ion signal. Next, correlations between ignition variables and ion current are performed. Another subject includes the valve opening strokes related to ion current. Correlations between secondary voltage and ignition variables, especially ion current, are explored. The dynamometer is not connected to the engine, hence the “N.A.”, not applicable, is correspondingly set on load. Indeed, the motoring pressure waveform constitutes a crucial input for combustion tests, but also to correlate with the ignition variables. Still, the motoring tests are limited to some specific ion factors and promoter voltages, given that the biggest part of tests belongs to combustion, and the finding of some limitations during the development of this work. The preliminary findings in tests performed when motoring are analyzed in section 4.2.2. Combustion tests are dominant, looking for correlations between ion current and pressure. The effect of the shunt resistor of the ammeter, the ion promoter source, the detection of abnormal combustions, are explored in section 4.4. Initial combustion tests were performed ignoring or not considering the small prechamber designed to vary the compression ratio of the original engine. In that condition, the spark plug was 9 mm inward with respect of the plane of cylinder head. Then, another group of tests were performed with the spark plug protruding to the combustion chamber 2 mm with respect of the plane of cylinder head. Another limitation of the test campaigns comprises the spark gap distance, held constant at 0,9 mm. Environmental conditions are not controlled, yet observing an absolute pressure of

86kPa, room temperatures between 22 °C and 27 °C, and relative humidities between 58% and 66%. The AFR was kept between 14,0 and 15,0 by tuning the MicroSquirt.

Figure 78. Test campaign schematic

Table 14. Motoring tests. (*) indicates no prechamber.

Table 14 illustrates the parameters successfully evaluated in motoring conditions. There were more motoring tests performed. However, data were invalid in some cases, or the motoring was made with VFD, having poor SNR, which is explained in section 4.3. All the motoring tests shown in this chapter were performed at cold start, with an ambient temperature of 25 °C ± 2 °C, 86 kPa absolute atmospheric pressure, and a relative humidity of 64 % ± 4 %. Table 15 comprises the values successfully assessed in combustion conditions, with a prechamber. Most of the tests were performed at 1900 min-1, with some samples at lower speeds and low loads. The seven settings of U were swept, as well as Rion and G, having factors from 103 to 106 and tackling various combinations of the same factor. This approach is done to find the best setting candidates of the microammeter and power source, to have additional observations. Unfortunately, 23 combustion tests with prechamber were found invalid due to abnormalities in interpretation of encoder signal by the coding, not detected at the initial visual validation of waveforms after acquisition.

The load settings are light (1 N•m to 6 N•m) and loaded (13 N•m and 15 N•m). Table 16 indicates the values successfully tested in combustion conditions, this time without the prechamber. Again, the most frequent regime is 1900 min-1, with minor explorations at lower speeds. No abnormalities in encoder signal were found. The promoter U settings are narrowed as well as Rion and G, being more frequent the 200 V position and the factor 104. The load settings are again light and loaded.

Throttle

(%)

Spark Adv. (°) Load (N•m) Regime (min-1) Rion () Gain U (V) Var

87

No Sp.

N.A.

500

100k

10

200

Valv*

87

No Sp.

N.A.

500

100k

10

300

Valv*

87

20

N.A.

500

100

100

100

Valv

87

20

N.A.

500

100

100

100

Prim

87

20

N.A.

500

10k

1

-100

Prim

87

20

N.A.

500

1k

1

100

Valv

87

20

N.A.

500

1k

1

100

Prim

87

20

N.A.

500

10k

1

200

Prim

87

20

N.A.

500

10k

1

200

Valv

87

20

N.A.

500

100

100

200

Valv*

87

20

N.A.

500

1k

10

200

Valv*

87

20

N.A.

500

10k

1

200

Valv*

87

20

N.A.

500

100k

1

200

Valv*

87

20

N.A.

500

10k

10

300

Valv

87

20

N.A.

500

10k

10

300

Prim

Table 15. Combustion tests with prechamber Throttle

(%)

Spark Adv. (°) Load (N•m) Regime (min-1) Rion () Gain U (V) Var

8

20

6

1200

1k

1

100

Prim

8

20

6

1200

1k

1

100

Valv

12

20

1

1600

100k

1

200

Prim

12

20

1

1600

100k

1

200

Valv

12

20

1

1600

10k

1

200

Prim

12

20

1

1600

10k

1

200

Valv

11

10

5

1900

100k

10

200

Valv

11

10

5

1900

100k

1

200

Valv

11

10

5

1900

10k

1

200

Valv

11

10

5

1900

1k

1

200

Valv

11

20

3

1900

100k

1

0

Prim

11

20

3

1900

100k

1

0

Valv

11

20

3

1900

10k

100

0

Valv

11

20

3

1900

10k

100

0

Prim

11

20

3

1900

10k

10

0

Valv

11

20

3

1900

10k

10

0

Prim

11

20

3

1900

10k

1

0

Valv

11

20

3

1900

10k

1

0

Prim

11

20

3

1900

1k

10

0

Valv

11

20

3

1900

1k

10

0

Prim

11

20

3

1900

100k

1

-100

Valv

11

20

3

1900

100k

1

-100

Prim

11

20

3

1900

10k

100

-100

Valv

11

20

3

1900

10k

100

-100

Prim

11

20

3

1900

10k

10

-100

Valv

11

20

3

1900

10k

10

-100

Prim

11

20

3

1900

10k

1

-100

Valv

11

20

3

1900

10k

1

-100

Prim

11

20

3

1900

100

100

100

Valv

11

20

3

1900

100

100

100

Prim

11

20

3

1900

100k

1

100

Valv

11

20

3

1900

100k

1

100

Prim

11

20

3

1900

10k

100

100

Valv

11

20

3

1900

10k

100

100

Prim

11

20

3

1900

10k

10

100

Valv

11

20

3

1900

10k

10

100

Prim

13

20

3

1900

1k

100

0

Valv

13

20

3

1900

1k

100

0

Prim

13

20

3

1900

1k

100

-100

Valv

13

20

3

1900

1k

100

-100

Prim

13

20

3

1900

1k

10

-100

Valv

13

20

3

1900

1k

10

-100

Prim

Table 15. (continued) End of table 15.

Throttle

(%)

Spark Adv. (°) Load (N•m) Regime (min-1) Rion () Gain U (V) Var

13

20

3

1900

1k

1

-100

Prim

13

20

3

1900

1k

1

-100

Valv

13

20

3

1900

10k

1

100

Prim

13

20

3

1900

10k

1

100

Valv

13

20

3

1900

1k

100

100

Prim

13

20

3

1900

1k

100

100

Valv

13

20

3

1900

1k

10

100

Prim

13

20

3

1900

1k

10

100

Valv

14

20

5

1900

100

100

0

Prim

14

20

5

1900

100

100

0

Valv

14

20

4

1900

1k

1

0

Prim

14

20

4

1900

1k

1

0

Valv

14

20

5

1900

100

100

-100

Prim

14

20

5

1900

100

100

-100

Valv

14

20

4

1900

1k

10

100

Prim

14

20

4

1900

1k

1

100

Prim

14

20

4

1900

1k

1

100

Valv

18

20

4

1900

100

100

100

Valv

46

10

15

1900

100k

10

200

Valv

46

10

15

1900

100k

1

200

Valv

46

10

15

1900

10k

10

200

Valv

46

10

15

1900

1k

100

200

Valv

46

10

15

1900

100k

1

300

Valv

46

15

13

1900

100k

1

300

Valv

50

10

13

1900

100k

10

0

Prim

50

10

13

1900

100k

10

0

Valv

50

10

13

1900

100k

1

0

Prim

50

10

13

1900

100k

1

0

Valv

50

10

13

1900

10k

10

0

Prim

50

10

13

1900

10k

10

0

Valv

50

10

13

1900

100k

1

100

Prim

50

10

13

1900

100k

1

100

Valv

50

10

13

1900

10k

10

100

Prim

50

10

13

1900

10k

10

100

Valv

50

10

13

1900

1k

100

100

Prim

50

10

13

1900

1k

100

100

Valv

50

10

13

1900

100k

10

-200

Prim

50

10

13

1900

100k

10

-200

Valv

50

10

13

1900

100k

1

-200

Prim

50

10

13

1900

100k

1

-200

Valv

50

10

13

1900

10k

10

-200

Prim

50

10

13

1900

10k

10

-200

Valv

Table 16. List of tests under combustion conditions without prechamber

Throttle

(%)

Spark Adv. (°) Load (N•m) Regime (min-1) Rion () Gain U (V) Var

13

10

3

1900

100

100

200

Prim

13

10

3

1900

100

100

200

Valv

13

10

3

1900

1k

10

200

Prim

13

10

3

1900

1k

1

200

Prim

13

10

3

1900

1k

1

200

Valv

15

10

3

1900

100k

1

200

Prim

15

10

3

1900

100k

1

200

Valv

15

10

3

1900

10k

1

200

Prim

15

10

3

1900

10k

1

200

Valv

15

10

3

1900

1k

10

200

Valv

40

20

4

1900

10k

1

200

Valv

42

10

12

1900

10k

1

200

Valv

42

10

12

1900

100k

1

200

Valv

42

12

12

1900

10k

1

200

Valv

42

12

12

1900

1k

10

200

Valv

42

12

12

1900

1k

1

200

Valv

42

15

12

1900

10k

1

200

Valv

42

17

12

1900

10k

1

200

Valv

42

20

12

1900

10k

1

200

Valv

0

20

3

720

100

100

200

Valv

0

20

3

800

10k

1

200

Valv

33

10

8

1800

100

100

-200

Prim

10

20

4

1300

1k

10

300

Prim

10

20

4

1300

10k

1

300

Prim

8

20

4

1400

100k

1

0

Prim

16

20

4

2000

1k

10

300

Prim

16

20

9

2100

1k

10

-300

Prim

16

20

9

2100

10k

10

-300

Prim

16

20

9

2100

10k

1

-300

Prim

10

12

5

1700

100k

1

200

Valv

14

20

4

1800

10k

1

200

Valv

100

12

14

1900

100

100

200

Valv

100

12

14

1900

100k

1

200

Valv

100

12

14

1900

10k

1

200

Valv

42

12

12

1900

100k

1

200

Valv

42

12

12

1900

1k

10

200

Valv

42

20

12

1900

10k

1

200

Valv

100

12

14

1900

100

100

200

Valv

100

12

14

1900

100k

1

200

Valv

100

12

14

1900

10k

1

200

Valv

4.2 Description of the software

The NI-DAQ acquisition system is interfaced with Matlab R2021b® for data gathering, followed by two offline postprocessing stages.

The first part is the “acquisition.m” script, which contains the code that sets the initial configuration of the number of channels, the number of active channels, their respective input format, the sampling frequency and sampling duration. This configuration step is performed only once. With all the channels configured correctly, the data gathering begins by pressing the RUN button, as shown in Figure 79. The code is intended to be as simple as possible, to prioritize the data capture over other processing operations.

After pressing the RUN button, acquisition takes place and the stacking begins, obtaining: three data matrices for eight fast channels, three mid-speed channels and four thermocouple channels, as explained earlier. Once the acquisition has ended, a .mat file, with approximately 69MB is generated.

Figure 79. Matlab® user interface An additional part of script “acquisition.m” allows viewing the gathered waveforms with individual figures for each channel, after the acquisition. The database file comprises the information of all channels as samples, and the amplitudes are given by the NI-DAQ in volts, except for the thermocouple channels, which provide the information directly in degrees celsius.

Figure 80 shows a sample of raw ZZZ (“Z larga”), AFR (“Lambda”) and load cell (“Celda carga”) signals obtained, all expressed in volts in the vertical axis. The horizontal axis is expressed in number of samples.

Given that the sampling time for each channel is known, an appropriate time vector can be generated with the linspace command of Matlab® and the signal can be shown as a function of time. The coding script, present into a file named “Codigo_Completo.m”, begins with the unit conversions. Figure 81 shows a sample of primary current, scaled to ampere units in the vertical axis, and time units in horizontal axis.

Figure 80. Sample of gathered signals.

Figure 81. Sample of a scaled primary current The display of each variable, whether in its original form or scaled, is done to check for appropriate operation of the instruments.

The coding script, present in “Codigo_Completo.m” comprises an initial offline postprocessing of the data, in which each signal is separated in individual cycles, where the horizontal axis is now the crank angle, with a resolution of 0,1 crankshaft degrees, resulting in 7200 values per cycle.

The encoder signal, which has the information of TDC and angle combined, is used to form a 3-D array for each signal, with one axis for crank angle, a second one for the cycle number and the third axis dedicated to the variable amplitude. This is shown conceptually in Figure

82.

Figure 82. Scheme of the conversion of a sampled signal to cyclic signal arrays The original signal can be any of the variables acquired and scaled to proper units, described earlier. The rotational speed is given by the crank angle signal and the time vector.

4.2.1 Conditioning and referencing of the pressure signal data

The piezoelectric sensor used to measure combustion chamber pressure, in conjunction with a charge amplifier, provides a manometric signal. However, as noted by Lancaster et al. [191] and Armas [189], piezoelectric sensors are subjected to drift, particularly under steady conditions. To obtain accurate pressure-volume (P-V) diagrams, pressure measurements must be referenced to absolute values. The cyclic variability that introduces uncertainty is mitigated by averaging the pressure traces over multiple cycles [187]. Furthermore, Abbad [188] highlights that the resulting average requires additional filtering to eliminate spurious peaks. Therefore, additional signal processing routines are necessary to ensure accurate and reliable pressure information.

The work of Brunt and Pond [192] mentions two referencing methods: using the intake manifold pressure and polytropic index referencing. In Lee et al. [193], five methods are discussed and a modified least-squares method based on a variable polytropic coefficient is proposed.

For this work, the MAP signal helps to obtain a correction and referencing for the chamber pressure. The signal from MAP sensor at certain angle is added to all values of chamber pressure. It is assumed that chamber and intake manifold pressures are in equilibrium at that crank position. An initial criterion for such crank angle is the intake BDC (180°), however, due to the compressibility of gases, a further value of crank angle is required. Tests on the present engine led to conclude that the suitable crank angle for pressure referencing is 190°. On the other hand, a polytropic indexing approach was performed, obtaining an error of 30 mbar, validating the methods of Brunt and Pond and the acceptance of the 190 degree value.

Cycles with abnormal combustion events can have effect on pressure, rotational speed, and torque. A detection of irregular pressure development is programmed for this work. This allows discarding cycles with atypical waveforms, based on a combination of mean, maximum and integral error. By removing these cycles, the pressure signal features are closer to each other, resulting in reduced uncertainty.

Additionally, a filtering of the remaining cycles is performed, based on Martin’s work approach [187]. The rotational frequency array is required, as instantaneous engine speed is used to determine the sampling frequency for pressure signal screening. The 720-degree cycle-windowed pressure signal is converted to frequency domain by means of Fast Fourier Transform (FFT). Figure 83.a) shows a sample log-log plot of a transformed pressure signal obtained from a mean motored cycle, with all original components in red. The filter considers the mean rotational frequency as the first harmonic component for the frequency-domain. Then, only certain components of pressure signal are retained and the rest are removed from the original signal, as shown in Figure 83.b). Next, the time-domain and subsequently the crank angle-domain pressure signals are reconstructed, using an inverse Fourier transform operation, with only the selected remaining frequency components. According to [187], the number of harmonics required to reconstruct the filtered signal must be greater than the number of cycles of the whole signal acquired. For the sample motored pressure signal previously referred, the comparison between original and referenced waveforms is shown in Figure 84.a), and a detail of the unfiltered and reconstructed waveform is presented in Figure 84.b).

Figure 83. Log-log plots of a) transformed pressure signal, in red. Transformed pressure with removed components, in blue. b) transformed pressure with removed components, cleaned. Original pressure signal obtained in motored condition at 500min-1.

Figure 84. a) Plots of motored pressure at 500min-1. Original signal in red, referenced signal in blue. And b), detail of unfiltered motored pressure signal in black and reconstructed (filtered) signal from inverse FFT in cyan.

A conceptual summary scheme of the code used to obtain the corrected pressure waveform is shown in Figure 85.

Figure 85. Scheme of the code for pressure signal correction.

4.2.2 Preprocessing of the ion current signal.

The amplifier of the microammeter (see Figure 48) has a non-inverting configuration, and a minor difference between the inverting and non-inverting inputs. To solve the issue, particularly when gain settings are 10 and 100, an offset correction is performed, subtracting the most frequent value of the 720-degree ion signal, that is, the mode. In addition, the inductive probe for secondary voltage provides two slopes, one rising slope to detect the primary charging and another one for the spark release. This helps to calculate the area under the ion curve starting from the spark release. A conceptual summary scheme of the code used to obtain the corrected ion current waveform, and the area calculation, is shown in Figure 86.

Figure 86. Scheme of the coding for correction of ion current signal and area calculation

4.3 Preliminary findings on motoring tests

Initial approaches for motoring were tackled using an AC induction motor, driven by a variable frequency drive, and then a compound-type DC motor. The AC motor or the DC motor were connected through the belt system to the engine under test.

Figure 87. Initial setup for motoring with AC induction motor, a) Initial mounting of belt drive and b) Additional instrumentation.

Figure 87.a) illustrates the initial setup, using a variable frequency drive (VFD) (1), an AC induction motor (4) connected to the engine (3). The eddy brake (2) is reserved for combustion tests.

It was found initially that the VFD produced an intense electromagnetic interference that affected the signals of the instrumented engine with high frequency spikes. An initial attempt to solve the problem consisted on shielding the cable between VFD and motor with several layers of aluminium foil. That approach was insufficient to reduce the interference.

The explored literature related to VFD’s applied to motors [194], [195], [196], [197] suggests adding toroids to each phase inside the VFD or a common mode filter. The cable should have a higher isolation capability and improved shielding. Additionally, the power supply lines of the VFD should include a line filter or, preferably, an isolation power transformer.

With pure sinusoidal supply, the phases are balanced. However, with pulsed signals from VFD, there is no balance of phases and circulating currents take path to the motor chassis [198], [199], causing not only erosion and damage of the surfaces of the bearings, but also ground currents that, for the case of the present work, reached the engine and the instrumentation, affecting all signals.

An attempt to mitigate the ground current that comes from the motor was to isolate electrically the chassis and motor stator base with teflon (8), as shown in Figure 87.b). The

cable was improved, toroids were added, but still the interference, although reduced, remained. Then, a consideration arised: The magnetic field of the stator is described as a rotational vector, mostly shown pointing toward the squirrel cage rotor, which is inside the stator [47]. Additionally, the magnetic lines of force are usually depicted forming loops inside the stator and passing through the rotor.

Now, what about the magnetic loops outside the stator projecting themselves to the air? Along with such loops, the pulses from VFD are square waves, having a fundamental frequency of 8 kHz, but with additional odd harmonics multiples of that frequency, especially the fifth, seventh, eleventh, thirteenth and beyond [200].

Thus, a steel mesh cage that surrounds the motor, (5) in Figure 87.b), covered the motor. The instrumentation cables were wrapped with aluminium foil (6), and the acquisition system was put inside a metal cabinet (7). That reduced the interference of the VFD that affected the instrumentation even more.

The exploration of requirements of starting torque for the engine, shown in [201], having in account the displacement, compression ratio, dimensions and other parameters of the engine, led to consider the AC motor as sufficient to turn the crank, which indeed happened in practice, with a rotational speed of 2400 min-1.

Nevertheless, the interference to the instruments continued to be a serious issue in terms of poor signal-to-noise ratio. Therefore, the AC motor was substituted with a compound-type DC motor (2), shown in Figure 88. The teflon isolation (3) remained, to avoid undesired contributions from the AC supply.

Figure 88. Setup for motoring with DC compound motor The DC supply consisted on two independent variable power sources based on VARIAC and full wave recitifier, that in turn are fed by 120 V AC, as shown in the schematic of Figure 89. DC1 feeds the series field and rotor and DC2 feeds the parallel field.

Figure 89. Schematic of the power supply for the DC compound motor Unfortunately, it was found that the rotor had an abnormal electrical continuity with respect to the ground, and the torque of the old motor was barely enough to rotate the engine at 450 min-1. That same speed is achievable with the starter motor.

All previous difficulties led to limit further motoring tests to the rotational speed given by the starter motor, integrated with the engine in the first place, which is a DC motor with permanent-magnet stator type, powered by battery, avoiding paths of high-frequency or 60 Hz interferences that could affect the instrumentation signals.

4.3.1 Motoring tests without spark

Literature refers the ion current signal limited to a crank angle range, particularly after spark onset and end of combustion [18], [25], [103]. In the present work, an initial approach tackles to take a look at the secondary loop current waveform for the 720-degree cycle, looking for additional capabilities of the ion sensor.

The first attempt aims to detect ion current signal without ignition excitation, at motoring speed. Figure 90 illustrates the ion current waveform obtained with microammeter sensitivity set to Rion = 100 k and G = 10, resulting in a total gain factor of 106. Rotational speed was

490 min-1.The promoter power source was set to U = 300 V, with 87% engine throttle

opening. The signal shown is a function of time. The period Tu corresponds to the commutation of the relay K2 of the promoter power source U (recall Figure 45) that causes the charging and discharging of the capacitor C3 when connecting to C2. Figure 91 shows the same ion current signal as function of crank angle, covering a total of 17 consecutive cycles.

Figure 90. Sample of ion current signal as a function of time during motoring without spark.

Figure 91. Sample of cyclic ion current during motoring without spark

Figure 92. Sample of mean ion current, pressure and valves displacement as function of crank angle during motoring without spark.

Figure 92 illustrates the mean of the aforementioned 17 cycles of ion current signal (red), its smoothed version (green), as well as the mean chamber pressure (blue) and intake (orange) and exhaust valve (purple) stroke, as functions of crank angle. From Figure 90, Figure 91, Figure 92, it is observed a noisy current signal, making inconclusive any attempt to distinguish features of ion current signal as function of intake valve opening, exhaust valve opening or compression without spark.

4.3.2 Motoring tests with ignition system on

The ignition system is coupled to the ion current measurement loop. As a result, the ion signal will contain features influenced by ignition events, making it necessary to identify and separate these contributions within the ion current waveform.

Relationships with primary and secondary windings variables

Figure 93 illustrates the average of 17 cycles of ion current signal (in this case secondary winding current, reminding the impossibility to show the real magnitudes of secondary current due to the engaging of the limiter described in section 3.1.5) related to pressure, primary winding voltage and current, as well as secondary voltage read by both probes. The waveforms are obtained using Rion = 10 k and a gain of G = 1, resulting in an amplification factor of 104, at a rotational speed of 472 min-1.The promoter power source was set to U =

200 V, and with the throttle opened to 87%. The secondary current signal is shown with

inverted polarity.

The crank angle points A1 and A2 correspond to the engaging of the ignition module transistor, leading to the magnetic field charging of the primary winding by means of gradual increases in primary winding current and voltage. Points B1 and B2 mark the disengaging of the transistor, in which the reaction of the primary winding is seen as a negative voltage spike when primary current drops to zero. The primary voltage signal indicates a lower negative spike in wasted spark (B2), in comparison with the compression stroke (B1). This is explained by Paschen law [37], having more voltage in the secondary winding at higher pressure, which in turn is reflected to primary winding voltage accordingly.

Figure 93. Primary winding signals related to ion current (secondary current) and pressure The secondary current signal rises in magnitude during the intervals of primary charging A1- B1 and A2-B2, showing oscillations due to the coupling between primary and secondary windings. The first peaks in the secondary current on A1 and A2 allow to detect the primary charging without looking at primary current or voltage.

Figure 94 focuses on the details of points B1 and B2 of Figure 93. The collapse of primary winding current can be inferred from the secondary current signal. The sustainments of primary and secondary winding voltages from B1 to C1 and B2 to C2 are followed with coarse approximation by the saturation of secondary current signal.

However, the secondary current exhibits oscillations at points D1 and D2 that originate from the L-C coupling between the secondary winding inductance and the spark gap capacitance. The oscillations visible at D1 and D2 of secondary current are absent in the primary voltage waveform. It can be noted also that, based on Paschen law, the secondary current oscillation D1 shows a higher initial amplitude of oscillation than D2. This behaviour suggests that higher pressure results in a greater secondary current amplitude.

Figure 94. Details of B1 (compression) and B2 (wasted spark) In addition, as shown, the ion current waveform remains zero both before the primary charging and after the D1 and D2 events, keeping no direct relation with variations in combustion chamber pressure during these intervals under motoring conditions.

As seen in Figure 94, the inductive probe presents the oscillations D1 and D2. The resistive probe has a voltage damping that, for the particular test performed, lasts up to the 346-degree mark on compression, and up to the 706-degree coordinate on wasted spark. The oscillations D1 and D2 are not evident in the primary voltage signal. The waveforms of the primary voltage and the secondary inductive probe should have the same features, according to the transformer principle and Faraday’s law. However, the events at the spark plug are better observed with the inductive sensor, instead of the primary winding one. The resistive probe and microammeter of secondary winding side are within the electrical loop of the spark plug, capturing its free-run happenings of the last one. Thus, the secondary voltage probes and the ion current sensor appear more promising for detecting events in the combustion chamber.

Effect of the prechamber on motoring waveforms

Figure 95 illustrates the average of 17 cycles of secondary current and secondary voltage waveforms, using resistive and inductive probes, related to motoring pressure. Figure 95.a) shows the graphs obtained for motoring with prechamber, whereas Figure 95.b) displays the waveforms without the prechamber. Both tests were conducted with the parameters Rion = 10 k and G = 1, resulting in a gain factor of 104, and a rotational speed of 472 min-1. Power source was set to U = 200 V, and 87% throttle opening.

The variations in the amplitude of the secondary voltage are explained by Paschen law, with higher secondary voltage observed at higher pressure, as indicated by both probes.

Although there are slight differences in the secondary current waveforms, the free-run oscillation, indicated by Tf , has a duration of 5 crank angle degrees for both cases, and since the regimes are identical, the oscillations have the same duration in time.

Figure 95. Motoring waveforms for pressure, secondary current and secondary voltage measured with both probes, being a) test with prechamber and b) test without prechamber.

Motoring waveforms referenced to valve displacements

Figure 96. Motoring waveforms for pressure, secondary current and secondary voltage measured with both probes, without prechamber Figure 96 illustrates the average of 17 cycles for intake and exhaust valves displacements related to in-cylinder pressure, secondary current and voltage, without prechamber. The settings Rion = 10 k and G = 1, factor of 104, rotational speed of 472 min-1, power source U = 200 V, and 87% throttle opening are maintained.

As the picture shows, neither the ion current (which corresponds to the secondary current in this case) nor the secondary voltage varies at any crank angle corresponding to intake valve actuation. The same observation applies to the exhaust valve displacement waveform.

Nevertheless, changes in capacitance of spark plug do occur during the exhaust valve opening and the compression stroke. When the exhaust valve is open, the chamber pressure is close to the atmospheric value. The primary charging, during which the processes of ignition-coil magnetic-coupling and saturation of the transistor occur - far from a perfect short-circuit -, provides a somewhat AC-like excitation to the spark plug capacitance. The charging during compression generates the oscillation C1, while the charging C2 occurs during the exhaust valve opening, having a variation in amplitude of oscillations due to the different pressures the spark plug is subjected to in each case. With the prechamber, the difference that can be noted in the coil charging oscillation amplitude becomes less noticeable.

Motoring waveforms and ion factor

The ion factor is defined as the product of Rion and the amplifier gain G of the microammeter used for ion current detection.

Figure 97. Comparisons between ion factors a) 104 and b) 105 for the microammeter.

Figure 97 illustrates the effect on the ion current (secondary current) when the Rion shunt resistor is changed.

Figure 97.a) shows the waveform for Rion = 100  and G = 100 having a factor of 104 [V/A], whereas Figure 97.b) displays the combination Rion = 100 k and G = 1, having a factor of 105 [V/A]. With the latter combination, both the primary charging oscillations and the freerun oscillations exhibit greater amplitude, yet still remaining within the ±67μA window of the factor 105 [V/A].

To evaluate whether an ion signal fits within the window, the maximum ion current Imax that can be displayed before the signal limiter fully clamps is:

|𝐼௠௔௫| = 10଺|𝑈ீ௠| 𝑅௜௢௡𝐺 [𝜇𝐴]

Eq. 49 Where: UGm = 6,7 V is the maximum output of the amplifier, either positive or negative, given by the power supply of the amplifier.

On the other hand, since the limiter exhibits a gradual decrease in impedance proportional to the input voltage, the shunt resistor Rion begins to be affected in its ohmic value by the limiter when the amplifier output reaches |UAmp | = 4 V. This is supported by the experimental approach presented in section 3.1.5 and Table 9. Thus, the ion current Iion is accurately represented if:

|I୧୭୬| ≤ 10଺|U୅୫୮| R୧୭୬G [μA]

Eq. 50 Table 17 shows the limits of the accurate current indication Iion and maximum current Imax.

Table 17. Accurate and maximum ion current indications for each factor In the 105-factor free-run oscillation shown in Figure 97.b), there is also a remnant lasting for two additional crank angle degrees before the waveform settles to zero.

Rion () Gain Factor (V/A) |Iion| (μA) |Imax| (μA)

100

100

104

400

670

1k

1

1000

4000

6700

1k

10

104

400

670

1k

100

105

40

67

10k

1

104

400

670

10k

10

105

40

67

10k

100

106

4

6,7

100k

1

105

40

67

100k

10

106

4

6,7

Figure 98. Comparisons between combinations of Rion and G for the microammeter.

On the other hand, Figure 98 illustrates three variations of the same ion factor of 104 [V/A], being a) the combination Rion = 100  , G = 100; b) the combination Rion = 1 k , G = 10; and c) the combination Rion = 10 k , G = 1. It can be noted that the differences in ion current (secondary current) signal are negligible, allowing any of the three combinations to be used. The secondary voltage waveforms, gathered with both probes, of Figure 97 and Figure 98 are very similar, indicating that those measurements are independent of the ion factor and combinations selected. For the tests shown in both figures, the power source is U = 200 V and the cranking speed is 472 min-1. In addition, the tests were performed without prechamber. The waveforms shown correspond to the average of 17 cycles.

Motoring waveforms and negative power supply

The ion promoter power source U is another parameter of the ion current detections system to be considered. Figure 99 shows the waveforms obtained with U = -100 V, Rion = 10 k , G = 1, cranking speed 472 min-1. Compared to U = 200 V, Figure 98.c), the free-run oscillation is very similar, since this portion depends on the temporary contribution of the ignition coil rather than the power source U.

Figure 99. Motoring waveforms for pressure, secondary current and secondary voltage measured with both probes. U = -100V

4.4 Preliminary findings on combustion tests

Once the motoring tests are completed, during which the contributions to the ion signal caused by the DC power source U and the ignition system are identified, the analysis of the added input due to the air-fuel mixture combustion becomes the next step. For all subsequent tests, the fuel used is commercial regular 87-octane gasoline, with a 10% ethanol content, in accordance with Colombian regulations.

Initial tests performed led to detect strange operation noises. It was found that the GM ignition module, at certain rotational speeds, produced 4 pulses in 2 turns. Two of them were triggered correctly at the desired advance angle. Therefore, a filter with 1 k resistor and a 1000 pF capacitor in parallel was added to the input of the ignition module (see Figure 100), restoring the one-pulse-per-turn normal triggering.

Figure 100. Modification to the ignition module

4.4.1 Findings with prechamber

The prechamber, provided to eventually vary the engine compression in future tests caused that the ignition strike occurred outside of the main cylindrical volume of combustion chamber.

The work of Shah et al. [202] deals with the effect of a prechamber on ion current waveform, using a spark plug with the addition of a hemispherical tip and four nozzles to form the prechamber. It is stated in the work that the presence of a prechamber causes a significant reduction on the thermal peak of ion current signal, although getting a bigger amplitude of the first peak, that is, the chemo-ionization peak.

The crank angle of the thermo-ionization peak is known to be used to correlate it with the incylinder pressure peak, since they tend to coincide [18], [21], [25]. Thus, the reduction of the thermal peak amplitude impairs the possibility of using the ion signal for correlation with the pressure peak. In addition, both the area under the ion signal curve and the crank angle position of its centroid are affected. .

Henein et al. [103] indicate that ion current sensor measures local properties of combustion products, rather than global properties of the combustion chamber, resulting waveforms that depend - in shape and peak magnitude - on the sensor’s design, its location within the chamber, as well as the flow patterns that characterize the propagation of flame front.

In the present work, unlike the work of Shah et al. [202], the prechamber is a cylinder, with

12 mm in diameter and 9 mm in length, with one side open to the volume of the combustion

chamber, and the other closed with the spark plug. Indeed the prechamber can have an

adverse effect on the ion signal waveforms. Nevertheless, some of the tests performed allowed to parameterize the ion current detection system.

Effect of the absence of DC bias U

Figure 101 displays the waveforms obtained for ion current a) and pressure b) for a total of seventy nine 720-degree cycles, having an average load torque of 14,06 N•m, mean AFR of 15,00, a mean rotational speed of 1942 min-1, and an engine temperature of 87,5°C. The settings of ion current sensor are U = 0 V, Rion = 10 k , G = 10, spark advance of 10° BTDC, and 50% throttle opening.

Figure 101. Cyclic plots of a) ion current, and b) pressure.

Figure 101.a) shows that near the 360 degree mark, there are no features that correspond to the evident variations in pressure, seen in Figure 101.b). The most prominent peaks of pressure appear atypical when compared to the average of the 79 cycles. Thus, without the biasing of U, the ion waveform is strongly dependent on the excitation provided by the ignition system, and appears unable to reflect pressure variations.

Figure 102.b) illustrates the average of 60 of the 79 cycles of Figure 101. The number 60 corresponds to the exclusion of 19 cycles, as shown in Figure 102.a).The excluded cycles are atypical pressure waveforms, outside of the average by ±1 bar or with abnormal pressure rise, usually preignition.

It can be noted from Figure 102.b) the spark release B1, followed by the saturation of the ion sensor, until the remnant free-run oscillation C1. The resistive secondary voltage waveform shows a decay to zero that has the same crank angle duration as the oscillation C1. However, its correlation with pressure is not evident.

Figure 102. a) Exclusion of atypical values of pressure, and b) average of 60 cycles of pressure, ion current and secondary voltage (both probes).

The ion factor used is 105 [V/A], which is equivalent to using a 100 k microammeter shunt resistor.

Even with such sensitivity, closer to the values seen in HCCI studies [114], the peaks of chemi-ionization and thermo-ionization, shown in literature of SI engines, are not evident. Thus, the promoter power source U is needed to observe variations of ion current, in response to fuel breakdown during combustion, after the free-run oscillation.

Effect of a negative DC bias

Figure 103. Cyclic plots of a) pressure, b) ion current. c) average of ion current, pressure and secondary voltage (both probes). U = -300V.

Figure 103 illustrates the waveforms obtained for pressure a) and ion current b) for a total of ninety one 720-degree cycles, having an average load torque of 10,52 N•m, mean AFR of 13,53, a mean rotational speed of 2245 min-1, and engine temperature of 80,2°C. The settings of ion current sensor are U = -300 V, Rion = 10 k , G = 10, spark advance of 20° BTDC, and 40% throttle opening.

Figure 103.c) shows the average of 65 cycles (27 non typical were discarded) for ion current, pressure, and secondary voltage.

In the cyclic plot shown in Figure 103.b), the ion signals are very noisy (ion factor 105 [V/A]). However, with such ion factor and -300V of excitation, the average ion current after the freerun oscilation, shown in Figure 103.c), barely reaches -2 μA. Nevertheless, the ion signal persists until the pressure reaches its maximum.

Therefore, the negative polarity for U, combined with a 105 [V/A] factor, are not ideal settings for the ion current sensor due to low signal amplitude and poor SNR. Reducing the ion factor by one order of magnitude decreases the amplitude of the ion signal. Moreover, the magnitude of U represents the maximum available excitation to generate the ion current.

Positive bias of U.

Figure 104. Cyclic plots of a) pressure, b) average of ion current, pressure and secondary voltage (both probes). U = 100V.

Figure 104.a) illustrates the waveforms obtained for pressure, for a total of seventy nine 720degree cycles, having an average load torque of 13,60 N•m, a mean AFR of 15,14, a mean rotational speed of 1943 min-1, and an engine temperature of 82,4°C. The settings of ion current sensor are U = 100 V, Rion = 10 k , G = 10; spark advance of 10° BTDC, and 50% throttle opening. Figure 104.b) displays the average of 64 cycles of ion current, pressure and secondary voltage. The ion current amplitude after the free-run oscillation reaches 4 μA. With more than 100 V and positive polarity, the promotion of ion current can be improved.

Combustion waveforms measuring valves’ displacements and considering the prechamber

Figure 105 illustrates the average of 91 cycles for intake and exhaust valves displacements related to pressure, ion current and secondary voltage, with prechamber. The tests were performed at an average load torque of 10,81 N•m, a mean AFR of 14,59, a mean rotational speed of 2294 min-1, and an engine temperature of 82,7°C.

The ion settings were Rion = 100  and G = 100, factor of 104, power source U = 200 V, 20 degrees BTDC of spark advance, and 87% throttle opening.

As the picture shows, the ion current or secondary voltages do not exhibit any changes at the crank angle positions corresponding to intake valve actuation. The same observation applies for the exhaust valve displacement waveform.

Figure 105. Combustion waveforms for pressure, secondary current and secondary voltage measured with both probes, with prechamber The ion current waveform after the free-run oscillation has a very low amplitude, although the ion promoter source is set at 200V. This is explained by the location of the spark plug, which is not exposed to the winds of the flow patterns.

The motoring tests and the ones in combustion with prechamber, the last ones not providing a useful ion signal that can be correlated directly with pressure, led to narrow the settings of ion current to factors of 104 [V/A] and 105 [V/A], and a promoter power supply to U = 200 V for the tests to be performed in combustion without prechamber.

4.4.2 Findings without prechamber

The work of Burgett et al. [203] presents the effects of spark plug gap projection, gap size, electrode size on the work produced by the engine and reduction of cyclic variation, remarking best results with gaps that protrude into the combustion chamber, bigger gap and electrode size, particularly with lean mixtures. The protruding gap helps to get a faster burning of the charge.

Therefore, in this work, realized the presence of the prechamber feature, the correction aimed for a ground electrode protruding 2 mm inside the combustion chamber. This way, the characteristic flow patterns get in touch with the spark plug electrodes.

Figure 106 illustrates the array of 77 cycles, showing pressure, ion current, and secondary voltage, without prechamber. The tests were performed at an average 13,16 N•m load torque, a mean AFR of 13,97, a mean rotational speed of 1929 min-1, and an engine temperature of 80,8°C. The ion settings were Rion = 10 k and G = 1, factor of 104, power source U = 200 V, 42% throttle opening, and 10 degrees of spark advance.

As the picture shows, there are variations in the pressure peak, Figure 106.a), represented in the box plot b). Similarly, variations in ion current, Figure 106.c), limited to the crank angle interval [365°, 410°], are observed, both for chemi-ionization and thermal ionization peaks. Variations in peak secondary voltage are also seen in Figure 106.d).

The average of 74 cycles (3 cycles have a higher pressure peak) is presented in Figure 107. It can be noted the engaging of the ignition transistor in B1 and B2, the primary charging chg, the transistor release in C1 and C2. The wasted spark caused a high-amplitude free-run oscillation A1, whereas the ignition of fuel caused a barely distinguishable free-run, followed by a chemi-ionization peak D and a thermo-ionization peak E.

The thermal peak E rises up to 60 μA, and the chemical peak D up to 379 μA, having a significant improvement in the magnitude of ion current with respect to the tests with prechamber.

The saturation on ion current in F occurred during the primary charging in the wasted spark. This saturation is abnormal, because it should have the small oscillations seen between B1 and C1. The algorithm to detect atypical cycles checks for variations in pressure during the compression cycle. Now, during the opening of the exhaust valve, some residual hot gases make contact with the spark plug, causing the saturation F of ion current sensor.

Therefore, the statement that ion current don’t change in any crank angle coordinate on which the intake valve is actuated, still holds. However, for the exhaust valve, the ion current, powered by a DC promoter, can not provide information of its opening stroke, but the presence of hot ionized gases can be detected on exhaust stroke by the ion current signal.

Figure 106. Cyclic plots for combustion without prechamber, a) Pressure, b) box plot of pressure deviation, c) ion current, d) secondary voltage (clamp)

Figure 107. Average of cycles for combustion without prechamber, showing pressure, ion current, valve displacements and secondary voltage (both probes)

Atypical pressure waveforms due to preignition.

Figure 108. Cyclic plots for combustion without prechamber, showing abnormal pressures and ion currents Figure 108 illustrates the plots of 77 cycles of pressure and ion current, having abnormal features for both signals. The tests were performed at an average load torque of 15,42 N•m, a mean AFR of 13,66, a mean rotational speed of 1843 min-1, and an engine temperature of 96,8°C. The ion settings were Rion = 1 k and G = 10, factor of 104, power source U = 200 V, 42% throttle opening, and 12 degrees of spark advance.

As the picture shows, there are abnormal pressure peaks A, that occur sooner than expected. Ion current plots have abnormal saturations C, that correspond to the presence of hot gases during the exhaust stroke, mentioned earlier. In addition, there are spurious saturations B that occur during the primary charging.

The average of 73 cycles is shown in Figure 109.a), excluding four atypical pressure waveforms (cycles 9, 46, 62 and 69) from the calculation. A detail of a sample abnormality C is shown in Figure 109.b), corresponding to the ninth cycle. The features of ion current during engaging of ignition transistor in A, and its releasing in B, are common to both mean and atypical curves. The mean waveform has the chemi-ionization peak C and the thermoionization peak D. However, the ion signal during primary charging in Figure 109.b) has the saturations E. Around the second saturation E, the pressure signal rises abruptly (F) before the transistor release B, that is, preignition.

The secondary voltage waveforms for the mean and atypical cases are very similar, making them unable to detect a rise of pressure attributable to preignition, at the time that the ion current can catch such anomaly during the primary charging.

Figure 109. a) Average of cycles for combustion without prechamber, b) Atypical cycle. Waves: pressure, ion current, valve displacements and secondary voltage (both probes). The saturations of ion current E, shown in Figure 109.b) that indicate a developed preignition, match with the findings of the work of Kumano et al. [204], performed on a gasoline directinjection engine.

4.5 Conclusions

For motoring tests with AC motors, driven by VFD, colossal electromagnetic interferences can appear, converting the VFD, in the terminology of electromagnetic compatibility, into an offender, and the instrumentation becomes the victim. It was found to be extremely difficult to eliminate that interference that eventually reached the measurement signals. Therefore, additional devices, further tests or better digital filtration techniques are needed.

The primary voltage signal is captured using an analog isolation amplifier before entering the acquisition module. In contrast, the inductive probe sends its signal directly to the acquisition module. Observations of the captured signals indicate that the remnant free-run oscillations after the spark pulse discharge are not clearly visible in the primary voltage signal displayed on the computer screen. The frequency response of the analog isolation amplifier might be from DC to 230kHz, but still there is a time-delay interval, inherent to the sigmadelta modulation of the TLP7920 chip, in which certain short spikes could not be captured correctly. Therefore, although the rough features of the primary current and voltage can be distinguished, better isolation amplifiers, preferrably analog in principle of operation are required to more faithfully capture the spikes predicted by Faraday’s law predict at the instants of ignition transistor release.

The ion current signal has been analyzed based on their features present across the total 720degree 4-stroke cycle, recurring to intake and exhaust valve displacement sensing, as well as primary voltage and current, looking for additional capabilities of the ion sensor. From the motoring tests performed using the DC power source U, the ion current is not affected by the changes in combustion chamber pressure, either caused by the intake and exhaust valves displacements, or compression stroke. Nevertheless, AC variations caused by the ignition system, that is, primary charging by transistor saturation, strike by transistor release or residual free-run ringing, excite the capacitor inherent to the spark plug. Then, variations of the interelectrode dielectric of spark plug in motoring, caused by changes in pressure of the air on chamber, are reflected in variations of the ion waveform, in this case secondary current waveform, as there is no fuel to ignite.

With negative bias of the promoter power supply U, on combustion tests with prechamber, it is required a considerable magnitude of voltage to observe an appreciable ion current. For U = -100 V and U = -200V, the chemical and thermal features of ion current attributed to fuel were not detectable. Using U = -300V, a small 2 μA current was observed, confirming the findings of the tests performed by Abdel-Rehim et al. [89], who used -380V, requiring a strong inverse electric field to promote ion current. This is because the spark gap, biased negatively, has a similar behaviour as an inversely polarized diode, in which it is necessary a high threshold of voltage to cause conduction, this also supported by the works of Wilstermann et al. [132] and Arrigoni et al. [91].

On the other hand, the positive biasing of the promoter U eases the ion current generation. Nevertheless, a minimum magnitude of U is required. The present work used 100 V as a minimum, slightly above the 70 V threshold value indicated by literature shown in Table 12, which turned out to be not enough to promote ion current with tests performed with prechamber. As for the maximum value, the work of Song et al. [205] shows a deformed ion

signal when the bias exceeds 400 V. In the present work, the limit of U = 300 V avoided deformation of ion signal caused by such a strong electric field.

The product of Rion shunt resistor and amplification G constitute the so-called ion factor for this work. Variations of that factor were explored in motoring and combustion conditions, finding that similar combinations Rion -G that provide the same factor lead to similar ion waveforms for the cases of 104 [V/A]. On the other hand, the factor 105 [V/A] is achieved with combinations in which the magnitude of shunt resistor Rion could affect significantly the resistive parameter of the R-L-C electrical loop formed by the secondary winding and its inherent ohmic resistance, the spark plug and its internal resistance, as well as the ohmic specification of the high-tension wire, and with this the response of the ion current detection system. This requires additional analysis.

The factors 106 [V/A] and 107 [V/A] are not suitable for spark ignition, due to the lower SNR and the fact that the ionization is strong enough to saturate the sensor. On the other hand, factors of 103 [V/A] and lower, provide a very small ion signal, hard to detect.

The presence of a prechamber lowers the magnitude of ion current, especially affecting the thermo-ionization peak. The form of the prechamber used in the present work is a cylinder, open on one side to the combustion chamber, differing from the hemisphere portrayed in the work of Song et al. [205]. They state that the chemical ion peak has a bigger magnitude, in comparison with the case of no prechamber. Nevertheless, the form of the prechamber of the present work turned to attenuate the peaks, even the chemical one, one order of magnitude, forcing the use of a promoter with a higher voltage and higher ion factors.

The primary voltage waveform can be inferred by the secondary inductive probe one because, although both sensors operate by electromagnetic induction, the intermediate isolation amplifier coupled across the primary winding is unable to capture certain short spikes. On the other hand, the resistive secondary voltage has a more smoothed curve that the inductive one, having a damping that ends approximately after the chemi-ionization peak (see Figure 109.a), detail C).

It can be noted that the secondary voltage probes can not offer the information that the ion current does, for example preignition or presence of hot ionized gases during exhaust stroke. Nevertheless, the ion current signal can benefit from secondary voltage signals and their insensitivities to confirm points of ignition transistor engaging or releasing. In addition, the motoring tests referred in the present work were performed at room temperature, whereas the combustion tests were performed at engine block temperatures between 52 °C and 89 °C. Both secondary voltage probes indicated spikes on the ignitiontransistor release events. The magnitude of the spikes at room temperature were lower than the ones with hot engine block, confirming the results of the work of De Zoysa et al. [206] of breakdown voltage as a function of temperature.

The difference in features of the ion current between wasted spark, compression without combustion, and combustion itself, after the ignition pulse, confirm the possibility of misfire diagnosis and camshaft timing detection indicated in Eriksson [86].

Analysis and correlations between ion current and pressure The previous chapter described the testing procedures, preliminary considerations used to define the experimental limits, including rotational speed, spark advance, microammeter settings, magnitude and sign of promoter power source, the amount of cycles to capture, and subsequently the test campaigns in motoring and combustion. Initial findings were also presented, which helped refine the settings of promoter U and microammeter, as well as characterize the waveform features both with and without ignition, and the contributions of fuel breakdown. The combustion campaigns were classified with the prechamber and without it. The prechamber helped shed light on the signals features at different conditions and the best fitted settings to use without prechamber.

This chapter advances the analysis stage by focusing on tests conducted without the prechamber. It explores correlations between measured pressure, ionization current, and secondary voltage signals, captured using both currrent and voltage probes.

Section 5.1 addresses the modeling of the ignition system used in the present work, which employs a Low-side scheme, where the ignition coil is a true transformer, including the coupling between primary and secondary side, their natural and damped frequencies, the damping factors associated to each loop, as well as the permanent and transient power sources that contribute to the free-run oscillations that appear at the end of the ignition strike. Particular attention is given to the relationship between this damped oscillation (previous to the chemi-ionization and thermo-ionization), and the capacitance of the spark gap, aiming to find a correlation with the working gases present in the gap that lead to a posterior successful flame initiation, its absence, or a wasted spark. Additionally, the effect of the value of the microammeter shunt resistor Rion in the damped oscillation and further ion signal features is analyzed.

Section 5.2 provides an introductory explanation of the relationship between the area under the ionization current curve and the pressure, mentioning the approaches described in the literature, and the magnitude and angular features of the ionization current that are employed in the subsequent analyses.

The third section is divided into three subsections.

 The first subsection focuses on estimating the cyclic variation of pressure from the damped frequency (that is, the final contribution of the ignition system) present in the secondary current signal and the magnitude of the area of ionization current as variables that help quantify the magnitude of pressure peak, as key predictors.  The second subsection addresses the estimation of the angular position of peak pressure. It examines the use of the centroid of the ionization current area, the percentage of the cumulative ion current integral associated with the peak pressure angle, and combined indicators based on damped frequency, ionization area, its centroid to determine the angular position of the occurrence of maximum pressure and to enhance angular prediction accuracy.

 The third subsection (5.3.3) presents the application of the cyclic variation estimation of the aforementioned pressure features, under different engine conditions, evaluating their performance and limitations.

Section 5.4 presents ignition delay estimations from cyclic ion signal. The last section addresses the crank angular relationships between averaged ionization current features and mass fraction burned, as well as heat release rate, last ones calculated from in-cylinder pressure.

5.1 The ion current signal before chemi-ionization and its relationship

with peak of combustion chamber pressure.

The ion current signal is mainly studied after the ignition pulse [21], [25], [49], [89], [111], [131]. However, for the Low-side circuit, given that the impulse contribution of the ignition and the ion promoter source are within the same electrical loop, there is a source of information previous to the chemi-ionization and thermo-ionization promoted by U, that is not explored in the literature.

Figure 110 illustrates the core net of the ignition system coupled with the Low-side ionization current measuring system. It is added to the core net the commuter transistor (switch S1) in parallel with a capacitor Cs on the primary loop. Cs stands for the small capacitance between collector and emitter in off-state of the transistor. Represented are the impulse power sources e2 and e12 that arise from the contribution after the releasing of the commuter transistor (switch S1).

Figure 110. Schematic diagram of the electrical nets of the ignition core module. Ignition core module comprises the spark plug, the secondary winding of the ignition coil, the resistance of the high tension wire (and internal spark plug resistor), the promoter power source and the microammmeter. The model of the primary and secondary electrical loops, includes the inductive effects of both coils.

The analysis described in [40] includes a capacitor Cs with a typical 220 nF value, proper of early contact breaker ignition systems, that leads to a primary resonant frequency ωn1 explained in section 2.4 with Eq. 12, repeated below for convenience: 𝜔௡ଵ= ට ଵ ௅భ஼ೞ Eq. 51 The capacitor Cs and L1 play a dominant role for ωn1.

There is also a natural frequency ωn2, referred previously in section 2.4, corresponding to Eq. 14, rewritten below:

𝜔௡ଶ= ඨ

1

𝐿ଶ((𝐶ଶ+ 𝐶ଷ+ 𝐶ସ+ 𝐶ହ) Eq. 52 As it is stated in [40], the free-run damped oscillation frequency after ignition pulse is a combination of ωn1 and ωn2 due to the strong magnetic coupling and values of capacitors and inductances involved. The absence of capacitor Cs leads to a reduced ignition pulse for contact-breaker ignition systems. Modern ignition systems lack that significant capacitance, making the natural frequency ωn2 a significant feature of the oscillation (ringing) of the secondary winding, just after the ignition pulse.

The loops shown in Figure 110 comprise series R-L-C circuits excited by temporary impulse inputs e1, e21 on primary side and e2, e12 on secondary one. The promoter U is a voltage sustained on the secondary loop causing the ion current features that appear during fuel breakdown reactions.

The differential equations that characterize the behaviour of the primary winding are:

𝑣௦+ 𝑒ଶଵ= 𝑅ଵ𝑖ଵ+ 𝐿ଵ 𝑑𝑖ଵ 𝑑𝑡+ 𝑈஼௦

Eq. 53 𝑖ଵ= 𝐶௦ 𝑑𝑈஼௦ 𝑑𝑡

Eq. 54 𝑒ଶଵ= 𝑀𝑑𝑖ଶ 𝑑𝑡

Eq. 55 For the secondary winding:

𝑈−𝑒ଵଶ= ൫𝑅௜௢௡+ 𝑅௪+ 𝑅௣+ 𝑅ଶ൯𝑖ଶ+ 𝐿ଶ 𝑑𝑖ଶ 𝑑𝑡+ 𝑈௚ Eq. 56 𝑖ଶ= 𝐶௦௣ 𝑑𝑈௚ 𝑑𝑡

Eq. 57 𝑒ଵଶ= 𝑀𝑑𝑖ଵ 𝑑𝑡

Eq. 58 Where: vs is the ignition power supply,tipically a 12 V battery, R1 and L1 are the resistance and inductance of primary winding respectively, Cs is the equivalent capacitance of the ignition transistor in off-state, e21 is the secondary-to-primary induced voltage, whereas e12 represents the primary-to-secondary induction. Rion is the resistance of the microammeter or the low ohmic value of the limiter when it clamps due to the ignition strike, Rw is the resistance of the high tension wire, Rp the internal resistance of the spark plug, R2 the resistance of the secondary winding, L2 its inductance, Ug the voltage of the electrodes of spark plug, Csp its capacitance and i2 the secondary current.

A better modeling of the transistor should include its clamping in off-state when the primary winding “kicks back” after the collapse of i1, as well as its leakage current. This is illustrated experimentally in [40]. Yet, the purpose of the next lines is to find the residual damped oscillations that occur after the ignition pulse, in which the primary winding voltage is around -20 V, having no clamping, as indicated by the details “C1” and “C2” of Figure 94. Therefore, the ignition transistor is modeled as a switch S1 in open-circuit condition.

Using the Laplace transform, and combining Eq. 53 to Eq. 58:

𝑉௦(𝑠) −𝑀𝑖ଶ(𝑡௥ି) + 𝐿ଵ𝑖ଵ(𝑡௥ି) − ଵ ௦𝑢஼௦(𝑡௥ି) = 𝑅ଵ𝐼ଵ(𝑠) + 𝐿ଵ𝑠 𝐼ଵ(𝑠) + ଵ ஼ೞ ௦𝐼ଵ(𝑠) −𝑀𝑠𝐼ଶ(𝑠)

Eq. 59 𝑈(𝑠) + 𝑀𝑖ଵ(𝑡௥ି) + 𝐿ଶ𝑖ଶ(𝑡௥ି) −1 𝑠𝑢௚(𝑡௥ି) = (𝑅௜௢௡+ 𝑅ଶ+ 𝑅௣+ 𝑅௪)𝐼ଶ(𝑠) + 𝐿ଶ𝑠 𝐼ଶ(𝑠) +

1

𝐶ௌ௣ 𝑠𝐼ଶ(𝑠) + 𝑀𝑠𝐼ଵ(𝑠)

Eq. 60 The inductive initial condition terms are:

L1 i1(tr-) = e1 ; L2 i2(tr-) = e2 ; M i1(tr-) = e12 and M i2(tr-) = e21 They contribute to the free-run oscillation of secondary current from the instant tr-. The capacitive initial voltage condition terms are: uCs(tr-) for the capacitance of transistor; ug(tr-) for the spark gap capacitance. The primary natural frequency is: 𝜔௡ଵ= ඨ1 𝐿ଵ𝐶௦

Eq. 61 Its associated damping factor is:

𝜁ଵ= 𝑅ଵ

2 ඨ𝐶௦

𝐿ଵ

Eq. 62

The secondary natural frequency is: 𝜔௡ଶ= ඨ

1

𝐿ଶ𝐶ௌ௣

Eq. 63 Its associated damping factor is:

𝜁ଶ= (𝑅௜௢௡+ 𝑅ଶ+ 𝑅௣+ 𝑅௪)

2

ඨ𝐶ௌ௣ 𝐿ଶ

Eq. 64 The transformed secondary current is written as:

𝐼ଶ(𝑠) = 𝑁ଵ(𝑠) 𝐷(𝑠) 𝑈(𝑠) + 𝑁ଶ(𝑠) 𝐷(𝑠) 𝑉௦(𝑠) + 𝑁ଷ(𝑠) 𝐷(𝑠) 𝑖ଵ(𝑡௥ି) + 𝑁ସ(𝑠) 𝐷(𝑠) 𝑖ଶ(𝑡௥ି) + 𝑁ହ(𝑠) 𝐷(𝑠) 𝑢௚(𝑡௥ି) + 𝑁଺(𝑠) 𝐷(𝑠) 𝑢஼௦(𝑡௥ି)

Eq. 65 Where:

𝐷(𝑠) = ቆ1 + 𝑀ଶ 𝐿ଵ𝐿ଶ ቇ𝑠ସ+ 2(𝜁ଵ𝜔௡ଵ+ 𝜁ଶ𝜔௡ଶ)𝑠ଷ+ (𝜔௡ଵ ଶ+ 4𝜁ଵ𝜔௡ଵ𝜁ଶ𝜔௡ଶ+𝜔௡ଶ ଶ)𝑠ଶ + 2(𝜁ଵ𝜔௡ଵ𝜔௡ଶ ଶ+ 𝜁ଶ𝜔௡ଶ𝜔௡ଵ ଶ)𝑠+ 𝜔௡ଵ ଶ𝜔௡ଶ ଶ

Eq. 66 D(s) is of a fourth order and includes the natural frequencies of primary and secondary windings. This polynomial is common to all the power sources of Eq. 65. Therefore, the oscillatory behaviour of the current i2 is strongly determined by the location within the complex plane of the polynomial poles [207]. The numerator for the ion promoter source U(s) is: 𝑁ଵ(𝑠) = ൬1 𝐿ଶ ൰(𝑠ଷ+ 2𝜁ଵ𝜔௡ଵ𝑠ଶ+ 𝜔௡ଵ ଶ𝑠)

Eq. 67 The numerator for the battery supply Vs(s) is: 𝑁ଶ(𝑠) = −൬𝑀 𝐿ଵ𝐿ଶ ൰(𝑠ଷ)

Eq. 68 The numerator for the contribution of the initial primary current i1(tr-) is: 𝑁ଷ(𝑠) = ൬𝑀 𝐿ଶ ൰(2𝜁ଵ𝜔௡ଵ𝑠ଶ+ 𝜔௡ଵ ଶ𝑠)

Eq. 69

The numerator for the contribution of the initial secondary current i2(tr-) is: 𝑁ସ(𝑠) = (ቆ1 + 𝑀ଶ 𝐿ଵ𝐿ଶ ቇ𝑠ଷ+ 2𝜁ଵ𝜔௡ଵ𝑠ଶ+ 𝜔௡ଵ ଶ𝑠)

Eq. 70 The numerator for the contribution of the initial spark gap voltage ug(tr-) is: 𝑁ହ(𝑠) = ൬1 𝐿ଶ ൰(𝑠ଶ+ 2𝜁ଵ𝜔௡ଵ𝑠+ 𝜔௡ଵ ଶ)

Eq. 71 And the numerator for the contribution of the initial off-state transistor capacitance voltage uCs(tr-) is:

𝑁଺(𝑠) = ൬𝑀 𝐿ଵ𝐿ଶ ൰(𝑠ଶ) Eq. 72

Using the expressions Eq. 61 to Eq. 72, a simulation is performed with Matlab ®, evaluating: -Step response of N1(s ) / D(s), having the promoter U as excitation. -Step response of N2(s ) / D(s), having vs as excitation. -Impulse response of N3(s ) / D(s), having i1(tr-) as excitation. -Impulse response of N4(s ) / D(s), having i2(tr-) as excitation.

- Impulse response of N5(s ) / D(s), having ug(tr-) as excitation.

-Impulse response of N6(s ) / D(s), having uCs(tr-) as excitation. And summing up the six expressions to get the transient response..

The time t = tr- corresponds to the beginning of the free-run oscillation D1 after the collapse of primary current B1, as presented in Figure 94. In that position, i1(tr-) = 0.

The parameters of the simulation are:

R1 = 1,5 , R2 = 13 k , Rion = 10 k , Rw = 5 k , Rp = 5 k , Cs = 80 pF , CSp = 50 pF , L2 = 32 H , L1 = 8 mH , U = 200 V, vs = 12 V, i2(tr-) = 1 mA, ug(tr-) = 200V, uCs(tr-)= 12 V.

The values presented were previously measured on the actual Low-side circuit, except for CSp, which is taken from [41]. Cs corresponds to the capacitance of the igntion transistor in off-state, taken from the Cres parameter of a typical FGA25N120ANTDU IGBT [208].

Figure 111. Results of the simulations of free-run: a) with primary coupling , b) without coupling.

Figure 111.a) presents the outcome of the simulation, including all the coupling expressions from Eq. 61 to Eq. 72, whereas Figure 111.b) shows the waveforms obtained without the coupling with primary winding. The uncoupled expression is:

𝑈(𝑠) + 𝐿ଶ𝑖ଶ(𝑡௥ି) −1 𝑠𝑢௚(𝑡௥ି) = (𝑅௜௢௡+ 𝑅ଶ+ 𝑅௣+ 𝑅௪)𝐼ଶ(𝑠) + 𝐿ଶ𝑠 𝐼ଶ(𝑠) +

1

𝐶ௌ௣ 𝑠𝐼ଶ(𝑠) Eq. 73 The transformed secondary current, from Eq. 73 is:

𝐼ଶ(𝑠) = 1 𝐿ଶ

൬

𝑠 𝑠ଶ+ 2𝜁ଶ𝜔௡ଶ𝑠ଶ+ 𝜔௡ଶଶ൰(𝑈(𝑠) + 𝐿ଶ𝑖ଶ(𝑡௥ି) −1 𝑠𝑢௚(𝑡௥ି)) Eq. 74 Since the free-run oscillation includes an exponential attenuation envelope, there is a damped frequency:

𝜔ௗଶ= 𝜔௡ଶට1 −𝜁ଶ ଶ= ඨ

1

𝐿ଶ𝐶௦௣ −(𝑅௜௢௡+ 𝑅௪+ 𝑅௣+ 𝑅ଶ)ଶ 4𝐿ଶ ଶ

Eq. 75 And the time between crests Td [207], highlighted in Figure 111, is calculated as: 𝑇ௗ= 2𝜋 𝜔ௗଶ

Eq. 76

As it is shown in Figure 111, the oscillation frequencies remain unchanged with or without primary coupling, confirming that the secondary damped frequency fd = 1 / Td, expressed in hertz, is dominant, while the contribution of primary is scarce. The capacitor Cs, due to the low off-state capacitance value of the IGBT, is incapable of tuning the primary and secondary windings for coupled interchange.

The damped frequency d2 depends on the self-inductance of secondary winding L2 and on the spark capacitance Csp. The value of Csp in turn, is influenced by the dielectric properties of the working gases present in the spark gap.

A general expression for the capacitance is [46], [48]:

𝐶= 𝜀∙𝜀଴ 𝐴 𝑑 Eq. 77.

Where ε is the dimensionless relative permittivity of the substance in the gap, ε0 = 8,8542•10-12 [ F / m ] is the permitivity of vacuum [48], A is the surface area of electrodes, assumed equal for both electrodes, and d is the distance between plates.

During the ringing after the ignition pulse, if the substance in the spark gap is air, the relative dielectrical permitivity is 1,00059, resulting in a relatively low spark plug capacitance Csp. However, if the spark plug is filled with a compressed air-fuel mixture, the effective capacitance Csp increases. This increase in capacitance reduces the damped frequency ωd2 of the oscillatory response.

Figure 112. Free run frequencies observed on ion current signal as function of time. Figure 112 corresponds to a single engine cycle, displaying the measured ionization current and the unfiltered in-cylinder pressure as functions of time. The ionization signal exhibitss four distinct saturation regions, each associated with a corresponding damped oscillation frequency. The test conditions are a load torque of 13,16 N•m, a mean AFR of 13,97, a mean rotational speed of 1929 min-1, and an engine temperature of 80,8°C, ionization circuit parameters Rion = 10 k, G = 1 (resulting in an amplification factor of 104), power source adjusted to U = 200 V, throttle opening of42%, and spark advance of 10 degrees crank angle.

Oscillations f1 and f3 correspond to the initial phase of primary coil charging, while the transistor is in the on-state. f4 corresponds to the damped oscillation of the wasted-spark event, occurring after the transistor is switched off, preceded by f3, with air present in the spark gap. On the other hand, f2 is the damped oscillation occurring after the transistor is switched off following f1, prior to ignition, when a compressed air-fuel mixture is present between the spark plug electrodes.

Figure 113. Details of the ionization current waveform: a) and c) Oscillations of transistor engaging, b) free-running oscillation in wasted spark and d) free-running oscillation preceding the ionization of the air-fuel mixture.

The damped frequencies obtained for this particular test are as follows: f1 = 10602 Hz, f3 = 10546 Hz, f4 = 3512 Hz and f2 = 2753 Hz, portrayed in detail in Figure 113. Frequencies f1 and f3 correspond to forced oscillations induced by the gradual conduction of the ignition transistor as it approaches saturation. These oscillations are characterized by relatively high frequencies.

The oscillation frequencies were quantified using the meanfreq function of Matlab®, applied to the ionization current signal as a function of the discrete samples. The channel’s sampling frequency of 500 kS/s, was used as an input parameter, yielding the frequency estimates in hertz. It was observed that, with the preignition depicted in Figure 109.b) of section 4.4.2, the meanfreq function failed to reliable compute frequencies f1 and f2 due to substantial deformation and saturation of the ionization signal under these conditions.

The wasted spark frequency f4 is higher than the free-running oscillation frequency f2, observed during the combustion phase. This behavior is consistent with Eq. 75 and Eq. 76, which confirm that the effective spark gap capacitance increases in the presence of the airfuel mixture. In this sample test, the total capacitance Csp during the wasted spark event is approximately 63 pF, while it increases to 103 pF during the pre-combustion phase. These results suggest that monitoring the variation in damped frequencies between the wasted-spark and combustion phases could be a promising method for detecting misfire events.

However, Appendix 1 provides an opposing trend when a prechamber is used, resulting in a higher frequency of oscillation during combustion than that observed during the wasted spark phase. This difference can be attributed to the pre-chamber’s shielding effect, which isolates the spark plug electrodes from the bulk flow structures (swirl, tumble, squish), present in the main chamber. As a result, the effective spark gap capacitance perceived by the circuit is altered.

Effect of Rion in free-run damped frequency.

The damped frequency d2, as highlighted by Eq. 75, depends on the secondary inductance L2 and the total capacitance Csp, which together define the secondary circuit’s natural frequency. However, the total secondary loop resistance has an additional effect on d2. Particularly, the ohmic value of the microammeter resistor Rion, can significantly affect the system’s damping and therefore must be considered as a parameter of influence.

To investigate this effect, simulations were performed in Matlab®, using Eq. 74 and Eq. 75, to assess sensitivity of d2 to variations in Rion. The common simulationparameters were as follows:

R1 = 1,5 , R2 = 13 k , Rw = 5 k , Rp = 5 k , Cs = 80 pF , CSp = 63 pF , L2 = 32 H , L1 = 8 mH , U = 200 V, vs = 12 V, i2(tr-) = 1 mA, ug(tr-) = 200V, uCs(tr-)= 12 V.

Figure 114 presents the simulation results for Rion = 100 , Rion = 1 k, Rion = 10 k, Rion = 30 k, Rion =100 k and Rion = 300 k. The sum of the remaining resistors of the secondary loop R2, Rp and Rw is 23 k. When Rion is lower than (R2+Rp+Rw), the resulting damped oscillation is characterized by a frequency near 3544 Hz and exhibits a prolonged duration.

Conversely, when Rion excedes (R2+Rp+Rw), the oscillation frequency decreases, and the duration of the oscillation is reduced due to increased damping. When Rion = 30 k, comparable to (R2+Rp+Rw), a frequency around 3544 Hz and a moderate oscillation duration is observed.

Figure 114. Results of the simulation of free-run without primary coupling for diverse values of Rion.

Based on the results of these simulations, it can be concluded that very low values of Rion cause a prolonged free-running oscillation, which can eventually mask the chemical and thermal ionization contributions in the ionization current signal. Conversely, high values of Rion result in shorter oscillations that solves the masking of fuel contributions on ionization waveform. At the same time, with high Rion, the amplitude of oscillation can activate the

clamping of the limiter circuit of the microammeter, as it is shown in the actual tests portrayed in Figure 115.

Figure 115. Comparison of average ionization current signals for varying Rion. Common engine settings: throttle opening = 100% ; spark advance = 12° BTDC. Ion amplifier gain G = 1, except for Rion = 100 .

The values tested, as shown in Figure 115, correspond to an engine speed of 1900 min-1 and a load torque of 14 N•m. For Rion = 100 , a prolonged oscillatory waveform is observed in comparison with the other Rion settings, along with significant masking of the chemiionization signal.

Therefore, in the present work, although lower ohmic values were initially considered in order to minimize the burden voltage across the microammeter, it was found that, for ion current detection using Low-side circuits, the best performance is achieved by selecting Rion to strike an appropriate balance between the duration of the free-running oscillation and the amplitude of the detected signal. For this particular configuration, setting Rion close to the sum of the other resistances in the secondary loop (R2+Rp+Rw), -that is, between 10 k and 30 k- yields the most favorable results.

5.2 The area of the ionization current curve during combustion and its

relationship with pressure.

In his work, Eriksson [25] analyzes the thermal peak of ionization current to look for a correlation with the maximum in-cylinder pressure; he explores by the area under the ion signal curve, focusing on the angular position of its centroid for the same purpose. Additionally, he proposes fitting the ionization signal with a sum of two gaussian functions to model the contours of the ion signal, particularly its peaks.

Shimasaki et al. [58] relies on an electronic hardware integration of the ionization current, combined with an electronic gating for the integration to detect misfire conditions, based on the magnitude of the ionization area signal. A misfire event is detected and highlighted when the cumulative ionization signal rises to a value below a predefined magnitude threshold. Their work uses a CDI ignition with a Low-side scheme, applied to a Formula One engine.

Budko et al. [149] compute the definite integral of the ionization current with respect to the crank angle to correlate it with the pressure peak, stating that, irrespective of the variations in the original ionization signal on each cycle, the crank angle at which the integral reaches 80% to 90% of its maximum value tends to coincide with the crank angle of the pressure peak. The engine and fuel type in their study are not specified.

Song et al. [209] apply a similar to Budko’s criterion, stating that the 80% of maximum value of the time-integrated ionization current concurs with the pressure peak moment. They employ a constant volume chamber, fueled with natural gas enriched with hydrogen, using a High-side electrical configuration with silicon diode stacks connected to both sides of the “Tee” that joins the ignition coil, the promoter power source and the electrodes.

The present work opts for tackling an offline software-based calculation of the ionization current integral signal with respect to the crank angle, using a Low-side scheme within a TCI ignition system, to correlate the integral with the angular position of pressure peak.

The integration of ionization signal requires well-defined angular limits for the beginning and the end of its computation. Gürbüz [18] uses an independent circuital scheme for ionization current detection, to circumvent the direct contribution of ignition pulse within the electrical loop. Nevertheless, the ignition pulse comprises a strong electrical field that reaches the microammeter by induction. Thus, Gürbüz subtracts the ignition contribution (starting point of integration) that appears during motoring from the total area under the curve of ion current, resulting in a net ion area value. The end of integration is flagged as the point where the ionization signal reaches zero microamperes after the thermo-ionization phase.

Eriksson [25] sets the start of residual ignition-induced oscillations as the starting point of integration, while the end of the accumulation is taken similarly to the one previously referred criterion of [18]. In addition, Eriksson applies an electronic low-pass filter tuned to 15 kHz to the ionization signal, resulting in a reduction of the residual damped ignition oscillations, that in turn leads to a lower ignition contribution to the total magnitude of ion area.

Shimasaki et al. [58] set the start of integration immediately after the end of residual ignition oscillations, avoiding such contribution to the ion area. The end of integral is flagged in the same way as Gürbüz did, with the additional precaution of resetting the amplifier integrator before the next oscillations, related to the charging of CDI’s capacitor, take place.

Laganá et al. [113] quantify the difference in ion area with and without combustion, integrating from the beginning of the residual ignition oscillations up to the zero current flag following the thermo-ionization phase.

Budko et al. [149] include the initiation of the residual ignition oscillations in the ionization current area calculation and the zero-current as its end. .

In the present study, the integral of the ion current is computed within a window that starts at the end of residual ignition oscillations and ends at the zero-current point after thermoionization. This choice is justified by the fact that most of the aforementioned literature consider those limits of integration as the interval in which the combustion and ionization processes take place. Additionally, the free-running damped oscillation frequency is analyzed to extract information about spark gap capacitance. The ionization signal is not filtered, as preserving the frequency content is necessary for this analysis.

Figure 116 displays the captured waveforms for a sample cycle of a test performed at a target speed of 1700min-1 and a throttle opening of 42%. The signal c) is the windowed ionization signal, removing the contribution of the ignition pulse and most of the free-running oscillations. The centroid of the area of the windowed ionization current signal is indicated in the graphs. For the particular sample, the angular position of the centroid does not coincide with the pressure peak, and the majority of the area under the curve of ion signal is located to the left of the centroid.

The angular position of the thermo-ionization for the sample cycle depicted is not entirely clear, although it appears that the thermal ionization peak precedes the pressure peak, in contrast with the frequent coincidence, between these events reported in the literature.

Similarly, the angular position corresponding to the 100% of ionization integral signal appears after the pressure peak. The criterion referred by Budko et al. [149], in which the 80% - 90% interval of the maximum ionization integral signal value occurs, appears more promising to correlate with the angular position of the maximum pressure. In this particular case, the 87,3% point of the ionization integral aligns closely with the pressure peak.

The criteria of residual free-running frequency, magnitude of ion area, percentage of maximum ion integral signal coinciding with pressure peak position, are applied in the following sections to further investigate the relationship between ionization current and incylinder pressure.

Figure 116. Sample waveforms of a) full ionization current signal, b) in-cylinder pressure,

c) windowed ionization current, d) ionization current integral.

5.3 Correlations between free-run damped frequency, area of ion

current and pressure

5.3.1 Cyclic variation: pressure peak magnitude

Figure 117. a) magnitude of maximum pressure compared with ionization current area under the curve, and b) regression model of the variables.

Figure 117.a) presents the maximum in-cylinder pressures and total ionization areas for a test performed under a rotational frequency of 1673 min-1, AFR of 14,7, load torque of 15,93 N•m, engine temperature of 74,8 °C, spark advance of 20 ° BTDC, U = 200V, Rion = 10 k, and G = 1, comprising a total of 68 engine cycles. The regression model shown in Figure 117.b) provides the correlation coefficient between ionization curent area and the pressure, quantifying the capability of the total ion signal area to estimate the cyclic variations.

Figure 118 presents the results of the comparison between maximum pressures and freerunning damped frequencies, obtained under the aforementioned conditons. The coefficient of correlation is indicated in the graph (R2 = 0,5648).

Figure 119 illustrates the results of the comparison between maximum pressures and magnitudes of thermo-ionization peaks of ion current signal, obtained under the aforementioned conditons. It was found a weak correlation (R2 = 0,16).

Figure 118. Magnitude of maximum pressure compared with free-run damped frequency and regression model of the variables.

Figure 119. Magnitude of maximum pressure compared with magnitude of thermoionization peak of ionization current signal and regression model of the variables.

In all the aforementioned cases, outlier cycles were excluded by removing data associated with pre-ignition events. These cycles were identified by detecting abnormal waveforms during dwell time, which corresponded to observed pressure anomalies.

The ionization current area and the free-running damped frequency of the ionization signal, are now used as inputs to improve the correlation with the pressure peak.

The vectors of ionization current area magnitude, damped frequency and pressure peak, as functions of the cycle number, are first normalized by dividing each vector by its corresponding maximum value. Using the “CFTool” toolbox in Matlab®, the normalized ionization current area is assigned to the x-axis, the normalized damped frequency is assigned to the y-axis, and the normalized peak pressure is assigned to the z-axis. A bilinear surface fit (first-degree polynomial in both x and y) is performed, using least absolute residuals (LAR) robustness, and enabling the ‘center and scale’ option, to enhance the fit stability and robustness to potential outliers.

Figure 120. Three-dimensional surface fit modeling normalized peak pressure as a function of normalized ion signal area and damped natural frequency Figure 120 presents the surface fitting obtained from the correlation between normalized magnitude of ion area, free-run damped frequency and peak pressure.

CFTool reports a linear model:

𝑓(𝑥, 𝑦) = 𝑝00 + 𝑝10 ∗𝑥+ 𝑝01 ∗𝑦 Eq. 78 That corresponds to a 3-D plane surface characterized by the expression:

𝑝௡(𝑎௡, 𝑓௡) = 𝑘ଵ+ 𝑘ଶ∗𝑎௡+ 𝑘ଷ∗𝑓௡ Eq. 79 For the particular test performed, the resulting coefficients were: k1 = 0,6132, k2 = 0,2936, k3 = -0,1102. The input variables are the normalized ionization signal area an, and the

normalized damped frequency fn., while the output variable pn is the normalized peak pressure. As specified in Figure 120, R2 = 0,9260. This result indicates a more promising correlation between features of the secondary (and ionization) current and the magnitude of the peak pressure.

5.3.2 Cyclic variation: determination of pressure peak position

Figure 121. Peak pressure position compared with the centroid of ion area. Figure 121 illustrates the results of the comparison between the crank angle of pressure peak and the crank angle of centroid of ionization current area under the curve. The test conditions are the same as those described in section 5.3.1. For this particular test, the box plot indicates that the angular difference between peak pressure point and ionization area centroid has an average delay of 15 degrees, with an interquartile range between 12,87 and 16,57 degrees, and a coefficient of determination of R2 = 0,8.

Following the criterion proposed by Budko et al. [149], the percentage of the maximum ionization integral can be evaluated to determine the crank angular position of pressure peak.

Figure 122 presents the percentage of maximum ionization integral at which the pressure peak occurs. This result corresponds to the test performed at 1700 min-1 referred previously in section 5.3.1. It can be observed that, on average, 86,4% of ionization integral signal corresponds to the crank angle position of the peak pressure.

The box plot indicates a median value of 88,1% with an interquartile range between 85,2% and 90,7%. Deviations extend from 82% up to 98%, after excluding outliers identified through the frequency screening described in section 5.3.1. Additionally, the 61th cycle exhibits an atypical value, not captured by the damped frequency analysis, forcing a division by zero. This was corrected by assigning a zero-percent value of the maximum ionization integral signal to that cycle. These observations underline the need for a more robust and narrower criterion for estimating the crank angle of pressure peak position.

The crank angle position of the thermo-ionization peak, present in the averaged ionization current signal, is used in the literature to correlate it with the pressure peak position [18], [86], [103], exhibiting a close proximity between both angular positions.

Figure 123.a) presents the comparison between PPP and position of thermo-ionization peak, sweeping the individual cycles with the test conditions described in section 5.3.1. For the particular test, there is an average delay of 22 degrees of crank angle between PPP and thermo-ionization angular peak position. There are multiple variations that take place on the non-averaged ionization current signal, likely due to the bulk flows, that impair the softwaredetection of thermo-ionization peak positions, and with it, the expected close correspondence between the aforementioned angular positions.

In addition, Figure 123.b) shows the correlation between the normalized angular position of thermo-ionization peak in the x axis, and the normalized crank angle of in-cylinder pressure peak position in the y axis. Although the coefficient of correlation appears promising (R2 = 0,78), there is a circular dispersion of data shown in the upper left corner of the graph. The remaining points in the graph are outliers, corresponding to the high points of the thermal peak angle in Figure 123.a), not detected by the damped frequency screening. These findings suggest caution with the correlation.

Figure 122. Percentage of maximum ionization current integral signal corresponding to the crank angle position of peak pressure.

Figure 123. a) Peak pressure position compared with angular positon of thermo-ionization peak, and b) correlation between them.

Figure 124 highlights that the pressure peak and its corresponding angular position exhibit a correlation in which a higher pressure tends to occur earlier in crank angle position. The test conditions correspond to those described in section 5.3.1. Thus, this finding can be used to formulate a two-variable correlation, where one ionization current characteristic is related to the magnitude of pressure peak on one axis, and another ionization current feature is related to the angular position of pressure on another axis.

Figure 124. Correlation between normalized pressure peak and its normalized angular position

Figure 125 presents a comparison of results obtained using “CFTool” toolbox in Matlab® to perform a two-input surface fit, following a similar procedure to that described in 5.3.1. In one case, Figure 125.a), the input vectors are the normalized damped frequency, assigned to the y-axis, and the normalized centroid of the ionization current area, assigned to the xaxis. Now, Figure 125.b) changes the assignment of the y-axis to the normalized ionization current area, while the x-axis remains the normalized centroid of the ionization current area. Both the damped frequency and ionization current area exhibit certain correlation with the magnitude of pressure peak, as explained in the results of section 5.3.1. On the other hand, the centroid of the ion current area correlates with the crank angle position of pressure peak, which is assigned to the z-axis.

The fit options are maintained as first-degree polynomials, both for x and y axes, with least absolute residuals (LAR) robustness, and center and scale option enabled.

The polynomial fit expressed in Eq. 78 can be applied to both cases a) and b) of Figure 125, For case a), this results in a three-dimensional planar surface described by the following expression:

𝜑௣௣௣൫௖௡, 𝑓௡൯= 𝑘ଵ+ 𝑘ଶ∗௖௡+ 𝑘ଷ∗𝑓ௗ௡ [𝑑𝑒𝑔] Eq. 80 For the particular test, the fitted coefficients are: k1 = 376,3, k2 = -0,1583, k3 = 1,3630. As inputs, cn is the normalized centroid of ionization current area, fdn is the normalized damped frequency, and the output ppp is the angle of pressure peak. As indicated in Figure 125.a), the coefficient of determination is R2 = 0,9105.

Figure 125. Comparison of three-dimensional surface fitting results for predicting crank angle position of pressure peak: a) normalized damped natural frequency and centroid of ion current area; b) normalized ion current area and centroid of ion current area. The zaxis represents the crank angle position of pressure peak

The second approach leads to a 3-D plane surface characterized by the expression:

𝜑௣௣௣൫௖௡, 𝑎௡൯= 𝑘ଵ+ 𝑘ଶ∗௖௡+ 𝑘ଷ∗𝑎௡ [𝑑𝑒𝑔] Eq. 81 For the particular test, the fitted coefficients are: k1 = 376,5, k2 = 0,3473, k3 = -0,9830, and the second input an is the normalized ionization current area. As indicated in Figure 125.b), the coefficient of determination is R2 = 0,8786.

Both models achieve good R2 values; however, the configuration using normalized damped frequency as the second input exhibits a higher degree of correlation.

5.3.3 Effect of engine variables on the ionization current based estimation of

pressure peak and its crank angular position Table 18 summarizes the correlation results obtained from multiple engine operating conditions, highlighting the performance of various ion current features in estimating peak pressure and its crank angle position.

Test

1

Engine conditions Ion sensing conditions Throttle Load Eng. spd Eng temp Spark adv U Rion [%] [N.m] [min-1]

AFR

[°C]

[°] BTDC

[V] [] G

15

6,18

1957

15,31

56,2

10

200

10k

1

In-cylinder pressure magnitude correlations Area ion pn = m an+b Frequency pn = m fdn+b Thermo-ion pn = m ithn+b .m .b R2 .m .b R2 .m .b R2

0,0009

0,9862

0,04

0,0291

0,9691

0,06

-0,00074

0,9900

0,02

Area ion-frequency pn=k1+k2 an+k3 fdn p-ppp pn = m pppn+b k1 k2 k3 R2 .m .b R2 Cycles

0,9866

-0,0002

0,0022

0,07

-0,7781

1,7534

0,23

78

In-cylinder pressure peak position correlations ppp-cent % integral ppp -th Centroid pppn= m cn+b Thermo-ion  pppn= m thn+b [°] [%] [°] .m .b R2 .m .b R2

14,45

x

11,8

0,0002

0,9835

0,89

0,00058

0,9833

0,86

Centroid-frequency ppp = k1+k2 cn+k3 fdn Centroid-Area ppp = k1+k2 cn+k3 an k1 k2 k3 R2 k1 k2 k3 R2

0,9868

-0,0009

0,0022

0,09

0,9871

-0,0009

-0,0024

0,24

Test

2

Engine conditions Ion sensing conditions Throttle Load Eng. spd Eng temp Spark adv U Rion [%] [N.m] [min-1]

AFR

[°C]

[°] BTDC

[V] [] G

12

7,09

1695

15,4

52,7

20

200

10k

1

In-cylinder pressure magnitude correlations Area ion pn = m an+b Frequency pn = m fdn+b Thermo-ion pn = m ithn+b .m .b R2 .m .b R2 .m .b R2

-0,0121

0,3738

0,94

-0,2267

0,565

0,94

-0,0026

0,3703

0,93

Area ion-frequency pn=k1+k2 an+k3 fdn p-ppp pn = m pppn+b k1 k2 k3 R2 .m .b R2 Cycles

0,3745

0,0014

-0,0462

0,92

1,0295

-0,6232

0,94

68

In-cylinder pressure peak position correlations ppp-cent % integral ppp -th Centroid pppn= m cn+b Thermo-ion  pppn= m thn+b [°] [%] [°] .m .b R2 .m .b R2

1,7

58,7

8,3

-0,1187

1,0698

0,09

0,0299

0,9364

0,1

Centroid-frequency ppp = k1+k2 cn+k3 fdn Centroid-Area ppp = k1+k2 cn+k3 an k1 k2 k3 R2 k1 k2 k3 R2

370,3

-1,59

0,7108

0,01

371

-7,307

7,891

0,22

Test

3

Engine conditions Ion sensing conditions Throttle Load Eng. spd Eng temp Spark adv U Rion [%] [N.m] [min-1]

AFR

[°C]

[°] BTDC

[V] [] G

14

7,24

1769

15,74

46,2

20

200

10k

1

In-cylinder pressure magnitude correlations Area ion pn = m an+b Frequency pn = m fdn+b Thermo-ion pn = m ithn+b .m .b R2 .m .b R2 .m .b R2

-0,0183

0,3774

0,95

-0,3028

0,5808

0,94

0,1084

0,3236

0,95

Area ion-frequency pn=k1+k2 an+k3 fdn p-ppp pn = m pppn+b k1 k2 k3 R2 .m .b R2 Cycles

0,3756

-0,0043

-0,0227

0,91

1,2799

-0,8616

0,94

70

In-cylinder pressure peak position correlations ppp-cent % integral ppp -th Centroid pppn=mcn+b Thermo-ion  pppn= m thn+b [°] [%] [°] .m .b R2 .m .b R2

3

57,4

10,1

-0,1718

1,1209

0,03

-0,0831

1,0371

0,05

Centroid-frequency ppp = k1+k2 cn+k3 fdn Centroid-Area ppp = k1+k2 cn+k3 an k1 k2 k3 R2 k1 k2 k3 R2

370,8

-1,877

-2,252

0,32

371,4

-8,117

7,247

0,29

Test

4

Engine conditions Ion sensing conditions Throttle Load Eng. spd Eng temp Spark adv U Rion [%] [N.m] [min-1]

AFR

[°C]

[°] BTDC

[V] [] G

40

11,51

1979

14,06

63,7

20

200

10k

1

In-cylinder pressure magnitude correlations Area ion pn = m an+b Frequency pn = m fdn+b Thermo-ion pn = m ithn+b .m .b R2 .m .b R2 .m .b R2

0,2498

0,7813

0,02

-0,3449

1,0752

0,52

-0,1329

0,922

0,05

Area ion-frequency pn=k1+k2 an+k3 fdn p-ppp pn = m pppn+b k1 k2 k3 R2 .m .b R2 Cycles

0,8551

0,0023

-0,0535

0,54

-16,2716

16,9351

0,89

79

In-cylinder pressure peak position correlations ppp-cent % integral ppp -th Centroid pppn=mcn+b Thermo-ion  pppn= m thn+b [°] [%] [°] .m .b R2 .m .b R2

15,68

89,6

20

-0,1451

1,1294

0,04

-0,1647

1,1453

0,15

Centroid-frequency ppp = k1+k2 cn+k3 fdn Centroid-Area ppp = k1+k2 cn+k3 an k1 k2 k3 R2 k1 k2 k3 R2

379,2

0,0668

1,264

0,5

379,1

-0,6159

-0,6499

0,04

Test

5

Engine conditions Ion sensing conditions Throttle Load Eng. spd Eng temp Spark adv U Rion [%] [N.m] [min-1]

AFR

[°C]

[°] BTDC

[V] [] G

42

13,16

1928

13,97

80,8

10

200

10k

1

In-cylinder pressure magnitude correlations Area ion pn = m an+b Frequency pn = m fdn+b Thermo-ion pn = m ithn+b .m .b R2 .m .b R2 .m .b R2

0,0913

0,5546

0,44

-0,1933

0,6723

0,63

-0,0935

0,6347

0,58

Area ion-frequency pn=k1+k2 an+k3 fdn p-ppp pn = m pppn+b k1 k2 k3 R2 .m .b R2 Cycles

0,5995

-0,0092

-0,0377

0,92

-10,8172

11,3043

0,84

74

In-cylinder pressure peak position correlations ppp-cent % integral ppp -th Centroid pppn=mcn+b Thermo-ion  pppn= m thn+b [°] [%] [°] .m .b R2 .m .b R2

13,5

80,9

14,6

-0,1079

1,0942

0,59

-0,0959

1,0804

0,66

Centroid-frequency ppp = k1+k2 cn+k3 fdn Centroid-Area ppp = k1+k2 cn+k3 an k1 k2 k3 R2 k1 k2 k3 R2

385,6

1,414

0,5868

0,94

385,9

1,673

-0,1784

0,93

Test

6

Engine conditions Ion sensing conditions Throttle Load Eng. spd Eng temp Spark adv U Rion [%] [N.m] [min-1]

AFR

[°C]

[°] BTDC

[V] [] G

100

14,28

1835

14,05

62,1

12

200

10k

1

In-cylinder pressure magnitude correlations Area ion pn = m an+b Frequency pn = m fdn+b Thermo-ion pn = m ithn+b .m .b R2 .m .b R2 .m .b R2

0,1036

0,8519

0,07

-0,1506

0,979

0,18

-0,0389

0,9487

0,009

Area ion-frequency pn=k1+k2 an+k3 fdn p-ppp pn = m pppn+b k1 k2 k3 R2 .m .b R2 Cycles

0,9234

0,0061

-0,0155

0,05

-11,7174

12,5539

0,78

72

In-cylinder pressure peak position correlations ppp-cent % integral ppp -th Centroid pppn= mcn+b Thermo-ion  pppn= m thn+b [°] [%] [°] .m .b R2 .m .b R2

7,8

76,7

5,75

0,0109

0,9811

0,003

-0,0437

1,0345

0,04

Centroid-frequency ppp = k1+k2 cn+k3 fdn Centroid-Area ppp = k1+k2 cn+k3 an k1 k2 k3 R2 k1 k2 k3 R2

380,3

0,0095

0,0971

0,014

380,3

0,0643

-0,1871

0,005

Test

7

Engine conditions Ion sensing conditions Throttle Load Eng. spd Eng temp Spark adv U Rion [%] [N.m] [min-1]

AFR

[°C]

[°] BTDC

[V] [] G

42

14,55

1812

14,09

83,3

15

200

10k

1

In-cylinder pressure magnitude correlations Area ion pn = m an+b Frequency pn = m fdn+b Thermo-ion pn = m ithn+b .m .b R2 .m .b R2 .m .b R2

0,13

0,539

0,93

-0,103

0,6419

0,92

0,0047

0,5715

0,88

Area ion-frequency pn=k1+k2 an+k3 fdn p-ppp pn = m pppn+b k1 k2 k3 R2 .m .b R2 Cycles

0,6128

0,0893

-0,0033

0,95

-6,6331

7,165

0,98

72

In-cylinder pressure peak position correlations ppp-cent % integral ppp -th Centroid pppn=mcn+b Thermo-ion  pppn= m thn+b [°] [%] [°] .m .b R2 .m .b R2

13,7

76,55

19,6

0,0621

0,9331

0,92

-0,0474

1,0388

0,92

Centroid-frequency ppp = k1+k2 cn+k3 fdn Centroid-Area ppp = k1+k2 cn+k3 an k1 k2 k3 R2 k1 k2 k3 R2

379,1

3,376

1,145

0,94

376,5

0,3473

-0,983

0,95

Test

8

Engine conditions Ion sensing conditions Throttle Load Eng. spd Eng temp Spark adv U Rion [%] [N.m] [min-1]

AFR

[°C]

[°] BTDC

[V] [] G

42

15,93

1673

14,7

74,8

20

200

10k

1

In-cylinder pressure magnitude correlations Area ion pn = m an+b Frequency pn = m fdn+b Thermo-ion pn = m ithn+b .m .b R2 .m .b R2 .m .b R2

0,3632

0,5443

0,76

-0,6134

0,9074

0,56

-0,0093

0,6889

0,16

Area ion-frequency pn=k1+k2 an+k3 fdn p-ppp pn = m pppn+b k1 k2 k3 R2 .m .b R2 Cycles

0,6853

0,04854

-0,01353

0,93

-14,9962

15,5886

0,77

65

In-cylinder pressure peak position correlations ppp-cent % integral ppp -th Centroid pppn=mcn+b Thermo-ion  pppn= m thn+b [°] [%] [°] .m .b R2 .m .b R2

14,55

86,4

21,6

-0,2175

1,2085

0,8

-0,11

1,0984

0,79

Centroid-frequency ppp = k1+k2 cn+k3 fdn Centroid-Area ppp = k1+k2 cn+k3 an k1 k2 k3 R2 k1 k2 k3 R2

376,3

-0,1583

1,363

0,91

376,5

0,3473

-0,983

0,88

Test

9

Engine conditions Ion sensing conditions Throttle Load Eng. spd Eng temp Spark adv U Rion [%] [N.m] [min-1]

AFR

[°C]

[°] BTDC

[V] [] G

42

17,19

1575

14,08

86,9

17

200

10k

1

In-cylinder pressure magnitude correlations Area ion pn = m an+b Frequency pn = m fdn+b Thermo-ion pn = m ithn+b .m .b R2 .m .b R2 .m .b R2

0,2148

0,6463

0,62

-0,2302

0,8722

0,57

0,1046

0,6577

0,39

Area ion-frequency pn=k1+k2 an+k3 fdn p-ppp pn = m pppn+b k1 k2 k3 R2 .m .b R2 Cycles

0,7342

0,0146

-0,0353

0,45

-10,0311

10,65

0,97

60

In-cylinder pressure peak position correlations ppp-cent % integral ppp -th Centroid pppn=mcn+b Thermo-ion  pppn= m thn+b [°] [%] [°] .m .b R2 .m .b R2

9,47

76,32

19,2

0,052

0,9375

0,71

-0,1012

1,0855

0,68

Centroid-frequency ppp = k1+k2 cn+k3 fdn Centroid-Area ppp = k1+k2 cn+k3 an k1 k2 k3 R2 k1 k2 k3 R2

375,7

2,815

1,544

0,84

375,9

4,137

-1,533

0,85

Test

10

Engine conditions Ion sensing conditions Throttle Load Eng. spd Eng temp Spark adv U Rion [%] [N.m] [min-1]

AFR

[°C]

[°] BTDC

[V] [] G

13

6,66

1918

15,43

63,6

10

200

100

100

In-cylinder pressure magnitude correlations Area ion pn = m an+b Frequency pn = m fdn+b Thermo-ion pn = m ithn+b .m .b R2 .m .b R2 .m .b R2

-0,0014

0,9866

0,15

0,0465

0,9441

0,01

-0,0048

0,9876

0,12

Area ion-frequency pn=k1+k2 an+k3 fdn p-ppp pn = m pppn+b k1 k2 k3 R2 .m .b R2 Cycles

0,9859

-0,0007

0,0027

0,11

-0,3758

1,3578

0,09

77

In-cylinder pressure peak position correlations ppp-cent % integral ppp -th Centroid pppn=mcn+b Thermo-ion  pppn= m thn+b [°] [%] [°] .m .b R2 .m .b R2

17

X

12

-0,0005

0,9881

0,89

0,0077

0,9802

0,87

Centroid-frequency ppp = k1+k2 cn+k3 fdn Centroid-Area ppp = k1+k2 cn+k3 an k1 k2 k3 R2 k1 k2 k3 R2

358,7

-0,0833

-0,1057

0,91

358,6

-0,0452

0,000008

0,89

Test

11

Engine conditions Ion sensing conditions Throttle Load Eng. spd Eng temp Spark adv U Rion [%] [N.m] [min-1]

AFR

[°C]

[°] BTDC

[V] [] G

15

6,36

1991

15,14

59,6

10

200

1000

10

In-cylinder pressure magnitude correlations Area ion pn = m an+b Frequency pn = m fdn+b Thermo-ion pn = m ithn+b .m .b R2 .m .b R2 .m .b R2

-0,0042

0,9864

0,002

0,01321

0,9747

0,03

-0,0064

0,9873 8,00E-04 Area ion-frequency pn=k1+k2 an+k3 fdn p-ppp pn = m pppn+b k1 k2 k3 R2 .m .b R2 Cycles

0,9852

-0,0007

0,0012

0,037

-0,8179

1,7925

0,16

79

In-cylinder pressure peak position correlations ppp-cent % integral ppp -th Centroid pppn=mcn+b Thermo-ion  pppn= m thn+b [°] [%] [°] .m .b R2 .m .b R2

17

X

12,6

-8E-06

0,9854

0,8

0,0007

0,9848

0,8

Centroid-frequency ppp = k1+k2 cn+k3 fdn Centroid-Area ppp = k1+k2 cn+k3 an k1 k2 k3 R2 k1 k2 k3 R2

359

0,1371

-0,5929

0,92

358

0,04309

-0,0023

0,9

Test

12

Engine conditions Ion sensing conditions Throttle Load Eng. spd Eng temp Spark adv U Rion [%] [N.m] [min-1]

AFR

[°C]

[°] BTDC

[V] [] G

42

15,48

1833

13,72

94,7

12

200

100k

1

In-cylinder pressure magnitude correlations Area ion pn = m an+b Frequency pn = m fdn+b Thermo-ion pn = m ithn+b .m .b R2 .m .b R2 .m .b R2

0,0757

0,4717

0,904

-0,0656

0,5577

0,92

-0,3092

0,8077

0,91

Area ion-frequency pn=k1+k2 an+k3 fdn p-ppp pn = m pppn+b k1 k2 k3 R2 .m .b R2 Cycles

0,5321

0,0032

-0,0254

0,91

-6,6202

7,0979

0,98

73

In-cylinder pressure peak position correlations ppp-cent % integral ppp -th Centroid pppn=mcn+b Thermo-ion  pppn= m thn+b [°] [%] [°] .m .b R2 .m .b R2

1,8

65

19,7

0,0753

0,9256

0,93

-0,0649

1,0546

0,93

Centroid-frequency ppp = k1+k2 cn+k3 fdn Centroid-Area ppp = k1+k2 cn+k3 an k1 k2 k3 R2 k1 k2 k3 R2

381,6

4,209

0,626

0,93

381,8

4,287

-0,2383

0,92

Test

13

Engine conditions Ion sensing conditions Throttle Load Eng. spd Eng temp Spark adv U Rion [%] [N.m] [min-1]

AFR

[°C]

[°] BTDC

[V] [] G

100

14,36

1957

14,05

78

12

200

100k

1

In-cylinder pressure magnitude correlations Area ion pn = m an+b Frequency pn = m fdn+b Thermo-ion pn = m ithn+b .m .b R2 .m .b R2 .m .b R2

0,0992

0,8433

0,002

-0,0498

0,9489

0,09

-0,2971

1,207

0,02

Area ion-frequency pn=k1+k2 an+k3 fdn p-ppp pn = m pppn+b k1 k2 k3 R2 .m .b R2 Cycles

0,9224

0,0138

-0,0043

0,11

-11,653

12,4763

0,78

78

In-cylinder pressure peak position correlations ppp-cent % integral ppp -th Centroid pppn=mcn+b Thermo-ion  pppn= m thn+b [°] [%] [°] .m .b R2 .m .b R2

0,4

54,7

1,6

0,0514

0,9407

0,18

0,0195

0,9722

0,13

Centroid-frequency ppp = k1+k2 cn+k3 fdn Centroid-Area ppp = k1+k2 cn+k3 an k1 k2 k3 R2 k1 k2 k3 R2

381,1

0,0661

0,1978

0,9

381,1

0,2721

-0,4241

0,11

Table 18. Summary of correlation results between ion current features and pressure peak characteristics across engine operating conditions.

Table 18 is structured to present, for each test, the following information:  Actual engine operating conditions: throttle opening, load torque, engine speed, airfuel ratio (AFR), engine temperature and spark advance.

 Ion current sensing configuration, promoter power source voltage, ion sensing resistance Rion and amplifier gain G.

 Correlations with peak pressure magnitude: linear regression coefficients m and b, and coefficients of determination R2 obtained by comparing normalized ionization current area an (Area Ion), normalized damped frequency fdn (Frequency), normalized peak magnitude of thermo-ioinzation ithn (Thermo-ion), and a linear combination of parameters (Area ion-frequency) with the normalized peak pressure pn.  Correlation (p-ppp) between peak pressure magnitude and its corresponding PPP, including the linear regression coefficients m, b and R2 for the normalized angle pppn and normalized peak pressure pn.

 Timing relationship to peak pressure: the angular difference between the crank angle of pressure peak ppp and the centroid of the ionization current area cent (ppp-cent). Next, it shows the percentage of the integrated ionization current signal reached at the crank angle of pressure peak (% integral). An “x” indicates that the calculation did not converge for that test. In addition, it is presented the angular difference between the crank angle of pressure peak and the angular position of the thermoionization peak of ion signal th (ppp - th). Here also, there are grouped the coefficient of correlation between the normalized centroid of the ion area cn and normalized PPP, pppn (Centroid), the coefficient of correlation between the normalized angular position of thermo-ionization peak thn and normalized PPP pppn (Thermo-ion ), and coefficients of correlation of PPP with two-variable models combining (i) the normalized centroid cn and the normalized damped frequency fdn (Centroid-frequency), and (ii) the normalized centroid cn and the normalized ion area an (Centroid-Area).

Although more tests were performed, no convergence was achieved in certain cases when attempting the respective correlations.

The correlations between normalized peak magnitude of thermo-ioinzation and normalized peak pressure (Thermo-ion) exhibit low determination coefficients, except for tests 2, 3 and

7. In all cases, except test 13, the angular difference ppp -th turned out to be between 8,3 and

21,6 degrees, resulting in a poor approximation to the PPP. The angular difference th - cn is negative in tests 1, 6, 11 and 12, which is inconsistent, since the thermo-ionization peak should not be located prior to the centroid of ion area. That inconsistencies take place due to the multiple peaks that appear on individual cyclic ionization current signals, leading to false estimations of thermo-ionization peak. Therefore, the thermo-ionization peak features, both normalized magnitude and angle, are discarded for further analyses.

The results summarized in Table 18 highlight that the accuracy of ion current-based estimations of both pressure peak magnitude and its crank angle position varies significantly with engine operating conditions. In general, better correlations are observed under part-load conditions ( tests 2, 3, 5, 7, 8, 9, 12), where the ionization current signal features exhibit stronger relationships with combustion parameters. Under wide-open throttle ( tests 6 and

13) and at certain low-load points (tests 1 and 4), the correlations deteriorate, likely due to

increased variability in flame propagation and ionization signal quality. The centroid of ion

area consistently provides a robust indicator of pressure peak position in many cases, while the combination of ion area and damped frequency offers the best results for pressure magnitude estimation when ion signal quality is sufficient. These findings demonstrate both the potential and the limitations of ionization current-based combustion diagnostics across a wide range of operating conditions for the particular engine tested.

To further elucidate these trends and assess the diagnostic robustness of the ionization current features, the following sections present a detailed analysis of individual parameters – namely, the damped frequency, ionization current signal area and its correspondent centroid – in relation to both the magnitude and angular position of peak pressure. This stepwise examination across varying engine conditions aims to clarify the specific contributions and limitations of each feature in the context of combustion monitoring.

Prediction of maximum pressure value

The damped frequency serves as an indicator of the instantaneous spark gap capacitance at the moment the flame kernel develops, prior to the onset of significant chemi-ionization. It is aimed to use here as a predictive parameter for the maximum pressure that would occur a few crank angle degrees later.

Table 19 presents the correlation coefficients between the normalized damped frequency and the normalized maximum pressure for the tests listed in Table 18. Across the range of engine conditions tested, the use of damped frequency generally provides an acceptable prediction of maximum pressure, although certain cases exhibit significant dispersion. While some low correlation coefficients might suggest otherwise, the overall trends to the fitted curves still support the predictive capability of this parameter.

Test 1 was performed under low load torque conditions, as inferred from the pressure trace shown in Figure 126. This results in significant dispersion of peak pressure values around the fitted line, visible in the figure.

Table 19. Correlation coefficients for the tests performed (normalized damped frequency and normalized maximum pressure) Test Throttle

(%)

Spark Adv. (°) Load (N•m) Regime (min-1)

AFR

microam meter Frequency R2

1

15

10

6,18

1957

15,31

10k, G1

0,06

2

12

20

7,09

1695

15,4

10k, G1

0,94

3

14

20

7,24

1769

15,74

10k, G1

0,94

4

40

20

11,51

1979

14,06

10k, G1

0,52

5

42

10

13,16

1928

13,97

10k, G1

0,63

6

100

12

14,28

1835

14,05

10k, G1

0,18

7

42

15

14,55

1812

14,09

10k, G1

0,92

8

42

20

15,93

1673

14,7

10k, G1

0,56

9

42

17

17,19

1575

14,08

10k, G1

0,57

10

13

10

6,66

1918

15,43

100, G100

0,01

11

15

10

6,36

1991

15,14

1k, G10

0,03

12

42

12

15,48

1833

13,72

100k, G1

0,92

13

100

12

14,36

1957

14,05

100k, G1

0,09

Figure 126. Correlation between normalized magnitude of pressure and normalized frequency, and superposition of pressure cycles for test 1.

In contrast, tests 2 and 3, yielded very high R2 values. This can be attributed to the more consistent pressure patterns observed in these tests. Test 2 shows a stronger correlation (see Figure 127). The pressure waveforms in this test exhibit a different character compared to those of test 1, especially in the descending portion after TDC. Although the test 2 pressure waveforms show three atypical overshoots (suggesting preignition events), these transient oscillations were not captured by the damped frequency analysis, yet not degrading the frequency-pressure correlation.

To further explore this predictive capability, Table 20 compares the correlation of maximum pressure with both the damped frequency and the ionization signal area, as well as their combined effect.

Figure 127. Correlation between normalized magnitude of pressure and normalized frequency, and superposition of pressure cycles for test 1.

Table 20. Comparisons of correlations with normalized peak pressure between normalized damped frequency, normalized ionization current area and the combination of both. A comparison of criteria of goodness of correlation is presented in Table 20. In general, ion area-maximum pressure correlations yield comparable performance to the damped frequency-maximum pressure ones. A preliminary observation is that combining two well-performing individual predictors tends to improve overall correlation. However, this is not always the case, as illustrated by test 9. Conversely, tests 5 and 8 show examples of “bootstrap” improvement, where the combined model significantly enhances the goodness of fit beyond the individual contributions.

Tests 10 to 13 were conducted using microammeter resistors outside the previously recommended value of 10 k (as discussed in section 5.1). Specifically, test 10 employed a 100  resistor, and test 11 used 1 k. Under these conditions, the free-running oscillations are prolonged, which complicates achieving a reliable correlation. Mixed results were observed with the use of a 100 k resistor in tests 12 and 13. Although test 12 yielded a high correlation coefficient, it must be noted that at such a high Rion value, the ionization current tends to saturate to the clamping limit, masking the true amplitudes of both chemi-ionization and thermo-ionization phases, thus warranting caution when interpreting this result.

Test Frequency

R2 Area ion

R2 Area ionfrequency R2

1

0,06

0,04

0,07

2

0,94

0,94

0,92

3

0,94

0,95

0,91

4

0,52

0,02

0,54

5

0,63

0,44

0,92

6

0,18

0,07

0,05

7

0,92

0,93

0,95

8

0,56

0,76

0,93

9

0,57

0,62

0,45

10

0,01

0,15

0,11

11

0,03

0,002

0,037

12

0,92

0,904

0,91

13

0,09

0,002

0,11

Relation between the centroid of ion area cent and position of peak pressure ppp

Table 21. Results of the angular difference ppp - cent In tests 12 and 13 (see Table 21), the previously mentioned signal saturation is expected to cause an apparent increase in the ionization area, which in turn shifts the centroid toward later crank angles. Thus, resulting in a reduced angular difference ppp - cent .

Tests 2 and 3 conducted under low-load conditions, exhibit ionization current signals with longer angular duration. This explains the smaller ppp - cent values observed in these tests. Test 6, which showed poor correlation between damped frequency and peak pressure magnitude, similarly exhibits atypical behaviour for ppp - cent . Across the remaining tests, the angular difference generally falls between 10 to 17 degrees, suggesting that the ionization current area centroid provides a mediocre basis to calculate the timing of peak pressure, PPP.

Relation between the percentage of maximum ion current integral value and position of peak pressure ppp

Table 22 summarizes the percentage of the maximum ionization current integral corresponding to the crank angle at which pressure peak (PPP) occurs. In test 1, the calculated percentage was zero, reflecting an extremely weak ionization signal throughout the cycles – consistent with near-misfire conditions-. In low-load cases (tests 2 and 3), the gradual buildup of the ionization current integral resulted in low percentages associated with PPP. Percentage calculations failed to converge in tests 10 and 11, due to irregular or atypical ion current signals. In tests 12 and 13, signal saturation introduced distortion into the integral calculation, producing an artificially low percentage value.

Among the tests with valid results, most fall within a range of 76% to 89% of the ion integral at PPP -consistent with the 80%-90% criterion reported by Budko et al. [149].

Test Throttle

(%)

Spark Adv. (°) Load (N•m) Regime (min-1)

AFR

microam meter ppp-cent [deg]

1

15

10

6,18

1957

15,31

10k, G1

14,45

2

12

20

7,09

1695

15,4

10k, G1

1,7

3

14

20

7,24

1769

15,74

10k, G1

3

4

40

20

11,51

1979

14,06

10k, G1

15,68

5

42

10

13,16

1928

13,97

10k, G1

13,5

6

100

12

14,28

1835

14,05

10k, G1

7,8

7

42

15

14,55

1812

14,09

10k, G1

13,7

8

42

20

15,93

1673

14,7

10k, G1

14,55

9

42

17

17,19

1575

14,08

10k, G1

9,47

10

13

10

6,66

1918

15,43

100, G100

17

11

15

10

6,36

1991

15,14

1k, G10

17

12

42

12

15,48

1833

13,72

100k, G1

1,8

13

100

12

14,36

1957

14,05

100k, G1

0,4

Table 22. Results of the percentage of maximum ion current integral related to PPP.

Comparison between the coefficient of correlation centroid-PPP, the combined centroid-damped frequency-PPP and the combined centroid-ion area-PPP

Table 23. Comparison of correlations: centroid-PPP, combined centroid-damped frequency-PPP and combined centroid-ion signal area-PPP. The comparison in Table 23 highlights cases where combined indicators enhance the prediction of the crank angle position of peak pressure (PPP), explained by the strong correlations found between normalized pressure and its corresponding normalized crank angle, presented in the column “R2” of “ p-ppp” .

Test Throttle

(%)

Spark Adv. (°) Load (N•m) Regime (min-1)

AFR

microam meter Percent integral [%]

1

15

10

6,18

1957

15,31

10k, G1

0

2

12

20

7,09

1695

15,4

10k, G1

58,7

3

14

20

7,24

1769

15,74

10k, G1

57,4

4

40

20

11,51

1979

14,06

10k, G1

89,6

5

42

10

13,16

1928

13,97

10k, G1

80,9

6

100

12

14,28

1835

14,05

10k, G1

76,68

7

42

15

14,55

1812

14,09

10k, G1

76,55

8

42

20

15,93

1673

14,7

10k, G1

86,4

9

42

17

17,19

1575

14,08

10k, G1

76,32

10

13

10

6,66

1918

15,43

100, G100 Fail

11

15

10

6,36

1991

15,14

1k, G10 Fail

12

42

12

15,48

1833

13,72

100k, G1

65

13

100

12

14,36

1957

14,05

100k, G1

54,7

Test p-ppp .m

R2 Centroid

R2 Centroidfrequency R2 Centroid- Area R2

1

-0,7781

0,23

0,89

0,09

0,24

2

1,0295

0,94

0,09

0,01

0,22

3

1,2799

0,94

0,03

0,32

0,29

4

-16,2716

0,89

0,04

0,5

0,04

5

-10,8172

0,84

0,59

0,94

0,93

6

-11,7174

0,78

0,003

0,014

0,005

7

-6,6331

0,98

0,92

0,94

0,95

8

-14,9962

0,77

0,8

0,91

0,88

9

-10,0311

0,97

0,71

0,84

0,85

10

-0,3758

0,09

0,89

0,91

0,89

11

-0,8179

0,16

0,8

0,92

0,9

12

-6,6202

0,98

0,93

0,93

0,92

13

-11,6530

0,78

0,18

0,9

0,11

Tests 5, 7, 8, and 9 exhibit strong correlations and can be highlighted as cases with promising prediction capability. The high R2 of “Centroid”, observed in test 1, is misleading, as Table 22 already indicated a null ionization integral percentage for this case; hence this result must be disregarded.

There are consistently negative slopes “m” of the correlations “ p-ppp”, between -6 and -16, except in tests under low-load conditions (tests 1, 2, 3, 10 and 11), in which p-ppp gets deteriorated. Those tests under low-load conditions, sistematically present poorer correlations, reflecting the inherent challenges in predicting PPP under such combustion conditions.

Test 4 presents a significant dispersion in the “Centroid” correlation, indicated by its low coefficient of determination (R2 = 0,04). Notably, the combined “Centroid-frequency” slightly improves the correlation to R2=0,5, helped by “p-ppp” and the normalized damped frequency related with the in-cylinder pressure magnitude, as shown in Figure 128.

Figure 128. Three-dimensional surface fitting result for predicting crank angle position of pressure peak from normalized damped natural frequency and centroid of ion current area (Test 4). The z-axis represents the crank angle position of pressure peak. Test 6 is characterized by its poor correlation between damped frequency and peak pressure magnitude, as well as the significantly disperse “Centroid” correlation, impairing the combined centroid-frequency determination.

For tests using Rion values outside the recommended 10 k settings some apparently good correlations were obtained; however, given the previously noted distortions and saturation effects in the ionization signal, these results should be disregarded.

A summary of the variability in correlation coefficients as functions of engine operating parameters is presented in Appendix 2. It is shown that the correlation coefficients vary significantly across conditions, highlighting the need for additional mappings that incorporate throttle opening, engine speed, load, air-fuel ratio (AFR), engine temperature and spark advance as inputs. Such mappings would be necessary to improve the estimation of cycle-to-cycle variability in both magnitude and crank angle position of peak pressure.

5.4 Ignition delay estimation

The crank angular positions of spark strike ign and chemi-ionization peak ch can be detected from the ionization current signal. The angular difference ch - ign constitutes an estimation the ignition delay. Figure 129 presents the cyclic variation of ignition delay and PPP under the conditions of test 8.

Figure 129. Ignition delay and angular position of pressure peak. Table 24. Ignition delay and PPP estimation from ignition delay, for tests 1 to 9. Table 24 presents the ignition delay from ion signal and the estimation of crank angle of pressure peak for the tests with Rion = 10 k. The remaining tests are discarded due to the impairment of ch estimation. The PPP estimation from ch presents poor fitting, improving on tests 7, 8, 9, characterized by the highest values of load torque. It can be noted a reduction in ignition delay from 17 to 10 crank angle degrees when increasing the load torque. Test Throttle

(%)

Spark Adv. (°) Load (N•m) Eng. spd (min-1)

AFR

Ignition delay [°] PPP est.

R2

1

15

10

6,18

1957

15,31

17±1

0,87

2

12

20

7,09

1695

15,4

15±1

0,11

3

14

20

7,24

1769

15,74

15±1

0,02

4

40

20

11,51

1979

14,06

16±1

0,05

5

42

10

13,16

1928

13,97

14±1

0,57

6

100

12

14,28

1835

14,05

13±1

0,01

7

42

15

14,55

1812

14,09

13±1

0,92

8

42

20

15,93

1673

14,7

12,5±1,5

0,82

9

42

17

17,19

1575

14,08

11±1

0,77

5.5 Mass fraction burned and heat release rate estimations

Mass fraction burned (MFB) can be computed from in-cylinder pressure and volume. As expressed by Mittal et al. [210], MFB shows the combustion progression along the crank angle. The rate of combustion (heat release rate, HRR) affects engine thermal efficiency, peak cyclic temperature and pressure, as well as exhaust emissions.

Mittal et al. [210] compare the net pressure method with the Rassweiler-Withrow one. They determine first the start of combustion (SOC) and end of combustion (EOC) from a logarithmic pressure indicator diagram, using least-squares fitting, to evaluate the polytropic indices during compression and expansion. The compression polytropic coefficient calculation is used to improve the net pressure method, finding better agreement with EOC.

Daniels [110] presents MFB estimations from both the in-cylinder pressure and volume, and the ionization current signal. Daniels uses the mass fraction burned and their first and second angular derivatives with respect of crank angle to indicate the maximum heat release point and the maximum acceleration point. The first derivative of ion signal is used to obtain the maximum acceleration point after the chemi-ionization peak of ion signal, and the EOC after the thermo-ionization peak of ion signal. With both approaches, Daniels obtains Вибе (Wiebe) functions to determine the corresponding mass fraction burned curves. Notably, Daniels use the high-side scheme for ion current detection.

Fiedkiewicz et al. [131] analyzes the relationship between the thermo-ionization peak point th and PPP. Additionally, they correlate with the maximum HRR position. Fiedkiewicz et al. use the approach of Daniels, leveraging both the in-cylinder pressure and the first derivative of ion signal to obtain the crank angular position of maximum heat release rate, and with it the pressure peak point ppp. They find strong relationships between both th - ppp and the angle of maximum HRR and the derivative of ion current evaluated after th .

The previous considerations are applied to obtain the MFB and HRR, leveraging the end of free-running oscillation as SOC and the crank angle position at which MFB = 1 during calculation as EOC.

Although the cyclic in-cylinder pressure and volume data vectors can be used to calculate MFB, the cyclic ion signals can not be used, since it has been proven that the determination of the thermal peak angle th is impaired, due to the additional peaks along the cyclic ionization current waveforms caused by the bulk flows. Thus, averaged in-cylinder pressure, volume and ionization current signals are used for the computation.

Figure 130 displays simultaneously the normalized and averaged waveforms of in-cylinder pressure pn, ion current signal without the ignition and free-run oscillation in, mass fraction burned MFB and heat release rate HRR, under the conditions of test 8.

Figure 130. Plots of normalized in-cylinder pressure, ionization current, MFB and HRR as functions of crank angle for test 8.

In Figure 130, the crank angle positions of ignition strike ign , chemi-ionization peak ch , thermo-ionization peak th, pressure peak point ppp, and zero ion current f, are highlighted, as well as the maximum heat release rate point HRRM , the 10% of MFB point MFB10, the 50% of MFB point MFB50 , and the 90% of MFB MFB90.

Table 25. Crank angular positions of ion current

Table 25 presents the results of ignition delay from ignition strike to chemi-ionization ch -ign , the crank angle duration between chemi-ionization and thermo-ionization th-ch , the total ionization duration f-ch , for tests 4, 5, 7, 8, 9. The remaining tests failed to Test Spark Adv.

(°) Load (N•m)

AFR

Ign.

Delay ch-ign [°] Chem-therm delay th-ch [°] Ionization duration  f -ch [°]

4

20

11,51

14,06

18

16

42

5

10

13,16

13,97

14,2

19,8

54,8

7

15

14,55

14,09

14

16

40

8

20

15,93

14,7

15

15,6

38,3

9

17

17,19

14,08

12,8

15,5

44,2

converge during computation. Test 5 exhibits the longer duration, both for th-ch and total ionization duration f-ch , with AFR = 13,97. Test 8 for contrast presents shorter ionization durations, with AFR = 14,7. This preliminary observation points to a direct relationship between the richness of air-fuel mixture and the ionization duration, although additional tests are needed to establish more conclusive statements. Test 9 represents a case of short ignition delay, as well as a short chemical-thermal delay th-ch, that is, a fast combustion that favours the torque production.

Table 26. Angular position estimations Table 26 shows the goodness of estimation of PPP based on the averaged thermo-ionization peak position, 10% of MFB based on the averaged chemo-ionization peak position, and the 50% of MFB based on the averaged thermo-ionization peak position. Noticeably, the 10% MFB occurs between 3 to 4 degrees of crank angle after the chemi-ionization peak, whereas the 50% MFB takes place between 3 to 4 degrees before the thermo-ionization peak. Test 5 exhibits the best PPP estimation based on thermo-ionization peak, whereas the remaining tests value the PPP between 3 to 5 degrees after thermo-ionization peak. The estimation of the 90% MFB, est90, is performed as follows:

𝜑௘௦௧ଽ଴= 𝜑௧௛+ 0,33(𝜑௙−𝜑௧௛)

Eq. 82 Observing that MFB90 takes place between the thermal peak th and the zero current f marks. As it is shown in Table 26, tests 5 and 9 deviate 2,55 and 4,77 crank angle degrees respectively from the MFB90 point. Nevertheless, the remaining tests exhibit closer approximations.

For the tests 4, 5, 7, 8, 9, it is observed that the end of ionization f takes place after the 100% of MFB, indicating that the end of ionization current and the end of combustion are not necessarily linked. This goes in agreement with the comments of Daniels [110] related to the inflection point after the thermo-ionization peak: “…The inflection point right after the third peak of the ion signal indicates that the heat release near the spark plug loses its influence to keep the temperature high enough for ion formation. This point is the end of ion formation and can be defined as the end of combustion. As opposed to the end of combustion used in the Rassweiler-Withrow method and many other mass fraction burned calculation models, this end of combustion does not necessarily mean that the mass fraction burned reaches 100% mass fraction burned…” Test Spark Adv.

(°) Load (N•m)

AFR

PPP-

thermpk ppp-th [°]

10%

MFB

MFB10-ch [°]

50%

MFB

MFB50-th [°]

90%

MFB

est90 -MFB90 [°]

4

20

11,51

14,06

4,6

3

-3

0,28

5

10

13,16

13,97

0,8

4

-4

2,55

7

15

14,55

14,09

3,4

3,7

-2,8

-1,08

8

20

15,93

14,7

5,1

3,4

-3

0,091

9

17

17,19

14,08

3,7

3,7

-2,7

4,77

5.6 Conclusions of the chapter

The modeling presented and the experiments provided useful criteria for selecting the value of Rion, between 10 k and 30 k in low-side ionization current detection circuits for spark ignition applications. This range is similar to the sum of the resistances of the secondary winding of the ignition coil, high tension wire, and spark plug internal resistance. This finding contrasts with conventional in-series current measurement approaches, where the value of the shunt resistor should be much lower than the total resistance of the electrical loop. In the context of ionization current detection, the value of Rion affects the damping of the ignitioninduced contribution.

A novel method based on the damped frequency that takes place when the electrical energy of the ignition coil is almost fully dissipated is introduced, to infer variations in spark gap capacitance. This method can be instrumental to estimate an instantaneous capacitance, not possible to measure with the much slower linear capacitor charging, to assess the content of the working gases present in combustion chamber. This technique was evaluated to correlate the damped oscillation frequency with the cyclic variation of maximum pressure that will be reached after, during ionization.

This technique shows promising for fitting results, though it requires to be accompanied by another indicator of pressure, such as the ion signal area under the curve, since the damped frequency measurement depends on spark gap capacitance, which is a local property of the combustion chamber.

The estimation of the thermo-ionization peak on a cyclic ionization current signal is impaired in comparison with the averaged ion signal, due to the multiple peaks that appear after the chemi-ionization, especially for low-load conditions. This leads to a poor estimation of the PPP from cyclic ion signal.

The combined use of damped frequency and ionization signal area resulted in improved correlations with peak pressure in many cases, although engine variability in combustion behavior remains a limiting factor. Specifically, at low load torque, increased pressure variability degraded the correlation between ionization-based indicators and both peak pressure magnitude and crank angle of occurrence (PPP), reducing the R2 values obtained.

Accurate estimation of cyclic variation in both the magnitude and angular timing of pressure peak from ionization current features and damped frequency remains challenging, particularly under low load conditions. Achieving more robust predictions will require enhanced signal processing, including consideration of features present during primary coil charging, as well as more rigorous outlier detection and data screening to manage cycle-tocycle variability.

Final conclusions and future work

6.1 Introduction

The development of this work led to the implementation of a non-intrusive in-cylinder pressure inference system, based on ionization current signal detection, applied to the combustion chamber of a single-cylinder engine, adding the secondary winding voltage, primary winding current and voltage, to diagnose the quality of combustion under diverse conditions of load torque, rotational frequency and parameter settings. This work was motivated by the ceaseless quest of using an ionization sensor as a surrogate device for pressure features estimation, as well as the scarcity of works on the subject found in the local academic community. The exploration of literature, in conjunction with the implementation and validation of the instrumentation, acquisition campaigns, ignition modeling and experimental validation, data analysis and summarization were key to the achievement of the goals of the present thesis.

6.2 Main conclusions

Following the sequential order of the chapters, there can be stated the following general conclusions from the development of this treatise:

Chapter 2 addressed an extensive stacking of key documentation on engine diagnostics, ignition systems, flame kernel development, schemes of ionization current detection, basics of chemistry of combustion, main models of ion generation, main features of ionization current signals, and their associated diagnostics capabilities.

Related to ignition systems, it is presented the basic dynamic modeling of primary coil charging and switch-off phases, applied to autotransformer-type ignition coils, from [40]. The transient power sources generated by their inductances and capacitances during such phases are explicitly included, being them responsible for spark voltage generation. Additionally, the primary and secondary loops comprise coupled series R-L-C circuits, where the inductances and capacitors determine the loops’ natural frequencies of oscillation, primordial parameters of the behavior of such dynamic systems. Those natural frequencies and transient sources can be further applied to the modeling of low-side schemes conjoined to transformer-type ignition coil systems.

From the basic dynamic model of autotransformer ignition system, this work presents the fundamental circuital models of high-side and low-side schemes, applied to transformer-type ignition coil, including the contribution of transient ignition sources, not explicit in literature, primordial sources for the modeling shown in chapter 5.

Regarding the high-side scheme, this work summarizes electrical configurations, in which the elements associated to the “Tee” connection vary, recurring to single diode on promoter side (Eriksson [25]), points gap in the secondary side (Anderson [92], Daniels [110]), double diode (Yoshiyama et al. [111]), diodes in series (Laganá et al. [113]), to limit the energy that arrives to the microammeter, due to the several orders of magnitude of difference between

the promoter U and the transient ignition sources. The complexity of configurations, the high specification of voltage of diodes, along with the “blindfolding” of the ionization sensor caused by the inherent capacitance of a high-voltage diode that appears during inverse biasing, referred in Eriksson [25], in addition to the likelihood of short-circuit diode failure [54], discouraged to employ the high-side scheme.

Additionally, the literature exploration led to identify the following voids:  There is not a criterion to choose a value, or at least a range, of the shunt resistor Rion on ionization current sensing.

 In low-side schemes, both the secondary winding current and ionization current cause voltage drop on resistor Rion. Additionally, the promoter power source U maintains an e.m.f. during the whole 4-stroke cycle. Thus, the ionization current signal features analysis, limited to a certain crank angle range, can benefit from complementary contributions coming from secondary winding current and other angular position ranges, aiming to improve the correlations with the in-cylinder pressure.  The secondary and primary winding voltages could add diagnostics capabilities to ionization current signal.

Chapter 3 addressed the implementation of several especially designed electronic equipments used to measure ionization current, secondary winding voltage, primary winding voltage and current, intake and exhaust valve opening, crank angle, torque, air-fuel ratio (AFR), and incylinder pressure, based on the voids found in previous chapter. The technical specifications of the subsystems come from the remarks gathered from literature. The choice of selecting a low-side ionization current measurement circuit, inserted into the TCI ignition system of the modified Changfa engine, was justified by leveraging the complementary contributions of secondary winding current to ionization current, and a simpler setup compared to the highside scheme, especially, the complexities associated to the high-tension diode.

A resistive high tension probe was implemented to capture the DC and low-frequency components of secondary winding voltage, complemented with an inductive probe that reacts to the remaining AC components. To obtain the static characterization of the resistive probe, a Cockroft-Walton-based DC high voltage source was built, finding in the process the true input impedance of the probe, below the initial 1 G specification. A high-voltage sinusoidal AC variable power source was also built to perform a sweep of frequencies to determine the actual frequency response of the resistive probe. This is an important outcome of this work, since that information is seldom found on probe’s manufacturer specifications.

The measurement of the primary current and primary voltage required two electrical nodes for each one, and none of them is the reference node of the negative of ignition power supply, the battery. Differential voltage measurements with respect to chassis ground were discarded, due to the limited number of available acquisition channels. In addition, the powerful transient negative voltage excursion that is generated during the ignition transistor switchoff, turned into another reason to discourage the differential setup. Thus, analog isolator amplifiers were specially built, offering a tested and verified flat frequency response from DC to 100 kHz, and the capability of withstanding a difference in ground potential between the input and output of 2 kV. This made possible to acquire with confidence the primary winding voltage and current during the tests. Moreover, to perform simultaneous acquisitions of primary and secondary variables of ignition coil, as well as ionization current.

On the subject of ion current detection, two fundamental technological outcomes of this work are highlighted: the subsystems of promoter power source U and the microammeter. The ionization promoter U comprises an alternate switching and charging of capacitors, resulting in a power supply with triple galvanic isolation that minimizes the saturation of the microammeter due to displacement current, inherent to AC supply. In addition, it is possible to promote ion current covering a test range between -300V and 300V for U. This allowed to compare and verify the statements of literature for the engine tested, particularly related to positive and negative DC bias.

The microammeter comprises a set of discrete values of shunt resistor Rion, that allowed to test outside the stated ohmic ones in literature. The signal limiter built was tested to ensure that the ionization current (rated in microamperes) had significant preference of circulation through Rion after the ignition damping, and high priority of circulation through the semiconductor limiter during the ignition strike (rated in milliamperes, with short spikes in the order of amperes), verifying that the signal limiter clamped only if the incoming signal surpassed a certain voltage threshold. Additional tests of the microammeter ensured flat frequency response.

The intake and exhaust valves opening measurement system was motivated by the graph depicted in the work of Martychenko et al. [36], which highlights secondary voltage variations during motoring that tend to follow pressure traces.

The valve opening sensing, both for intake and exhaust valves, recurred to proprietary displacement feelers, based on variable coupling coefficient transformers, similar in principle of operation to an LVDT, with the difference that the transformers built for this work have only one secondary winding. The characterization included the nonlinearities between r.m.s. voltage and displacement. Direct outcomes of the application of the sensors are the actual crank angles of opening and closing of both valves, providing input to initial correlation with ionization variables, adding the potential of further analyses related to the engine distribution mechanism.

The torque sensing included the static calibration of the dynamometer load cell and its corresponding signal conditioning module, which includes the presentation of an instantaneous load torque value on an LCD to ease the tests. In addition, the signal conditioning module sends a voltage signal, proportional to the torque, into the acquisition system, for further improved averaging.

An incremental encoder was selected for crank angle position measurement. Given the limited number of acquisition channels, an especially designed circuit was used to combine the “A” and “Z” signals of the encoder into one that generates a single higher pulse on the geometric TDC of the piston.

Chapter 4 presents the preliminary findings on motoring and combustion tests, defining the ranges of variables for the acquisition campaigns, and describing the software implementations, particulary in referencing and filtering the in-cylinder pressure, based on the work of Martin [187].

Motoring tests were performed initially with an induction motor driven by an VFD, encountering a significant interference in all sensors that led to discard its use. Later, a DC compound motor was selected as a motoring surrogate, finding cleaner signals, utterly discarding it for insufficient power and windings isolation issues.

Motoring tests without spark, maximum sensitivity of ionization sensing, and the DC promoter power source U active, showed that neither the secondary current nor the secondary voltage vary at any crank angle corresponding to intake or exhaust valves’ actuation. On the other hand, with the ignition active, variations in oscillation amplitudes on secondary current were found during the primary coil charge phases, differentiating the compression stroke and the exhaust one, due to the distinct conditions of in-cylinder pressure that affect the spark gap capacitance.

The presence of a prechamber or its absence has little effect on the secondary current signal during the primary coil charging and ignition transistor switch-off stages.

Combustion tests with prechamber presented lower amplitudes of ionization current, at least one order of magnitude in comparison with the tests performed without prechamber. The prechamber shields the spark plug electrodes from the bulk flows, deteriorating the arrival of ionized species to the electrodes, especially affecting significantly the thermo-ionization peak magnitude. This confirms the shortcoming of measurement of local properties of the combustion products by ionization current signal, referred by Henein et al. [103], and the requirement of ionized mass transport by the bulk flows to improve a correlation with incylinder pressure.

The negative polarity of the DC promoter power source U causes a significant reduction in ionization signal amplitude. The tests performed with negative biasing confirm the diodelike current-blocking action of the spark plug electrodes, proposed in Wilstermann et al. [132].

Abnormal ionization patterns located during the interval of primary coil charging in compression stroke (dwell time) are strongly and repetitively correlated with events in which the in-cylinder pressure rises before the ignition strike (preignition).

Fifth chapter focuses on cycle-by-cycle variations in in-cylinder pressure, using the ionization current signal and its features, as well as the damped natural frequency that come from the last contribution of the ignition strike prior to chemi-ionization. For this purpose, based on the dynamic modeling of the autotransformer ignition system presented in chapter 2, a novel model of a transformer ignition system is presented, pertaining to a low-side ionization current sensing system, restricting the modeling to the final oscillation stage, to circumvent the difficulty of modeling the initial clamping of the ignition transistor that appears during the switch-off, due to the “kick-back” of the ignition coil.

Related to the criterion used to choose a value, or at least a range, of the shunt resistor Rion required to perform ionization current sensing, diverse authors indicate the ohmic value of the shunt used, without any further explanation. With the aforementioned ignition coil-lowside-ion sensing model, this work contributes with a criterion of ohmic value based on a balance between the duration of the remaining free-running oscillation and the amplitude of the ion signal. This duration is strongly conjoined with the damping factor ζ2 of the secondary loop of ignition system, and in turn this factor ζ2, as Eq. 64 highlights, depends in part on the value of Rion, the resistance of the secondary winding, and the values of high-tension wire and internal resistor of spark plug. For SI engines with TCI ignition that employ the low-

side scheme of ion current detection, this work proposes a criterion with a recommended range between 10 k and 30 k, close to the sum of the remaining resistors involved in the electrical loop.

One shortcoming of the low-side and high-side schemes of ionization current detection, in comparison with the independent scheme, is the presence of the free-running oscillations that can dwell in the crank angle, up to the point of masking the chemi-ionization part of the ion signal, or even further. The contribution of this criterion of range of Rion allows to select such resistor to minimize the duration of the free-running oscillation.

The magnitude of ion current integral constitutes a method for misfire detection, as shown in Shimasaki et al. [58] and Laganá et al. [113]. An initial approach of the present work aimed to suppress the free-running oscillations prior to chemi-ionization from the ion signal. This approach tackled the capture of the free-running oscillation during wasted spark event, and subsequently subtract it from the total ion waveform. Differences in damped frequencies between them were found that, although impaired such subtraction, provided a differentiation of spark plug capacitance between wasted spark, compression and combustion phases, that is, a novel misfire detection by the estimation of the damped frequency. This allows to flag a misfire before it is detected by the traditional method of ion signal integral threshold referred in Shimasaki et al. [58] and Budko et al. [149], leveraging an additional feature of the secondary loop current signal, shared with the ion current.

The quantification of the residual damped frequency oscillation, occurring prior to chemiionization, is proposed as a novel method that can be used to estimate the instantaneous spark plug capacitance, calculated from the damped frequency. The tests performed under different conditions indicate a direct correlation between instantaneous spark gap capacitance and peak pressure magnitude, that will be reached after.

The estimation of damped frequency was compared with the computing of the area under the ionization signal curve, to assess the cycle-by-cycle variation of the peak pressure magnitude. Both methods tend to achieve similar results under partial load, between 7,1 N•m and 17,2 N•m, rotational speeds between 1670 min-1 and 1980 min-1, air-to-fuel ratios between 14 and 15,4, engine temperatures between 46 °C and 87 °C, and spark advances between 10 and 20 crank angle degrees, for the particular engine used. When both estimators are combined, the correlations with peak pressure tend to improve in many cases, exhibiting a positive slope in the relationship pressure versus ion area, and a negative gradient for the pressure versus damped frequency. Under the different tests conditions the aforementioned slopes vary, requiring further mappings of pressure magnitude estimations.

The calculation of the centroid of the area under the averaged ionization signal curve, proposed by Eriksson [25], the criterion of 80%-90% of maximum ion signal integral, proposed by Budko et al. [149], and the angular position of the thermo-ionization peak present in the averaged ion signal, proposed by Gürbüz [18], Eriksson [86], Henein et al. [103], were compared in the present work to estimate the angular position of peak pressure, and its variation, from the individual cyclic ionization current signals. For the engine tested, under the partial load conditions previously referred above, narrowing the load torque conditions between 12 N•m and 17 N•m , the angular difference between centroid and PPP generally fell between 10 to 17 crank angle degrees, and the percentage of maximum ionization signal integral between 75% - 89%. Angular inconsistencies were found on the estimation of the cyclic thermo-ionization peaks, due to the multiple high contours of ion

signal, especially during low-load conditions, discarding the thermal peak positioning method.

The angular position of peak pressure and its magnitude were correlated. From the tests performed, it was found that a higher pressure tends to occur earlier in crank angle. Leveraging that finding, the combination of centroid-damped frequency and the combination centroid-ion signal area were compared, resulting in improved correlations in both cases, compared to the centroid method alone, with the crank angle position of peak pressure. Combining two estimators, the correlations with PPP exhibit a positive slope in the relationship PPP-centroid, positive slope for PPP-damped frequency, and a negative decline for PPP versus ion signal area. Under the different tests conditions the aforementioned slopes vary, requiring further mappings of PPP estimations.

Mass fraction burned, heat release rate estimations and correlations with ionization current features were performed using the methods proposed by Mittal et al. [210] and Daniels [110], restricting the tests conditions between 11,5 N•m and 17,2 N•m, rotational speeds between

1575 min-1 and 1980 min-1, air-to-fuel ratios between 14 and 14,7, engine temperatures

between 64 °C and 87 °C, and spark advances between 10 and 20 crank angle degrees, for the particular engine used. From the averaged ion signal, the ignition delay can be determined using the crank angular positions of chemi-ionization peak and ignition strike. The start of combustion is marked from the end of free-run oscillation. The position of 10% fraction of MFB is found within an interval between 3 to 4 crank angle degrees after chemi-ionization peak. The angular position of 50% MFB ranged between 2,7 and 4 crank angle degrees before the thermo-ionization peak. The proximity between PPP and thermal peak yielded between 1 to 5 crank angle degrees. An interval of one third of the angular span between the thermal peak position and zero ion current was used to estimate the angular position of 90% of MFB, ranging the findings between -1 and 5 crank angle degrees.

6.3 Future work

From the development of this work, although promising results were found, there still space for improvement in the following aspects:

 From the implementations, there were found limitations in the resistive high tension probe related to its limited frequency response. Similarly, the transient response shortcomings of the isolated analog amplifiers were identified, that impaired the detection of very short spikes.

 The microammeter module can benefit from further implementations with more Rion values, exploring intermediate values between 10 k and 100 k for SI engine setups.

 Valve opening measurement systems can be used to assess actual crank angles of opening and closing, feeding experimental data to further computational models related to intake and exhaust processes, as well as indicator diagrams and heat release quantifications for more refined combustion diagnostics.

 Torque sensing can be explored on a different setup, implementing other mechanical transmission systems between engine and dynamometer, aiming to capture cyclic engine torque waveforms without the shortcomings of the belt-pulley transmission.

 It was challenging to perform motoring tests with AC induction motor driven by VFD, due to strong interference that deteriorated the SNR of many sensors. Further EMI reduction techniques emerge as a necessity for AC-machine engine starting and motoring implementations.

Reviewing the preliminary findings of this work, additional engine conditions of load torque, engine speed, AFR, spark advance, as well as additional ion current measurement settings, can be performed to extend the fitting goodness span database of the ionization current methods described.

The most common promoter power source U comprises a constant DC level. AC power sources have been explored in literature [132], [133]. Novel hypotheses and results can emerge from implementations of U with a DC level, added to an AC sinusoidal component, exploring the relative AC/DC amplitudes and frequency range of the AC component.

The free-running secondary oscillations allow to capture the spark gap capacitance, that in turn is related to the in-cylinder pressure and the content of the working gases. Additional oscillations along the non-explored crank angles, particularly during intake, compression and exhaust could help infer additional points of the in-cylinder pressure through spark gap capacitances estimations, carrying the quest of piezoelectric pressure sensors surrogates on.

It was not possible to conclusively detect knock on the tested engine, due to its construction tolerances. Additional assessment of knock, using the implemented ionization current measurement system on different engines and fuels, is encouraged from this work.

As the comparison between tests with and without prechamber highlighted, the ionized mass transport, promoted by the bulk flows, is primordial to approach the local property of ionization current signal, as well as the local spark gap capacitance, to the general character of in-cylinder pressure. Nevertheless, under certain engine operating conditions, in particular low-load torque, the variations in mass transport result in disperse correlations between damped frequency, area under the ionization current signal curve, its centroid, the percent of maximum ion integral related to PPP, thermo-ionization peak, and in-cylinder pressure peak magnitude along with its corresponding crank angle position.

Additional input variables and ion current signal features could contribute to circumvent the dispersions, such as the angular position of chemi-ionization peak, the delay between the ignition strike and the chemi-ionization peak.

6.4 Contributions

From the present work, a summary of outcomes is presented as follows:

Technological implementations:

 Static and frequency sweep characterization of a resistive high tension probe.  Measurement of primary winding current and voltage through proprietary analog isolation amplifiers, including their corresponding static and frequency response characterizations.

 Variable power source, both magnitude and polarity, for ionization current promotion with triple galvanic isolation, to minimize the effect of displacement current.  Microammeter for ionization current measurement, with fixed discrete impedance settings during ionization and low-impedance signal limiting during ignition strike.  Intake and exhaust valves opening measurement system.

 Combined signal generation system for crank angle measurement.  Torque signal conditioning with indication of instantaneous value, including its corresponding static load cell characterization.

Theoretical contributions:

 Contact-breaker ignition system model for autotransformer configuration, including mechanical contact bouncing [40].

 TCI Ignition system model for transformer configuration, including low-side ionization current measurement, to assess natural damped frequency, and from it the spark gap capacitance.

 Criterion to establish a best fitted range of Rion shunt resistor values for SI engine.

Experimental and data analysis outcomes:

 Identification of clamping of TCI ignition transistor during the switch-off and spark strike, and comparison with.contact-breaker ignition waveforms [40].  Detection of preignition from saturation of ionization current signal during dwell time.

 Identification of wasted spark, misfire, combustion from natural damped frequency analysis.

 Estimation of cyclic variation of pressure, in particular, peak pressure magnitude and its corresponding crank angle position, both correlated from natural damped frequency analysis, compared and combined with ion current signal area, its centroid, percentage of maximum ion current integral, in partial load conditions.

Publications:

Experimental Study for Energy Balance and Exergy Analysis of a Single-Cylinder, Spark- Ignition, Air-Cooled Engine. May 2025. SAE International Journal of Engines 18(4).

DOI: 10.4271/03-18-04-0023

“Experimental and numerical approach to assess the dynamic performance of an inductive ignition system”. International Journal of Engine Research. ISSN: 1468-0874. Online ISSN:

2041-3149. 2024. DOI: 10.1177/14680874241233812

Test bench development and implementation for experimental determination of mechanical losses in single cylinder internal combustion engines. November 2023. EUREKA Physics and Engineering. DOI: 10.21303/2461-4262.2023.002847

Performance and In-Cylinder Pressure-Based Diagnosis of a Single-Cylinder Variable Compression Engine Developed for Education and Research. SAE Technical Paper 2023-

01-1667. 2023. https://doi.org/10.4271/2023-01-1667.

Considerations for starting combustion engines with AC machines”. Diagnostyka 24 (1):1-

10. 2022. e-ISSN: 2449-5220. DOI: 10.29354/diag/156748

Assembly and instrumentation of a didactic bench for testing of starters of internal combustion engines, “Ensamble e instrumentación de un banco didáctico para pruebas de arranque en motores de combustión interna”. Revista UIS Ingenierías 19 (3). 37-48. 2020. DOI: https://doi.org/10.18273/revuin.v19n3-2020004

Divulgation:

“Instrumentación del sistema de ignición de un motor monocilíndrico incluyendo medición de corriente de ionización”. XII Congreso Internacional de Ingeniería Mecánica, Mecatrónica y Automatización. CIMM2025. May 2025. Montería. Colombia.

A Appendix 1 Combustion tests with prechamber

Test conditions are: torque load of 14,06 N•m, mean AFR of 15,01, mean rotational speed of

1941 min-1, and engine temperature of 87,5°C. The ion settings were Rion = 10 k and G =

10, factor of 105, power source U = 0 V, 50% throttle opening, and 10 degrees of spark advance.

Figure A-1. a) wasted spark and b). combustion

Figure A-1 shows details of the ion waveforms of the 25th cycle in wasted spark and combustion. The absence of promoter power source U, as expected, causes the vanishing of the chemi-ionization and thermo-ionization peaks due to combustion. At the same time, the damped frequency in combustion is higher than in wasted spark, revealing an opposed tendency in comparison with the test without prechamber, and promoter U providing a potential difference. This means that it is required to the bias of the promoter power source to have consistent results in combustion tests without prechamber. However, this test was performed with prechamber. Unfortunately, it was not possible to recover an archive with U = 0 V and without prechamber, to confirm the tendency of Figure A-1.

Another test with prechamber was performed having: load torque of 15,42 N•m, a mean AFR of 15,2, a mean rotational speed of 1926 min-1, and an engine temperature of 90,96°C. The ion settings were Rion = 10 k and G = 10, factor of 105, power source U = 200 V, 46% throttle opening, and 10 degrees of spark advance. In this case, the damped frequency on wasted spark was 3694 Hz, and the frequency on combustion was 6047 Hz. Then , there is an opposed tendency with respect to the tests without prechamber. The prechamber guards the electrodes of the spark plug, protecting it of most of the flow characteristics of air-fuel mixture. Thus, the capacitance of the gap is also dependent of such flow characteristics.

B Appendix 2

Summary of correlations of peak pressure magnitude and angle

C Appendix 3 Figure C-1 presents the combined correlation results between the angular position of chemiionization (Aquimo), the area of ion current signal (AreaIon) and maximum pressure (Pmax), evaluated for the tests 2 to 9, referred in section 5.3.3. This analysis was boosted by the use of the data analysis software Orange ®, that helped to find additional features of ion signal, removing more atypical data, and returning the screened database to MATLAB ®.

Figure C-2 presents the combined correlation results between the angular position of chemiionization (Aquimo), the damped frequency (Flibre) and crank angular position of maximum pressure (Apres).

Figure C-1

Bibliography

[1] L. Lešnik, B. Kegl, E. Torres-Jiménez y F. Cruz Peragón, «Why we should invest further in the development of internal combustion engines for road applications,» Oil&Gas Science and Technology-Rev. IFP Energies nouvelles 75, 56 (2020), 2020. [2] G. Kalghatgi, «Is it the end of combustion and engine combustion engine research? Should it be?,» Transportation engineering, vol. 10, p. 11, 2022. [3] T. Sinigaglia, M. E. Santos y J. C. Mairesse, «Technological evolution of internal combustion engine vehicle: A patent data analysis,» Applied Energy, vol. 306, nº A,

2022.

[4] J. Thema y M. C. Roa García, «The Energy Transition in Colombia,» Wuppertal Papers-Universidad de los Andes, p. 68, 2023.

[5] D. Vollberg, D. Wachter, T. Kuberczyk y G. Schultes, «Cylinder pressure sensors for smart combustion control,» Journal of Sensors and sensors systems, vol. 8, p. 11,

2019.

[6] Kurtz, A D; Kane, A; Goodman, S; Landmann, W; Geras, L; Ned, A A;, «High Accuracy Piezoresistive Internal Combustion Engine Transducers,» Kulite Semiconductor Products, Inc, 2004.

[7] M. Kondo, A. Niimi y T. Nakamura, «Indiscope-A New Combustion Pressure Indicator with Washer Transducers,» SAE Technical Paper 750883, p. 10, 1975. [8] M. Sellnau, F. Matekunas, P. Battiston, C.-F. Chang y D. Lancaster, «Cylinder- Pressure-Based Engine Control Using Pressure-Ratio-Management and Low-Cost Non-Intrusive Cylinder Pressure Sensors,» SAE Technical paper 2000-01-0932, p.

22, 2000.

[9] N. Fiorini, L. Romani, A. Bellisima, G. Vichi, A. Bianchini y G. Ferrara, «An indirect in-cylinder pressure measurement technique based on the estimation of the mechanical strength acting on an engine head screw: development and assessment,» Energy Procedia, vol. 148, p. 8, 2018.

[10] M. Wlodarczyk, «Fiber optic-based in-cylinder pressure sensor for advanced engine control and monitoring,» COMBUSTION ENGINES, nº 4, 2012.

[11] Houben, Hans; Marto, Arno; Pechhold, Frank; Haussner, Michael; Borgers, Marc;, «Pressure sensor glow plug (PSG) for diesel engines,» MTZ Worldwide, vol. 65, p.

8, 2004.

[12] E. Henao, Aportación constructiva, tecnológica y experimental al estudio nacional de los motores de combustión interna, Pereira: Universidad Tecnológica de Pereira,

2020.

[13] V. Naoumov, A. Demin, I. Andersson y A. Sokolov, «Modeling of Combustion and Non-Equilibrium Ionization in Spark Ignition Engines,» SAE technical paper 2002-

01-0009, 2002.

[14] H. F. Calcote, «Mechanisms for the formation of ions in flames,» Combustion and Flame, vol. 1, nº 4, pp. 385-403, 1957.

[15] L. A. Mejía, C. A. Romero y M. Monroy, «Pasado y presente en el diagnóstico de los motores en los talleres de servicio automotor. Del vacuómetro a los sistemas basados en la nube,» Informador Técnico. 85(1), 2020.

[16] R. Isermann, Combustion engine diagnostics. Model-based condition monitoring of gasoline and diésel engines and their components, Springer, 2017. [17] T. Yamada, Y. Matsubara y N. Kondo, «Characteristics of multispark ignition system,» SAE technical paper 1999-01-0205, 1999.

[18] H. Gürbüz, «Experimental evaluation of combustion parameters with ion current sensor integrated to fast response thermocouple in SI engine,» Journal of Energy Engineering.ASCE, 2016.

[19] J. Merkisz, P. Bogus y R. Grzeszczyk, «Overview of engine misfire detection methods used in on board diagnostics,» Journal of Kones. Combustion engines, vol. 8, nº 1-2, 2001.

[20] J. Chang, M. Kim y K. Min, «Detection of misfire and knock in spark ignition engines by wavelet transform of engine block vibration signals,» Measurement Science and Technology 13(7):1108, 2002.

[21] A. Gazis, D. Panousakis, R. Chen y W. H. Chen, «Computationally inexpensive methods of ion current signal manipulation for predicting the characteristics of engine in-cylinder pressure,» International Journal of Engine Research 7 (3), 2006. [22] F. Şahin, «Effects of engine parameters on ionization current and modeling of excess air coefficient by artificial neural network,» Applied Thermal Engineering 90, pp. 94-101, 2015.

[23] D. Panousakis, A. Gazis, J. Paterson, W. H. Chen, R. Chen, J. Turner y N. Milovanovic, «Ion Current Signal Interpretation via Artificial Neural Networks for Gasoline HCCI Control,» SAE Technical paper 2006-01-1088, 2006. [24] Z. J. Wu y A. Lee, «Misfire detection using a dynamic neural network with output feedback,» SAE Technical Paper 980515, 1998.

[25] L. Eriksson, Methods for ionization current interpretation to be used in ignition control. Examensarbete utfört i Fordonssystem vid Tekniska Högskolan i Linköping, Linköping University, 1995.

[26] M. Hellring y U. Holmberg, «A comparison of ion current based algorithms for peak pressure position control,» SAE technical paper 2001-01-1920, 2001. [27] A. Deptula, D. Kunderman, P. Osinski, U. Radziwanowska y R. Wlostowski, «Acoustic diagnostics applications in the study of technical condition of combustion engine,» Archives of acoustics, vol. 41, nº 2, 2016.

[28] R. E. Hanson, «Apparatus for engine compression testing». United States Patent Patente 3839906, 1974.

[29] RCA Corporation y R. E. Hanson, «Determining engine compression from starter motor current». United States Patent Patente 3952586.

[30] V. Mahieu, P. Duponcheele, B. Leduc, J. L. Vanhemelrijk, M. Morenville y P. Dehout, «Misfire detection on S.I. Engines, by instantaneous torque analysis,» SAE Technical Paper 2000-01-0367, 2000.

[31] J. Förster, A. Lohmann, M. Mezger y K. R. Müller, «Advanced engine misfire detection for SI Engines,» SAE Technical Paper 970855, 1997.

[32] M. Willimowski y R. Isermann, «A time domain based diagnostic system for misfire detection in spark-ignition engines by exhaust-gas pressure analysis,» SAE Technical Paper 2000-01-0366, 2000.

[33] B. Nybäck, Exhaust pressure-based misfire detection. A method for detecting misfire in diesel combustion engine based upon measured exhaust manifold pressure, Stockholm. Sweden: KTH Royal Institute of Technology. School of Electrical Engineering and Computer Science, 2018.

[34] F. Taglialatela, N. Cesario, M. Porto, S. S. Merola, P. Sementa y B. M. Vaglieco, «Use of accelerometers for spark advance control of SI engines,» SAE Int. Journal. Engine, vol. 2, pp. 971-981, 2009.

[35] N. Cavina, A. Businaro, N. Rojo, M. De Cesare, L. Paiano y A. Cerofolini, «Combustion and intake/exhaust systems diagnostics based on acoustic emissions of a GDI TC engine,» 71st. Conference of the Italian Thermal Machines Engineering Association. ATI2016, vol. Energy Procedia. 101(2016), 2016. [36] A. Martychenko, J. K. Park, Y. S. Ko, A. A. Balin, J. W. Hwang y J. O. Chae, «A study on the possibility of estimation of in-cylinder pressure by means of measurement of spark-gap breakdown voltage,» SAE Technical paper, 1999. [37] N. Pashley, R. Stone y G. Roberts, «Ignition System Measurement Techniques and Correlations for Breakdown and Arc Voltages and Currents,» SAE Technical paper.

2000-01-0245, 2000.

[38] K. H. Billings, Switchmode power supply handbook, vol. 2, McGraw-Hill, 1999. [39] J. Thurman, Engine Analysis measurement. Measurement concepts, Tektronix Inc.,

1970.

[40] M. Monroy, J. D. Ramirez, C. A. Romero y J. C. Mejía, «Experimental and numerical approach to assess the dynamic performance of an inductive ignition system,» International Journal of Engine Research, pp. 1-21, 2024. [41] R. C. Stevenson, R. Palma, C. S. Yang, S. K. Park y C. Mi, «Comprehensive modeling of automotive ignition systems,» SAE Technical Paper 2007-01-1589,

2007.

[42] Q. Wang, Y. Zheng, J. Yu y J. Jia, «Circuit model and parasitic parameter extraction of the spark plug in the ignition system,» Turk J Elec Eng & Comp Sci, vol. 20, nº 5,

2012.

[43] G. J. Rohwein y L. S. Camilli, «Automotive ignition transfer efficiency,» SAE Technical paper series. 2002-01-2839, 2002.

[44] N. Khan, Z. Saleem y A. A. Mirza, «Ignition study of a petrol/CNG single cylinder engine,» Intelligent Systems in Design and Manufacturing. VI. Proc. of SPIE, vol. 5999, nº 59990J.

[45] S. Rózowicz, «Use of the mathematical model of the ignition system to analyze the spark discharge, including the destruction of spark plug electrodes,» Open Phys, vol. 16, pp. 57-62, 2018.

[46] J. R. Reitz, F. J. Milford y R. W. Christy, Foundations of Electromagnetic Theory, vol. 4, Pearson/Addison-Wesley, 2009.

[47] J. Cathey, Máquinas eléctricas. Análisis y diseño aplicando Matlab, McGraw-Hill,

2001.

[48] A. Zangwill, Modern Electrodynamics, Cambridge University Press, 2012.

[49] I. Andersson, Cylinder pressure and ionization current modeling for spark ignited engines. Thesis 962., Linköping: Division of Vehicular Systems. Department of Electrical Engineering. Linköpings Universitet, 2002.

[50] L. Sforza, T. Luccini, A. Onoratti, X. Zhu y S. Y. Lee, «Modeling ignition and premixed combustion including flame stretch effects,» SAE technical paper 2017-

01-0553, 2017.

[51] S. Pischinger y J. B. Heywood, «How heat losses to the spark plug electrodes affect flame kernel development in an SI-engine,» SAE technical paper 900021, 1990. [52] Günther, M; Sens, M;, Ignition systems for gasoline engines. 3rd international conference, Berlin: IAV. Springer, 2016.

[53] Günther, M; Sens, M;, Ignition systems for gasoline engines. 4th international conference, Berlin: IAV. Expert, 2018.

[54] Y. Shimasaki, M. Kanehiro, S. Baba, S. Maruyama, T. Hisaki y S. Miyata, «Spark Plug Voltage Analysis for Monitoring Combustion in an Internal Combustion Engine,» SAE Technical paper. 930461, 1993.

[55] C. Poggiani, M. Battistoni, C. Grimaldi y A. Magherini, «Experimental Characterization of a Multiple Spark Ignition System,» Energy Procedia, vol. 82,

2015.

[56] C. Forte, G. M. Bianchi, E. Corti y S. Fantoni, «Evaluation of the Effects of a Twin Spark Ignition System on Combustion Stability of a High Performance PFI Engine,» Energy Procedia, vol. 81, 2015.

[57] D. Jung y N. Iida, «An investigation of multiple spark discharge using multi-coil ignition system for improving thermal efficiency of lean SI engine operation,» Applied Energy, vol. 212, 2018.

[58] Y. Shimasaki, H. Maki, J. Sakaguchi, K. Nishizawa, A. Kato, H. Suzuki, N. Kondo y T. Yamada, «Study on Combustion Monitoring System for Formula One Engines Using Ionic Current Measurement,» SAE Technical paper series. 2004-01-1921,

2004.

[59] B. S. Goulart, C. Bringhenti, J. T. Tomita y A. C. Oliveira, «Technical evaluation of vehicle ignition systems: conduct differences between a high energy capacitive system and a standard inductive system,» Acta Scientiarum. Technology. [60] J. D. Dale, M. D. Checkel y P. R. Smy, «Application of high energy ignition systems to engines,» Prog. Energy Combustion Sci., vol. 23, 1997. [61] H. Gao, M. Hall, O. Ezekoye y R. Matthews, «Railplug design optimization to improve large-bore natural gas engine performance,» Proceedings of ICES2005. ASME Internal Combustion Engine Division 2005 Spring Technical Conference.

ICES2005-1031, 2005.

[62] O. Lang, K. Habermann, K. K. Hortmann, A. Sehr, M. Thewes, H. Kleeberg y D. Tomazic, «Potential of the spray guided combustion in combination with turbocharging,» SAE Technical paper 2008-01-0139, 2008.

[63] S. Yu y M. Zheng, «Future gasoline engine ignition: a review on advanced concepts,» International journal of Engine Research 00/(0). IMechE, 2020. [64] M. Balmelli, R. Färber, L. Merotto, P. Soltic, D. Bleiner, C. M. Franck y J. Biela, «Experimental analysis of breakdown with nanosecond pulses for spark ignition engines,» IEEE Access, vol. 9, 2021.

[65] J. B. Heywood, «Combustion and its modelling in SI engines,» International symposium Commodia 94. C94-P001, 1994.

[66] R. Herweg y R. Maly, «A fundamental model for flame kernel formation in SI engines,» SAE Technical paper 922243, 1992.

[67] J. Song y M. Sunwoo, «Flame kernel formation and propagation modelling in spark ignition engines,» Proc Instn Mech Engrs, vol. 215, 2001. [68] D. Arcoumanis y T. Kamimoto, Flow and combustion in reciprocating engines, Springer, 2009.

[69] M. Zadeh, H. Schmidt, W. Atkinson y J. Naber, «Spark mechanism in high speed flow,» SAE Technical paper 2019-01-0729, 2019.

[70] H. Shen, P. Hinze y J. B. Heywood, «A model for flame initiation and early development in SI engine and its application to cycle-to-cycle variations,» SAE Technical paper. 942049, 1994.

[71] T. Lucchini, L. Cornolti, G. Montenegro, G. D'Errico, M. Fiocco, A. Teraji y T. Shiraishi, «A Comprehensive Model to Predict the Initial Stage of Combustion in SI Engines,» SAE Technical paper 2013-01-1087, 2013.

[72] R. Maly, «Spark ignition: its physics and effect on the internal combustion engine,» de In: Hilliard. J.C., Springer, G.S.(eds) Fuel Economy., Boston, Springer, 1984. [73] C. R. Ferguson y A. T. Kirkpatrick, Internal combustion engines. Applied Thermosciences, John Wiley and sons Ltd., 2016.

[74] J. Hunicz, P. Filipek y A. Sobiesiak, «Application of ion current measurement to identification of combustión parameters in a homogeneous charge compression ignition engine,» Metrology and measurement systems, vol. 25, nº 1, 2018. [75] L. Nielsen y L. Eriksson, «Ionization current interpretation for ignition control in internal combustion engines,» IFAC 13th Triennial World Congress, 1996. [76] C. J. Savant, M. S. Roden y G. Carpenter, Diseño electrónico. Circuitos y sistemas, Prentice Hall, 2000.

[77] F. A. Soldera, F. T. Mücklich, K. Hrastnik y T. Kaiser, «Descriptions of the discharge process in spark plugs and its correlation with the electrode erosion patterns,» IEEE transactions on vehicular technology, vol. 53, 2004. [78] S. Rózowicz, «Use of the mathematical model of the ignition system to analyze the spark discharge, including the destruction of spark plug electrodes,» Open Physics,

2018.

[79] M. Sebok, M. Kucera, M. Gutten y D. Korenciak, «Thermal and electrical analysis of automotive ignition system,» The Archives of Automotive Engineering-Archiwum Motoryzacji, vol. 83, nº 1, 2019.

[80] C. A. Amman, «Classical combustion diagnostics for engine research,» SAE Technical paper 850395, 1985.

[81] D. MacKenzie y R. Honaman, «The velocity of flame propagation in engine cylinders,» SAE Technical Paper 200010, 1920.

[82] K. Schnauffer, «Engine-Cylinder Flame-Propagation Studied by New Methods,» SAE Transactions, vol. 29, pp. 17-24, 1934.

[83] A. Saitzkoff, R. Reinmann, T. Berglind y M. Glavmo, «An Ionization Equilibrium Analysis of the Spark Plug as an Ionization Sensor,» SAE Technical Paper 960337,

1996.

[84] J. Förster, A. Günther, M. Keterer y K. J. Wald, «Ion current sensing for spark ignition engines,» SAE Technical Paper 1999-01-0204, 1999. [85] Keithley. A Tektronix company, «Low current/high resistance measurements,»

2024. [En línea]. Available: https://www.tek.com/en/documents/technical-

article/technical-info-low-current-high-resistance-measurements.. [Último acceso: 17 04 2024].

[86] L. Eriksson, Spark Advance modeling and control. PhD. Dissertation No. 580, Linköping: Linköping University, 1999.

[87] S. Yoshiyama, E. Tomita y Y. Hamamoto, «Fundamental study on combustion diagnostics using a spark plug as ion probe,» SAE Technical paper 2000-01-2828,

2000.

[88] T. Pfeffer, P. Bühler, D. E. Meier y Z. Hamdani, «Influence of Intake Tumble Ratio on General Combustion Performance, Flame Speed and Propagation at a Formula One Type High-Speed Research Engine,» SAE Technical paper 2002-01-0244,

2002.

[89] A. A. Abdel-Rehim, N. A. Henein y E. VanDyne, «Ion Current in a Spark Ignition Engine using Negative Polarity on Center Electrode,» SAE Technical paper 2007-

01-0646, 2007.

[90] S. Labuda, J. Sotton, M. Bellenoue, A. P. Chernukho y A. N. Migoun, «Application of ionization probes for diagnostics of knocking combustion,» de Chia Laguna, Cagliari, Sardinia, Italy, Sardinia, Italy, 2011.

[91] V. Arrigoni, F. Calvi, G. M. Cornetti y U. Pozzi, «Turbulent flame structure as determined by pressure development and ionization intensity,» SAE Technical Paper

730088, 1973.

[92] R. L. Anderson, «In-cylinder measurement of combustion characteristics using ionization sensors,» SAE Technical Paper 860485, 1986.

[93] P. O. Witze, «Cycle-resolved multipoint ionization probe measurements in a spark ignition engine,» SAE technical paper 892099, 1989.

[94] P. O. Witze y S. C. Bopp, «Investigation of in-cylinder fluid motion using a head gasket instrumented with ionization probes,» SAE technical paper 910719, 1991. [95] Z. Hu, J. H. Whitelaw y C. Vafidis, «Flame propagation studies in a four-valve pentroof-chamber spark ignition engine,» SAE technical paper 922321, 1992. [96] Z. Hu, J. R. Laker, E. Mastorakos, C. Vafidis y H. M. Xu, «The development and application of a fast-response DC-coupled flame ionization current sensor,» Experiments in fluids, vol. 14, pp. 316-320, 1993.

[97] D. E. Nicholson y P. O. Witze, «Flame location measurements in a production engine using ionization probes embodied in a printed-circuit-board head gasket,» SAE technical paper 930390, 1993.

[98] R. Meyer, J. T. Kubesh, S. M. Shahed y J. K. Davis, «Simultaneous application of optical spark plug probe and head gasket ionization probe to a production engine,» SAE technical paper 930464, 1993.

[99] S. Russ, G. Peet y W. Stockhausen, «Measurements of the effect of in-cylinder motion on flame development and cycle-to-cycle variations using an ionization probe head gasket,» SAE technical paper 970507, 1997.

[100] D. K. Park, H. S. Kim y W. T. Kim, «A study of flame propagation for different combustion chamber configurations in an SI engine,» SAE Technical paper 970876,

1997.

[101] S. Yoshiyama, E. Tomita, N. Tabuchi, K. Matsumoto y K. Matsuki, «Combustion diagnostics of a spark ignition engine by using Gasket ion sensor,» SAE Technical paper. 2003-01-1801. JSAE 20030066, 2003.

[102] M. Glavmo, P. Spadafora y R. Bosch, «Closed Loop Start of Combustion Control Utilizing Ionization Sensing in a Diesel Engine,» SAE Technical paper 1999-01-

0549, 1999.

[103] N. Henein, W. Bryzik, A. Abdel-Rehim y A. Gupta, «Characteristics of Ion Current Signals in Compression Ignition and Spark Ignition Engines,» SAE International Journal of Engines, vol. 3, nº 1, 2010.

[104] F. Estefanous, Ionization in diesel combustion: Mechanism, new instrumentation and engine applications. Thesis 272, Wayne State University Dissertations, 2011. [105] R. M. George, Investigation Of Ion Current Signal In A Modern Automotive Diesel Engine. Thesis 298., Wayne State University, 2014.

[106] K. A. Assaad, Combustion Phasing Through Combustion Ionization In A Caterpillar C7 Compression Ignition Engine, Wayne State University, 2017. [107] P. Mehresh, J. Souder, D. Flowers, U. Riedel y R. W. Dibble, «Combustion Timing in HCCI engines determined by ion-sensor: experimental and kinetic modeling,» Proceedings of the Combustion Institute, vol. 30, 2005.

[108] T. M. Phan, J. H. Mack, R. H. Butt y R. Dibble, «A comparison of three ion sensing circuits in a homogeneous charge compression engine,» Combustion, Science and Technology, 2017.

[109] G. Dong, Y. Chen, L. Li, Z. Wu y R. Dibble, «A skeletal gasoline flame ionization mechanism for combustión Timing prediction on HCCI engines,» Proceedings of the Combustion Institute, vol. 36, pp. 3669-3676, 2017.

[110] C. F. Daniels, «The Comparison of Mass Fraction Burned Obtained from the Cylinder Pressure Signal and Spark Plug Ion Signal,» SAE Technical paper 980140,

1998.

[111] S. Yoshiyama y E. Tomita, «Combustion diagnostics of a spark ignition engine using a spark plug as an ion probe,» SAE Technical paper series. 2002-01-2838. Powertrain&Fluid systems conference& exhibition, 2002.

[112] Z. Gao, B. Li, C. Li, B. Liu, S. Liu, X. Wu y Z. Huang, «Investigation of characteristics of ion current in a metanol direct-injection spark-ignition engine,» Fuel, vol. 141, pp. 185-191, 2015.

[113] A. A. Laganá, L. L. Lima, J. F. Justo, B. A. Arrudab y M. M. Santos, «Identification of combustion and detonation in spark ignition engines using ion current signal,» Fuel, vol. 227, p. 469–477, 2018.

[114] Z. Gao, Y. Deng, Y. Wen, J. Lu, Z. Du, C. Tang, E. Tomita y Y. Tan, «Deeplearning based in-cylinder pressure modeling and resolution of ion current signals,» Fuel, vol. 282, 2020.

[115] N. Rivara, P. B. Dickinson y A. T. Shenton, «A neural network implementation of peak pressure position control by ionization current feedback,» Journal of Dynamic Systems, Measurement and Control, vol. 131, 2009.

[116] Y. Chen, G. Dong, J. H. Mack, R. H. Butt, J. Y. Chen y R. W. Dibble, «Cyclic variations and prior-cycle effects of ion current sensing in an HCCI engine: A timeseries analysis,» Elsevier, 2016.

[117] A. Dittrich, J. Laurin y T. Zvolsky, «Device for detecting the level of ionization of the fuel mixture in the cylinder petrol engine,» Applied mechanics and materials, Vols. %1 de %2799-800, pp. 880-884, 2015.

[118] P. Gillbrand, H. Johansson y J. Nytomt, «Method and arrangement for detecting ionising current in an internal combustion engine ignition system». Patente EP, C,

188180, 1989.

[119] J. Auzins, H. Johansson y J. Nytomt, «Ion gap sense in misfire detection, knock and engine control,» SAE technical paper 950004, 1995.

[120] K. Terada y S. Suzuki, «Measuring ignition and combustion by detecting ionization current,» Internal combustion engine, vol. 17, nº 212, 1978. [121] M. Asano, T. Kuma, M. Kajitani y M. Takeuchi, «Development of new ion current combustion control system,» SAE Technical Paper 980162, 1998. [122] S. Miyata, Y. Ito y Y. Shimasaki, «Flame ion density measurement using spark plug voltage analysis,» SAE Technical Paper 930462, 1993.

[123] A. Lee y J. S. Pyko, «Engine misfire detection by ionization current monitoring,» SAE technical paper 950003, 1995.

[124] A. Lee y J. S. Pyko, «Combustion detection via ionization current sensing for a Coilon-plug ignition system». United States Patent Patente US5534781, 1996. [125] J. Gatowski, E. Balles, K. Chun, F. Nelson, J. Ekchian y J. B. Heywood, «Heat release analysis of engine pressure data,» SAE Technical paper 841359, 1984. [126] Y. Huang, J. Grogan, J. Wang y J. S. Sarlashkar, «Effects of Engine Operating Conditions on In-Cylinder Air/Fuel Ratio Detection Using a Production Ion Sensing Device,» SAE Technical paper 2004-01-0515, 2004.

[127] G. Zhu, I. Haskara y J. Winkelman, «Closed-loop ignition timig control for SI engines using ionization current feedback,» IEEE Transactions on control systems technology, vol. 15, nº 3, 2007.

[128] D. Molina, F. Restrepo y I. Bedoya, «Combustion monitoring system on a natural gas fuelled spark ignition engine with high compression ratio using ionization current sensors,» Energy and sustainability VI. WIT Transactions on Ecology and the Environment, vol. 195, 2015.

[129] D. Molina y F. Restrepo, Monitoreo del proceso de combustión de un motor de encendido provocado utilizando sensores de corriente de ionización, Universidad

EIA, 2015.

[130] R. Butler y Delphi Technologies, Inc, «Ion sense biasing circuit». United States Patent Patente US6186129B1, 2001.

[131] Ł. Fiedkiewicz y I. Pielecha, «Assessment of posible use of the ionization signal for the combustion process diagnostics in a spark ignition combustion engine powered by natural gas,» Eksploatacja i Niezawodnosc-Maintenance and reliability, vol. 20, nº 4, pp. 630-637, 2018.

[132] H. Wilstermann, A. Greiner, P. Hohner, R. Kemmler, R. Maly y J. Schenk, «Ignition system integrated AC ion current sensing for robust and reliable online engine control,» SAE Technical Paper 2000-01-0553, 2000.

[133] J. Nilsson, AC jonströmsinterface. AC ion current interface. Examensarbete. Degree Project. Elektro-och datoringenjörsprogrammet., Karlstadts universitet, 2008. [134] M. S. Jovaj y G. S. Maslov, Motores de automóvil. Teoría, cálculo y estructura de los motores de combustión interna, MIR Moscú, 1982.

[135] J. B. Heywood, Internal combustion engine fundamentals, New York: Mc. Graw- Hill, 1988.

[136] C. F. Taylor y E. S. Taylor, The internal combustion engine, International Textbook Company, 1962.

[137] G. D’Errico y T. Lucchini, «A Combustion Model with reduced kinetic schemes for SI engines fuelled with compressed natural gas,» SAE Technical Paper 2005-01-

1123, 2005.

[138] D. E. Ramajo y N. Nigro, «Modelización numérica de la combustión en motores de combustión interna,» Mecánica Computacional, vol. XXIII, pp. 2233-2248, 2004. [139] J. Zheng, W. Yang, D. L. Miller y N. Cernansky, «Prediction of pre-ignition reactivity and ignition delay for HCCI using a reduced chemical kinetic model,» SAE Technical Paper 2001-01-1025, 2001.

[140] T. Kravchik, E. Sher y J. B. Heywood, «From Spark Ignition to Flame Initiation,» Combustion Science and Technology, vol. 108, 1995.

[141] C. F. Forigua y J. M. Mantilla, «Modelo de combustión de motores de encendido por chispa acoplando Cantera a KIVA 4,» Boletín de Matemáticas, vol. 23, nº 1, pp. 35-

69, 2016.

[142] H. F. Curran, P. Gaffuri, W. J. Pitz y C. K. Westbrook, «A comprehensive modeling study of iso-octane oxidation,» Combustion and Flame, vol. 129, pp. 253-280, 2002. [143] N. Atef y et al, «A comprehensive iso-octane combustion model with improved thermochemistry and chemical kinetics,» Combustion and Flame, vol. 178, pp. 111-

134, 2017.

[144] H. J. Curran, P. Gaffuri, W. J. Pitz y C. K. Westbrook, «A comprehensive modeling Study of n-heptane oxidation,» Combustion and Flame, vol. 114, pp. 149-177, 1998. [145] H. J. Curran, «Developing detailed chemical kinetic mechanisms for fuel combustion,» Proceedings of the Combustion Institute, 2018. [146] K. Mohanesen, D. Ramasamy, K. Kadirgama, M. M. Noor y M. M. Rahman, «Flame ionization testing in an internal combustion engine to measure the speed of the flame for gaseous fuels,» 1st International Postgraduate Conference on Mechanical Engineering (IPCME2018) IOP Conf. Series: Materials Science and Engineering 469, 2019.

[147] A. Vressner, Studies on the load range of an HCCI engine using in-cylinder pressure, ion current and optical diagnostics. Doctoral Thesis, Lund: Lund University. Division of Combustion engines, 2007.

[148] A. Franke, Characterization of an electrical sensor for combustion diagnostics. Doctoral Thesis, Lund: Lund University. Division of Combustion engines. Lund reports on combustion physics, 2002.

[149] A. Y. Budko, M. Y. Medvedev, R. Y. Budko, P. V. Ivashin, A. Y. Tverdokhlebov, D. N. Gerasimov y V. V. Rakhmanov, «Analysis of Ion Current Integral Characteristics for Estimation of Combustion Process Parameters in Internal Combustion Engines,» International Journal of Mechanical Engineering and Robotics Research, vol. 6, nº 3, 2017.

[150] T. Suzuki, H. Ohara, A. Kakishima, K. Yoshida y H. Shoji, «A study of knocking using ion current and light emission,» SAE Technical Paper 2003-32-0038, 2003. [151] Abhijit, G. George y J. Naber, «Correlation of air fuel ratio with ionization signal metrics in a multicylinder spark ignited engine,» SAE Technical Paper 2009-01-

0584, 2009.

[152] D. Zhu, J. Deng, J. Wang, S. Wang, H. Zhang, J. Andert y L. Li, «Development and application of ion current/cylinder pressure cooperative combustion diagnosis and control system,» MDPI Energies, vol. 13, nº 5656, 2020.

[153] ST Microelectronics, «TDA2030A 18w hi-fi amplifier and 35W driver,» 2011. [154] Texas Instruments, «LM386 Low Voltage Audio Power Amplifier,» 2023. [155] Analog devices, «Low Cost Low Power Instrumentation Amplifier AD620,» 2011. [156] Texas Instruments, «LM136-2.5-N, LM236-2.5-N, LM336-2.5-NV Reference Diode,» 2013.

[157] Texas Instruments, «LM340, LM340A and LM7805 Family Wide VIN 1.5-A Fixed Voltage Regulators,» 2016.

[158] Texas Instruments, «LM79XX Series 3-Terminal Negative Regulators,» 2013. [159] Fluke Corporation, «287/289 True-rms Digital Multimeters Users Manual,» 2007. [160] A. R. Thakare, S. B. Urkude y R. P. Argelwar, «Analysis of Cockcroft-Walton Voltage Multiplier,» International Journal of Scientific and Research Publications, vol. 5, nº 3, 2015.

[161] E. M. Ali, N. Z. Yahaya, N. Perumal y M. A. Zakariya, «Development of Cockcroft‐ Walton voltage multiplier for RF energy harvesting applications,» Journal of Scientific Research and Development, vol. 3, nº 3, pp. 47-51, 2016. [162] J. R. Eoff, «Voltage multipliers - Part 7 Impedance of the Cockcroft-Walton voltage multiplier,» 2019. [En línea]. Available: https://josepheoff.github.io/posts/diodecapacitors-volts-pt7. [Último acceso: 04 05 2024].

[163] M. H. Rashid, Power Electronics Handbook, 4th ed., Butterworth-Heinemann, 2017. [164] ON Semiconductor, «Zener Theory and Design Considerations Handbook HBD854/D Rev. 1,» 2017.

[165] Texas instruments, «CD4047B CMOS Low-Power Monostable/Astable Multivibrator,» 2003.

[166] Texas instruments, «CD4073B, CD4081B,CD4082B CMOS AND Gates,» 2003. [167] Diodes Incorporated, «1N5817-1N5819 1.0A Schottky Barrier Rectifier,» 2023. [168] Vishay General Semiconductor, «1N4001, 1N4002, 1N4003, 1N4004, 1N4005, 1N4006, 1N4007 General purpose plastic rectifier,» 2020.

[169] Vishay Semiconductors, «1N4728A to 1N4761A Zener diodes,» 2021. [170] Vishay General Semiconductor, «1N4148 Small Signal Fast Switching Diodes,»

2017.

[171] Silicon Laboratories, «AN614 A simple alternative to isolation amplifiers,» 2011. [172] Texas Instruments, «ISO124 +/-10V input, precisión isolator amplifier.,» 2018. [173] Analog Devices, «AD215. 120 kHz Low distortion, isolation amplifier,» 1996. [174] Toshiba corporation, «TLP7920, TLP7920F. Photocouplers Optically isolation amplifiers.,» Toshiba corporation.

[175] Renesas Corp., «PS8551AL4. Analog output type optical coupled isolation amplifier.,» 2014.

[176] H. Ocampo, Diseño de accionamientos y transmisiones de máquinas, Pereira: Universidad Tecnológica de Pereira. Facultad de Ingeniería Mecánica, 1993. [177] B. Cholango, Sistema de separación de residuos sólidos domésticos por medio de cribado mecánico (tamaño de partícula), Ibarra, Ecuador: Universidad Técnica del Norte. Facultad de Ingeniería en Ciencias Aplicadas, 2022. [178] Burr-Brown, D. Jones y M. Stitt, «Application bulletin. Precision Absolute Value Circuits.,» Burr-Brown, Tucson, Arizona, 1997.

[179] K. Songsuwankit, W. Petchmaneelumka, V. Riewruja y A. Rerkratn, «Linear Variable Differential Transformer Signal Conditioning Circuit Based on Phase- Locked Loop,» Sensors and Materials, vol. 36, nº 4, p. 1473–1486, 2024. [180] L. E. Avendaño, Fundamentos de instrumentación, Publicaciones UTP, 2004. [181] EXAR, «XR-2206 Monolithic function Generator,» 2008. [182] R. Yañez-Valdez, R. Alva-Gallegos, A. Caballero-Ruiz y L. Ruiz-Huerta, «Selection of Soft Magnetic Core Materials Used on an LVDT Prototype,» Journal of Applied Research and Technology, vol. 10, pp. 195-205, 2012.

[183] H. Guo y R. Zhang, «Dynamic analysis of valve system of gasoline engine based on AVL-Excite,» IOP Conference series: Earth and Environmental Science 632, vol. 632, nº 3, p. 10, 2021.

[184] X. Zheng, X. Luo, Y. Qiu y Z. Hao, «Modeling and NVH Analysis of a Full Engine Dynamic Model with Valve Train System,» Applied Sciences, p. 24, 2020. [185] R. Pallás, Adquisición y distribución de señales, Barcelona: Marcombo, 2008. [186] T. Wescott, «Sampling: What Nyquist Didn't Say, and What to Do About It,» Wayne StateUniversity , Detroit, 2016.

[187] J. Martin, Aportación al diagnóstico de la combustión en motores Diesel de inyección directa, Valencia: Universidad Politécnica de Valencia, 2007. [188] A. Abbad Andaloussi, Metodología para la parametrización en tiempo real parametrización en tiempo real del proceso de combustión en MCIA, U. P. d. Valencia, Ed., Valencia: Universidad Politécnica de Valencia, 2013. [189] O. Armas, Diagnóstico experimental del proceso de combustión en motores Diesel de inyección directa., Valencia: Universidad Politécnica de Valencia, 1998. [190] A. Cartwright y R. Fleck, «Cylinder pressure analysis in high performance two stroke engines,» SAE Technical paper 962535, 1996.

[191] D. R. Lancaster, R. B. Krieger y J. B. Lienesch, «Measurements and analysis of engine pressure data,» SAE Technical paper 750026, p. 18, 1975. [192] M. Brunt y C. Pond, «Evaluation of Techniques for Absolute Cylinder Pressure Correction,» SAE Technical Paper 970036, p. 14, 1997.

[193] K. Lee, M. Yoon y M. Sunwoo, «A study on pegging methods for noisy cylinder pressure signal,» Control Engineering Practice, vol. 16, nº 8, p. 7, 2008. [194] ABB, «ABB Drives. Reference manual. Grounding and cabling of drive systems,»

2013.

[195] Belden. Phillips, L. Bulington E., «Evaluating Critical VFD Cable Parameters,»

1999.

[196] N. Hanigovszki, J. Landkildehus, G. Spiazzi y F. Blaabjerg, «An EMC evaluation of the use of unshielded motor cables in AC adjustable speed drive applications,» IEEE Transactions on Power Electronics, vol. 21, nº 1, pp. 273-281, 2006. [197] Pacific Gas and Electric Company, «Solution of EMI Problems from Operation of Variable-Frequency Drives,» 2018.

[198] J. Ma, Y. Xue, Q. Han, X. Li y C. Yu, «Motor Bearing Damage Induced by Bearing Current: A Review,» Machines, vol. 10, nº 1167, 2022.

[199] AEGIS. Bearing Protection rings. Oh, W. Willwerth, A., «VFD-Induced Bearing Currents,» ASHRAE, 2012.

[200] TMEIC; Tessendorf, R; Hosoda, H; Kodama, S; Fukuma, K; Ishida, T;, «AC drive cable selection,» Roanoke, 2009.

[201] M. Monroy, C. Romero y E. Henao, «Considerations for starting combustion engines with AC machines,» Diagnostyka, vol. 24, nº 1, pp. 1-10, 2023. [202] A. Shah, P. Tunestal y B. Johansson, «Applicability of Ionization Current Sensing Technique with Plasma Jet Ignition Using PreChamber Spark Plug in a Heavy Duty Natural Gas Engine,» SAE Technical Paper 2012-01-1632, p. 11, 2012. [203] R. Burgett, J. Leptich y K. Sangwan, «Measuring the effect of spark plug and ignition system design on engine performance,» SAE Technical Paper 720007, p.

19, 1972.

[204] K. Kumano, Y. Akagi, S. Matohara, Y. Uchise y Y. Yamasaki, «Using an ioncurrent sensor integrated in the ignition system to detect precursory phenomenon of pre-ignition in gasoline engines,» Applied Energy, vol. 275, nº 115341, p. 9, 2020. [205] C. Song, J. Li y D. Qu, «Analysis and research on the influence factors of the spark plug ion current in spark ignition engine,» Advanced Materials Research, Vols. %1 de %21070-1072, pp. 1831-1834, 2014.

[206] M. De Zoysa, R. Howlett y S. Walters, «Effect of Varying Engine Block Temperature on Spark Voltage Characterization for The Measurement of Air-Fuel Ratio in Internal Combustion Engines.,» KES'2000. Fourth International Conference on Knowledge-Based Intelligent Engineering Systems and Allied Technologies., 2000.

[207] K. Ogata, Modern Control Engineering, 5th ed., Prentice Hall, 2010. [208] On Semiconductor, «NPT Trench IGBT 1200V, 25A FGA25N120ANTDTU,» On semiconductor, 2024.

[209] Z. Song, X. Zhang y B. Zhang, «Study on the correlation between ion current integral signal and combustion pressure,» International Journal of Hydrogen Energy, vol. 47, pp. 8060-8070, 2022.

[210] M. Mittal, G. Zhu y H. Schock, «Fast mass-fraction-burned calculation using the net pressure method for real-time applications,» Proc. IMechE Part D: J. Automobile Engineering JAUTO1006, vol. 223, p. 6, 2008.

--------------------------------------