Design and development of an ignition system for heavy-duty transport applications through a novel AI-based approach

Dipartimento di Ingegneria “Enzo Ferrari”

Master’s Degree in Electronics Engineering

Design and development of an ignition system

for heavy-duty transport applications

through a novel AI-based approach

Mentor: Candidate:

Prof. Claudio Bianchini Michele Faccone

Company tutors:

Ing. Simone Daniele, Ph.D

Ing. Federico Ricci, Ph.D Academic year 2024/2025

Contents

1 Introduction 1

1.1 Evolution of Ignition Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1.1 Lenoir, 1860: the first ICE . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1.2 Honold, 1902: High-Tension Magneto Ignition . . . . . . . . . . . . . 2

1.1.3 Kettering, 1910: Battery & Coil Ignition System . . . . . . . . . . . . 3

1.1.4 Spielberg, 1953: Electronic Transistor Ignition System . . . . . . . . . 4

1.2 Modern Era Ignition Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.2.1 Coil-on-Plug Ignition System . . . . . . . . . . . . . . . . . . . . . . 5

1.2.2 Smart Ignition Coils implementation . . . . . . . . . . . . . . . . . . . 7

1.3 Next-Generation Ignition Systems . . . . . . . . . . . . . . . . . . . . . . . . 8

1.3.1 Ion-Sensing Ignition System . . . . . . . . . . . . . . . . . . . . . . . 8

1.3.2 Plasma Ignition Systems . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.4 Introduction to the SuperCoil Ignition System . . . . . . . . . . . . . . . . . . 11

2 Theory 13

2.1 Magnetizing Inductance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.2 Mutual Inductance, Turn Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.4 Intrinsic Resistance of Windings . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.5 Skin Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.6 Proximity Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.7 Parasitic Capacitances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.8 Leakage Inductances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.9 Core Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.10 Fundamental Relation on Ignition Systems . . . . . . . . . . . . . . . . . . . . 25

3 Design and Simulation 26

3.1 Transformer design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.1.1 Core Selection and Physical Dimensions . . . . . . . . . . . . . . . . . 27

3.1.2 Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.2 Simulation of the proposed design . . . . . . . . . . . . . . . . . . . . . . . . 28

3.2.1 Electrical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.2.2 FEMM Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4 Parameter Optimization 36

4.1 Python Algorithm Description . . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.1.1 Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.1.2 Design Loading and Test . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.1.3 Target Values Declaration, Custom Helpers, Data Extraction . . . . . . 37

4.1.4 SciPy Differential Evolution function . . . . . . . . . . . . . . . . . . 39

4.1.5 Python Optimizer Results . . . . . . . . . . . . . . . . . . . . . . . . 40

4.2 AI refinement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

4.2.1 Deep Learning Model . . . . . . . . . . . . . . . . . . . . . . . . . . 42

4.2.2 Genetic Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

4.2.3 Convergence Check and Validation . . . . . . . . . . . . . . . . . . . 44

5 Test Campaign 45

5.1 First trial: Tenneco’s SuperCoil Proposal with Parameter Optimization . . . . . 45

5.1.1 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

5.1.2 First Step: Python Research . . . . . . . . . . . . . . . . . . . . . . . 46

5.1.3 Second Step: AI refinement . . . . . . . . . . . . . . . . . . . . . . . 47

5.1.4 Final Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5.2 Second trial: Frenetic Prototype Simulation . . . . . . . . . . . . . . . . . . . 49

5.2.1 Measurements on Prototype: Electrical Measurements, X-Ray Exam

and Test Bench Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 50

5.2.2 Tailored Algorithm Analysis . . . . . . . . . . . . . . . . . . . . . . . 54

5.2.3 Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

5.3 Third trial: Corrective Design Proposal . . . . . . . . . . . . . . . . . . . . . 58

6 Conclusions 62

6.1 Summary of Findings and Contributions . . . . . . . . . . . . . . . . . . . . . 62

6.2 Future Work and Recommendations . . . . . . . . . . . . . . . . . . . . . . . 63

7 Acknowledgements 64

A Python Optimizer Code i

List of Figures

1 Lenoir motor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 High-tension magneto system . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

3 Battery and Coil ignition system . . . . . . . . . . . . . . . . . . . . . . . . . 3

4 Coil-on-plug ignition system . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

5 Example of AEM’s high output IGBT inductive smart coil . . . . . . . . . . . 7

6 Ion-Sensing Coil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

7 Conventional spark plug and Transient Plasma System ignition module . . . . . 10

8 3D render of SuperCoil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

9 Ideal transformer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

10 Magnetizing Inductance visualized . . . . . . . . . . . . . . . . . . . . . . . . 15

11 Interwinding capacitance measurement . . . . . . . . . . . . . . . . . . . . . . 21

12 Leakage flux representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

13 Laminated core example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

14 SIMBA scheme of the SuperCoil . . . . . . . . . . . . . . . . . . . . . . . . . 29

15 Spark event waveforms. SIMBA environment . . . . . . . . . . . . . . . . . . 31

16 Detail on discharge event waveforms. SIMBA environment . . . . . . . . . . . 31

17 FEMM 2D planar design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

18 Property table of NdFeB 39 MGOe. FEMM environment . . . . . . . . . . . . 33

19 Property table of Nanocrystalline. FEMM environment . . . . . . . . . . . . . 33

20 Density Plot of SuperCoil - FEMM environment . . . . . . . . . . . . . . . . . 34

21 Core saturation due to permanent magnets, perpendicular orientation - FEMM

environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

22 Input and Output waveforms of initial configuration - SIMBA SuperCoil model 38

23 Input and Output waveforms of optimized configuration - SIMBA SuperCoil

model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

24 Dataset division, given from the Python optimizer . . . . . . . . . . . . . . . . 42

25 Deep Learning Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

26 GA (input and output combination example) . . . . . . . . . . . . . . . . . . . 44

27 Input and Output waveforms of First Trial, Python Research - SIMBA SuperCoil

model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

28 Input and Output waveforms of First Trial, Converged result - SIMBA SuperCoil

model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

29 Voltech Test Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

30 KeySight Impedance Analyzer Test Setup . . . . . . . . . . . . . . . . . . . . 50

31 Primary Winding, 01 prototype . . . . . . . . . . . . . . . . . . . . . . . . . . 51

32 Primary Winding, 02 prototype . . . . . . . . . . . . . . . . . . . . . . . . . . 51

33 Secondary Winding, 01 prototype . . . . . . . . . . . . . . . . . . . . . . . . 51

34 Secondary Winding, 02 prototype . . . . . . . . . . . . . . . . . . . . . . . . 51

35 Interruped wire detected on coil 02. X-ray portrait . . . . . . . . . . . . . . . . 52

36 Damaged magnet detected on 01 prototype. X-ray portrait . . . . . . . . . . . 52

37 air gap measurement on 01 prototype. X-ray portrait . . . . . . . . . . . . . . 52

38 Detail on test bench performance. 0.8 ms charge time . . . . . . . . . . . . . . 53

39 Graphical representation of peak currents with respect to the charging times . . 54

40 Graphical representation of Table 9 . . . . . . . . . . . . . . . . . . . . . . . . 55

41 Input and Output waveforms of Tailored Research - SIMBA SuperCoil model . 57

42 Energy estimation of SIMBA model . . . . . . . . . . . . . . . . . . . . . . . 58

43 Energy estimation of test bench perforrmance . . . . . . . . . . . . . . . . . . 58

44 Input and Output waveforms of Corrective Design Proposal - SIMBA SuperCoil

model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

List of Tables

1 Design specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2 Parameters used in the simulation . . . . . . . . . . . . . . . . . . . . . . . . 30

3 Parameter optimization boundaries. First trial . . . . . . . . . . . . . . . . . . 45

4 Target output data of optimization process. First trial . . . . . . . . . . . . . . 45

5 Output of Python research. First trial . . . . . . . . . . . . . . . . . . . . . . . 47

6 Output of AI refinement procedure. First trial . . . . . . . . . . . . . . . . . . 47

7 Electrical parameters measurement, 01 prototype . . . . . . . . . . . . . . . . 50

8 Test bench measurement report on 01 prototype . . . . . . . . . . . . . . . . . 54

9 Estimation on magnetizing inductance. Comparison between approach results

and measurement tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

10 Parameter optimization boundaries. Third trial . . . . . . . . . . . . . . . . . 59

11 Target output data of optimization process. Third trial . . . . . . . . . . . . . . 59

12 Output of AI refinement procedure. Third trial . . . . . . . . . . . . . . . . . . 60

To my loved ones, to myself,

and to all those who believed in me.

Abstract

The Internal Combustion Engine (ICE) remains the dominant engine typology worldwide, converting

chemical energy from fuel–air mixtures into mechanical work. As the automotive and industrial

sectors face increasing pressure to reduce pollutant emissions and greenhouse gasses, significant

efforts are being directed toward the development of cleaner combustion solutions, among which

hydrogen represents a particularly compelling alternative. Hydrogen stands out because of its zero-

carbon combustion products and its suitability for heavy-duty applications such as tractors, busses,

and trucks, where battery-electric solutions often encounter limitations in terms of payload, range,

and charging time. However, hydrogen combustion presents its own challenges: ignition is highly

sensitive to in-cylinder conditions [15], particularly pressure and which dictate whether the mixture

,

is rich or lean. The operation of lean-burn hydrogen, in particular, requires a significantly higher

ignition voltage [8], leading to physically larger ignition systems and imposing additional packaging

restrictions. In addition, the ignition device must withstand the harsh environment of the engine,

where limited space and elevated temperatures impose strict design constraints.

This thesis addresses these challenges by focusing on the development of a novel high-power ignition

coil specifically optimized for hydrogen engines. The design process is structured into multiple stages.

First, a volumetric definition of the device is established in collaboration with Frenetic, providing opti-

mized material selections and feasible geometries for physical realization. Subsequently, the electrical

design is refined through the optimization of an equivalent circuit model using Python-based tools

and an AI-driven methodology combining Deep Learning and Genetic Algorithms. This approach

aims to achieve high output energy comparable to, or surpassing, that of traditional Capacitive Dis-

charge Ignition (CDI) systems, while delivering the long-duration sparks required for stable hydrogen

combustion.

The results demonstrate the successful development of a robust AI-assisted design workflow capable

not only of improving an existing prototype but also of providing comprehensive design guidelines

for entirely new ignition devices. The methodology significantly reduces development time and

enables the achievement of high-energy, long-duration spark characteristics that are essential for

reliable hydrogen ignition. The study concludes with three different validation attempts: the first

demonstrating the capability of the system to support the design of new devices, the second confirming,

through prototype measurements, the accuracy of the model in reproducing real product behavior,

and the third showcasing the ability of the method to guide corrective redesign of an existing device.

The conclusions discuss the success of this innovative approach and outline the next steps for further

refinement and development.

1 Introduction

1.1 Evolution of Ignition Systems

1.1.1 Lenoir, 1860: the first ICE

Ignition systems are a key part of internal combustion engines (ICE) and, over time, have evolved

and adapted to achieve greater performance, efficiency, and reliability. Historically, the first system

developed was patented in 1860 by Étienne Lenoir [4], who built one of the first successful internal

combustion engines. While rudimentary by today’s standards, this system laid the groundwork for

modern spark ignition engines. The system used a battery to provide electrical energy, while the spark

was generated by a jumping spark ignition system that is essentially an earlier version of today’s spark

plug. Using a breal-type contact within the combustion chamber, while the contacts were opened, a

spark jumped across the gap, igniting the fuel-air mixture. In that period, the engine was running

on coal gas and not gasoline, so clearly the energies involved in this historical mechanism were

completely different compared to the ones of today. The timing was extremely primitive: no timing

Figure 1: Lenoir motor

advance or sophisticated distributor. The spark generated was more or less continuous or, in some

versions, manually controlled. This system was inefficient and quite cumbersome, but an absolute

breakthrough. It was the first system to ignite a fuel-air mixture with the help of an electrical spark,

and it was also the precursor of modern spark plug and battery-coil ignition systems.

1

1.1.2 Honold, 1902: High-Tension Magneto Ignition

A step forward in the development of a proper spark ignition system occurred in 1902, when Bosch’s

engineering team, headed by Gottlob Honold, developed the High-Tension magneto ignition system

[3]. Introduced in 1902, it provided a self-contained and reliable solution capable of generating a

high-voltage spark without a battery. In fact, the electrical energy required to generate the spark was

obtained from the magneto attached to the crankshaft. The spin of the permanent magnets induce a

low voltage on the primary of the magneto. Since the magneto has to step up the voltage in order

to ionize the air, i.e. to generate the spark, the primary coil was composed of a few turns of thick

wires, while the secondary coil was composed of many turns of thin wire. Basically, the magneto was

employed as a transformer. Once the voltage has stepped up, a mechanical breaker point is set to open

Figure 2: High-tension magneto system

when ignition should occur, and, when it opens, the primary current suddenly stops, causing a rapid

collapse of the magnetic field. By reflection, this phenomenon induces a massive voltage surge in the

secondary coil. The high voltage travels through a distributor that routes it to the correct spark plug.

In the end, the spark plug ignites, igniting the fuel-air mixture at precisely the right time.

Although this system introduced an increase in ignition efficiency and precision, it showed some

improvements and downsides. First, the absence of a battery helped the system behave as a self-

sustained system, being always ready to ignite the mixture; second, it showed impressive reliability

at high camshaft rotation speeds, a key aspect for aircraft deployment; and last, it was durable, with

fewer points of failure. In contrast, starting the engine was difficult because magnetos do not produce

strong voltage at low RPM and, also, the engine suffers mechanical wear in breaker points and rotating

parts. All these aspects made this kind of ignition system suitable for aircraft applications, and

still it is implemented in modern solution due to redundancy, obtained by using two magnetos, and

independence from battery or electrical systems, a fundamental aspect when dealing with emergencies.

2

1.1.3 Kettering, 1910: Battery & Coil Ignition System

The 1910s and 1920s represent the age of mass-produced automobiles, and new ignition systems were

needed to be developed to be reliable, precise, and allow easy use in everyday automobiles. The

battery and coil ignition system, proposed by Charles F. Kettering and developed at Delco in 1910,

became the king of reliability [3, 4], thanks to Cadillac in 1912 and especially with the rise of Ford’s

Model T and beyond. This system fixed many of the drawbacks of magnetosis and paved the way

for a modern ignition architecture. An important difference from the system proposed by Honold is

that this solution is battery-powered, so it worked well even at low RPM or when starting. Also, it

standardized spark timing, was cheaper, easier to maintain than magnetos, and could be scaled for mass

production. A 6V or 12V lead-acid battery powers the ignition system that is charged by a generator

Figure 3: Battery and Coil ignition system

or alternator and allows consistent energy to be available at any engine speed, including idle. This

aspect represents a huge step forward with respect to the high-tension magneto ignition system that

instead required manual engine start. The ignition coil is basically a step-up transformer: the primary

winding is made with a thick wire turned around a few hundred times; the secondary winding is made

with tens of thousands of turns of thin wire. The battery current flows t