Artificial Intelligence and Data Driven Optimization of Internal Combustion Engines (eBook)

Front Cover 1
ARTIFICIAL INTELLIGENCE AND DATA DRIVEN OPTIMIZATION OF INTERNAL COMBUSTION ENGINES 2
ARTIFICIAL INTELLIGENCE AND DATA DRIVEN OPTIMIZATION OF INTERNAL COMBUSTION ENGINES 4
Copyright 5
Contents 6
Contributors 10
Foreword 12
Preface 16
1 - Introduction 18
1. Industrial revolution 18
2. Artificial intelligence, machine learning, and deep learning 19
3. Machine learning algorithms 20
4. Artificial intelligence-based fuel-engine co-optimization 21
4.1 Optimization of internal combustion engine 21
4.1.1 Design of experiments 22
4.1.2 Genetic algorithm 24
4.1.3 Machine learning-based algorithms 25
4.2 Optimization of fuel formulation 30
4.3 Mitigation of rare combustion events 32
5. Summary 33
References 33
1 - Artificial Intelligence to optimize fuel formulation 42
2 - Optimization of fuel formulation using adaptive learning and artificial intelligence 44
1. Introduction and motivation 44
2. Mixed-mode combustion and fuel performance metrics 45
3. A neural network model to predict fuel research octane numbers 48
4. Optimization problem formulation and description of solution approaches 49
4.1 Constrained optimization formulation 49
4.2 Genetic algorithm 50
4.3 Gaussian process–based surrogate model optimization algorithm 52
5. Numerical experiments and results 54
6. Discussion 57
7. Summary and concluding remarks 59
Acknowledgments 60
References 60
3 - Artificial intelligence–enabled fuel design 64
1. Transportation fuels 64
1.1 Fuel representation 64
1.2 Fuel formulation workflow 65
1.3 Artificial intelligence modeling approaches 66
2. Application of artificial intelligence to fuel formulation 69
2.1 High throughput screening: finding a needle in the haystack 69
2.2 Fuel property prediction by machine learning models 71
2.3 Reaction discovery 74
2.4 Fuel-engine co-optimization 75
3. Conclusions and perspectives 75
Acknowledgments 77
References 77
2 - Artificial Intelligence and computational fluid dynamics to optimize internal combustion engines 86
4 - Engine optimization using computational fluid dynamics and genetic algorithms 88
1. Introduction 88
2. Modeling framework and acceleration strategies 91
2.1 Computational fluid dynamics acceleration techniques 91
2.1.1 Adaptive mesh refinement 91
2.1.2 Detailed chemistry acceleration strategies 92
2.2 Engine geometry generation 93
2.2.1 Method of splines 93
2.2.2 Method of forces 94
2.3 Virtual injection model 95
3. Optimization methods 96
3.1 Fundamentals of genetic algorithms 96
3.2 Pioneering investigations 98
3.3 Multiobjective framework 101
3.4 Convergence acceleration 108
4. Summary and concluding remarks 114
References 115
5 - Computational fluid dynamics–guided engine combustion system design optimization using design of experiments 120
1. Introduction 120
2. Methodologies 123
2.1 Design space construction 124
2.2 Response surface model formulation 126
2.3 Model-based design optimization and verification 129
3. A recent application 130
3.1 Engine and fuel specifications 130
3.2 Computational fluid dynamic model setup and validation 130
3.3 Design variables 131
3.4 Objective variables and evaluation method 133
3.5 Data fitting and optimization 134
4. Recommendations for best practice 135
4.1 Adequate computational fluid dynamic model validation 135
4.2 Efficient geometry and mesh manipulation 136
4.3 Sample size 136
4.4 Optimization across full engine operation range 136
4.5 Computational efficiency 136
5. Conclusions and perspectives 137
Acknowledgments 138
References 138
6 - A machine learning-genetic algorithm approach for rapid optimization of internal combustion engines 142
1. Introduction 142
2. Engine optimization problem setup 144
3. Training and data examination 146
4. Machine learning-genetic algorithm approach 149
4.1 Optimization methodology 149
4.2 Repeatability of machine learning-genetic algorithm 151
4.2.1 Extension of variable domain 153
4.3 Postprocessing and robustness 156
5. Automated machine learning-genetic algorithm 158
5.1 Hyperparameter selection 159
5.1.1 Manual selection 159
5.1.2 Automated strategies for selecting hyperparameters 160
5.2 Problem setup 162
5.3 Results 163
6. Summary 173
Acknowledgments 173
References 173
7 - Machine learning–driven sequential optimization using dynamic exploration and exploitation 176
1. Introduction 176
2. Active ML optimization (ActivO) 177
2.1 Basic algorithm 177
2.2 Query strategies 178
2.3 Convergence criteria 180
2.4 Dynamic exploration and exploitation 181
3. Case study 1: two-dimensional cosine mixture function 182
4. Case study 2: computational fluid dynamics (CFD)-based engine optimization 188
5. Conclusions 196
Acknowledgments 197
References 197
3 - Artificial Intelligence to predict abnormal engine phenomena 200
8 - Artificial-intelligence-based prediction and control of combustion instabilities in spark-ignition engines 202
1. Introduction 202
1.1 Artificial intelligence applications to engine controls 202
1.2 Dilute combustion instability background 204
2. Case study: artificial-intelligence-enhanced modeling of dilute spark-ignition cycle-to-cycle variability 206
3. Case study: neural networks for combustion stability control 210
3.1 Artificial neural networks 210
3.2 Spiking neural networks 212
4. Case study: learning reference governor for model-free dilute limit identification and avoidance 216
4.1 Constrained combustion phasing control problem 216
4.2 Learning reference governor for avoiding misfire events 219
5. Summary 221
References 222
9 - Using deep learning to diagnose preignition in turbocharged spark-ignited engines 230
1. Introduction 230
1.1 Fault detection 230
1.2 Optimization and control 231
1.3 Predicting combustion parameters (phasing and cycle-to-cycle variation) and emissions 232
2. Preignition detection using machine learning algorithm 232
2.1 Feed forward multilayer neural networks 234
2.2 Convolutional neural networks 235
2.3 Recurrent neural networks 236
3. Activation functions 238
4. Experiments and data extraction 239
5. Machine learning methodology 241
6. Model 1: Input from principal component analysis 247
7. Model 2: Time series input 248
8. Model metrics 249
9. Results and discussion 250
9.1 Training and validation losses 250
10. Conclusions 251
References 252
Further reading 253
Index 256
A 256
B 257
C 257
D 257
E 258
F 258
G 258
H 258
I 258
K 259
L 259
M 259
N 259
O 260
P 260
Q 260
R 260
S 260
T 260
U 260
V 260
Z 260
Back Cover 262