A Data-Driven Fleet Service: State of Health Forecasting of Lithium-Ion Batteries (AutoUni Schriftenreihe, 170)

Given the limitations of state-of-the-art methods, this book presents a state of health (SOH) forecasting method that is suitable for lithium-ion battery (LIB) systems in real-world battery electric vehicle operation. Its histogram-based features can capture the higher operational variability compared to constant and controlled laboratory operation. Also, the transferability of a trained machine learning model to new LIB cell types and new operational domains is investigated. The presented SOH forecasting method can be provided as a cloud service via a web or smartphone app to fleet managers. Forecasting the SOH enables fleet managers of battery electric vehicle fleets to forecast and plan vehicle replacements. Preface Context of this Thesis Abstract Kurzfassung Contents Acronyms List of Figures List of Tables 1 Introduction 1.1 Motivation 1.2 Research Questions 1.3 Structure of this Thesis 2 Theoretical Background 2.1 Lithium-Ion Batteries 2.1.1 Operating Principle and Components of Lithium-Ion Battery Cells 2.1.2 Parameters for Characterization and Operation of Batteries 2.1.3 State of Health and Remaining Useful Lifetime 2.1.4 Fundamentals of Lithium-ion Battery Cell Aging 2.1.5 Battery Packs 2.1.6 Battery Operation 2.2 Supervised Machine Learning for Regression Problems 2.2.1 Artificial Neural Networks 2.2.2 Feature Scaling 2.2.3 Hyperparameter Tuning 2.2.4 Metrics for Regression Problems 2.2.5 Transfer Learning 3 Towards State of Health Forecasting of Lithium-Ion Batteries 3.1 State of Health Estimation, Prediction, and Forecasting 3.2 State of Health Forecasting vs. Remaining Useful Life Prediction 3.3 Scenarios in State of Health Forecasting 3.4 Battery Model Transfer 3.5 Model Key Criteria 4 Related Work 4.1 Screening Method 4.2 State of Health Forecasting Models 4.2.1 Models without Information about Future Load 4.2.2 Models with Information about Future Load 4.3 Transfer Learning for Battery Models 4.4 Gaps of Related Work 5 Data 5.1 Laboratory Battery Cell Data Sets 5.1.1 Stanford Battery Data Sets 5.1.2 RWTH Aachen ISEA Cyclic Aging Data Set 5.1.3 NASA Randomized Battery Usage Data Set 5.1.4 Oxford Degradation Data Set 5.2 Real-World Battery Electric Vehicle System Data Set 6 Battery Cell State of Health Forecasting 6.1 Method 6.1.1 Stressor Extraction 6.1.2 Decision on Signal Selection 6.1.3 Machine Learning Regression Model 6.2 Design of Experiments 6.3 Results 6.4 Summary and Contribution to the Research Questions 7 Transfer of Battery Cell State of Health Forecasting 7.1 Method 7.1.1 How to Transfer: Layerwise Freezing 7.1.2 When to Transfer: Data Availability 7.1.3 Benchmarks 7.1.4 Data 7.2 Design of Experiments 7.3 Results 7.4 Summary and Contribution to the Research Questions 8 Battery System State of Health Forecasting 8.1 Method 8.1.1 State of Health Forecasting 8.1.2 Stressor Data 8.2 Design of Experiments 8.2.1 Histogram-based Features 8.2.2 Accessible Features 8.2.3 General Settings 8.3 Results 8.3.1 Histogram-based Features 8.3.2 Accessible Features 8.4 Use Cases 8.5 Summary and Contribution to the Research Questions 9 Concept for a Technical Implementation 9.1 Battery Electric Vehicle Fleets 9.2 Stakeholder Roles of Fleets 9.3 State of the Art: Fleet Management 9.4 Fleet Management Concept for Manufacturers 9.4.1 Machine Layer 9.4.2 Fleet Operator Layer 9.4.3 Manufacturer Layer 9.4.4 Data Storage and Machine Learning Model Training 10 Limitations & Outlook 10.1 Method 10.2 Future Data 10.3 Battery Systems 10.4 Further Learning Paradigms 11 Conclusion Bibliography