Data-driven thermal modeling of a permanent magnet synchronous motor with machine learning / Wilhelm Kirchgässner, M. Sc. ; Erster Gutachter: Prof. Dr.-Ing. Oliver Wallscheid, Zweiter Gutachter: Prof. Dr.-Ing. Oliver Nelles, Dritter Gutachter: Prof. Dr.-Ing. Joachim Böcker. Paderborn, 2024
Inhalt
- Acknowledgments
- Abstract
- Zusammenfassung
- Acronyms, symbols, and notation
- Introduction
- Automotive electric traction drives
- Market and societal significance
- System overview of an electric drive train
- Requirements and challenges
- Electric motor types
- Power electronics
- Drive control
- Relevance of thermal stress and its monitoring
- Dynamic modeling of a permanent magnet synchronous motor
- Machine learning and its applications in the engineering field
- Current and past trends
- Supervised learning
- Black-box modeling
- Neural networks in state-space representations
- Data set description and design of experiments
- Relevant data sets
- Excitation planning – Design of experiments
- Means and metrics for measurement profile comparison
- Black-box temperature estimation with machine learning
- Feature engineering
- Cross-validation
- Static modeling overview
- Linear thermal modeling analysis
- ANN-based dynamic thermal modeling
- Further topological enhancements for sequence modeling
- Mini-batch training scheme over subsequences
- Performance overview
- Hyperparameter optimization
- Discussion
- Gray-box temperature estimation with state-space machine learning
- Related work
- Learning an explicit-Euler-discretized nonlinear function
- Thermal neural networks
- Hyperparameter optimization
- Model sparsification
- Input-to-state stability and the severity of skewed initialization
- Generalization to neural ordinary differential equations
- Optimization of a data-driven, expert-designed LPTN
- Conclusion
- Bibliography
- Own publications
- List of Figures
- List of Tables
- Appendix