Port-Hamiltonian Neural Networks for Surrogate Modelling and Uncertainty Quantification (D06*)

• Subject Area: Mechanics ; Electronic Semiconductors, Components and Circuits, Integrated Systems, Sensor Technology, Theoretical Electrical Engineering
• Term: since 2026
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 492661287

Efficient performance evaluation, design exploration, optimisation and uncertainty quantification for design and digital twins of electric machines require computationally fast, low-dimensional nonlinear surrogate and reduced-order models Thus, the objective of this project is to develop data-driven but physics-aware surrogate modelling techniques that can provide accurate, efficient, reliable and robust predictions, are trainable in small data regimes, and include parameter dependencies. For this purpose, we combine the physical interpretability and mathematical rigour of port-Hamiltonian systems with the approximation capabilities and flexibility of machine learning. These models will be able to replace high-fidelity simulations and facilitate uncertainty quantification and design optimisation.

• DFG Programme: CRC/Transregios
• Subproject of: TRR 361: Computational Electric Machine Laboratory: Thermal Modelling, Transient Analysis, Geometry Handling and Robust Design
• Applicant Institution: Technische Universität Darmstadt
• Project Heads: Professor Dr. Sebastian Schöps; Professor Dr. Oliver Weeger