Physics-Informed Neural ODEs for IBR Dynamics
Physics Constraints and Residual Backbones
C3-05 · Module outline
Place physical knowledge in the loss or in the model architecture.
Learning objectives
- Distinguish soft penalties from a physical backbone.
- Explain the bias introduced by incorrect nominal physics.
Concept outline
- Pure vector field with a physics penalty
- Nominal dynamics plus learned residual
- Parameter mismatch and residual placement
Experiment direction · Planned
Simulation and code experiment
Vary the physics weight and compare against a residual architecture.
Model explanations, parameter settings, runnable code, and result interpretation will be developed here.
Practice and discussion
State what the residual is allowed to change.
Worked solutions and feedback will accompany the full lesson.
Material preparation
Related notes, models, or notebooks are available and need adaptation into a web lesson and experiment.
Adapt the existing material into a lesson, a reproducible experiment, and worked practice.