Lab
IdeaIdea
Physics-informed models beyond feature engineering
Next steps after the GSM paper: constraints in the loss, other fabric structures, and models that refuse to predict outside physical bounds.
Our paper encoded physics as a feature. Open directions I want to explore:
- Constraints in the objective — penalise predictions that violate known monotonic relationships between stitch length and GSM.
- Beyond Single Jersey — rib and interlock structures, and data from more than one mill, to test generalisation.
- Bounded predictions — a model that says "outside what I've seen" instead of extrapolating confidently.
- A shop-floor tool — a calculator that shows the prediction and the physics behind it.