Casting solidification GNN
A graph neural network predicts when every point of a casting solidifies in a green-sand mold, and where the last-to-solidify regions (hot spots) are. Pick a part and an alloy to explore precomputed predictions in 3D.
The outer surface is drawn see-through and coloured by solidification time. The solid body inside is the 10% of the casting that solidifies last. Drag to rotate, scroll to zoom.
Run it on your own part. This page shows precomputed results
for three examples. The valve body is a real casting; the other two are synthetic shapes made
for this demo. To predict a STEP file of your own, install the pipeline from
GitHub and run
python -m src.predict_casting part.step; it downloads the model automatically.
A version of this page that accepts uploads is in preparation.
The predictions are a screening aid. They do not replace a calibrated process simulation or physical validation. Training data, evaluation and limitations are on the model card.