05Scientific machine learning
SURGE · IIT KanpurMicrostructure evolution forecasting
Simulation accelerated by 50%
An autoencoder and ConvLSTM forecasting model that predicts how binary-alloy microstructure evolves over time, using a sequence of simulation images as input.
- Encode simulation frames into a compact representation
- Forecast temporal evolution with ConvLSTM
- Accelerate the phase-field simulation workflow by 50%
HIGH LEVEL DESIGNSELECT A STEP TO EXPLORE
STEP 01 / 05Simulation frames
Binary alloy microstructure
RUNTIME / SYSTEM LAYER
10k+ simulation imagesReproducible training pipeline50% faster simulation workflow
From the constraint
to the system.
Developed the fusion forecasting model and reproducible training pipeline. The research work is publicly documented in the project repository.
END-TO-END PATH
01Microstructure image sequence
02Autoencoder representation
03ConvLSTM temporal model
04Forecast next states
05Compare with simulation