| license: openrail | |
| implements: | |
| 1. Multi-output property surrogate | |
| Predicts resistivity, TCR, Seebeck coefficient, resistance drift, hardness, strength, ductility, and stability score. | |
| 2. Uncertainty-aware prediction | |
| Uses ensemble spread as prediction uncertainty, matching your document’s idea that property prediction models can provide candidate estimates and uncertainty bounds for robust feasibility testing | |
| 3. Scenario robustness testing | |
| Perturbs composition, cold work, annealing, test temperature, cyclic strain, and measurement noise. | |
| 4. Robust feasibility scoring | |
| Checks whether each alloy route remains inside target property windows across scenarios. | |
| 5. Matroid-style greedy partitioning | |
| Assigns candidates into buckets such as: | |
| * electrical stability | |
| * high resistivity | |
| * formability | |
| * robust manufacturing | |
| * experimental validation | |
| 6. Closed-loop update path | |
| The code is compatible with the workflow in your PDF: Generate → Predict → Partition → Validate → Update, where new experiments update the database and retrain the surrogate model | |
| Important: the script includes synthetic data only for testing the pipeline. For real conclusions, replace it with actual experimental, CALPHAD, phase-field, or published Cu–Ni property data. | |
| https://matroid.streamlit.app/ |