ICFormer
ICFormer applies attention over incremental capacity curves to estimate loss of lithium inventory and loss of active material. It is trained on 'alawa synthetic curves, which provide exact labels for every degradation path.
Model details
- Architecture
- Transformer encoder
- Input
- IC curve (dQ/dV vs. voltage)
- Output
- LLI, LAM_PE, LAM_NE (%)
- Training data
- Synthetic
- Uncertainty
- Not reported
- Physics constraints
- None
Evaluation results
| Task | Split | Score |
|---|---|---|
| Degradation-mode diagnosis | In-distribution | RMSE 1.10 |
| Degradation-mode diagnosis | Cross-chemistry | RMSE 3.90 |
| Degradation-mode diagnosis | Noisy and partial curves | RMSE 2.40 |