Battery foundation model
Pretrained on large volumes of 'alawa synthetic data, starting with IC curves and adding DV, partial voltage curves and full V/I/T series. One pretrained backbone, fine-tuned per task, with the 'alawa consistency check as an uncertainty signal.
Model details
- Architecture
- Transformer (pretraining in progress)
- Input
- IC curves (first stage), then DV, partial V(Q) and V/I/T
- Output
- Task-specific heads
- Training data
- Synthetic
- Uncertainty
- Planned ('alawa consistency check, conformal prediction)
- Physics constraints
- Planned (bounds on degradation modes, monotonic fade)