DTW-CNN battery degradation diagnosis

DTW-CNN battery degradation diagnosis

Original DTW-CNN checkpoints from Costa et al. (2022) that estimate LLI, LAM_PE and LAM_NE from slow-charge curves of LFP, NCA and NMC cells.

Model details

Architecture

CNN on dynamic-time-warping images, one checkpoint per chemistry

Input

C/25 charge V(Q) curve, turned into a 128 × 128 DTW image of its IC curve against a pristine reference

Output

LLI, LAM_PE, LAM_NE (%)

Training data

Synthetic

Uncertainty

Not reported

Physics constraints

None

Limitations

• Needs a pristine reference curve; references are supplied only for the C0–C3 configurations.
• DTW images are scaled by one maximum per group of curves, so predictions depend on the group; preparing one curve at a time doesn't reproduce the paper's preprocessing.
• Original checkpoints, not retrained on the current Hugging Face splits; the published checks don't reproduce the paper scores.