Swin3D Transformer Predicts Battery Discharge Dynamics
Researchers introduced a deep learning surrogate pipeline using the Swin3D Transformer to predict spatiotemporal discharge dynamics in lithium-ion batteries from volumetric data. The model integrates Gaussian Positional Encoding for spatial feature adaptation to electrode microstructures and a Temporal Encoding module for non-linear time-series evolution. Validated on an Electrochemical Simulation dataset, it outperforms state-of-the-art point cloud baselines in accuracy while reducing computational overhead by orders of magnitude. The work was published on arXiv with ID 2607.20577.
Key facts
- arXiv ID: 2607.20577
- Uses Swin3D Transformer
- Gaussian Positional Encoding enhances spatial features
- Temporal Encoding module captures non-linear evolution
- Validated on Electrochemical Simulation dataset
- Outperforms point cloud baselines
- Reduces computational cost by orders of magnitude
- Published as a replace-cross announcement
Entities
Institutions
- arXiv