joonkiBattery CT defect screening
Ultralytics models that screen battery CT images for internal defects (porosity and resin overflow in module pouches, porosity in cells, swelling), tuned for recall to prevent in-flight fires.
Model details
- Architecture
- Ultralytics detection, segmentation and classification models; 5-fold ensemble for swelling
- Input
- CT images (224–640 px)
- Output
- Defect boxes and masks (porosity, resin overflow), swelling class
- Training data
- Experimental
- Trained on
- AI-Hub 103 battery defect images (CT only)
- Uncertainty
- Not reported
- Physics constraints
- None
- License
- mit