One-Shot Segmentation of Battery Cells with Gradual Learning
Due to the growing importance of electric vehicles and battery energy storage systems, battery safety must be ensured during and after production. One aspect is the visualization of inner structures, which can be achieved by computed tomography (CT) as a non-destructive testing (NDT) method. Deep-l...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | deu |
| Published: |
NDT.net
2025-02-01
|
| Series: | e-Journal of Nondestructive Testing |
| Online Access: | https://www.ndt.net/search/docs.php3?id=30722 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Due to the growing importance of electric vehicles and battery energy storage systems, battery safety must be ensured during and after production. One aspect is the visualization of inner structures, which can be achieved by computed tomography (CT) as a non-destructive testing (NDT) method. Deep-learning tools can learn and perform different image processing tasks in the obtained volume quickly. However, in most settings, generating labeled data needed to train these tools is expensive. Therefore, this work addresses the segmentation of anodes and cathodes in CT volumes through Gradual Learning (GL), a technique that requires a single annotated slice of the volume only. The technique exploits the high similarity between adjacent slices and is applied to both battery stack cells and cylindrical cells. For the stack cells, translational similarity is used, which resulted in an average gain of 0.09 Intersection over Union (IoU) points. For cylindrical cells, sequential segmentation along the rotated center slices is proposed. This led to a higher initial IoU of 0.78 vs. 0.73 for stack cells before GL application. While the IoU gain of 0.01 IoU points through GL was smaller for cylindrical cell types, the qualitative samples showed improvement as the remaining artifacts were removed.
|
|---|---|
| ISSN: | 1435-4934 |