The lithium-ion battery (LiB) is seen as an important component for the future of energy and mobility. The demand for these batteries has been rising in recent years. To keep up with this growth, production needs to be scaled efficiently on a large scale. With the increasing use of LiBs in electromobility, the need for safety is even more critical, as even a small defect in a LiB can pose a significant hazard. Many problems in LiBs can be identified and eliminated during production, but with today's rapid and large-scale manufacturing processes, quality control procedures are mainly dependent on either single 2D X-ray image inspection or human inspection, which can affect productivity; this indicates the crucial demand for a robust and automated defect detection system.

For quality assessment of LiB, image-based non-destructive quality control techniques, such as computed tomography (CT) and X-ray, are becoming increasingly popular as they allow the inspection of LiB cells without causing damage to the cell. However, these methods produce a large amount of data that requires a significant amount of manual inspection, demanding the need of an automated system. Recent advancements in deep learning have demonstrated the capabilities and scope of deep learning in defect detection, which can improve LiB quality control.

This research proposes an unsupervised anomaly detection method for CT images of cylindric Li-ion battery cells. A deep learning architecture called Autoencoder is utilized to find abnormalities in LiBs. The advantage of this method is that it eliminates the laborious task of data labeling, as it is unsupervised. The workflow is as follows: first, LiB 2D slices (images) are extracted from 3D CT data, then fed into a trained Autoencoder architecture to detect abnormalities. The final result are images indicating any abnormalities present. The approach can further be applied for automated quality assessment for LiB, saving significant time by making quality control faster and more precise.