Lithium-ion batteries (LiBs) are essential for the future of energy storage and mobility, making their reliability and safety a critical concern. Even minor defects in LiBs can pose significant hazards. To develop better and more efficient batteries, it is crucial to understand their internal geometry. Depending on the use case, materials scientists often rely on non-destructive techniques such as computed tomography (CT) and X-ray imaging to analyze LiB cells without compromising their integrity. These methods are particularly useful for studying aging effects and structural changes by comparing pristine and used cells. However, while non-destructive techniques provide valuable insights, they generate vast amounts of data that require extensive manual inspection, making the process both time-consuming and prone to human error. This highlights the need for automated systems to enhance efficiency and accuracy in defect detection.

This research explores an unsupervised deep learning approach for analyzing 3D CT data of cylindrical LiB cells. Various unsupervised deep learning architectures are applied and compared to identify structural anomalies. Additionally, a key objective is to evaluate the capabilities and limitations of deep learning in detecting LiB anomalies in CT data, given the inherent resolution constraints of CT imaging. The trained model enables automated detection of anomalies, which can be further quantified in a 3D volume. This helps in understanding the patterns and occurrences of anomalies, such as deformations.

A major advantage of the unsupervised approach is that it eliminates the need for extensive data labeling, reducing the manual effort required for training deep learning models. The proposed method enhances LiB quality assessment and helps LiB research by providing an efficient and time-saving, abnormality detection, which ultimately contributes to the development of safer and more reliable energy storage solutions.