Lithium-ion batteries (LiBs) are central to the advancement of clean energy technologies, powering everything from electric vehicles to grid storage systems. Ensuring their reliability and safety remains a priority for both materials researchers and industry alike. Internal structural defects—such as pores, foreign particles, cracks, deformations, and delamination—though often rare and subtle, can significantly compromise battery performance, efficiency, and lifespan. Therefore, detecting and understanding such defects is important for both quality control and fundamental research. Non-destructive imaging techniques like X-ray computed tomography (CT) have become invaluable for inspecting the internal geometry of LiB cells. These methods allow researchers to analyze pristine and aged batteries without compromising their integrity. However, 3D CT data are complex and large in volume, making manual inspection labor intensive, slow, and potentially inconsistent. This has driven growing interest in machine learning (ML) solutions for defect detection and classification.

This talk provides insights into approaches to develop deep learning assisted automated defect detection tailored for CT data of LiB cylindrical as well as pouch cells. The focus is on known defect types that are challenging yet possible to identify within the resolution limits of CT imaging, such as pores, foreign particles, cracks, deformations, etc. The study critically examines the limitations of previously developed unsupervised deep learning models for defect quantification. Additionally, it evaluates multiple supervised deep learning-based approaches and architectures, such as segmentation and object detection, for their efficacy in identifying and localizing these defects. The talk also addresses the key limitations in both the CT imaging modality and ML-based detection methods when applied to LiB CT data, highlighting which defects are likely to be detectable, which remain out of scope at available CT resolutions, and where further research is needed to bridge the gap.

The developed approach helps battery researchers to accelerate material characterization and failure analysis by significantly reducing analysis time and enhancing defect interpretability. Ultimately, this work supports the broader goal of advancing LiB reliability through faster and more scalable diagnostic tools.