Mask-to-CT : Translating Electrode Masks to Synthesize CT-Style Battery Images

Industrial Computed Tomography (CT) has become a vital tool in battery manufacturing, enabling non-destructive and high resolution inspection of internal electrode structures. With the rapid diversification and advancement of battery powered applications such as Electric Vehicles (EV), manufacturers are under huge pressure to accelerate the massive development of novel battery types. Despite the increasing adoption of AI-based battery inspection systems, training robust inspection models remains a critical bottleneck due to the severe imbalance between abundant normal data and scarce defective samples, especially for newly developed battery architectures. The problem becomes worse during the initial stages of production, where the lack of labeled defective data significantly hinders AI-model performance, often resulting in delays of the production lines. To address this issue, we integrated a GAN-based training scheme to design a new architecture for synthesizing structurally diverse yet stylistically consistent CT-style battery images from explicitly masked input electrode mask data. These findings indicate that mask-to-CT generative augmentation could effectively mitigate data shortage and opens opportunities for flexible and faster battery manufacturing system. In conclusion, our method enables stable generation of synthetic battery images, potentially overcoming data scarcity and enhancing AI inspection performance for new battery types