Image-matching in electrode production of lithium-ion batteries for marker-free tracking and tracing applications

The high complexity of lithium-ion battery production is defined by many process steps and numerous parameters. Unknown cause–effect relationships lead to production deviations and costly scrap. Therefore, data-driven approaches are increasingly being used to uncover these complex interactions. Creating comprehensive, transparent data sets based on production data, enables the identification of correlations and relevant parameters, thus facilitating efficiency improvements and the optimization of the lithium-ion production. Currently, tracking and tracing of production data to individual electrode sections is performed with physical markers on uncoated electrode parts or coils. Interactions within the processing of the battery electrodes and mechanical deformation of the current collector foil can lead to illegibility of the physical markings and thus make tracking and tracing impossible. Furthermore, the resolution of the data recording is currently tied to the discrete markings. However, advances in computer vision and computing power offer a non-invasive, marker-free solution using camera images. This paper presents an approach of unique identification of electrode sections using image-matching for marker-free tracking and tracing applications. High-resolution electrode images were acquired in-line, allowing the evaluation of artificial intelligence to demonstrate the applicability of different classical and learning-based image-matching approaches for the identification of electrode sections. The results provided a clear recommendation for image-based tracking and tracing using neuronal networks. Marker-free tracking using electrode images allowed for immediate identification of electrode sections, storage of production data in a virtual electrode model, and inclusion of additional quality information derived from the images.