Material-resolving computed tomography of lithium-ion batteries using deep learning

Le résumé fourni par la source

The demand for batteries as portable energy storages increases drastically. Especially for electric mobility, battery safety is crucial which begins at seamless quality control during and after manufacturing. Recent developments in high-speed computed tomography (high-speed CT) enable scan times around 10 s, roughly matching the speed of a typical battery production line. While the majority of defects in batteries can be detected using the CT scan data directly, data post-processing such as material identification can reveal further insights. As the complexity of modern battery production grows, traditional material-resolving CT methods face challenges in delivering the precision and efficiency required. To meet these demands, more advanced, data-driven approaches are becoming essential. This has led to an ongoing paradigm shift in material-resolving CT, introducing deep learning techniques that promise enhanced accuracy and processing speed. In the scope of this paper, we propose an end-to-end deep learning approach, which is designed to resolve materials in CT scans in presence of heavy CT artifacts by exploiting context knowledge with a convolution-based neural network. The model computes atomic numbers and densities directly from the dual-energy CT volume slices for each pixel. Our approach uses simulation-generated training data, thereby avoiding the need for manual annotation. CT scans from two fundamentally different systems — one providing slow, high-quality scans and the other fast, medium-quality scans — are compared in terms of material identification performance. Especially for high-speed CT, increasing the scanning time can influence the data quality drastically. We believe, that the combination of a high-speed scanner for pre-screening together with a slower high-quality scanner provides comprehensive in-line inspection, where only critical candidates, revealing anomalies in the high-speed scan, will be send to the high-quality scanner.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.