The electrical properties of lithium-ion batteries (LIBs) are highly dependent on the production process. Process variations during production lead to a deviation of physical product parameters, inhomogeneities, and defects. While defects should be avoided due to their significant impact on the cell’s performance and safety, inhomogeneities are often acceptable in tolerances [1]. As a result, LIBs of the same type often show a distribution of electrical properties such as internal resistance and capacity [2]. However, identifying and localizing the origins of deviations in electrical performance parameters is often challenging due to the overlapping effects of inhomogeneities at the cell, electrode, and particle levels.

In contrast to electrical measurements, image-based methods such as computed tomography (CT) allow the direct detection of defects and measurement of inhomogeneities [1]. However, when applied with reduced scanning times [3] and for full-cell scans of larger cells, the resulting image quality and resolution decrease. A decrease of resolution and image quality leads to difficulties in applying machine learning-based measurement algorithms. In addition, low amounts of training data hinder the development of machine learning-based algorithms prior to mass production. To address these issues, we present a novel approach for advanced image processing based on transfer learning, synthetic training data, and a generative adversarial network (GAN) approach for image improvement. Imaging and measurement results are shown for prismatic battery cells and presented with their electrical performance.

[1] Evans, D., Luc, P.-M., Tebruegge, C., & Kowal, J. (2023). Detection of Manufacturing Defects in Lithium-Ion Batteries-Analysis of the Potential of Computed Tomography Imaging. Energies, 16(19), 6958.
[2] Evans, D., Brieske, D. M., Tebruegge, C., & Kowal, J. (2024). Analysis of the impact of manufacturing induced cell-to-cell variation for high-power applications. J. Power Sources, vol. 614, p. 235001.
[3] Dreier, T., Nilsson, D., & Espes, E. (2024). In-line and at-line battery CT enabled by MetalJet sources. 13th Conference on Industrial Computed Tomography (iCT) 2023, 6 – 9 February 2024 in School of Engineering, Wels Campus, Austria. e-Journal of Nondestructive Testing.