Lithium plating is a safety-critical side reaction in lithium-ion batteries that lowers efficiency, reduces usable capacity, and increases the risk of failure. Its assessment is still often based on manual post-mortem inspection, which is slow, subjective, and difficult to scale. This poster presents an automated computer-vision approach for the systematic characterization of lithium plating from high-resolution flatbed scans across a condition matrix spanning 1C to 4C and -20 °C to -5 °C.
The method combines a convolutional attention-based variational autoencoder with k-means clustering to identify, organize, and quantify plating-related surface features. The model was trained on the full imaging dataset of more than 5000 scans from over 60 cells to construct a comprehensive morphological atlas of the experimental matrix. A fixed subset of 13 cells with 1092 scans, covering the full range of conditions, was used exclusively for convergence monitoring and for the quantitative trend analyses presented here.
The learned two-dimensional latent space captures the main operating-condition effects within the matrix. It separates both C-rate and temperature while also reflecting plating severity and spatial uniformity. In addition, the latent coordinates correlate monotonically with the stripping capacity, Q_strip, establishing a quantitative link between visible surface morphology and electrochemical indicators. Spearman correlation analysis confirms that the latent dimensions encode the relevant stress factors, showing strong relationships with both temperature and C-rate.
Overall, the results demonstrate that computer vision can provide an objective, scalable, and quantitative alternative to manual inspection. The proposed framework complements non-invasive electrical diagnostics by extracting condition-dependent plating information directly from post-mortem images. This enables high-throughput screening, systematic risk mapping, and improved understanding of lithium plating behavior under 1C–4C and -20 °C to -5 °C conditions, supporting the development of safer and longer-lasting lithium-ion battery systems.