In large-scale production, lithium-ion battery cells are usually classified only according to the two characteristics of capacity and internal resistance, provided there are no optical defects, geometric tolerance violations, high self-discharge rates or weight overruns.
However, since cells are almost exclusively connected in groups (XsYp) in the application, inhomogeneities in the voltage profile must be avoided explicitly, which can lead to string limitations and temperature deviations without complex balancing. Particularly in dynamic high-performance applications such as automotive traction batteries, this is of utmost importance.
Investigations on commercial cells show that there are more sensitive parameters than capacity and internal resistance of a cell, thus being better suited for high-precision matching. In literature and an own series of measurements, the voltage profile differences prove to be a viable indicator of scattering for previously identically charged and series-connected cells. By using a partial differential voltage analysis (DVA) and partial incremental capacity analysis (ICA), the differences become even more apparent, allowing good and bad cells to be selected using simple classification techniques such as k-means clustering.
Based on twelve series-connected cells, an approach for pass/fail decision and sorting of cells at the end of a production line is presented, which identifies the inhomogeneities of the stress profile resulting from the inevitably tolerance-limited production process. By introducing an adaptive individual cell homogeneity score, different characterization methodologies can be weighted into an overall assessment.
The method shown proves to be scalable and enables cost reductions in the process due to the smaller number of current channels. The use of extensive analysis criteria in the individual cell homogeneity score achieves a higher sensitivity compared to conventional grading methods.
Future research will shed light on the use of dynamic load profiles, alternative clustering algorithms, and sensitivity to additional production defects.