Detection of inhomogeneities in serially connected lithium-ion battery packs

Serially connected battery packs present a cell-to-cell variation in their electrochemical behavior due to deviations during production, different temperature distributions or mechanical stress that tends to increase over the lifetime. This can lead to individual cells having significantly different electrical properties than the cell group, described in this work as inhomogeneity. This single outlier can already lead to the end-of-life criterium of the pack (or system) being reached. Since the battery pack remains to have a high economic value, the replacement of defective or accelerated-aged individual cells after its first use becomes attractive. However, detecting those inhomogeneities remains a challenging task, which has to the author’s knowledge, not been addressed in the literature so far.

This work presents a preliminary study for the impedance-based detection of singular inhomogeneities in serially connected lithium-ion battery packs. For the representation of cell-to-cell variations, a novel quantitative and qualitative analysis is given through a combined representation in the Nyquist plot. For this, impedance measurements are done with twelve Samsung NCA INR21700-50E cells at a state of charge of 50%. Virtual battery packs are built based on this impedance data to analyze a different number of cells in the serial connection. Through a comparative analysis approach, four impedance-based features are extracted from the Nyquist plots of the cells. The performance of the features is analyzed based on the number of serially connected cells and sensitivity analysis. Here the influence of the cell-to-cell variations and the ageing mechanism are examined based on newly introduced evaluation criteria. The advantages and disadvantages of the features are discussed, and the low-frequency minimum of the Nyquist plot is found to be the best-performing feature. With this, Inhomogeneities could be detected in serial connections of up to 12 cells. A flexible scalable battery module test bench is developed to prove the concept of this approach. Results show slightly lower performance than simulation, with detection possible for up to 10 serially connected cells.

Since the method allows the determination of inhomogeneities based on clamp information, it should be developed into a more application-oriented environment in the future. Therefore, the frequency range could be significantly reduced, drastically reducing the measurement time. Additionally, artificial intelligence can be used for further feature extraction or classification in the future.