Cluster Analysis of Cell Health and Characterization of Lithium and Active Material Loss in a Large Format LiFePO4 Hybrid Vehicle Battery Pack

Abstract

Listen

Translate article

Currently, the second-life Li-ion battery (LIB) landscape lacks a widely accepted safety standard for determining the viability of retired batteries for reuse.1 While the impact of abusive conditions on battery failures, such as toxic gas releases, fires, and explosions, has been studied extensively, the understanding of how aging affects the likelihood of thermal runaway under normal use conditions remains limited. This knowledge gap underscores the critical need for a deeper understanding of material degradation throughout the LIB lifecycle and its direct safety implications. In response, researchers are exploring advanced characterization methods to deconvolve and identify degradation behaviors, typically using lab-aged cells. These methods, which range from non-destructive evaluation to post-mortem analyses, have revealed various degradation modes (such as active material loss and Li inventory loss) and mechanisms (including Li plating, SEI growth and decomposition, transition metal dissolution, and particle cracking).2 However, the variability of these degradation modes in real-world battery packs remains poorly understood, making it challenging to estimate the likelihood of thermal runaway. A precise statistical understanding of these degradation mechanisms is essential for developing robust safety standards and maximizing the economic benefits of reusing LIBs.In this work, we apply differential voltage analysis to assess the statistical distribution of active material loss in a retired 1,536-cell hybrid-vehicle battery pack (see Ref. 3 for pack details).3 Initially, we measured the capacity and ohmic resistance of 1,500 cells (98% of the pack) using galvanostatic cycling at a rate of 1C (Fig. 1A-D). We then implemented k-medoids clustering to categorize the cells based on their capacity and resistance (Fig. 1E). The representative medoid cell from each cluster was then selected for further analysis at a slower C/10 cycling rate to gather dV/dQ data. Using open-source software provided by Dahn et al.,4 we fit the measured dV/dQ curves to reference curves collected from a LiFePO4 cathode vs. Li/Li+ and a graphite anode vs. Li/Li+ half-cells (the latter was provided by with the dV/dQ software). The fitting process yields 4 parameters: mG (active mass of the graphite anode), mLFP (active mass of the LFP cathode), 𝛿 LFP (LFP cathode slippage), and 𝛿 G (graphite anode slippage). Active material loss and slippage are then calculated relative to an uncycled cell with mG = 8.1 g and mLFP = 13 g.Fig. 1 summarizes the SOH of 1,500 cells (A-C), the cell categories and representative cells determined via k-medoid clustering (D), the dV/dQ curve-fitting method (E), and example dV/dQ curve-fitting results for 2 cells (F). The cell capacities (nominally 2.3 to 2.5 Ah) ranged from 0.0 to 2.2 Ah and their direct-current internal resistances (DCIR), which started at 10 mΩ when newly manufactured, ranged from 30 to 1000 mΩ. For two example cells (Fig. 1F), a -24% and +22% difference in % initial capacity (PIC) and DCIR, respectively, corresponded to a 150% increase in active graphite loss, a 99% increase in active LFP loss and a 6.7% difference in slippage (where 𝛿 = 𝛿 G - 𝛿 LFP ). In this presentation, we will expand on these initial results and will summarize our findings on the distribution of capacity, resistance, active material loss, and Li inventory loss across the entire pack.References Christensen, P. A., Mrozik, W. & Wise, M. S. A Study on the Safety of Second-Life Batteries in Battery Energy Storage Systems. UK BEIS/OPSS Report (2021).Li, A. G., West, A. C. & Preindl, M. Applied Energy 316, 119030 (2022). DOI: 10.1016/j.apenergy.2022.119030.Ramirez-Meyers, K., Rawn, B. & Whitacre, J. F. J. Energy Storage 59, 106472 (2023). DOI: 10.1016/j.est.2022.106472.Dahn, H. M., Smith, A. J., Burns, J. C., Stevens, D. A. & Dahn, J. R. J. Electrochem. Soc. 159, A1405–A1409 (2012). DOI: 10.1149/2.013209jes. Figure 1: A-D summarize the SOH of 1,500 cells. A) Charge and discharge curves, V(Q), at a 1C rate (that is, current = 2.0 A). B and D) Capacity and DCIR distributions. C) K-medoids clustering using z-scores (normalized capacity and DCIR values) shows representative cells as medoids of each cluster. E) An example of fitting measured dV/dQ (black) to dV/dQ calculated from the LFP cathode (blue) and graphite anode (purple) references. F) Comparative example of electrode mass and slippage calculations obtained from dV/dQ analysis. * = outliers are exclud ed. DCIR = direct-current internal resistance. LAM = Loss of active material. LFP = LiFePO4. PIC = Percent of initial capacity. Z-score = (Value – Mean)/Standard Deviation. Figure 1