The growing market of electric vehicles in recent years has increased the demand for lithium-ion batteries (LIBs) as well as the potential large amount of retired vehicle batteries. The concept of repurposing used EV batteries as second-life energy storage units has been accordingly developed to bolster sustainable energy systems while reducing waste and environmental impact. To optimize the performance, safety, and reliability of such second-life batteries (SLBs), some research gaps still need to be addressed. Firstly, SLBs are likely to come from different manufacturers, vehicle models, and usage scenarios, leading to significant variations in their internal characteristics and behavior. This heterogeneity makes it challenging to develop accurate and robust models for their state estimation. Furthermore, the estimation accuracy of battery states is highly influenced by aging mechanisms, which arise from a complex interplay of physical and chemical mechanisms affected by environmental conditions, usage patterns, and operational history. SLBs have already undergone a certain level of degradation during their first life, and this degradation may continue during their second life due to further usage and environmental conditions. In these cases, the model and algorithms need to be more robust to cover different scenarios (e.g., different capacities and/or resistances at the same temperature). Such challenges would be harder to overcome when the usage history of SLBs is not available.
This study focuses on developing an adaptive state-of-charge (SoC) estimation framework for retired EV LIBs. Two types of commercial NMC batteries (type A and B, with nominal capacities of 65Ah and 141Ah, respectively) have been dismantled from automotive battery packs after 1st life usage. To better understand the aging status of the SLBs, a bunch of characterization experiments (e.g., the quasi-open circuit voltage test, the pulse charging test, etc.) have been conducted. Then, the SoC estimation framework is proposed by considering the hardware and software requirements of SLBs management systems on a grid scale.
The Worldwide harmonized Light-duty vehicles Test Cycles (WLTC) is employed to validate the method, as it mimics real-world usage scenarios. For type A battery (with state-of-health around 93.54%), the validation results show that the adaptive filter converges quickly from the initialization error (20% SoC difference) and tracks the measured SoC accurately with the root mean square error (RMSE) as 0.02. Noticeably higher SoC estimation errors have been seen when the SoC is lower than 20%. This higher error could be introduced by the higher nonlinearity of OCV-SoC correlation at those regions, which further suggests that a lower SoC range should be better avoided in SLBs applications. The proposed framework was also applied to type B batteries, and similar estimation accuracies were achieved under the WLTC dynamic profile.