Abstract
Electrochemical Impedance Spectroscopy (EIS) is a powerful non-destructive tool for evaluating the health and sorting of lithium-ion batteries. However, its practical application is hindered by the requirement that batteries be adjusted to a consistent State of Charge (SOC) before testing, a process that significantly reduces efficiency. This paper presents a novel method for EIS curve reconstruction based on geometric feature transformation and migration. By modeling the EIS curve as a combination of three circular arcs and a straight line, we define key geometric parameters that exhibit linear relationships with SOC. By establishing these relationships with a small pilot batch, we can "migrate" this knowledge to reconstruct the EIS curves of a large batch of batteries from any arbitrary SOC to a target reference. Experimental results demonstrate that this method maintains high accuracy—with root mean square errors under 1.2 mΩ—while reducing testing time by over 98% compared to traditional methods.
The Challenge of Battery Sorting for Second-Life Applications
As the first generation of electric vehicle (EV) batteries reaches retirement, "cascade utilization"—repurposing these batteries for less demanding energy storage roles—has become a critical sustainability goal. To group retired batteries effectively, engineers must assess their State of Health (SOH) and internal resistance. EIS is ideal for this because it captures the internal electrochemical dynamics of the cell without damaging it.
However, a major bottleneck exists: a battery’s impedance profile is highly sensitive to its SOC. To compare two batteries fairly, they must be at the same SOC. Traditionally, this requires charging or discharging each cell to a specific level and allowing it to reach equilibrium, a process that can take several hours per cell. In high-volume recycling environments, this delay makes EIS-based sorting economically unfeasible. This research addresses this by allowing batteries to be tested at their current SOC and mathematically "mapping" that data to a standard reference.
Methods: Geometric Modeling and Feature Migration
1. Geometric Decomposition
Instead of relying on complex equivalent circuit models (ECM), which often suffer from computational intensity and fitting instabilities, this method treats the EIS Nyquist plot as a geometric object. The curve is decomposed into four distinct segments: three circular arcs representing the Solid Electrolyte Interphase (SEI) resistance, charge transfer resistance, and transition zones, followed by a straight line representing Warburg diffusion in the low-frequency range.
2. Feature Extraction
From this geometric model, seven key features are extracted. These include the coordinates of the arc centers ($O_1, O_2, O_3$), the intersection points with the real axis ($A_1, A_2$), and the slope of the diffusion tail. These features encapsulate the electrochemical state of the battery in a simplified, manageable format.
3. Linear Mapping and Migration
The core innovation lies in the discovery that these geometric features change linearly with SOC, particularly in the low SOC range (<30%). By testing a small "pilot" batch of batteries (e.g., 3 cells) across multiple SOC levels, the researchers established the linear coefficients for how each feature moves as SOC changes. These coefficients are then "migrated" to the rest of the population. If a new battery is measured at 10% SOC, the system uses the established linear slopes to calculate where those geometric features would be if the battery were at 20% SOC, effectively reconstructing the target curve.
Experimental Validation
The study utilized 15 Foster 18650 ternary lithium-ion batteries with varying health levels (SOH ranging from 75.89% to 99.53%). EIS measurements were taken at five SOC levels (10% to 30%) using an AutoLab workstation.
As shown in the correlation analysis, the geometric features (labeled 1-10 in the study's expanded feature set) show Pearson correlation coefficients very close to 1 or -1 for different cells. This high degree of linearity confirms that the migration of these features across different SOCs is a robust and predictable process, regardless of the individual battery's aging history.
Key Results and Efficiency Gains
The proposed reconstruction method achieved remarkable accuracy and efficiency:
- Accuracy: The reconstructed curves closely matched actual measurements, with a Root Mean Square Error (RMSE) below 1.2 mΩ and a Symmetric Mean Absolute Percentage Error (SMAPE) within 3.5%.
- Time Savings: In a typical scenario, adjusting a battery from 15% to 20% SOC (including charging and the necessary rest period) takes approximately 7.76 hours. The geometric reconstruction method provides the same data in roughly 10 minutes—the time required for a single EIS scan. This represents a time reduction of over 98%.
Conclusion
By shifting the focus from complex electrochemical modeling to geometric feature transformation, this method provides a practical solution to the SOC bottleneck in battery testing. The ability to reconstruct EIS curves accurately from arbitrary SOCs enables high-throughput battery sorting, which is essential for the economic viability of battery recycling and second-life applications. Future work will look to expand this linear model into non-linear machine learning frameworks to cover the full 0-100% SOC range and adapt the technique for different battery chemistries like Lithium Iron Phosphate (LFP).
Summary updated