New Method Enhances Sorting Accuracy of Retired EV Batteries
As the global electric vehicle (EV) market surges forward, the lifecycle management of lithium-ion batteries has become a pivotal challenge for sustainable development. With millions of EVs reaching end-of-life, the volume of retired batteries is growing at an unprecedented rate. While battery recycling remains a critical solution, the concept of “second-life” or echelon utilization—repurposing used batteries for less demanding applications such as energy storage systems—is gaining traction as a more economically viable and environmentally responsible alternative. However, a major bottleneck in realizing this potential lies in the inconsistent performance of retired battery cells, which vary widely in capacity, internal resistance, and aging characteristics due to their diverse usage histories. To ensure safety, longevity, and efficiency in second-life applications, it is essential to group these cells into highly consistent clusters—a process known as consistency sorting.
A groundbreaking study published in Battery Bimonthly introduces a novel and highly effective approach to this challenge. Led by Zhao Guangjin from the State Grid Henan Electric Power Research Institute, in collaboration with Meng Gaojun, Dong Ruifeng, Su Ling, and Zhang Zheng from Nanjing Institute of Technology, the research team has developed a precision sorting methodology that leverages discharge curve analysis and an enhanced fuzzy clustering algorithm. Their work, titled Research on the consistent sorting method of retired batteries based on discharge curves and improved fuzzy C-means algorithm, presents a scalable, accurate, and efficient solution that could significantly advance the commercial viability of battery reuse.
The urgency of the issue cannot be overstated. As Zhao Guangjin points out, “The rapid adoption of electric vehicles, driven by national ‘dual carbon’ goals, has led to a looming wave of battery retirements. Without an efficient and reliable sorting mechanism, the promise of second-life utilization remains largely theoretical.” Current sorting practices often rely on simple metrics such as remaining capacity or internal resistance. While these single-parameter methods are fast and easy to implement, they fail to capture the full spectrum of a battery’s health and aging behavior. Two batteries with identical capacities, for instance, may have vastly different internal degradation patterns, leading to mismatched performance when grouped together. This inconsistency can accelerate aging, reduce system efficiency, and even pose safety risks in large-scale energy storage deployments.
Multi-parameter approaches attempt to address this limitation by incorporating several metrics—such as open-circuit voltage, impedance, and capacity fade rate—into the sorting process. While more comprehensive, these methods are often time-consuming and costly, requiring extensive cycling tests and sophisticated diagnostic equipment. This makes them impractical for large-scale industrial applications where thousands of batteries need to be processed efficiently. Moreover, many existing techniques overlook the dynamic behavior of batteries during operation, focusing instead on static snapshots of performance.
The research team’s innovation lies in shifting the focus from isolated parameters to the entire discharge profile—a rich source of information that reflects both the static and dynamic characteristics of a battery. The discharge curve, which plots voltage against time during a controlled discharge, serves as a unique fingerprint of a battery’s electrochemical behavior. As batteries age, their internal resistance increases and active materials degrade, causing subtle but measurable shifts in the shape of the discharge curve. These changes are particularly evident in the voltage plateau duration, the slope of the initial and final discharge phases, and the overall voltage decay profile.
To translate these complex curves into actionable data, the team devised a streamlined feature extraction process. Instead of analyzing the entire curve—which would generate massive datasets and lead to computational inefficiencies—they identified seven key feature points that best represent the curve’s morphological characteristics. These include the start and end points of discharge, the 50% time midpoint, and critical inflection points marking transitions between different discharge phases. However, identifying precise inflection points through tangent analysis is computationally intensive and not suitable for high-throughput sorting.
To overcome this, the researchers introduced a simplified yet highly effective approximation. They replaced the complex geometric calculations with fixed percentage-based time markers: 3% for the first inflection point, 10% for the start of the voltage plateau, 90% for its end, and 97% for the final transition point. This pragmatic adjustment drastically reduces processing time while preserving the discriminative power of the original method. As Meng Gaojun explains, “Our goal was to balance accuracy with practicality. By using time-based reference points, we enable rapid, automated feature extraction without sacrificing the ability to distinguish between batteries with different aging trajectories.”
Once the feature points are extracted, they are fed into an advanced clustering algorithm to group similar batteries together. The team employed an improved version of the Fuzzy C-Means (FCM) algorithm, a machine learning technique that assigns each battery a degree of membership to multiple clusters rather than forcing a hard classification. This is particularly useful in the context of retired batteries, where performance degradation is a gradual and overlapping process. A battery may exhibit characteristics of both moderately and heavily aged groups, and the fuzzy approach allows for a more nuanced and realistic grouping.
However, traditional FCM algorithms are sensitive to initial conditions and can converge to suboptimal solutions, especially when dealing with heterogeneous datasets like retired batteries. To address this, the researchers integrated a pre-processing step known as Subtraction Clustering (SUB), which analyzes the density distribution of data points to identify optimal initial cluster centers. This hybrid SUB-FCM approach ensures faster convergence and higher clustering accuracy, making it robust even when faced with noisy or irregular data.
The effectiveness of the proposed method was validated through extensive experiments on a dataset of 96 retired 18650 lithium-ion cells, all of which had undergone real-world use in electric vehicles. The batteries were first categorized into two capacity ranges: 2.0–2.5 Ah (high capacity) and 1.75–2.0 Ah (medium capacity). Within each group, the SUB-FCM algorithm was applied to perform consistency sorting based on the simplified feature points.
The results were striking. In the high-capacity group, the 41 batteries were successfully clustered into three distinct groups. Post-sorting analysis showed a significant reduction in voltage dispersion across all feature points. For example, the maximum voltage difference (ΔUmax) at the initial inflection point dropped from 0.132 V before sorting to below 0.061 V after sorting. Similarly, the standard deviation (SDU) decreased from 0.048 V to as low as 0.005 V in the best-performing cluster. These improvements indicate a much tighter grouping of batteries with similar electrochemical behavior, which is crucial for ensuring uniform current distribution and thermal management in a battery pack.
In the medium-capacity group, the algorithm identified four distinct clusters among the 55 batteries. Again, the post-sorting metrics demonstrated a marked improvement in consistency. The average ΔUmax across all feature points was reduced by over 30%, and the SDU values showed a consistent downward trend. Notably, the mean voltage (Uave) of each cluster varied significantly, reflecting different aging levels and internal resistances. This allowed for clear differentiation between groups, enabling users to assign sorted batteries to appropriate applications based on performance requirements.
Perhaps most importantly, the method proved capable of distinguishing batteries with similar capacities but different aging patterns—a capability that traditional sorting techniques often lack. For instance, two batteries with the same nominal capacity might have one that aged primarily due to calendar aging (time and temperature) and another due to cycle aging (frequent deep discharges). These different degradation mechanisms leave distinct signatures on the discharge curve, which the feature-based approach can detect and classify.
The implications of this research extend far beyond the laboratory. For battery recyclers and second-life operators, the method offers a scalable and cost-effective solution for improving the quality and reliability of repurposed battery systems. By ensuring higher consistency within each group, the risk of premature failure is reduced, warranty costs are minimized, and overall system performance is enhanced. This, in turn, boosts consumer confidence in second-life products and accelerates market adoption.
From a sustainability perspective, the impact is equally significant. Efficient sorting enables higher utilization rates of retired batteries, reducing the volume of waste sent to recycling or landfills. It also lowers the demand for new raw materials, thereby decreasing the environmental footprint of battery production. As global demand for energy storage continues to rise—driven by renewable integration, grid stabilization, and backup power needs—the ability to deploy reliable, low-cost storage solutions based on second-life batteries becomes increasingly valuable.
Industry experts have welcomed the findings. “This work represents a significant step forward in battery lifecycle management,” said an independent energy storage analyst. “The integration of discharge curve analysis with intelligent clustering algorithms is both innovative and practical. It addresses a real-world problem with a solution that is ready for industrial deployment.”
The research team is already exploring ways to further refine the method. One area of focus is the impact of temperature on discharge characteristics. As noted in the paper, all experiments were conducted under constant temperature conditions. In real-world applications, however, batteries are exposed to varying thermal environments, which can influence voltage profiles and, consequently, sorting accuracy. Future work will investigate temperature compensation models to enhance the robustness of the algorithm across different operating conditions.
Another promising direction is the integration of machine learning for real-time sorting. By training deep neural networks on large datasets of discharge curves, it may be possible to automate the entire process—from data acquisition to final classification—without the need for manual feature engineering. This could further reduce processing time and improve adaptability to different battery chemistries, such as lithium iron phosphate (LFP) or nickel manganese cobalt (NMC).
The study also highlights the importance of standardization in the battery reuse industry. As more players enter the market, there is a growing need for common protocols and performance metrics to ensure interoperability and quality control. The feature points and evaluation indices proposed in this paper—such as ΔUmax, SDU, and Uave—could serve as foundational elements for future technical standards.
In conclusion, the work by Zhao Guangjin, Meng Gaojun, and their colleagues offers a powerful and practical solution to one of the most pressing challenges in the EV and energy storage sectors. By combining physical insight with advanced data analytics, they have developed a method that not only improves sorting accuracy but also paves the way for a more circular and sustainable battery economy. As the world transitions to clean energy, innovations like this will play a crucial role in ensuring that every kilowatt-hour is used to its fullest potential.
Research on the consistent sorting method of retired batteries based on discharge curves and improved fuzzy C-means algorithm by Zhao Guangjin, Meng Gaojun, Dong Ruifeng, Su Ling, Zhang Zheng, published in Battery Bimonthly, DOI: 10.3969/j.issn.1002-087X.2024.09.016