New AI-Free Method Accurately Tracks Battery Aging in Electric Vehicles
A groundbreaking method for tracking the capacity degradation of lithium-ion batteries in electric vehicles has been introduced by researchers at Jiangsu University, offering a promising solution to one of the most persistent challenges in battery management: accurately estimating how much a battery has aged without relying on extensive laboratory testing or pre-trained models.
The study, led by Dr. Pei Lei from the Automotive Engineering Research Institute at Jiangsu University, presents a novel approach that reconstructs a three-dimensional surface—known as the OCV surface—using real-world operational data collected by vehicle battery management systems (BMS). This method enables the continuous, online extraction of battery capacity trajectories, which are essential for assessing battery health, predicting remaining useful life, and ensuring the long-term reliability and safety of electric vehicles.
Unlike conventional techniques that depend heavily on offline experiments and large datasets to train machine learning models, this new framework operates entirely without training. It leverages only the natural voltage behavior of batteries during idle periods, specifically the open-circuit voltage (OCV), which reflects the thermodynamic state of the cell when no current is flowing. By analyzing how OCV evolves with both state of charge (SOC) and cycle count over time, the researchers have developed a self-contained system capable of mapping the full aging journey of a battery from its first use to end-of-life.
The innovation lies in the way scattered OCV measurements—often incomplete and noisy due to real-world driving patterns—are transformed into a coherent, evolving surface. In typical vehicle operation, batteries are rarely fully charged or discharged, and rest periods are short and irregular. As a result, BMS systems collect only fragmented snapshots of the OCV curve at various points in the battery’s life. These fragments, while individually insufficient, collectively contain rich information about the battery’s aging process.
Pei and his team recognized that these fragments, though sparse, retain local shape characteristics of the underlying SOC-OCV relationship. More importantly, fragments from the same aging stage share consistent shape features, while those from different stages diverge due to capacity fade and electrode degradation. By exploiting this principle—termed “shape compatibility” within the same aging level and “shape exclusivity” across different levels—the algorithm can intelligently align and reposition each fragment in both the SOC and cycle count dimensions.
This alignment is achieved through an iterative optimization process that enforces the fundamental monotonic relationship between SOC and OCV. For most lithium-ion chemistries, including the widely used lithium iron phosphate (LFP) cells tested in the study, higher OCV always corresponds to higher SOC. Any deviation from this order indicates misalignment due to estimation errors or aging differences. The algorithm corrects these deviations by adjusting the SOC offset of entire fragments—rather than individual points—to preserve their intrinsic shape while improving global consistency.
Once all fragments are properly aligned, they form a continuous 3D surface where the x-axis represents cycle count, the y-axis SOC, and the z-axis OCV. This reconstructed OCV surface becomes a powerful diagnostic tool. A horizontal slice through the surface at the maximum OCV value reveals the battery’s capacity trajectory over time—essentially showing how much capacity is lost with each charge-discharge cycle. Meanwhile, vertical slices along the cycle axis yield precise SOC-OCV curves at specific aging stages, which are critical for accurate state-of-charge estimation in aging batteries.
One of the most compelling advantages of this method is its independence from prior knowledge. Traditional data-driven models require thousands of charge-discharge cycles under controlled conditions to generate training data. Even then, their performance often degrades when applied to different battery types, temperatures, or usage patterns. In contrast, the OCV surface reconstruction method needs no such preparation. It learns directly from the vehicle’s own operational history, making it inherently adaptable to diverse real-world conditions.
To validate their approach, the research team designed a comprehensive simulation framework based on actual aging experiments conducted on 32650-type LFP/GIC cells with a nominal capacity of 5 Ah. These cells were cycled at 2C (10 A) under room temperature conditions, and high-resolution OCV curves were recorded at regular intervals. From this experimental baseline, they generated a synthetic dataset mimicking the fragmented and error-prone nature of real BMS data—adding random SOC estimation noise and limiting each OCV fragment to just 5–8 data points, consistent with typical parking durations.
When tested on this realistic dataset, the algorithm demonstrated exceptional accuracy. The reconstructed OCV surface matched the true surface with a mean absolute error (MAE) below 4.2 mV and a root mean square error (RMSE) under 7.9 mV across the entire lifespan of the battery. More importantly, the extracted capacity trajectory showed a maximum absolute percentage error of only 3.51%, with a mean absolute percentage error (MAPE) under 0.35% and a root mean square percentage error (RMSPE) below 0.70%. These figures indicate that the method can reliably track capacity fade within a margin of error that is acceptable for both consumer applications and fleet management systems.
The implications of this work extend beyond simple capacity estimation. By providing a continuous, high-fidelity view of the battery’s SOC-OCV evolution, the method opens new avenues for understanding aging mechanisms. For instance, shifts in the OCV curve’s inflection points can reveal changes in electrode kinetics, such as lithium plating or solid electrolyte interphase (SEI) growth. Such insights are invaluable for battery designers seeking to improve longevity and for operators aiming to optimize charging strategies.
Moreover, the method’s computational efficiency makes it suitable for onboard implementation. The algorithm converges within 30 iterations, requiring minimal processing power—critical for integration into embedded BMS platforms with limited resources. Because it does not store or transmit large volumes of raw data, it also aligns well with privacy and bandwidth constraints in connected vehicles.
Another key strength is its universality. The method is not tied to any specific battery chemistry, size, or manufacturer. As long as the battery exhibits a monotonic SOC-OCV relationship—which holds true for nearly all commercial lithium-ion cells—the algorithm can be applied. This cross-compatibility is a significant advantage in an industry where multiple battery types coexist and new chemistries are constantly emerging.
The research also addresses a common limitation in existing diagnostic tools: the need for periodic recalibration using full charge-discharge cycles. In practice, such cycles are rarely performed in daily driving, rendering many offline methods impractical. The OCV surface reconstruction method circumvents this issue by working with partial and intermittent data, effectively turning every parking event into a potential data point for health assessment.
From a system integration perspective, the method complements existing BMS functionalities rather than replacing them. It enhances the accuracy of state-of-health (SOH) estimation without interfering with real-time state-of-charge (SOC) calculations or thermal management systems. In fact, the refined SOC-OCV curves produced by the method can be fed back into the BMS to improve SOC estimation precision as the battery ages—a critical factor in preventing overcharging and deep discharging, which accelerate degradation.
While the current implementation focuses on cycle aging, the theoretical foundation of the method suggests it could be extended to calendar aging as well. Calendar aging, driven by time and temperature rather than usage, is a major contributor to capacity loss in vehicles that are driven infrequently. Future work could incorporate temperature data and storage duration into the OCV surface model, enabling a more comprehensive health assessment.
The team acknowledges one limitation: the method assumes that the cycle count or equivalent aging metric is available for each OCV fragment. In some cases, especially with second-life or repurposed batteries, this information may be incomplete or missing. However, the researchers suggest that alternative aging indicators—such as cumulative charge throughput or time-in-use—could serve as proxies, allowing the method to function even in data-scarce scenarios.
This work stands out in the growing field of battery diagnostics for its rigorous theoretical foundation and practical applicability. Rather than relying on black-box AI models whose decisions are difficult to interpret, the method is grounded in electrochemical principles and transparent logic. Every step—from fragment alignment to surface slicing—is explainable and verifiable, enhancing trust in its outputs.
In an era where artificial intelligence is often hailed as the solution to every engineering challenge, this study demonstrates the enduring value of physics-informed, rule-based approaches. It proves that deep domain knowledge, combined with clever algorithm design, can outperform data-hungry models in complex, real-world environments.
For automakers and battery manufacturers, the technology offers a pathway to more reliable warranties, smarter battery recycling, and enhanced customer confidence. For fleet operators and charging infrastructure providers, it enables predictive maintenance and optimized charging schedules. And for individual drivers, it means greater transparency about their vehicle’s battery condition—without the need for specialized equipment or service visits.
As electric vehicle adoption continues to rise, the demand for accurate, low-cost, and scalable battery diagnostics will only grow. Methods like the one developed by Pei Lei and his colleagues at Jiangsu University represent a critical step toward sustainable, data-driven mobility. By transforming routine operational data into actionable health insights, they bring us closer to a future where battery degradation is no longer a mystery—but a measurable, manageable part of the driving experience.
The research was supported by several national and regional funding programs, including the National Natural Science Foundation of China, the Jiangsu Provincial Natural Science Foundation, the China Postdoctoral Science Foundation, and the Zhenjiang Municipal Science and Technology Program. These investments underscore the strategic importance of advanced battery management in the transition to clean transportation.
Looking ahead, the team plans to test the method on real-world vehicle fleets and explore its application in second-life battery evaluation and grid-scale energy storage systems. With further refinement, the OCV surface reconstruction technique could become a standard feature in next-generation BMS architectures, setting a new benchmark for battery health monitoring.
In summary, this study introduces a robust, training-free method for extracting battery capacity degradation trajectories by reconstructing the cycle count–SOC–OCV surface from discrete operational data. The approach is universal, accurate, and computationally efficient, making it highly suitable for online implementation in electric vehicles and other battery-powered systems.
Pei Lei, Chen Bin, Wang Tiansi, Li Huanhuan, Jiangsu University, Journal of Power Sources, DOI: 10.3969/j.issn.1002-087X.2024.12.017