Abstract
Unsorted retired batteries with varied cathode materials hinder the adoption of direct recycling due to their cathode-specific nature. The surge in retired batteries necessitates precise sorting for effective direct recycling, but challenges arise from varying operational histories, diverse manufacturers, and data privacy concerns of recycling collaborators (data owners). Here we show, from a unique dataset of 130 lithium-ion batteries spanning 5 cathode materials and 7 manufacturers, a federated machine learning approach can classify these retired batteries without relying on past operational data, safeguarding the data privacy of recycling collaborators. By utilizing the features extracted from the end-of-life charge-discharge cycle, our model exhibits 1% and 3% cathode sorting errors under homogeneous and heterogeneous battery recycling settings respectively, attributed to our innovative Wasserstein-distance voting strategy. Economically, the proposed method underscores the value of precise battery sorting for a prosperous and sustainable recycling industry. This study heralds a new paradigm of using privacy-sensitive data from diverse sources, facilitating collaborative and privacy-respecting decision-making for distributed systems.
Introduction
As the first generation of electric vehicle batteries reaches retirement, the industry faces a massive recycling challenge. Direct recycling—which restores the battery material's structure rather than melting it down or dissolving it in acid—is the most environmentally friendly and profitable route. However, direct recycling is "cathode-specific"; mixing different chemistries (like LFP and NMC) in the same batch ruins product quality. Currently, the industry lacks an efficient way to sort these batteries because labels are often unreliable and manufacturers are hesitant to share sensitive operational data. This "data islanding" creates a bottleneck where recyclers lack the data to build accurate sorting models.
The Federated Learning Framework
To overcome these privacy barriers, this research proposes a Federated Machine Learning (FL) framework. In this architecture, data owners (manufacturers, fleet operators, etc.) train local models on their own servers. Instead of sending raw data to a central recycler, they only send model parameters, preserving proprietary secrets while contributing to a robust global model.
A core technical innovation in this framework is the Wasserstein-Distance Voting (WDV) strategy. Standard federated learning often struggles with "heterogeneous" data, where different clients have different types of batteries. WDV uses a mathematical distance metric to weight contributions from different clients, making the global model significantly more robust to biased or noisy local datasets compared to traditional majority voting.
Feature Engineering and the "One-Cycle" Approach
A significant breakthrough of this work is the ability to identify a battery's cathode chemistry using data from just one end-of-life charge-discharge cycle. This eliminates the need for a battery's entire operational history, which is often unavailable or proprietary. The model utilizes 30 features extracted from voltage-capacity and dQ/dV curves.
These features are physically grounded; for instance, the model identified specific voltage peaks (features F1 and F16) as the most important. These peaks represent phase transitions unique to specific cathode structures, providing a physical justification for the machine learning results based on the Gibbs phase law. The study utilized a diverse dataset of 130 batteries covering five cathode materials (LCO, NMC, LFP, NCA, and NMC-LCO blends) from seven different manufacturers.
Results and Economic Impact
The model achieved exceptional accuracy, with only a 1% sorting error in homogeneous settings and a 3% error in heterogeneous settings. In contrast, traditional voting methods saw errors jump to ~45% when faced with heterogeneous data.
Beyond technical accuracy, the study highlights the economic viability of this approach. A techno-economic analysis showed that direct recycling of NMC batteries using this precise sorting method yields a profit of approximately 29,944 ¥/t, which is 2.25 times higher than the next best recycling alternative (hydrometallurgy). By using FL, the authors turn a competitive disadvantage (proprietary data) into a collaborative strength for the circular economy.
Conclusion
This federated learning approach solves the technical and privacy barriers to large-scale battery sorting. While future work is needed to address State of Health (SOH) estimation and economic incentive structures for collaborators, this framework provides a scalable, privacy-preserving path toward a sustainable and profitable battery recycling industry.
Summary updated