Data-driven classification of lithium-ion batteries for second-life applications

As degraded lithium-ion batteries proliferate from ageing electric vehicles, we must develop methods of forecasting battery lifetime to increase profitability and safety in second-life applications. However, electric vehicle batteries are subjected to variable and generally unknown operating conditions that yield different degradation mechanisms, affecting their future health trajectory. We propose a data-driven method of classifying retired Lithium-ion batteries to determine whether they should be reused or recycled. This method only takes a few minutes of testing requiring one electrochemical impedance spectroscopy measurement. The model was tested across five different use cases where the classification boundary was adjusted accordingly, resulting in an average accuracy of 92%. The model was also trained and tested against another independent dataset, achieving 90% accuracy. This method shows promise as a tool for lithium-ion battery repurposing companies to identify batteries that will likely exhibit rapid capacity degradation if repurposed to avoid expending resources on full battery re-certification. • Proposed a framework for rapid second-life battery classification using EIS-informed machine learning. • Accuracy of Impedance-based sorting is compared to capacity-based sorting. • The benefits of using data augmentation in a data-limited field is tested with variational autoencoders.