A machine learning model predicts the cycle life of lithium-metal batteries using features extracted from first-cycle charge–discharge data and impedance spectroscopy. Trained on 43 cells with varied chemistries and formation protocols, the model achieved up to 9.6% error, demonstrating a fast, cost-effective approach to early battery life prediction.

Battery cell manufacturing comprises numerous steps requiring co-optimization, making the development process time consuming and expensive. Lithium-metal batteries with ionic liquid electrolytes are a promising next-generation technology for applications demanding high specific energy and safety but currently suffer from limited cycle stability. Optimizing the manufacturing process can improve performance, and early cycle life prediction can accelerate this process, reducing cost and time. However, correlating early-stage behavior with long-term stability is challenging. Machine learning (ML) can assist in building these correlations, but feature extraction remains a key hurdle. A set of features manually extracted from the first cycle with high correlation with the battery cycle life is presented here. These features are then used as inputs to a ML model based on linear regression. While the dataset contains batteries that have reached end-of-life through two different mechanisms, the model can predict the cycle life with an error of 15.3%. The error decreases to 9.6% when the cells are first sorted by end-of-life mechanism. This work highlights the importance of the early charge–discharge behavior of lithium-metal batteries and how this data can be used to inform on the battery cycle life with a view to greatly reducing experimental workload.