Accurate battery lifetime prediction is important for maintenance, warranties, and cell design. However, manufacturing variability and usage-dependent degradation make life prediction challenging. Here, we investigate new features derived from capacity-voltage data in early life to predict the lifetime of cells cycled under varying charge rates, discharge rates, and depths of discharge. The early-life features capture a cell’s state of health and the change rate of component-level degradation modes. Using a newly generated dataset from 225 nickel-manganese-cobalt/graphite Li-ion cells aged under a wide range of conditions, we demonstrate a lifetime prediction of in-distribution cells with 15.1% mean absolute percentage error (MAPE). A hierarchical Bayesian model shows improved performance on extrapolation, achieving 21.8% MAPE for out-of-distribution cells. Our approach highlights the importance of using domain knowledge of battery degradation to inform feature engineering and model construction. Further, a new publicly available battery lifelong aging dataset is provided.
The major contributions of this work are four-fold:
1. Proposing a new idea to categorize early-life features into two hierarchical levels, the condition (upper) and the cell (lower) level, in order to capture an inherent hierarchical structure in the battery aging data and enable greater generalization especially to out-of-distribution data,
2. Creating a hierarchical Bayesian model (HBM) to address the hierarchical nature of the aging data and quantify the uncertainty in lifetime predictions,
3. Demonstrating a new method for extracting predictive features from incremental capacity curves, incorporating optimization of voltage window to improve correlations with lifetime, and,
4. Generating and publicly sharing a large battery aging dataset consisting of 225 NMC cells cycled under a wide range of operating conditions, enabling researchers without access to battery testing equipment to study lifetime modeling and other related topics.