Prediction of fast-charging capabilities in LiFePO₄/graphite lithium-ion batteries using internal resistance and machine learning

Abstract

The fast-charging capability of lithium-ion batteries is a crucial performance indicator, yet its accurate evaluation remains a significant challenge. The maximum chargeable state of charge (SOC) at a specific fast-charging rate is closely correlated with fast-charging capability; however, it is difficult to measure via non-destructive testing. In this study, we investigate the potential correlation between a highly measurable battery parameter (i.e., internal resistance) and this hard-to-measure parameter (i.e., maximum chargeable SOC) using machine learning. Given the difficulty in generating large datasets containing internal resistance and maximum chargeable SOC through experimental methods, a virtual dataset is created via the P2D electrochemical model using PyBaMM—an open-source Python-based P2D simulation tool. Using PyBaMM, we simulate 100 virtual LiFePO₄/Graphite lithium-ion batteries with distinct electrochemical kinetics, from which internal resistance data (at various SOC levels and times) and maximum chargeable SOC (at a 3 C charging rate) were extracted to form the dataset. Using this dataset, we employ machine learning to predict maximum chargeable SOC from internal resistance. Results show that the average prediction error between the machine learning-predicted maximum chargeable SOC and the dataset's P2D-simulated benchmarks is below 0.05, with a maximum error of 0.11. Furthermore, compared with experimental measurements, the machine learning-predicted maximum chargeable SOC is highly consistent with values measured via three-electrode experiments. This study demonstrates that a well-designed linear model, with carefully engineered features derived from domain knowledge, can achieve highly satisfactory performance while offering significant advantages in interpretability, reliability, and computational efficiency—attributes critical for real-world industrial applications in battery design. It thereby establishes the potential of using the highly measurable internal resistance of batteries to predict the hard-to-measure fast-charging capabilities in LiFePO₄/Graphite lithium-ion batteries, enabling faster and more efficient evaluation in the design process of fast-charging batteries.