Abstract
Accurate state of health estimation is crucial for the reliable operation of lithium-ion batteries in electric vehicles. The charging curve contains valuable features for health evaluation, but real-world charging often lacks sufficient data due to the users' early recharging habits. A smart method is proposed for accurate battery health estimation using super-short charging segments. This method combines a degradation mechanism-guided Scale-Invariant Feature Transform for smart health feature identification with machine learning for health evaluation. Validation with 87 batteries with various chemistries, formats, and capacities from 6 manufacturers demonstrates its efficacy. Regardless of battery specifications, health features can be identified automatically from the charging data. The method promises high accuracy (estimation error as low as 1.97%) even with super-short charging covering 10% state of charge span, where all the existing health feature extraction approaches fail. This method provides new avenues for battery health evaluation in uncertain real-world electric vehicle applications.