Tracking lithium-ion battery degradation with robust voltage–capacity signatures

This is the authors' abstract. We don't add key points for this paper.

The diagnostic and prognostic battery models are increasingly demanded by industry to extend battery lifetime, enhance control strategies, support second-life design, and meet emerging regulatory requirements such as battery passports. These models must be flexible in coping with practical challenges of real-world data, including frequent incomplete charge-discharge cycles, cycling at moderate-to-high C-rates, and overall data sparsity throughout the battery's lifetime. Here, we propose a feature-extraction method capable of deriving informative features from charge and discharge profiles without relying on specific physics-relevant signatures such as peak coordinates in derivative voltage or capacity curves. We develop and validate our methodology using a representative experimental aging dataset from a commercial 1.85 Ah lithium-ion cell subjected to partial charge and discharge. We identify a set of features encoding the relative change in charge exchange within a fixed voltage window as the most efficient indicator for the prediction of state-of-health, remaining-useful-life, and knee-point. Moreover, we perform a data sparsity analysis to demonstrate the high resilience of the framework against incomplete training sets from field operation. Our results demonstrate that the long-term battery degradation trends can be reliably tracked and predicted within the studied dataset using simple voltage–capacity descriptors and early-life baseline aging information.