Enhanced online detection of lithium plating via a dynamic threshold generation strategy with aging effect compensation

Abstract

Lithium plating (LP) is an abnormal side reaction during lithium-battery cycling that accelerates capacity loss and introduces safety risks. Existing online detection methods often suffer from LP-feature masking caused by gradual aging, baseline drift, and cell-to-cell inconsistency, which can delay warning and reduce adaptability. This study proposes a physically informed and experimentally validated cell-level online LP detection framework under controlled continuous charging conditions. The framework estimates a time-domain equivalent ohmic resistance using a second-order equivalent circuit model and an aging-aware OCV-SOC compensation strategy based on an iterative multi-objective Gaussian process. With the OCV-SOC tracking error controlled below 3.0% over the investigated aging stages, the compensated resistance sequence shows improved linear consistency with the EIS-based reference resistance. Based on this stabilized resistance sequence, sequential SOC-relative resistance (SRR) features are constructed to enhance localized high-SOC LP-associated mutations against macroscopic baseline offsets caused by condition variation, cell inconsistency, and residual polarization coupling. A semi-supervised dynamic K-means threshold strategy is further introduced to update battery-specific LP warning boundaries from historical and current high-SOC SRR features. Under the tested cell-level conditions, the proposed method advances the warning cycle by more than 15.0% relative to the selected baselines, while post-mortem SEM images show surface morphology consistent with the corresponding LP-associated warning states. The framework uses ordinary voltage-current cycling data and requires no specialized rest or pulse protocols under the tested continuous charging conditions. It therefore provides an experimentally validated cell-level algorithmic basis for future BMS-oriented validation.