Decoupling Reversible Dynamics from Irreversible Aging: A Physics-Aware Framework for Robust SOH Diagnosis under Diverse Conditions

Existing data-driven State of Health (SOH) diagnosis models often falter under realistic conditions due to the coupling of reversible fluctuations and irreversible degradation.To address this, we propose a physics-aware feature engineering framework integrated with a Random Forest regressor that decouples complex operating dynamics from intrinsic aging trends.Extensive validation across four datasets (CALCE, NASA, HUST, and TRI) demonstrates the framework's superior generalization capability.Specifically, in dynamic loading scenarios, the model leverages current variance to filter load-induced noise, while in fast-charging LFP regimes, voltage kurtosis captures structural instability via a distinct "Reversed-L" pattern, enabling robust pre-failure detection despite flat voltage plateaus.By achieving high estimation accuracy with a lightweight and interpretable architecture, this framework bridges the gap between theoretical data science and practical, real-time BMS implementation.