Generalizable Mechanism‐Anchored Health Management for Lithium‐Ion Batteries Across Complex Degradation Trajectories

ABSTRACT Complex degradation trajectories compromise the lifetime reliability and operational safety of lithium‐ion batteries in advanced energy‐storage systems. Aging pathways induced by temperature, rate, and voltage‐abuse stresses are often entangled in conventional health‐estimation models, limiting transferable degradation representation and future‐trajectory updating under distribution shifts. Here, a generalizable, mechanism‐anchored health‐management framework is developed for lithium‐ion batteries under heterogeneous aging conditions. Using aging‐condition identification as a proxy task, we identify a degradation‐sensitive 3.3–3.4 V charging window in LiFePO 4 /graphite cells, where statistically selected features are linked to incremental‐capacity fingerprints. These electrochemically anchored features are integrated with a condition‐aware soft mixture‐of‐experts model to disentangle heterogeneous feature–state‐of‐health (SOH) relationships with and without condition labels during inference. Cross‐dataset continual learning enables adaptation to an unseen data source while mitigating catastrophic forgetting. Finally, data‐driven SOH observations are assimilated into a physics‐inspired parametric degradation model through an adaptive unscented Kalman filter, converting pointwise health estimates into online‐updatable degradation trajectories. The framework achieves robust SOH estimation across multi‐stress aging conditions and stable 100–300‐step look‐ahead trajectory prediction, establishing an interpretable route from electrochemical fingerprints to trajectory‐level battery prognostics.