Prediction and control of copper plating in retired LFP battery discharge pretreatment via an explainable, lightweight, and robust electrochemical phase transition prediction and control architecture

Efficient recycling of retired lithium-ion batteries is bottlenecked by the safety risks posed by “Unknown Aging States” and “Manufacturing Heterogeneity” during discharge pretreatment. To address this, we propose a Physics-Supervised Adaptive Control Framework (A-X-LTP-GRU). Grounded in in-situ three-electrode analysis, we identified an empirically consistent phase-transition ordering pattern for control under the tested conditions, despite substantial parameter drift across cells. On this basis, the framework addresses copper plating indirectly by forecasting the late-stage dV/dQ trajectory and triggering discharge cut-off before the trajectory enters the hazardous region associated with cathodic copper plating. This is implemented through a lightweight trend predictor supervised by an Accurate Global Statistical Limit (A) and a Local Adaptive Baseline (X). Trained on 66 heterogeneous commercial LFP cells, the proposed architecture achieved an observed 100% success rate under the tested protocol (one-sided 95% Clopper–Pearson lower confidence bound of 90.8%) when validated on an independent set of 31 strictly isolated retired LFP batteries. It also remained robust under extreme industrial noise and packet loss and outperformed the traditional Savitzky-Golay filter baseline. Preliminary validation on an NCM battery suggested component-level transferability. Overall, this work provides a practical, lightweight, and explainable approach for integrating physical constraints into data-driven battery management for automated recycling.