Unlocking Ultrafast Diagnosis of Retired Batteries via Interpretable Machine Learning and Optical Fiber Sensors

Abstract

Retired batteries are of great economic and environmental importance, which are indispensable considerations in the life cycle of lithium-ion batteries. However, existing methods for evaluating retired batteries are time- and resource-consuming, hindering efficient screening for later recycling or reuse. Herein, combining optical fiber sensors and interpretable machine learning (ML), we establish a data-driven framework for retired battery datasets with 265 cells of different chemistries (LiFePO₄/graphite, LiMn₂O₄/graphite) and achieve ultrafast state of health diagnosis within 3 min, offering mean absolute errors of 1.17% and 2.78%, respectively. The proposed data-driven framework identifies the salient regions in the time-resolved multivariable data and helps to uncover underlying thermodynamic/kinetic aging mechanisms. We also demonstrate the incorporated thermal information obtained via optical fibers complements voltage signals by improving prediction accuracy and antinoise ability. This work not only showcases the potential of battery sensing in retired battery diagnosis but also unlocks the unexplored synergy between sensing and interpretable ML for diverse battery applications.