Interpretable\nLearning of Accelerated Aging in Lithium\nMetal Batteries

Lithium\nmetal batteries (LMBs) with high energy density are perceived\nas the most promising candidates to enable long-endurance electrified\ntransportation. However, rapid capacity decay and safety hazards have\nimpeded the practical application of LMBs, where the entangled complex\ndegradation pattern remains a major challenge for efficient battery\ndesign and engineering. Here, we present an interpretable framework\nto learn the accelerated aging of LMBs with a comprehensive data space\ncontaining 79 cells varying considerably in battery chemistries and\ncell parameters. Leveraging only data from the first 10 cycles, this\nframework accurately predicts the knee points where aging starts to\naccelerate. Leaning on the framework’s interpretability, we\nfurther elucidate the critical role of the last 10%-depth discharging\non LMB aging rate and propose a universal descriptor based solely\non early cycle electrochemical data for rapid evaluation of electrolytes.\nThe machine learning insights also motivate the design of a dual-cutoff\ndischarge protocol, which effectively extends the cycle life of LMBs\nby a factor of up to 2.8.