Deciphering failure paths in lithium metal anodes by electrochemical curve fingerprints.

Understanding anode failure mechanisms in lithium metal batteries (LMBs) is crucial for their use in energy storage, as the anode directly affects battery stability and electrolyte selection. Unfortunately, post-mortem methods reveal failure outcomes but often miss dynamic progressions, obscuring cause-and-effect relationships in failure evolution. Leveraging domain knowledge informed machine learning and a 4-year dataset of over 18 000 cycles and 12 million data points, from cells cycled to failure, we uncovered a correlation between initial lithium plating/stripping behavior and subsequent anode changes, enabling the identification of early indicators for distinct failure types. Our model accurately predicts failure types using only the first two cycles, less than 2% of the data, demonstrating the effectiveness of initial curve features as electrochemical fingerprints. Key electrochemical fingerprints describing lithium microstructure and its interphase with the electrolyte are validated to be critical to kinetics and reversibility degradation. Specifically, the fingerprints influence the formation of ineffective interphase regions (lacking intimate contact with the lithium metal) and inactive lithium, which in turn lengthen charge carrier (lithium-ion and electron) transport paths, leading to poorer kinetics and reversibility. The fingerprints and model generalize well across typical published electrolyte systems with low misidentification, demonstrating versatility and practicality. Broadly, this study using a pre-mortem prediction method deepens understanding of lithium metal anode failure mechanisms by uncovering the root causes of kinetics and reversibility degradation from fingerprints hidden in initial cycles instead of a post-mortem manner, facilitating the rapid assessment of battery reliability and development of electrolytes.