Data-driven insights into the reaction mechanism of Li-rich cathodes

Abstract

Lithium-rich layered oxides (LRLOs) hold great promise as cathode materials for lithium-ion batteries, but they face challenges due to their complex electrochemical behavior and structural instability. Here, we propose an unsupervised analysis framework that applies principal component analysis (PCA) to a large dataset of over 30 000 LRLO charge curves to identify fundamental degradation factors and enhance predictability. By incorporating ex situ Mn L-edge and O K-edge soft X-ray absorption spectroscopy (sXAS), along with electrochemical impedance spectroscopy (EIS), we connect each principal component to physical phenomena such as Mn reduction and rising charge-transfer resistance. Leveraging these insights, we demonstrate robust predictive models that can accurately reconstruct full charge curves and reliably detect outliers or abnormal cycling patterns. By bridging mechanistic domain knowledge with unsupervised learning, this framework underscores the value of combining data-driven methodologies with mechanistic insights, paving the way for more reliable and high-performance materials in next-generation battery systems.