Seeing the unseen: Real-time tracking of battery cycling-to-failure via surface strain

Passive failures in lithium-ion batteries, driven by cumulative degradation, present serious safety risks due to their spontaneity and asymptomatic progression. Conventional electrochemical diagnostics cannot differentiate passive failures from routine degradation, while temperature responds too late for early detection. Here, we introduce surface strain as a non-invasive indicator to track degradation-to-failure transitions in commercial 21700 batteries. Under accelerated degradation by depth-of-discharge stress, strain signals detect failure onset > 50% earlier than temperature. We develop two strain-derived metrics, the slope-based threshold and failure-proximity index and achieve 99.7% F1-score (100% true positive rate) and 3.82% normalized mean absolute error for failure detection and proximity estimation using machine learning on partial charging data with strain. Strain monitoring remains effective under fast-charge cycling. Using attachable sensors and lightweight computations, our approach allows easy integration with current battery systems, enabling real-time, onboard passive failure monitoring to enhance battery safety, reliability, and lifespan for future energy storage.