Accelerated commercial battery electrode-level degradation diagnosis via only 11-point charging segments

Accelerated and accurate degradation diagnosis is imperative for the management and reutilization of commercial lithium-ion batteries in the upcoming TWh era. This work proposes a framework combining both deep learning and physical modeling to extend traditional capacity degradation diagnosis to a rapid and accurate degradation diagnosis at the electrode level using only readily measurable charging current and voltage signals. Deep learning is used to rapidly and robustly predict polarization-free incremental capacity analysis (ICA) curves in minutes, which are traditionally obtained in a dozen hours, and the physical model is to quantitatively reveal the electrode-level degradation modes by decoupling them from the ICA curves. It is demonstrated that 11 points collected at any starting state-of-charge (SOC) in a minimum of 2.5 minutes are sufficient to predict reliable ICA curves with a mean root mean square error (RMSE) of 0.2774 Ah/V. Accordingly, batteries can be accurately elevated based on their degradation at both macro and electrode levels. Through transfer learning, such a method can also be adapted to different battery chemistries, emphasizing the enticing potential for rapid promotion of this work in battery degradation diagnosis. • A framework combining both deep learning and physical modeling to achieve a rapid and accurate degradation diagnosis at the electrode level. • A deep learning architecture is proposed to predict polarization-free ICA curves and can use 11-point charging segments starting from any SOC as input. • The proposed method shows good robustness under various influencing factors that could seriously affect the ICA curve. • Through transfer learning, the proposed method can be adapted to batteries with different chemistries and still show good reliability.