Rapid Assessing Cycle Life and Degradation Trajectory Based on Transfer Learning for Lithium-Ion Battery

Abstract

Before the certain model lithium-ion battery is mass-produced, continuous testing is typically needed to assess the degradation end points and trajectory of cycle life, determining if it satisfies the reliability requirements for the target applications. However, the traditional testing method, which typically takes an amount of time such as one year or more, significantly hinders the development of the battery industry. How to rapidly obtain the battery life is The significant challenge lies in finding how to obtain the battery life. Hence, a rapid cycle life assessment framework with transfer learning (TL) is proposed, which substitutes prediction for continuous test to obtain the end points and corresponding degradation trajectories. In this framework, the features extracted from charge-discharge process and other information stream are taken as inputs, and the different capacity degradation percentage points (CDPPs) are used as outputs. The similarities and joint-properties can be found in different battery features by TL, to avoid time-consuming operations such as fine-tuning. The effectiveness is demonstrated by hundreds of thousands of samples from four different manufacturers, which are able to accurately get the life end points and degradation trajectory. Experimental results present that the life assessment time reduced by at least 83%. The end-of-life (EOL) point prediction error is less than 11.2%, and the similarity between the predicted degradation trajectory and the real value is more than 90%. More importantly, the proposed method is expected to enable rapid life assessment across various working conditions and different types.