Extracting health indicators from a single cycle of data based on the aging mechanism to accurately predict the knee point and remaining useful life of spent lithium-ion battery

Abstract

Spent lithium-ion batteries will degrade quickly and unpredictably to failure after the knee point. Therefore, predicting the knee point is a key prerequisite to forecast the remaining useful life (RUL) for evaluating the reuse value of spent lithium-ion batteries (LIBs). Given the lack of historical data and the need for minimal testing time in the large-scale application scenarios of huge spent battery value assessments, this work proposes a feature engineering method based on battery aging mechanisms to identify health indicators (HIs). Furthermore, these HIs are used for the knee point and RUL prediction. The proposed method does not require historical data and can accurately predict the knee point and RUL using data from any one of cycle within the entire battery life cycles. Testing on two typical datasets with samples of Li(NiCoMn)O₂ and LiFePO₄ LIBs, the mean absolute error, MAE (mean absolute percentage error, MAPE) for predicting the knee point is only 47.15 cycles (10.60 %) and 13.63 cycles (6.38 %) in the two different datasets, respectively. The MAE (MAPE) reaches as low as 40.70 cycles (8.11 %) and 11.81 cycles (5.04 %) for the RUL prediction, respectively. The proposed method demonstrates excellent predictive accuracy and time-saving performance, and then a significant practical value.