Facilitating battery quality classification: Early life prediction with sequence-sampling data augmentation

Abstract

With the rapid development of electric transportation systems, early-stage quality classification of lithium-ion batteries (LIBs) is crucial for improving the overall performance of battery systems throughout their life-cycle. However, the complex degradation mechanisms of LIBs lead to significant differences in the aging rates of individual cells under identical conditions, which directly affects the accuracy of early-stage quality classification. To address this challenge, this paper proposes a novel framework for predicting the full life-cycle end of life (EOL) of LIBs, combining a sequence sampling-based virtual battery construction scheme with semi-supervised learning. The framework achieves high-precision EOL prediction by augmenting early-cycle data and leveraging the automated feature extraction capabilities of a masked autoencoder (MAE), using only minimal labeled data. Experimental validation demonstrates that the mean absolute percentage error (MAPE) on the validation set can be reduced to 2.6%. This research not only provides a new approach for early-stage battery quality classification utilizing minimal labeled data but also offers robust support for enhancing pack efficiency and enabling pre-screening of abnormal cells, through efficient data utilization and precise predictive capabilities.