The effectiveness of BMS is directly influenced by the quality of input data from operating lithium-ion cells and the real-time computational capabilities of state estimation algorithms. Machine learning (ML) models are the best choice nowadays.
This study focuses on two key stages in preparing experimental data for machine learning models: data acquisition and preprocessing.
In the present research, conventional battery measurements, typically limited to current, voltage, and surface cell temperature, were extended to include multipoint data acquisition of internal non-electrical parameters. Modified lithium-ion cells with embedded fiber Bragg grating (FBG) sensors were produced, and their internal temperature and “breath effect” were observed during the cells’ cycling. A detailed study of preprocessing strategies was conducted using the extended experimental dataset, resulting in data suitable for training ML models.
Based on the processed dataset, four ML algorithms were trained and evaluated: ANN, HGB, RF, and SVR.

This study was supported by Israel Science Foundation (ISF) grant 2979/23.