ML framework for RUL prediction of LFP cells from early ageing characteristics

In the recent past, lithium-iron-phosphate (LFP) has become the successful lithium-ion battery variant in the mobility business due to its high specific power, good thermal stability, excellent cycle life, safe operation, and most importantly its economic advantage over other chemistry. The inability to accurately estimate the lifespan of the battery is because of the lack of comprehensive understanding of the factors affecting the LFP cell’s non linear and unpredictable degradation. This research work focuses on methodology development to estimate life cycle from early ageing characteristics of LFP cells which are subjected to fast charging under high C-rate conditions. The methodology integrates best practices for feature extraction from raw data, feature engineering, feature importance analysis and modeling techniques to accurately predict the life cycle of LFP cells. Machine learning models with ability to learn the non-linear characteristic like support vector machines, gaussian processes, ensemble models and neural networks are used, and are evaluated and compared based on RMSE and R-squared metrics. LSBoost ensemble model outperformed other models in predicting the cycle life, yielding a root mean square error of 90 cycles and an R-squared value of 0.96. The developed methodology can be applied to identify early anomalous behavior, optimization of fast-charging protocol, BMS optimization for maximizing cell life cycle and accelerating the cell development.