State of Health Estimation using Machine Learning for Li-ion battery on Electric Vehicles

Prognostics in State of Health (SoH) of lithium-ion batteries (LIBs) holds high importance in ensuring the reliability and safe electrification of battery systems such as in mobile electronic devices, electric vehicles (EVs), and grid battery energy storage systems (BESS). However, due to reasons such as the SoH dependency on a chain of historical data, both internal and external to the battery systems, the SoH estimation is not directly measured. Indirectly, several electrochemical in-situ measurement methods are employed as non-destructive determination of SoH such as ICA, DVA, and EIS, yet forming more post-mortem type of analysis. Alternatively, the machine learning (ML) techniques in data mining for predictions shown high caliber in unambiguously identifying the multi-scale, multi-factorial and complex degradation patterns of LIBs. This work investigates a couple of ML techniques, FNN and LSTM models, to train on a dataset on LIB degradation profiles and improve them for SoH predictions on new battery aging cycles. Here, a publicly available dataset at NASA website has used to train the models. Further, it is expected to explore the performance and the reliability of the LSTM for SoH predictions and plans to extend the solution for other types of batteries.