A quantitative comparison of different Machine Learning and Deep Learning models for SOH and RUL Estimation of Lithium-Ion Batteries

The accurate estimation of State of Health (SOH) and Remaihetening Useful Life (RUL) of electrochemical energy storage devices (EESDs), such as batteries and hybrid supercapacitors, is a crucial task for predictive maintenance. Since traditional health assessments based on direct capacity measurements are not always feasible in online monitoring systems, indirect feature-based approaches have gained the attention of current research. However, existing methods are usually not able to take into account variable operating conditions and they lack validations in highly heterogeneous datasets, which is essential to prove their generalizability and reliability in real case scenarios. To address this issue, this paper proposes a robust method that extracts 28 thermal, electrical and time features from charge, discharge and rest phases during the cycle life of the cell. The extracted physical features are utilized to implement multiple machine learning (ML) and deep learning (DL) models, whose performances are evaluated via repeated k-fold cross-validation on a highly variable dataset that includes diverse EESD technologies and operating conditions, such as non-standard charging methods and distinct discharge rates. The obtained results demonstrate that the proposed method provides promising SOH and RUL predictions across different EESD cycling profiles, offering an innovative approach for robust diagnostics and prognostics.