Prediction of Remaining Useful Life of Second-Life Lithium-ion Batteries Considering Cell-to-Cell Variation

The electrification of the transportation industry requires increasing attention to battery life-cycle management. In particular, Lithium-ion batteries are often retired from Electric Vehicles (EVs) when they still have significant residual capacity. Therefore, they can be repurposed in less-demanding applications, where they are referred to as Second-Life Batteries (SLBs). A key challenge to the reliable use of SLBs is the accurate prediction of their Remaining Useful Life (RUL). This thesis proposes a Gaussian Process Regression (GPR) model for RUL prediction that explicitly accounts for Cell-to-Cell variation while requiring limited prior knowledge. The model is validated on both experimental and synthetic data. Due to the strong influence of input data on data-driven models, a comprehensive sensitivity analysis was conducted to evaluate the impact of different input features and data quantity on prediction accuracy. Among the various features considered, only normalized capacity and test duration consistently provided robust and reliable performance across diverse conditions. A key innovation is an attempt to develop a physics-informed data-driven model to predict the RUL of SLBs that learns from purely synthetic data and is validated over experimental data. The model was then adapted to module configuration using features extracted from half of the discharge test curve. This modified approach better suits the real-world application, where EV batteries are rarely dismantled to the cell level, while maintaining high accuracy.