State Estimation for Automotive Batteries through Extended Kalman Filter

The performance of electric and hybrid vehicles relies on an effective battery management strategy to ensure safety, driving range, and durability. Among all monitored parameters, the state of charge (SOC) plays an essential role in optimally implementing the control logic strategies. However, since SOC cannot be directly measured, its estimation relies on algorithms capable of guaranteeing a balance between accuracy and computational efficiency. This thesis proposes the design and implementation of an extended Kalman filter (EKF) for SOC estimation of a lithium-ion cell - LG INR18650 MJ1 - within Matlab and Simulink environments. The cell was first modelled adopting a Dual Polarisation Model (DPM) and characterised through experimental data, with model parameters optimised via Simulink Design Optimisation (SDO) toolbox. Subsequently, the EKF was developed, tuned, and validated under different operating conditions, demonstrating convergence, stability, and robustness against incorrect SOC initialisation and sensor offsets. Specifically, for both tests analysed under correct initialisation conditions, the filter achieved an estimation SOC error profile constrained within ±1% window, while in the incorrect initialisation scenario, even for the most dynamic and shortest test, the mean SOC error remained limited to −2.38%. The study was then further extended to a configuration composed of six identical cells, denoted as multiple cells EKF Simulink model, enabling simul- taneous SOC estimation of a battery module. The model was validated across different scenarios, proving its accuracy and convergence capability. Finally, the model was converted into C code and validated through software in the loop (SIL) approach, confirming its suitability for real-time implementation on production-level embedded hardware.