Impedance-Based State Estimation for Battery Systems

Transportation electrification has increased the demand for reliable lithium-ion batteries. Battery degradation remains unavoidable, making accurate diagnosis and prognosis essential for safe and efficient operation. Electrochemical impedance spectroscopy (EIS), which reflects internal electrochemical processes, has emerged as a promising technique for advanced battery management.

This thesis investigates impedance-based battery modeling, state-of-charge (SOC) estimation, and state-of-health (SOH) estimation at both cell and module levels using NMC, LFP, and LTO batteries. Both equivalent circuit models and data-driven methods are employed.

First, the accuracy and repeatability of EIS measurements are evaluated, and the minimum detectable aging cycle (MDAC) is proposed to quantify aging sensitivity. The results verify the reliability of the EIS measurement system.

A hybrid-domain parameter identification method for fractional-order equivalent circuit models is then developed by combining frequency- and time-domain information. Based on this framework, an impedance-based SOC estimation method is proposed using an aging-insensitive impedance indicator and cubature Kalman filter initialization.

For SOH estimation, both manually engineered and automatically learned impedance features are investigated using machine learning and deep learning techniques. A transformer-based model and a KNN-based impedance pattern recognition method are further proposed to improve robustness under uncertain SOC conditions.

Finally, the proposed methods are extended to battery modules by considering busbar impedance. Worst-cell- and module-impedance-based SOH estimation methods are developed and experimentally validated, demonstrating reliable module-level SOH estimation from a single EIS measurement.