Accelerated battery state-of-health estimation using conditional autoencoders for intelligent frequency band selection

Electrochemical Impedance Spectroscopy (EIS) is a widely adopted non-intrusive diagnostic technique in battery management systems (BMS) for evaluating battery state-of-health (SoH) and tracking degradation processes. However, conventional full-spectrum EIS measurements, spanning millihertz to kilohertz, are time-consuming and resource-intensive, limiting their practicality in real-time BMS applications. This work extends diagnostic research to Sodium-ion batteries (SIBs), an emerging technology with limited characterization. Despite their promising advantages in cost, safety, and resource availability, SIBs remain significantly less studied than lithium-ion chemistries. A Conditional Autoencoder, explicitly conditioned on operating mode, temperature, and cycle index, extracts latent features from SIBs’ EIS data, while a Random Forest regressor leverages these features for robust SoH prediction. This optimized framework achieves substantial reductions in required frequency points without sacrificing accuracy. Experimental validation on sodium-ion cells across both charge and discharge modes at and demonstrates clear advantages over full-spectrum baselines in capturing degradation trends with minimal SoH precision reduction. Beyond immediate performance gains, this approach supports the design of next-generation BMS firmware capable of adaptively probing only the most informative frequencies, achieving a 95% reduction in measurement points while preserving near full-spectrum accuracy ( ), enabling intelligent charging, dynamic balancing, and real-time compensation in energy storage systems.