This study aims to develop and validate a robust data-driven framework for the simultaneous estimation of State of Health (SoH) and State of Charge (SoC) in lithium-ion batteries for electric vertical take-off and landing (eVTOL) aircraft. It specifically addresses the lack of comprehensive dual-state estimation and external experimental validation in existing aerial vehicle literature.
The study benchmarks ten regression algorithms using a high-fidelity open-access eVTOL mission data set. To ensure robustness and overcome overfitting risks, a novel “dual-level” validation strategy is used: predictions are cross-validated against a MATLAB/Simulink equivalent circuit model and further verified experimentally using data collected from six independently sourced Murata VTC-6 cells under varying aging cycles.
Extreme Gradient Boosting emerged as the superior model, offering the best trade-off between accuracy and computational efficiency. The model achieved a mean squared error of 0.2199 and root mean square error of 0.4688 for SoH, keeping capacity prediction errors below 2%. Furthermore, SoC estimation errors remained below approximately 1% across different aging cycles, demonstrating robust generalization across diverse operating conditions.
Unlike previous studies that rely solely on internal data splitting for validation, this research introduces an independent experimental verification layer using separate battery cells. This approach provides a transferable pathway for safe and reliable battery management system integration in electric aviation, bridging the gap between theoretical machine learning performance and real-world applicability.
