ENHANCED STATE OF CHARGE ESTIMATION IN eVTOL BATTERY SYSTEMS USING FEDERATED LEARNING

Abstract

As electric vertical takeoff and landing (eVTOL) aircraft gain recognition as a next-generation mobility solution, the need for reliable, high-performance onboard battery systems has become progressively critical. At the center of safe and efficient operation lies the accurate estimation of battery state of charge (SOC), which governs energy availability, flight endurance, and operational planning. This thesis investigates SOC estimation in eVTOL systems, with a focus on enhancing accuracy, scalability, and data privacy. The study begins by exploring the architecture and energy requirements of eVTOL propulsion and power distribution systems, followed by an in-depth review of battery management technologies. Emphasis is placed on the evolving role of battery management systems (BMS) in electric aviation, including their functional safety requirements and the challenges of integrating SOC estimation under high energy and mission constraints. Existing SOC estimation methods, including traditional algorithms and machine learning-based approaches, are critically reviewed, revealing limitations in transitional centralized training and data sharing practices particularly within modular and privacy-sensitive platforms such as eVTOLs. To address these, a federated learning (FL) framework is proposed for SOC estimation, enabling decentralized model training while preserving data locality. The approach leverages a long short-term memory (LSTM) network trained across multiple clients under varying configurations. The best-performing configuration achieved an R² score of 0.9828, outperforming centralized benchmarks while significantly reducing training time by 44.4%. The model’s robustness was further validated on diverse mission profiles, including baseline, extended cruise, and power reduction scenarios, where it consistently maintained R² scores above 0.9570 with highest per mission score of 0.9921.The findings of this work demonstrate that FL can serve as an effective and scalable solution for SOC estimation in eVTOL applications, combining high predictive accuracy with computational efficiency and strong privacy guarantees. This positions FL as a viable approach for next-generation battery intelligence in electric aviation.