A review and bibliometric analysis of intelligent techniques for advanced battery state estimation in aviation propulsion systems

Abstract

This review assesses advanced battery state estimation techniques for electric aviation, focusing on machine learning (ML), filtering methods, and fuzzy-based energy management strategies. Aviation batteries face unique challenges, including extreme fluctuations in power demand and significant variations in temperature and pressure across flight phases. These conditions disturb battery behavior and complicate state of charge (SOC), state of health (SOH), and remaining useful life (RUL) estimations. Filtering methods such as Kalman and particle filters demonstrate resistance to noise and dynamic loads. However, their application remains mostly limited to unmanned aerial vehicles (UAVs), with minimal studies addressing hybrid-electric aircraft (HEA) and none focused on electric vertical takeoff and landing (eVTOL) aircraft. ML techniques, including deep and hybrid models, offer adaptability under harsh conditions. Nevertheless, most studies rely on non-benchmarked or static-temperature datasets, limiting real-world relevance, especially in eVTOL aircraft applications. Transformer models outperform traditional deep learning approaches in low temperatures, showing promise for HEAs. Fuzzy-based techniques, while less suited for regression tasks, are widely adopted for energy management due to their ability to incorporate expert-defined logic via membership functions and rule-based control. However, most hybrid fuzzy systems lack interpretability evaluation, which poses a barrier to certification and deployment. This review highlights critical gaps, including the insufficient aviation-specific datasets, the underutilization of lithium polymer batteries in intelligent models, and the need for adaptive, context-aware estimation architectures tailored to dynamic aviation missions.

Keywords: Long short-term memory, Neural networks, Physics-informed models, Random forest, Regression