Propeller Damage and Screw Loosening Detection in UAV Systems Based on Acoustic Signals: An Empirical Mode Decomposition Approach

In recent years, the early detection of mechanical faults in unmanned aerial vehicle systems has become critically important in terms of operational safety and system continuity. However, most of the existing studies do not adequately consider real-world environmental conditions or the behavior of unmanned aerial vehicles at different speed levels during both airborne and ground operations, and most of them are limited solely to detecting propeller-related faults. The early identification of mechanical faults in unmanned aerial vehicle systems such as propeller damage and screw loosening in rotors, is of great significance for ensuring system safety and enhancing operational efficiency. This study proposes a novel acoustic-based fault diagnosis framework that combines Empirical Mode Decomposition and Hilbert-Huang Transform to detect propeller damage and rotor screw loosening, offering an alternative to the traditional vibration sensor-based approaches frequently encountered in the literature. In contrast to conventional vibration-based methods or fixed-band spectral analysis techniques, the proposed approach adaptively decomposes non-linear and non-stationary acoustic signals into sub-frequency components known as Intrinsic Mode Functions, without requiring any predefined basis. These components are further processed by applying Hilbert-Huang Transform to extract instantaneous amplitude and frequency features. The extracted features are comparatively analyzed under various rotor speed conditions in both faulty and non-faulty scenarios. In terms of fault detection performance, the classification accuracy of Decision Tree and Support Vector Machine algorithms have been investigated. With a maximum performance accuracy of 98.27%, the proposed approach has demonstrated the capability to effectively detect not only propeller faults but also other rotor screw loosening-related problems.