Acoustic-Based MAV Propeller Inspection

Abstract

The growing adoption of MAVs, alongside mobile vehicles and robots, has advanced research in autonomous driving, particularly for applications such as search and rescue operations and surveillance. However, MAV component failures can lead to mission-critical consequences. To enable fully autonomous operations, hangars can be strategically deployed to ensure MAV readiness and reliability without human intervention. A critical requirement for successful missions is verifying that MAV components, especially propellers, are undamaged and flight-ready, posing a challenge for real-time, automated inspection systems. This thesis argues that acoustic-based fault detection offers a non-invasive, costeffective solution for real-time propeller health monitoring in Micro Aerial Vehicles (MAVs), addressing the limitations of traditional inspection methods. Focusing on autonomous hangar deployments for emergency response scenarios like search and rescue, the study develops a methodology that integrates statistical, Mel-Frequency Cepstral Coefficients (MFCC), and Short-Time Fourier Transform (STFT) features with a hybrid ensemble of classical machine learning models, optimized using the Tree-based Pipeline Optimization Tool (TPOT). The system achieves an accuracy and F1-score of 0.9965, surpassing baseline models, and demonstrates scalability across diverse UAV models and operational conditions. By eliminating labor-intensive manual checks and resource-heavy image-based methods, the approach enhances UAV safety and operational efficiency, shifting maintenance from a reactive to a predictive paradigm. Despite challenges such as the computational cost of optimization and reliance on controlled datasets, the framework paves the way for efficient, autonomous inspection systems, with potential for transfer learning to further generalize its applicability in real-world settings. Datei-Upload durch TIB