Deep learning-enhanced safety system for real-time in-situ blade damage monitoring in UAV using triboelectric sensor

Abstract

Unmanned aerial vehicles (UAVs) are being increasingly utilized in various applications, which necessitates the assessment of their safety status. While self-powered sensors utilizing triboelectric nanogenerators have advanced fault monitoring methodologies, the effective identification of damage to UAV blades remains an area that warrants further investigation. This study presents the UAV blade damage monitoring system (UBDMS), a novel system designed for the identification of UAV blade damage. The UBDMS incorporates a blade sensor mounted on the UAV motor to record rotational data, an Arduino for initial data acquisition, and a Raspberry Pi for subsequent data processing and damage evaluation. A comprehensive analysis and testing of the sensor's structure, operational principles, and electrical output characteristics were performed. The experimental findings demonstrate that the electrical signals generated by the sensor correspond to various blade damage types within the frequency domain. However, the development of a universal and precise judgment standard proves to be difficult. To overcome this challenge, deep learning technology was utilized to analyze and evaluate friction electric signals, resulting in a classification accuracy rate of 94.4 % for damage types. This research significantly enhances UAV flight safety and introduces a new methodology for the in-situ monitoring of UAV blade damage.