Real-time edge computing system for crack detection of UAV structures

Chien-Hung Lai, Zheng-Long Li, Yi Lin, Yuh-Shyan Hwang

Abstract: In this research, a real-time detection system for UAV structural cracks based on edge computing technology was proposed, aiming to improve the detection efficiency and flight safety. The system uses a 1D Convolutional Neural Network (1D-CNN) deep learning model running on a microcontroller platform to automatically extract features from vibration data and make real-time inferences without cloud computing. Compared with the traditional server-dependent fault diagnosis method, this method reduces the computing delay and network transmission pressure, and the experimental results achieves 98% detection accuracy on resource-constrained equipment, which effectively guarantees the operation safety of the UAV. This achievement can be directly deployed on the miniature hardware of the UAV to demonstrate convenience, low power consumption and high performance. Although faced with the limitations of memory and computing capability, this research successfully challenges by quantifying the model and optimizing the hardware design, providing a universal and scalable solution for UAV safety detection technology.