Abstract
Autonomous systems, including unmanned aerial vehicles and commercial airplanes, are increasingly integrated into modern aircraft to minimize pilot errors while enhancing flight control. Ensuring flight safety requires accurate detection of anomalies in sensor data that causes error. This study, AnoSense, proposes an autoencoder-based deep neural network designed to detect anomalies in an unmanned aerial vehicle. AnoSense processes 20 flight sensor parameters to identify irregularities that could compromise operational safety. The model is trained and evaluated using NASA’s DASHlink anomaly data set, achieving 97.07% precision, outperforming conventional deep learning methods. Additionally, AnoSense is optimized for deployment on resource-constrained edge devices, with implementation and performance validation conducted on a Raspberry Pi. The experimental results demonstrate the feasibility of real-time flight anomaly detection on embedded systems, making AnoSense a promising solution to improve aircraft safety through edge computing.