Data-Driven Anomaly Detection of UAV based on Multimodal Regression Model

Abstract

Unmanned Aerial Vehicles (UAVs) have been widely used in military and civilian applications. Meanwhile, anomaly detection as an essential part of UAV condition monitoring has become particularly critical for maintenance scheduling and mission re-planning in advance, especially for autonomous UAVs. Due to the issue of multiple flight modes dynamic switching in the actual operation, the adaptability of anomaly detection methods is always challenging when dealing with different flight trajectories. In this work, a data-driven method for flight data anomaly detection with enhanced adaptability is proposed based on multimodal regression model. Firstly, a complete flight trajectory is divided by flight mode recognition, and Relevance Vector Machine (RVM) regression is used as the basic anomaly detection method. Secondly, the dynamic input parameters of RVM models are automatically extracted by calculating the Pearson correlation coefficient between different flight parameters in each flight mode. Finally, RVM-based anomaly detection models in different flight modes are established. To deal with different flight trajectories, switching the anomaly detection model according to the flight mode can achieve adaptive anomaly detection. Experiments based on real flight data of UAV verify the effectiveness of the proposed method, and the adaptability of the anomaly detection approach can be improved.