Real-time Anomaly Detection System For Unmanned Aerial Vehicle Flight Data

With the rapid development of aerospace technology,UAV with its low cost and high flexibility and other advantages,in the military and civilian areas are more and more widely used,its function and structure are more and more complex.Real-time anomaly detection is one of the key technologies to ensure the safety and reliability o f UAVs,and can provide decision support for UAV system health management.However,most UAV systems do not have the capability of real-time anomaly detection.In this paper,the real-time anomaly detection system of UAV is developed to realize the real-time anomaly detection of UAV flight data and provide the required high performance,Low-power non-intrusive computing systems.Based on the analysis of the data interface and processing capability required for real-time anomaly detection,this paper considers the constraints of reliability,power consumption,volume and weight of UAV system,and proposes a non-invasive embedded real-time anomaly detection System solutions.The Zynq heterogeneous SoC as the core processor embedded computing platform,through the RS-422 data communication interface and Ethernet interface,UAV flight data to achieve real-time access and abnormal detection results of the feedback,the use of high-level integration(HLS)tool to achieve the prediction of LS-SVM anomaly detection algorithm and cluster-based k-means anomaly detection algorithm to accelerate the design of the module to improve the data processing capabilities to complete the UAV flight data real-time anomaly detection system development.First of all,the design and development of the hardware platform for airborne real-time anomaly detection system is completed.The hardware platform uses Zynq heterogeneous SoC as ARM and FPGA as the core processor.It adopts RS-422 and Ethernet interface to realize data interaction with UAV,and DDR3 SDRAM as data buffer.Secondly,based on the hardware platform,the modular design of LS-SVM and k-means algorithm is completed based on HLS tool,and the computational power of algorithm is improved by using the parallel computing characteristic of programmable logic(PL)part in Zynq.At the same time,to improve the data interface and exception algorithm to accelerate the data exchange rate between modules,the design of the acceleration module using standardized AXI4-Stream interface,making the UAV data can be used for rapid data transfer DMA.Finally,the real-time anomaly detection system is tested and validated with the real data of the UAV and the simulation data generated by Flight Gear.The tests of the anomaly detection processing performance and processing time have achieved good results.The feasibility and effectiveness of the real-time anomaly detection system proposed in this paper are verified by the real flight data of the UAV,which will lay a foundation for the application of the late system health management technology on the UAV.The research on the system health management of the aerospace vehicle also has a certain value and significance.