A Dynamic Threshold Approach to Fault Detection in Uninhabited Aerial Vehicles

The integration of UAVs into civil airspace is hampered by the inability to asses and guarantee safety and reliability properties of these systems. The paper outlines a software oriented approach for the real-time detection of faults in multi-modal hybrid systems. Online model generation of continuous variables enables the detection of multiple faults, while managing computational complexity. The aileron control loop and propulsive system of a UAV using a Piccolo Plus autopilot are examined. A theoretical upper bound for the detection threshold for the ”stuck aileron” fault is derived. This enables the dynamic selection of the optimal threshold, in terms of false alarms, over a range of maneuvers in an online manner. The development of several independent models for the calculation of the propulsion system residual that use diverse and redundant data sources, allows for the diagnosis of several distinct faults via multiple thresholds on a single residual in a concurrent fashion. The use of simplified models under enforceable assumptions enables the online implementation of the proposed FDI system. Simulation data is presented to illustrate that it is possible to eliminate missed detections while reducing the rate of false alarms. The performance of the proposed FDI algorithm for the propulsion system was validated experimentally for the engine failure fault.