A Framework for Diagnosis of Critical Faults in Unmanned Aerial Vehicles

Abstract

Unmanned Aerial Vehicles (UAVs) need a large degree of tolerance towards faults. If not diagnosed and handled in time, many types of faults can have catastrophic consequences if they occur during flight. Prognosis of faults is also valuable and so is the ability to distinguish the severity of the different faults in terms of both consequences and the frequency with which they appear. In this paper flight data from a fleet of UAVs is analysed with respect to certain faults and their frequency of appearance. Data is taken from a group of UAV's of the same type but with small differences in weight and handling due to different types of payloads and engines used. Categories of critical faults, that could and have caused UAV crashes are analysed and requirements to diagnosis are formulated. Faults in air system sensors and in control surfaces are given special attention. In a stochastic framework, and based on a large number of data logged during flights, diagnostic methods are employed to diagnose faults and the performance of these fault detectors are evaluated against flight data. The paper demonstrates a significant potential for reducing the risk of unplanned loss of remotely piloted vehicles used by the Danish Navy for target practice.