Simulating and Detecting Progressive Motor Thrust Degradation in Autonomous VTOL Missions

Abstract

UAVs are increasingly used to survey large and difficult-to-access areas. Vertical takeoff and landing (VTOL) platforms are especially suitable for these missions because they can operate without a runway while maintaining the efficiency of fixed-wing flight. However, a partial loss of motor effectiveness may remain hidden when the flight controller compensates for the reduced thrust. This paper presents a PX4–Gazebo framework for simulating and detecting progressive motor thrust degradation during autonomous VTOL survey missions. The fault represents the physical effect of hardware degradation or malicious actuator-command manipulation. Several degradation levels are applied during two lawnmower survey missions to generate normal and degraded flight logs. The collected telemetry is then synchronized and transformed into multivariate timeseries windows using recorded and path-relative features. Several unsupervised anomaly-detection methods are then trained using only normal flight data and evaluated on normal and degraded flight conditions. The evaluation examines their ability to detect motor degradation and compares their computational cost. The results demonstrate that progressive motor thrust degradation can be identified from onboard telemetry while the VTOL continues its survey mission.