Quadrotor actuator fault detection and isolation. a model-based approach

Abstract

Unmanned Aerial Vehicles (UAVs), particularly quad-rotors, have gained increasing importance across a wide range of civilian and industrial applications due to their maneuverability, vertical take-off and landing (VTOL) capabilities, and mechanical simplicity. However, their inherently nonlinear, unstable, and underactuated dynamics make them highly sensitive to actuator faults, especially those related to the brushless DC motors that directly influence flight control. Any degradation in motor performance, such as loss of effectiveness (LOE), can compromise flight safety and mission success. This thesis focuses on developing a robust model-based Fault Detection and Isolation (FDI) system capable of detecting and isolating actuator faults in real-time without requiring direct motor feedback. The proposed methodology integrates two complementary observers: a nonlinear Thau observer for estimating system states and detecting deviations in attitude dynamics and a dynamic inversion-based observer that reconstructs actuator behavior from measured accelerations and angular rates. The combination of these two observers enables dual-residual evaluation, enhancing detection sensitivity and fault isolation performance, even under noisy measurements and model uncertainties. The nonlinear Thau observer is designed using the full Newton-Euler dynamic equations of the quadrotor, avoiding linearization and enhancing estimation accuracy across the entire flight envelope. The dynamic inversion observer provides a secondary set of residuals by comparing estimated motor responses with expected motor behavior. An adaptive thresholding technique is implemented to handle noise and varying flight conditions, while a two-stage magnitude estimation mechanism enables both fast detection and reliable steady-state assessment of fault severity. Finally, a fault-tolerant control allocation strategy is incorporated to compensate for the detected faults using the estimated fault magnitudes. The complete system was tested in a simulation environment with a realistic quadrotor model. Various fault scenarios, including sudden and progressive LOE faults affecting different motors, were introduced. The results show that the proposed system effectively detects faults within a range of 0.1 to 0.3 seconds from the moment they occur. The fault magnitude was accurately estimated, enabling the controller to adjust thrust using a fault-tolerant control allocation strategy. This allowed the quadrotor to maintain stability and successfully complete its mission, even with partial motor faults. The simulation results validate that the system is capable of detecting both sudden and gradual Loss of Effectiveness (LOE) faults, estimating their magnitude, and ensuring consistent control performance. Overall, the approach developed in this thesis presents a real-time, lightweight FDI solution that enhances the reliability and safety of quadrotor UAVs, making it suitable for practical application in mission-critical scenarios.