Propeller Fault Detection and Isolation for Multirotor Drones with Adaptation to Battery Voltage Drop

This paper addresses the problem of propeller fault detection and isolation in multirotor aerial vehicles using inertial data, explicitly accounting for the impact of battery voltage drop to ensure reliable residual generation. A complete mathematical model is presented, including the vehicle’s kinematics, dynamics, and powertrain. From this model, an experimentally fitted static powertrain model is developed, which encompasses PWM commands, supply voltage, and blade faults. This model enables effective estimation of the lift force by incorporating battery voltage measurements, which is then used by a bank of observers designed for actuator fault detection and isolation. The resulting residuals are fed to a lightweight neural network classifier, achieving $$95.04%$$ 95.04 % fault isolation accuracy despite considering small faults (starting from a $$5%$$ 5 % reduction in one propeller blade length), varying operating conditions, sensor noise, and model mismatches. The proposed method is validated through Monte Carlo simulations, and its real-time feasibility is demonstrated using processor in the loop experiments on a standard flight controller.