Rotor Fault Detection and Identification on Multicopter based on Statistical Data-driven Methods: Experimental Assessment via Flight Tests

A robust framework for fault detection and identification of rotor faults in multicopters is validated with data from experiments with a quadcopter and a hexacopter. The rotor fault detection and identification methods employed in this study are based on excitation-response signals of the aircraft under atmospheric disturbances. A concise overview of the development of the statistical time series model for healthy aircraft using the aircraft attitudes as the output and controller commands as the input is presented. This model is utilized to extract quality features for training a simple neural network to perform effective online rotor fault detection and identification. A proper justification of choosing the method of time-series assisted neural network has been given. It is shown a statistical time-series assisted neural network employed for online monitoring in the quadcopter and hexacopter achieves accuracy over 96% and 95%, respectively. It is effective under gusts and experimental variability encountered during outdoor flight and is sensitive to even partial loss of rotor thrust.