Statistical time series methods for multicopter fault detection and identification

This work presents the use of statistical time series methods to detect and identify rotor failures in multicopters. A concise overview of the development of various time series models using scalar or vector signals, statistics, and fault detection and identification methods has been provided. The statistical methods employed in this study are based on parametric time series representations and response-only signals of the aircraft state, as the external excitation is non-observable. The comparative assessment of the effectiveness of scalar and vector statistical models and several residual-based fault identification methods are presented in the presence of external disturbances, such as various levels of turbulence and uncertainty, and for different rotor failure scenarios. Fault identification (classification) of different rotor failures has been performed upon post-failure controller compensated steady state signals. Vector models, being more elaborate models than their scalar counterparts, exhibit superior performance in fault identification. On the other hand, residual uncorrelatedness method have greater capability to differentiate between the different rotor failures than residual variance method.