Research on Anomaly Detection Method for the Executive Motor of Six-Rotor Unmanned Aerial Vehicle

In order to achieve real-time anomaly detection for the motors of multi-rotor UAV, this study proposes a anomaly detection method based on improved multivariate state estimation technique. The anomaly detection method proposed in this study includes two sections: offline and online. The offline section includes data preprocessing, memory matrix (MM) construction, and anomaly detection threshold calculation. The online section includes obtaining estimated states based on improved multivariate state estimation technique and anomaly detection based on residuals. By analyzing the flight logs of DJI Matrice600 and collecting monitoring data of the executive motor, a training set, a validation set, and a testing set were obtained. The effectiveness of the improved method was verified based on the validation set, and the results showed that estimated states obtained by the improved multivariate state estimation technique is closer to the actual states compared to traditional methods. Based on the test set, the anomaly detection method proposed in this study was tested. The experimental results showed that the proposed anomaly detection method based on improved multivariate state estimation technique can detect the operating status of the six-rotor unmanned aerial vehicle’s executive motor. Compared with the actual fault information recorded by the unmanned aerial vehicle’s executive motor, abnormal states can be detected 22 time points in advance. Compared to traditional anomaly detection method, the anomaly detection method proposed in this study can effectively reduce the error detection rate and improve the detection efficiency by 10%.