MEMS Sensor Data Anomaly Detection for the UAV Flight Control Subsystem

MEMS sensor is being applied more and more in Unmanned Aerial Vehicle (UAV), especially for the flight control of UAV. To enhance MEMS sensor reliability, a data-driven model based on the combination of Kernel Principal Component Analysis (KPCA) and flight mode is proposed. The raw data of MEMS sensor are classified by the flight mode. Then, the training and testing data for KPCA to detect the target data are determined accordingly. False positive rate is utilized as metric to weight the performance of the anomaly detection, which can be adopted to measure the MEMS sensor reliability. The evaluation experiments are implemented based on the practical MEMS sensor data of UAV flight control subsystem. Experimental results demonstrate the effectiveness of the proposed model.