Detection and identification of sensor anomaly for aerospace applications

This paper presents a data-driven approach based on mutual information (MI) and Gaussian Process Regression (GPR) for sensor anomaly detection and identification for aerospace applications. The proposed method not only detects and identifies the sensor which is anomalous, but also provides the anomaly detection accuracy, including false positive ration and false negative ration. First, the MI between sensors is calculated to detect the un-normal situation of sensors. According to the variation of MI among difference sensors, the anomalous sensor is preliminarily detected. Then, GPR is utilized to measure the actual anomalous situation. The practical power subsystem of the satellite is used to evaluate and verify the proposed approach, and the experimental results prove the effectiveness of the proposed method.