Early Fault Detection of Aircraft Components Using Flight Sensor Data

In this paper, a classification-based anomaly detection model is proposed to detect the aircraft component fault by exploring the historical flight sensor data. Detection of the aircraft component fault is formulated as a classification problem. Firstly, several sensors relevant to the fault are selected using statistical analysis. Secondly, flight phase-based statistical features are extracted using the selected sensors. Thirdly, several important features are selected using correlation analysis with the flight label. Finally, the random forest algorithm is applied to build the fault classification model based on the selected features. Experimental results show the proposed method can detect the component fault earlier than or as early as the current aircraft alarming system.