DBSCAN and SVDD Based Method for Civil Aircraft Flight Anomaly Identification  
Author Di Zhou


Co-Author(s) Xiao Zhuang; Hongfu Zuo; Xufeng Zhao; Jiawei Xiang


Abstract In order to address the problem that the currently adopted exceedance detection approach based on predefined criteria does not utilize active identification of unknown abnormal flights. An anomaly flight identification method of civil aircraft that combines Density-based spatial clustering of applications with noise (DBSCAN) and Support vector data description (SVDD) is proposed in this paper. The proposed method is driven by the data in the flight data recorder (FDR) without the need for predefined criteria. Firstly, the cluster analysis is performed on the data based on DBSCAN algorithm. The feature space of normal flight is generated to determine the common pattern of the FDR data. Subsequently, the common pattern is used for assigning pseudo-labels to the FDR data without status labels. Then, the pseudo-labelled FDR data is used to train the SVDD model, which is able to detect anomaly flights with unique pattern and obtain a quantitative flight anomaly score. Finally, the visual processing of anomaly flight is realized by using the established flight reference nominal profile. The accuracy of flight anomaly identification results can be intuitively shown by the visual results.


Keywords Civil Aviation Safety; Anomaly Detection Technology; DBSCAN; SVDD; Flight Anomaly Visualization
    Article #:  RQD27-122

Proceedings of 27th ISSAT International Conference on Reliability & Quality in Design
Virtual Event

August 4-6, 2022