Flight Data Adaptive Segmentation and Classification for Fleet-level Anomaly Detection

Flight data collected by the flight data recorder (FDR) are a series of parameters which can be used to reflect flight state and performance. Once an aircraft behaves abnormally, the flight data will certainly change. Then the flight anomaly can be detected if anomalous features are captured. Clustering methods are usually used to distinguish the abnormal data from the normal massive data. Most state-of-the-art clustering methods are oriented to the data sets that have equal length which requires the isometric division of flight sequences. However, the actual sampling rates of flight parameters are different, and the duration of each flight is diverse. A huge challenge is presented for fleet-level anomaly detection. In this work, an improved Density Based Spatial Clustering of Applications with Noise (DBSCAN) approach is proposed by integrating Dynamic Time Warping (DTW) to address this issue. Two types of actual flight data sets from fleet data that have unequal length are utilized to verify the proposed method. Experimental results indicate that the improved DBSCAN algorithm can detect potential abnormal flight behaviors in both climbing stage and descending stage.