The Similarity Recognition of Pilots’ Operational Action Sequence Based on Blocked Dynamic Time Warping during a Flight Mission

Human errors are the primary cause of aviation safety accidents. It is necessary to identify the potential operation errors of a crew during a flight task. In this work, we propose a flight mission recognition method using Blocked Dynamic Time Warping (BDTW) to detect the operational errors. DTW is a popular method to calculate the similarity of time series, and can also be used as an effective sequence alignment method for pilots’ operational sequence recognition. Traditional DTW has a high computational complexity, and cannot meaningfully align the time series when the outlier exists. In contrast, BDTW based on the encoded representation of time series to reduce calculation time can be applied to the alignment of the operational sequence consisting of a series of the areas of interest (AOI) of pilot’s actions. Since the same flight mission has the same operation process, we use BDTW to progressively align multiple AOIs under a mission, and then extract common operations as the pattern. In this way, the interference of noise on recognition is mitigated. The data with 10 flightcrews in 5 different missions was used for test and the results show high recognition accuracy of the proposed method.