Abstract: The analysis of QAR data is important to continuously improving the quality of flight operations. During the flight, the pilot controls the equipment, such as the rod, the plate, the rudder according to the dynamic changes of environmental conditions and the state of the aircraft. It is a process of constant adjustment and coordination, which increases the difficulty of data analysis. So whether pilots have similar operation patterns and what effects these operations hase on the QAR monitoring indexare of great interest to us.
In this paper, by studying the feature extraction method of the multivariate time series data of flight parameters, the definition of similarities of flight operations is analyzed.The piecewise linear fitting based multivariate Dynamic Time Warping distance is employed to depict the similarities of flight operations.The hierarchical clustering analysis is used to recognize the similar patterns of flight operations. And then, the descriptive statistics and the Kolmogorov-Smirnov test is adopted to quantify the relations between flight patterns and the QAR monitoring index. The judgement of risk levels is obtained. Finally, the validity of the model is verified by using the actual QAR data recorded during the landing stage of a specific aircraft.Other classifiers like BP Neural Networks and Support Vector Machine are used to compare with the proposed method.It turns out that the raised method provides an effective way to analyze flight operationsand the relationship between flight patterns and the QAR monitoring index.
In the future studies,focus will be put on the better description of multivariate time series and clustering methods for multivariate time series. The proposed approach could also be applied in the analysis of other vehicle driving, for example the monitoring of car driving. The method advocated couldhelp tofind the recurring patterns of drivers and how they affectsafety.

Key words: QAR, multivariate time series, operation pattern, risk analysis