Unsupervised Flight Phase Division Method Based on VMSD-DcTICC

The traditional unsupervised flight phase division methods have limited accuracy, failing to meet the requirements of general aviation and high maneuverability flight phase division. Therefore, this article proposes a novel flight phase division method based on vertical motion state decompositiondivide-and-conquer Toeplitz inverse covariance-based clustering (VMSD-DcTICC), which can perform autonomous correction of outlier state points, correction of flight transition phase, and precise positioning of take-off and landing points. First, according to the altitude change of adjacent points and the altitude rate in the time window, the flight data are decomposed into ascending, descending, and horizontal states by VMSD. Then, the flight data under different states are extracted and divided into different numbers of clusters by DcTICC. Finally, a feature-based cluster constraint mechanism is designed to achieve effective matching between flight phases and clusters, and sequence the time points to combine the flight phases coherently. The proposed method achieves an average accuracy of 95.79% on the NASA Tail653, which is better than other comparison algorithms, especially 15%~18% higher than traditional clustering methods. Notably, it remains applicable in typical flight and high maneuverability go-around flight, demonstrating robust generalization performance.