A Subdivision Method of Flight Phases Based on Divide-and-Conquer Gaussian Mixture Model

Monitoring the flight status of aircraft is crucial for ensuring safe and reliable flights. A global monitoring model is a commonly used method to adapt to the monitoring of whole flight process. However, the global monitoring model does not accurately capture the features when the distribution of flight data is dynamically transformed with flight phases, which affects the performance of the model. Therefore, it is essential to subdivide the data into different flight phases by analyzing distributions before building the monitoring model. In view of this, this article proposes the divide-and-conquer Gaussian mixture model (DcGMM) method for a subdivision of flight phases. To improve the division performance of complex flight phases, the proposed method divides the subdivision task into independent subtasks based on both longitudinal and lateral aircraft motion. Then, separately conquer problems of both subtasks based on Gaussian mixture model (GMM) models, not only to divide general flight phases by clustering but also to divide transition phases by using probabilities generated from GMM with optimization thresholds. Finally, combine division results of both subtasks, and a boundary-based flight phase correction method is designed for decreasing misclassified phases to achieve a high-performance subdivision. In summary, a new high-precision subdivision method for complex flight phases is proposed in this article. This article validates the performance of the proposed method experimenting on NASA public datasets from flight recorded data. Compared with state-of-the-art methods, the proposed method demonstrates higher performance by the accuracy and macro $F1$ score.