| Resumo : |
The pursuit of stringent targets on operational safety and efficiency in an increasingly complex aviation system has been driving the development of novel analytics capabilities for more proactive aviation performance management. Anomaly detection in flight operations data is a prominent approach to delivering actionable information, as anomalies are often related to critical safety events or inefficient operations. This study proposes a systematic flight data analytics framework for anomaly detection in flight operations in order to provide a comprehensive and reusable pipeline for model building, application, and explanation, applicable to both online and offline regimes and at multiple scales. We demonstrate the framework's applicability in three cases of flight operations monitoring considering both airline and Air Traffic Management (ATM) perspectives. In the first one, the framework is applied to aircraft performance data within an unsupervised learning setting with a density-based clustering approach for anomaly detection in landing operations at Minneapolis-Saint Paul International Airport. The results are compared with those obtained with exceedance-based methods used in the current practice, revealing the detection of operationally significant anomalies beyond the benchmark. In the second scenario, we apply the framework on flight tracking data within a supervised learning setting with the development of an autoencoder classifier for offline anomaly detection in approach operations at Sao Paulo/Guarulhos International Airport. Additionally, supervised learning models are developed for anomaly explanation. The autoencoder classifier detected operationally significant anomalies, while the explanatory models provided novel insights about contributing factors to the anomalies identified. For instance, we learned that anomalous flight trajectories are more likely to be associated with landing operations on runway 27 under wind scenarios, with an increase in the odds ratio of 62% and 48% for tailwinds and headwinds, respectively. In addition, we also observed a positive association between anomalies and wind gusts situations. Finally, in the third case, the framework is applied to streaming surveillance data in an online setting to predict the occurrence of go-around maneuvers for real-time air traffic control decision support. For this, two models are learned and compared in terms of predictive and computational performance: a Gaussian Mixture Model and a Temporal Convolutional Neural Network Model. The models are found to behave similarly, anticipating 67% of the go-arounds with a false positive rate of less than 8%. Moreover, we find that the inclusion of weather features in both models allows for earlier prediction of go-arounds during the approach. In particular, the Gaussian Mixture Model is able to predict 78% of the go-around maneuvers correctly as early as 55 seconds from the runway. The results from the three applications highlight the value of data-driven anomaly detection models for enhancing airline and ATM processes toward more efficient and safer flight operations. |