32th Congress of the International Council of the Aeronautical Sciences

10 - Safety and Security


M. Emara¹, M. dos Santos¹, N. Chartier¹, J. Ackley¹, T.G. Puranik¹, A. Payan¹, M. Kirby¹, O.J. Pinon¹, D.N. Mavris¹; ¹Georgia Institute of Technology, United States

The hazards posed by turbulence remain an important issue in commercial aviation safety analysis. Flight routes are subject to clear-air turbulence, mountain wave turbulence, and convectively-induced turbulence. In particular, clear-air turbulence is difficult to detect and predict and yet affects flight safety the most. Turbulence is among the leading cause of in-flight injury to passengers and airline employees, especially flight attendants. In severe cases it can also lead to more serious consequences such as loss of control (LOC) in flight. Current methods of turbulence detection may suffer from sparse or inaccurate forecast data sets, low spatial and temporal resolution, lack of in-situ reports, etc. The increased availability of recorded data in-flight offers an opportunity to improve the state of the art in turbulence detection. Eddy Dissipation Rate (EDR) is consistently recognized as a reliable measure of turbulence and is widely used in the aviation industry. In this paper, machine learning models (both classification and regression) are used in conjunction with flight operations quality assurance (FOQA) data collected from routine flights to estimate EDR (and thereby turbulence severity) in future time horizons. Data from routine airline operations that encountered different levels of turbulence is collected and analyzed for this purpose. A stratified sampling approach is used to ensure a balanced model is built even in the presence of data imbalance. The model consists of using a sliding window-based approach to collect hundreds of parameters from the FOQA data during the flight in order to predict EDR at a designated future time. Results indicate that the model is able to perform reasonably well in predicting EDR and turbulence severity between 10 to 20 seconds prior to encountering it. Continuous deployment of the model enables obtaining a near-continuous prediction of possible future turbulence and acts as an early warning system forrnpilots and f

View Paper