32th Congress of the International Council of the Aeronautical Sciences

10 - Safety and Security


H.K. Lee╣, T.G. Puranik╣, O.J. Pinon Fischer╣, D.N. Mavris╣; ╣Georgia Institute of Technology, United States

In recent years, due to the increased availability of data and improvements in computing power, application of machine learning techniques to various aviation safety problems has gained in popularity. Data-driven techniques are used for identifying, isolating, and reducing risk in aviation operations. Data collected from on-board recorders in commercial aircraft have thousands of parameters recorded in the form of multivariate time-series (continuous, discrete, categorical, etc.) which is used with machine learning models for retrospective safety analysis. While these retrospective insights are valuable, they offer limited utility for real-time risk identification. The performance and trajectory of the aircraft during the approach phase is a strong indicator of its landing performance which, in turn, is typically indicative of incident or accident probability such as runway excursions or hard landings. Energy state awareness and energy management are critical concepts in the characterization, detection, and prevention of safety-critical conditions, particularly in the approach and landing phase. Hence, landing performance is commonly measured using energy metrics such as landing kinetic energy, vertical speed, etc. In this work, a methodology for building a novel prediction model of aircraft future energy state and landing performance is developed using flight data from the approach phase. This information can provide a direct insight into the landing performance of the aircraft given the current state and history of the flight and enable proactively preventing unsafe situations. The methodology is demonstrated using publicly available flight data from NASA that contains thousands of flight records from commercial airline operations.

View Paper