In recent years, there has been a rapid growth in the application of data science techniques that leverage aviation data collected from commercial airline operations to improve safety. This paper presents the application of machine learning to improve the understanding of risk factors during flight and their causal chains. With increasing complexity and volume of operations, rapid accumulation and analysis of this safety-related data has the potential to maintain and even lower the low global accident rates in aviation. This paper presents the development of an analytical methodology called Safety Analysis of Flight Events (SAFE) that synthesizes data cleaning, correlation analysis, classification-based supervised learning, and data visualization schema to streamline the isolation of critical parameters and the elimination of tangential factors for safety events in aviation. The SAFE methodology outlines a robust and repeatable framework that is applicable across heterogeneous data sets containing multiple aircraft, airport of operations, and phases of flight. It is demonstrated on Flight Operations Quality Assurance (FOQA) data from a commercial airline through use cases related to three safety events, namely Tire Speed Event, Roll Event, and Landing Distance Event. The application of the SAFE methodology yields a ranked list of critical parameters in line with subject-matter expert conceptions of these events for all three use cases. The work concludes by raising important issues about the compatibility levels of machine learning and human conceptualization of incidents and their precursors, and provides initial guidance for their reconciliation.

View Video Presentation: https://doi.org/10.2514/6.2022-3612.vid The total number of civil aviation operations has returned to high levels despite the COVID-19 pandemic. In the coming years, aviation passenger travel is expected to surpass pre-COVID levels and increase steadily. With the increasing flight operations and improved sensor technology, there is an explosion of flight data available for safety analysis. This indicates an opportunity for analyzing flight safety with machine learning models that require a large amount of data. Recent studies on in-flight safety have focused on the use of machine learning tools to identify, analyze, and reduce threats. One such method is focused on predicting the flight parameters with a significant contribution to the landing performance. Improving landing performance is critical to runway safety, which is one of the prime concerns within civil aviation since the highest accident rate is observed during approach and landing operations. A common issue with predictive models is the difficulty in predicting flight parameters close to touchdown due to changes in aircraft behavior from ground effect and pilot behaviors that are difficult to predict during the flare maneuver. This study is focused on analyzing the sensitivity of the flight state prediction model to those effects. To do so, predictive machine learning models are trained with varying neural network model configurations. The models are trained with flight operational quality assurance (FOQA) data from 1000 ft to 300 ft height above touchdown (HAT) during the final approach to predict true airspeed, calibrated airspeed, vertical speed, and pitch angle from 200 ft to 10 ft HAT. Then test flight data is used to demonstrate the degradation in prediction accuracy closer to the touchdown or during the flare. The point in the final approach where a reduction in prediction accuracy is observed will provide a reference point for future studies to consider when developing predictive machine learning models. This study concludes by showing the effectiveness of a neural network for capturing the aircraft dynamics for predicted altitudes where no information from those altitudes is provided.