Safety is of paramount importance in aviation due to the catastrophic consequences of accidents. Consequently, efforts have been made over the years to research and improve flight safety analysis techniques. Recent improvements in data storage and processing capabilities enable new techniques in flight safety analysis. The predominant safety analysis technique used by airlines is event exceedance analysis, which combines Flight Operations Quality Assurance (FOQA) data and predefined parameter safety thresholds to identify risks to safety. However, there is a need to continuously enhance risk identification and improve flight safety analysis. There are several areas where event exceedance analysis is limited. Event exceedance analysis is constrained by a reliance on pre-determined safety thresholds, and the analysis of parameters at a single point in time with no consideration for the behavior of the aircraft prior to that point. Furthermore, the pre-defined safety thresholds often are not consistent across airports and airframes, and analysis is thus limited within airport or airframe type. This research addresses these gaps by developing and implementing a robust methodology for identifying anomalous flights, and providing further insight into risk and safety trends in the approach phase of flight. This was achieved by 1) leveraging a Hierarchical clustering algorithm to categorize airports into clusters, 2) detecting anomalous flights by leveraging the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm and varying the length of the approach phase across the National Airspace System (NAS) as well as on an airport cluster level, 3) combining clustering results with an anomaly scoring algorithm to compare the detection of anomalous flights with current exceedance analysis event definitions and finally 4) investigating the impact of the length of the approach phase on outlier identification. Flight cluster labels were obtained through DBSCAN and flight anomaly scores were obtained through an anomaly scoring algorithm. Results from these studies showed that event flights (designated by event exceedance analysis) on average had higher anomaly scores than non-event flights. Airport clustering was shown to have a non-negligible impact on flight anomaly score, and should be investigated further. Finally, identifying anomalies by varying the length of the approach phase from 10 to 600 seconds prior to touchdown showed significant variability in the flight outlier designation. The last 100 seconds of approach are shown to be particularly volatile. By evaluating risk at single point, it is possible that event exceedance analysis is missing potentially dangerous events occurring during final approach. Applying a method that utilizes time series data shows promise in improving anomaly detection capabilities.

The complexity of commercial aviation operations has grown substantially in recent years, together with a diversification of techniques for collecting and analyzing flight data. As a result, data-driven frameworks for enhancing flight safety have grown in popularity. Data-driven techniques offer efficient and repeatable exploration of patterns and anomalies in large datasets. Text-based flight safety data presents a unique challenge in its subjectivity, and relies on natural language processing tools to extract underlying trends from narratives. In this paper, a methodology is presented for the analysis of aviation safety narratives based on text-based accounts of in-flight events and categorical metadata parameters which accompany them. An extensive pre-processing routine is presented, including a comparison between numeric models of textual representation for the purposes of document classification. A framework for categorizing and visualizing narratives is presented through a combination of k-means clustering and 2-D mapping with t-Distributed Stochastic Neighbor Embedding (t-SNE). A cluster post-processing routine is developed for identifying driving factors in each cluster and building a hierarchical structure of cluster and sub-cluster labels. The Aviation Safety Reporting System (ASRS), which includes over a million de-identified voluntarily submitted reports describing aviation safety incidents for commercial flights, is analyzed as a case study for the methodology. The method results in the identification of 10 major clusters and a total of 31 sub-clusters. The identified groupings are post-processed through metadata-based statistical analysis of the learned clusters. The developed method shows promise in uncovering trends from clusters that are not evident in existing anomaly labels in the data and offers a new tool for obtaining insights from text-based safety data that complement existing approaches.

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.

Among the operations in the general-aviation community, one of the most important objectives is to improve safety across all flight regimes. Flight-data-monitoring or flight-operations-quality-assurance programs have percolated in the general-aviation sector with the aim of improving safety by analyzing and evaluating flight data. Energy-based metrics provide measurable indications of the energy state of the aircraft, and can be viewed as an objective currency to evaluate various safety-critical conditions. The use of data-mining techniques for safety analysis, incident examination, and fault detection is gaining traction in the aviation community. In this paper, a generic methodology is presented for identifying anomalous flight-data records from general-aviation operations in the approach-and-landing phase. Energy-based metrics, identified in previous work, are used to generate feature vectors for each flight-data record. Density-based clustering and one-class classification are then used together for anomaly detection using energy-based metrics. A demonstration of this methodology on a set of actual flight-data records from routine operations, as well as simulated flight data, is presented, highlighting its potential for retrospective safety analysis. Anomaly detection using energy metrics, specifically, is a novel application presented here.

In recent years, multilayered hierarchical compositions of the well-known and widely used Gaussian process models called deep Gaussian processes are finding use in the approximation of black-box fu...