ABSTRACT

The development of a broad array of human performance models varying in complexity has greatly enhanced the ability to examine human-system interactions and evaluate potential benefits and drawbacks of introducing new operational concepts and technologies envisioned for the next-generation air transportation system. However, despite the benefits, there are deficiencies in existing human performance models that affect the accuracy of the predictions being made from simulation platforms. In this dissertation, I attempt to address two identified deficiencies associated with human performance models of the air traffic control system.

The first study of the dissertation systematically investigates how air traffic controllers alter flight trajectories in response to predicted aircraft pair conflict using recorded air traffic control data. A deterministic simulation was performed with open-loop flight trajectories to identify aircraft pairs that were expected to engage in aircraft conflict by calculating the separation distance between aircraft. Then, a rule-based algorithm was developed and used to automatically identify and classify operationally significant deviations in recorded transportation data by analyzing the actual positions of an aircraft to its filed flight plan when the aircraft trajectories have been identified as having encounters in aircraft conflict. The proposed data-driven approach enables the analysis of controller strategies under a variety of different conditions that exist in real-world operations, which complement empirical experiments by ensuring rigor and adding validity to various aspects of controller performance currently being incorporated into human performance models.

The second study of the dissertation determines if adding cognitive multitasking processes to computational task-analytic models of human performance result in practically different predictions of system or human performance, specifically task-relevant performance metrics and workload. Event-driven discrete simulations were run using either a simple task analytic model, a more detailed GOMS task- analytic model, and a multi-tasking QN-MHP task analytic model; these models' predictions of the task-relevant performance metrics and workload were then compared to see if the models, which differ in terms of complexity and development effort, made meaningfully different predictions. The results from the study provide useful information for modelers and stakeholders in determining whether the benefits from integrating a more complex model with multitasking and cognitive processing capabilities, outweigh the additional time and effort required for developing such models.