Neural Network based Modeling and Simulation for the Optimization of Safety Logic

This thesis focuses on a knowledge based approach to reduce the incursion situation in an airport. In an effort to reduce the number of runway incursions the Federal Aviation Administration (FAA) is actively investigating automated systems to provide the air traffic controllers with early warnings about impending incidents. The Airport Movement Area Safety System (AMASS) is the first ground surveillance safety system that is being deployed at the 34 busiest airports in the United States. The optimization of the parameters, which control the safety logic algorithms that generate warnings of possible incursions, is a labor and time intensive endeavor. Currently AMASS uses over 200 safety parameter values for all airports independent of the organization or traffic flow of a specific airport. The thesis focuses on how neural networks are used to organize alert types in a semiautonomous process that will provide alerts to air traffic controllers as early as expediently. The methodology developed provides a generalized system that can be optimized independently for each airport and still generate warnings of possible incursions. Several networks exhibit good learning shown using different statistical parameters and could eventually be integrated with AMASS. Based on the current operations of neural networks it is projected that neural networks will use the AMASS log data and data filter software to extract necessary information. In the future the technologies discovered in these neural network models can help generate new versions of safety logic software for the advanced surveillance systems. Thesis Supervisor: Dr. Amar Gupta Title: Co-Director, Productivity From Information Technology (PROFIT) Initiative