Date of Award

Summer 2004

Document Type

Thesis - Open Access

Degree Name

Master of Science in Human Factors & Systems

Department

Human Factors and Systems

Committee Chair

Linda Trocine

Committee Member

Shawn Michael Doherty

Committee Member

Mehmet Sozen

Abstract

Expendable launch vehicles in the United States currently have a reliability of 92%. The failures that do occur cost millions of dollars in spacecraft replacement, lost revenue, and other expenses. These costs are passed on in higher insurance rates and launch vehicle price. If the launch outcome of the launch vehicles could be better predicted, the overall cost of launching payloads into space would decrease. This study used artificial neural networks to model the overall launch outcome of a launch vehicle so that the results of a launch could be predicted. Two neural network architectures--MLP and fuzzy ARTMAP--were trained on historical launch data of Atlas, Delta, and Titan launch vehicles. The networks were then tested on their ability to generalize to new data. Fuzzy ARTMAP performed slightly better than MLP overall, but neither network can be used during launch countdown today. Future application of the networks in real-time during the vehicle launch countdown will require the use of more launch specific data.

Share

COinS