Doctoral Dissertations

Date of Award

8-1998

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Nuclear Engineering

Major Professor

Lawrence W. Townsend

Committee Members

Marshall O. Pace, J. Wesley Hines, Arthur E. Ruggles

Abstract

Space radiation is an important issue for manned space flight. For long missions outside of the Earth's magnetosphere, there are two major sources of exposure. Large Solar Particle Events (SPEs) consisting of numerous energetic protons and other heavy ions emitted by the Sim, and the Galactic Cosmic Rays (OCRs) that constitute an isotropic radiation field of low flux and high energy.

In deep-space missions both SPEs and OCRs can be hazardous to the space crew. SPEs can provide an acute dose, which is a large dose over a short period of time. The acute doses from a large SPE that could be received by an astronaut with shielding as thick as a spacesuit may be as large as 500 cGy. GCRs will not provide acute doses, but may increase the lifetime risk of cancer from prolonged exposures in a range of 40-50 cSv/yr. Total doses received by astronauts during SPEs or from GCRs can be estimated using the measured intensity and energy dependence of the particle fluxes in combination with reliable, accurate space radiation shielding codes.

In this research, we are using artificial intelligence to model the dose-time profiles during a major solar particle event. Artificial neural networks are reliable approximators for nonlinear functions. Calculated dose-time profiles from past major solar particle events show a cumulative Weibull distribution shape. In this study we design a dynamic network. This network has the ability to update its dose predictions as new input dose data are received while the event is occurring. To accomplish this temporal behavior of the system we use an innovative Sliding Time-Delay Neural Network (STDNN). By using a STDNN one can predict doses received from large SPEs during the event. The parametric fits and actual calculated doses for the skin, eye and bone marrow are used. The parametric data set obtained by fitting the Weibull functional forms to the calculated dose points has been divided into two subsets. The STDNN has been trained using some of these parametric events. The other subset of parametric data and the actual doses are used for testing the trained network. This is done to show that the network can generalize properly. Results of this testing indicate that the STDNN is capable of predicting doses from events that it has not seen before. The percentage error for large events is as low as 0.22% for training and testing the smooth parametric data. The actual error is as low as ±3.47%, for more than 95% of the event period. From the prediction results, we can see that, at the very beginning of the transient, some fluctuation around the target value exists, but after approximately 5% of the event is elapsed, the prediction almost overlaps the target dose value.

Download
Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS