Neural-net-based imager offset estimation in fieldable associated particle imaging

Author

Justin R. Powers-Luhn, University of Tennessee, KnoxvilleFollow

Date of Award

5-2021

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Nuclear Engineering

Major Professor

Jason P. Hayward

Committee Members

Howard H. Hall, Paul A. Hausladen, Jens Gregor

Abstract

Associated particle imaging with deuterium-tritium generators has been demonstrated to be extremely useful in laboratory settings. The Oak Ridge National Laboratory Nuclear Materials Identification System is once such system. A portable imaging system with similar capabilities could be of value to non-destructive analysis of potentially hazardous items. A neural network modeled after the ResNet architecture was trained to predict the position of the detector array. The network was trained using an MCNP simulation of the NMIS system with the neutron array offset in two position and one rotation dimension. The final network was accurate to within 0.49cm, 0.66cm, and 0.66◦ for axial, lateral, and horizontal plane rotations, respectively. Images reconstructed with this predicted values are qualitatively comparable to the original object and significantly improved when compared to images without these corrections applied.

Recommended Citation

Powers-Luhn, Justin R., "Neural-net-based imager offset estimation in fieldable associated particle imaging. " PhD diss., University of Tennessee, 2021.
https://trace.tennessee.edu/utk_graddiss/6759