An Evolutionary Algorithm for Spatial Discretization Optimization

Abstract

A novel evolutionary algorithm for spatial discretization meshing using a Cartesian grid for radiation transport simulations is proposed. The generated Cartesian grid is used by ADVANTG to generate weight windows for MCNP. The algorithm was tested on the Ueki-35 benchmark and obtained a figure of merit 2.68, 3.39, and 2.25 times higher in the average, best, and worst cases respectively than the reference mesh provided with ADVANTG for the benchmark. The algorithm produced counter-intuitive solutions from a neutron shielding perspective that outperformed the reference mesh in all 15 trials. Additionally, the algorithm revealed that from the adjoint perspective, accurate geometric representation of the polyethylene collimator was more significant that then shield itself when generating optimal weight windows.

Department(s)

Nuclear Engineering and Radiation Science

Comments

This work is supported by the Nuclear Engineering Graduate Fellowship Grant from the U.S. Nuclear Regulatory Commission (NRC-HQ-13-G-38-0026).

Keywords and Phrases

Mesh generation; Optimization; ADVANTG; Co-evolution; MCNP; Mesh; Spatial discretizations; Evolutionary algorithms; Figure of merit

International Standard Serial Number (ISSN)

0149-1970

Document Type

Article - Journal

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2017 Elsevier, All rights reserved.

Publication Date

01 May 2017

Share

 
COinS