Speaker
Description
Radiation transport computations are an essential part of operating and maintaining an accelerator facility. However, these computations can be challenging because they require a significant amount of time and resources. Neutron transport analysis is an important computational requirement in heavy-ion facilities like the Facility of Rare Isotope Beams (FRIB) to support safe operation. At FRIB beam conditions (ion species, energy, power) vary widely, requiring frequent reevaluation of radiation environments and confirming shielding effectiveness. A Machine Learning (ML)-based approach promises to speed up neutron transport and shielding analysis.
The objective of this work is to predict the neutron differential flux at a given distance after shielding material of a certain thickness. To generate the required training dataset, we use the Monte-Carlo-based radiation transport code PHITS (Particle and Heavy-Ion Transport code System) to simulate the effects of mono-energetic and mono-directional narrow neutron beams impinging normally on a radiation shield of chosen thickness and material. The volume-averaged output differential neutron flux is used to generate the training set as a response function of a given incident neutron energy.
In view of the complexity of the problem, we have employed one-dimensional convolutional neural networks (1D-CNNs) as a candidate for our ML algorithm. The successful application of this tool in a proof of concept, albeit on a simplified problem, has provided us confidence about the suitability of this approach. We will present the required tool set and initial outcomes of our work.
This material is based upon work supported by the U.S. Department of Energy, Office of Science, Office of Nuclear Physics and used resources of the Facility for Rare Isotope Beams (FRIB), which is a DOE Office of Science User Facility, operated by Michigan State University, under Award Number DE-SC0000661.
