Thermal neutron beam optimization for PGNAA applications using Q-learning algorithm and neural network.

Area of Science:

• Nuclear Physics
• Materials Science
• Computational Physics

Background:

• Prompt gamma neutron activation analysis (PGNAA) is a non-destructive technique.
• PGNAA relies on thermal neutron capture reactions, necessitating thermalized neutron sources.
• Optimizing thermalization devices (TDs) for electron Linac-based neutron sources is crucial for maximizing thermal neutron flux.

Purpose of the Study:

• To develop an efficient method for optimizing the geometry of thermalization devices (TDs).
• To reduce the time-consuming nature of multi-parameter optimization using traditional Monte Carlo methods.
• To enhance thermal neutron flux for PGNAA applications.

Main Methods:

• Utilized a multilayer perceptron (MLP) neural network in conjunction with a Q-learning algorithm.
• Optimized collimator thickness and diameter using MLP for various electron energies and moderator configurations.
• Employed MCNPX2.6 code for calculating neutron flux and training the MLP model.

Main Results:

• Successfully optimized collimator geometry (thickness and diameter) using MLP.
• Determined optimal moderator thicknesses for different Linac electron energies via Q-learning and MLP.
• Achieved optimal TD setup with significantly fewer simulations compared to conventional Monte Carlo methods.

Conclusions:

• The combined MLP and Q-learning approach provides an efficient and faster method for TD optimization.
• This AI-driven optimization strategy accelerates the development of suitable neutron sources for PGNAA.
• The study demonstrates a significant reduction in computational cost for complex optimization problems in nuclear analysis.