A neutron spectrum unfolding code based on generalized regression artificial neural networks.

Area of Science:

• Nuclear Physics
• Computational Physics
• Machine Learning

Background:

• Neutron spectrometry relies on complex unfolding processes to derive spectral information from measurements.
• Artificial Neural Networks (ANNs) show promise, but Back Propagation Neural Networks (BPNN) face challenges in topology selection and training time.
• Generalized Regression Neural Networks (GRNN) offer faster training, global convergence, and improved accuracy, making them suitable for neutron spectrometry.

Purpose of the Study:

• To develop and present a computational tool based on GRNN for solving the neutron spectrometry unfolding problem.
• To automate the pre-processing, training, testing, statistical analysis, and post-processing stages of spectral analysis.
• To utilize Bonner spheres count rates as input data for the GRNN model.

Main Methods:

• Implementation of a Generalized Regression Neural Network (GRNN) computational tool.
• Automated data pre-processing, training, and testing stages.
• K-fold cross-validation (3 folds) for robust model evaluation.
• Utilizing 7 Bonner spheres count rates and a 60-energy bin response matrix from an IAEA compilation.
• Designed for a Bonner Spheres System with a 6LiI(Eu) neutron detector.

Main Results:

• The GRNN-based computational tool effectively automates the neutron spectrometry unfolding process.
• The system demonstrates efficient data handling and analysis from Bonner spheres measurements.
• The GRNN approach provides accurate spectral information with reduced training time compared to BPNN.

Conclusions:

• GRNN presents a superior alternative to BPNN for neutron spectrometry unfolding due to its speed and accuracy.
• The developed computational tool offers an automated and efficient solution for spectral analysis in nuclear applications.
• This work highlights the potential of GRNN in advancing neutron detection and measurement techniques.