Artificial neural networks in neutron dosimetry

Area of Science:

• * Nuclear physics and radiation protection.
• * Computational dosimetry and artificial intelligence applications.

Background:

• * Neutron dosimetry relies on accurate measurement of neutron radiation.
• * Bonner spheres spectrometers (BSS) are commonly used but present challenges in data interpretation.
• * Ill-posed problems in dosimetry hinder accurate dose determination.

Purpose of the Study:

• * To design an artificial neural network (ANN) for calculating neutron doses.
• * To utilize only count rates from a Bonner spheres spectrometer (BSS) as input.
• * To assess the ANN's performance and robustness against uncertainties.

Main Methods:

• * Developed an ANN using MATLAB for neutron dose calculation.
• * Utilized 181 neutron spectra to train and test the network.
• * Employed MCNP 4C code for spectral processing and response matrix calculations.

Main Results:

• * The ANN successfully calculated ambient, personal, and effective neutron doses.
• * Investigated the impact of +/-5% uncertainties in BSS count rates on dose calculations.
• * Demonstrated the ANN's capability to handle ill-conditioned dosimetry problems.

Conclusions:

• * ANNs provide a viable alternative for neutron dosimetry.
• * This approach simplifies dose determination using BSS count rates.
• * The ANN method offers a robust solution for complex dosimetry challenges.