Determination of isotopes activity ratio using gamma ray spectroscopy based on neural network model

The uranium isotopes activity-ratio was determined using in-situ γ-ray spectroscopic measurements and an artificial neural network model. The method was developed to use forward-learn multilayer algorithm. Each layer consists of a perceptron, that controls the forward-learn process, and a mean-square-error mapping for the spectral data from the set of fired perceptrons. The set of output parameters should represent a vector of coefficients for double logarithmic polynomial that distinguish the instrumental efficiency. The forward-learn is controlled by a rejection function which is based on an input set Ψ of parameters that tells the neural layer to accept or reject data points. Each layer maps to the next layer by reducing chi-square-difference with the experimental uncertainty as weight. There are two supervised controls to the network, the maximum deviation from interpolated curve and the assumed initial set of rejection parameters (Ψ₀). The model was tested on spectra of known enrichments and gave an excellent agreement with low enriched uranium samples ((1.38 ± 0.14)% and (20 ± 1.55)%). The use of the algorithm on natural uranium ore and association with radium-226 daughters causes increase of uncertainty and deviation of the results from the certified value. The current algorithm provides a practical solution to a wide range of gamma-ray measurement problems encountered for in-situ characterization of uranium-containing materials. These include security, safeguards, fuel assessment, decontamination and decommissioning operations with no collimation or special setup. It is also applicable for large-scale installations.