Artificial Neural Network for Unfolding Accelerator-based Neutron Spectrum by Means of Multiple-foil Activation Method

In medical radioisotope (RI) production by accelerator neutron, double-differential thick-target neutron yield (DDTTNY) is necessary to be measured to estimate production amount and its radioactive and isotopic purity. We adopted the multiple-foil activation method for the measurement. The DDTTNY should be derived by an unfolding technique from measured numbers of produced atoms via the activation reactions. We have developed an unfolding code using artificial neural network (ANN) which requires no initial guess spectrum and no human-inducible convergence condition which are required for conventional unfolding methods. To demonstrate the ability to derive DDTTNY by the ANN unfolding code, we input numbers of produced atoms obtained by a multiple-foil activation experiment conducted at Kyushu University Tandem Laboratory. The resultant DDTTNY is compared with that by GRAVEL code, which is one of the conventional codes. Since there is no large discrepancy, we found that the ANN unfolding code has same ability to GRAVEL code even no initial guess spectrum was used.