June 19, 2024
Gholam Hossein Roshani

Gholam Hossein Roshani

Academic rank: Associate professor
Education: Ph.D in Nuclear Engineering
Faculty: Faculty ofٍٍ Electrical Engineering


Combined application of neutron activation analysis using IECF device and neural network for prediction of cement elements
Type Article
IR-IECF device Artificial neural network MCNPX Neutron activation analysis Cement
Researchers Gholam Hossein Roshani، Ehsan Eftekhari zadeh، Farzin shama، A. Salehizadeh


Purpose Gamma spectrum of a cement sample is not straightforward to analyze as a result of peak overlapping produced by Compton effect of gamma rays radiated from activated elements in the neutron activation process and also because of the change into the neutron energy spectrum in the target sample during activation. Methods Artificial neural network (ANN) is an excellent solution for complex and nonlinear systems. Consequently, the use of ANN for quantitative analysis of major cement elements including Ca, Si, Al, and Fe would be very advantageous. Iranian inertial electrostatic confinement fusion device is a fast, monoenergetic and steady neutron generator. It was simulated as a high-energy neutron source for performing neutron activation analysis. In the present study, a library of 29 members of delayed gamma-ray spectra of knowing cement samples were generated via MCNPX version 2.7. Specific photo-peaks related to Ca, Si, Al and Fe obtained from these spectra were used as inputs for ANN (21 of them for training and 8 for testing). Then, using MLP architecture, an ANN model has been presented to model the system to predict the percentages of the elements in cement. The ANN model is optimized to have only a hidden layer with five neurons. Results The comparison between modeling data and results of proposed MLP network shows that there is a good consistency between them. The MAE of training set for Ca, Si and Fe outputs were 0.0351, 0.0656 and 0.0660, respectively, and the MAE of testing set for mentioned outputs were 0.0997, 0.3046 and 0.8699, respectively. Conclusion Overall, it can be concluded that the defined dispensable errors show that this modeling is a good one as a prediction tool for the mentioned purpose.