Hybrid combination of multi-layer perceptron and neutron activation analysis in cement prediction

Abstract

Determination of concentration of major elements such as Ca, Si, Al, and Fe in cement is very important for quality control during its production, correct classification according to the existing standards, and thus for appropriate use in the construction industry. For this purpose, neutron activation analysis is very suitable. In this preliminary theoretical work, the irradiation and consecutive measurement of the percentage of the constituent elements in different cement samples were done using MCNPX with γ-ray spectra as the output. Specific peaks of Ca, Si, Al, and Fe obtained from these spectra were used as input for artificial neural network (18 of them for training and 8 for testing) resulting in the determination of each element in the given sample. The mean absolute errors of the results are less than 0.4%, which is very promising for the future experimental work where the uncertainties are usually one order higher.