Prediction of chemical elements in cement using neutron activation analysis and artificial intelligence

This study presents a methodology for predicting the percentages of the key chemical elements (Calcium, Silicon, Aluminum, Iron and Oxygen) in Portland cement Class G samples. The approach integrates prompt gamma neutron activation analysis (PGNAA) and neutron activation analysis (NNA) and with artificial neural network (ANN) to enhance the characterization of these cement samples. A mathematical model was developed using the MCNP6 code to simulate both prompt and delayed gamma-ray emissions resulting from neutron activation in cement samples containing multiple elements. An ²⁴¹Am–Be neutron source was also simulated. To establish a relationship between the gamma radiation spectra from the neutron reactions and the elemental concentrations in the cement samples, a MultiLayer Perceptron (MLP) ANN was employed. This network, consisting of a hidden layer with three independent modules, was trained using the supervised error backpropagation algorithm. The results show exceptional accuracy in predicting the concentrations of the five main elements present in the cement, as well as its density, with an average relative error of less than 5 % for 97.92 % of the Validation test.