An artificial neural network model was developed to predict the density of polymer materials used in gamma scattering and gamma transmission measurements

Abstract

This paper presents a new approach using an artificial neural network (ANN) allows for the density of polymer materials based on input data obtained from gamma transmission and gamma scattering measurements. Besides the results of sample density, this approach also allows for the detection of any defects inside the sample. The conventional method for measuring material density involves determining the mass and volume of a sample, and the density measured in this way is the average density of the sample. This means that it is not possible to detect defects within the sample based on average density without destroying it. Unlike the conventional approach, this proposed method allows for the measurement of the detection of a material at a specified location of the sample, known as the local density. This allows for the detection of defects inside the sample without destroying it. To achieve this, the data used to train the ANN model, were generated using a Monte Carlo simulation. The parameters of the ANN model were investigated to find the optimal configuration for predicting the density of polymer materials. The experimental data was fed into the optimal ANN model to predict the density of 8 types of polymers, including polyoxymethylene (POM), Teflon (PTFE), polyvinylidene fluoride (PVDF), polyether ether ketone (PEEK), polybutylene terephthalate (PBT), acrylonitrile butadiene styrene (ABS), polyamide (PA) and polyurethane (PU). The results showed that the relative deviation between the reference density and the predicted value was less than 4%. Based on these results, the proposed method is reliable and highly feasible when applied to measuring the density for actual measurements, providing an additional solution besides previously available methods.