Neural network modeling of temperature dependences of elastic modulus of deformable aluminum alloys

The elastic modulus (EM) of metals and alloys are the characteristics used in strength calculations of any products. Many parts and assemblies of machines and mechanisms operate at temperatures above room temperature, so it is necessary to have an understanding of the temperature dependences of elastic modulus (TDEM). In the construction sector, for elements of various civil structures made of aluminum alloys, information about the EM value becomes critically important for calculating the bearing capacity in the case of effects of elevated temperatures. However, data on EM at high temperatures are not provided for all aluminum alloys or calculated models of TDEM are presented, and those available in the reference literature are not always confirmed by experimental data. In this regard, it is necessary to adjust the models for individual alloys. The fundamental possibility of using neural networks to predict the elastic modulus of aluminum alloys in the temperature range from 20 to 400 оC has been established. Deformable aluminum alloys of 20 domestic brands and foreign brands of the 5xxx and 6xxx grades have been studied. The array of EM values of aluminum alloys at different temperatures is divided by the neural network into three samples: training, test and control in the proportions of 50–25–25% of the total data volume. After completing the training, the neural network evaluated the accuracy of reproducing the EM values, which amounted to 88, 85 and 90% for the three studied samples. The evaluation of the obtained predictive neural network model was performed on aluminum alloys 6061 and 5083, which were not included in the training sample. The type of the resulting model is a “multilayer perceptron”. The average prediction accuracy of the neural network model was more than 90%.