Prediction of the boroaluminized layer thickness using an artificial neural network

Abstract

The application of mathematical models of artificial neural networks for predicting the properties of diffusion layers created by thermal-chemical treatment based on the boroaluminizing process was considered. Formalization and analysis of prediction of experimental results were carried out. It was established that the construction of computer prediction models based on experimental data of boroaluminizing was a solvable problem with high precision when using artificial neural networks of the multilayer perceptron type. Thus, testing the number of hidden layers and the number of neurons in them revealed the highest correlation coefficient R = 0.99993 of the artificial neural network using two hidden layers with ten and six neurons, respectively. The highest efficiency can be achieved using the “hyperbolic tangent” activation function.