The accuracy of various training algorithms in tribological behavior modeling of A356-B₄C composites

Abstract

In the present study, various artificial neural network (ANN) training algorithms were implemented for finite element technique (FEM) modeling of the composites wear behavior. The experimental results show that the weight losses of the composites are less than that of unreinforced alloy. It is believed that incorporation of hard particles to aluminum alloy contributes to the improvement of the wear resistance of the base alloy to a great extent. Hard particles take part in resisting penetration, cutting and grinding by the abrasive and protect the surface. It is noted that the increase in the weight fraction of B₄C particles improves the wear resistance of the composite. The wear resistance increases with increasing the size of reinforcing particles. The FEM method is used for discretization and to calculate the transient temperature field of quenching. During the ANN training process, the weights and biases in the network are adjusted to minimize the error and to obtain a high-performance in the solution. The test set was used to check the system accuracy of each training algorithm at the end of learning. It was observed that Bayesian regularization learning algorithm gave the best prediction.