Artificial neural network based estimation of wear trace width in hard-faced structural components

Abstract

Abstract: In this paper is presented a study aimed at developing a predictive model for estimating the wear trace width of hard-faced and base metal samples based on their chemical composition. The research motivation arises from the need to establish a correlation between the alloying elements and tribological performance, thereby enabling early-stage evaluation of material behavior without extensive experimental testing. The dataset was formed using characteristic experimental cases obtained from block-on-disk tribological tests. The wear trace width values were measured after experimental tests, while the chemical composition data were taken from material specifications. A feed-forward ANN with Bayesian regularization was implemented in MATLAB. The proposed model demonstrated low agreement between predicted and experimental values due to small number of training data-sets.