Construction of a prediction model for properties of wear-resistant steel using industrial data based on machine learning approach

Abstract

A prediction model leveraging machine learning was developed to forecast the tensile strength of wear-resistant steels, focusing on the relationship between composition, hot rolling process parameters and resulting properties. Multiple machine learning algorithms were compared, with the deep neural network (DNN) model outperforming others including random forests, gradient boosting regression, support vector regression, extreme gradient boosting, ridge regression, multi-layer perceptron, linear regression and decision tree. The DNN model was meticulously optimized, achieving a training set mean squared error (MSE) of 14.177 with a coefficient of determination (R²) of 0.973 and a test set MSE of 21.573 with an R² of 0.960, reflecting its strong predictive capabilities and generalization to unseen data. In order to further confirm the predictive ability of the model, an experimental validation was carried out, involving the preparation of five different steel samples. The tensile strength of each sample was predicted and then compared to actual measurements, with the error of the results consistently below 5%.