A steel property optimization model based on the XGBoost algorithm and improved PSO

Abstract

Exploring the relationships between the properties of steels and their compositions and manufacturing parameters is extremely crucial and indispensable to understanding the science of materials, and subsequently developing new materials. Tensile strength and plasticity, as two important properties of steels, are key to the improvement and optimization of the mechanical properties of steels. In the present paper, we propose an optimization model combining XGBoost algorithm with improved PSO to address the continuous multivariable optimization problem. The main goal is to determine the mapping functions between the tensile strength and plasticity and their influencing factors, based on a diversity of machine learning models such as Linear Regression, SVM, XGBoost, etc. After evaluating the performance these models, we then select the XGBoost model with highest accuracy as the mapping function, which has not been done in previous studies. Moreover, the determined mapping function serves as the fitness value of particle swarm optimization, after which the tensile strength and plasticity optimization with many variables is realized. Finally, the experimental results are analyzed theoretically, and proven to be effective and reliable.