Intelligent optimization of steel composition for continuous casting steel towards data-driven and machine learning

Abstract

Longitudinal cracks are common surface defects of continuous casting slabs, particularly in steels susceptible to severe peritectic reactions during solidification. Reasonable optimization of the chemical composition of casting steels can enhance the high-temperature mechanical properties of the slabs and reduce the risk of longitudinal cracks, which is crucial for improving slab quality and caster productivity. The present work focuses on the microalloyed peritectic steel with a high probability of longitudinal crack occurrence (C content 0.15%–0.17%), and a machine learning-based optimization model of steel composition is established by constructing a sample database of steel composition and slab quality information. The cyclic K-means clustering algorithm is used to obtain the distribution structure and characteristics of the composition and quality data, explore the mapping relationship between steel composition and longitudinal cracks, and extract the chemical composition intervals of 12 chemical elements with a lower probability of longitudinal crack occurrence. The results of continuous casting experiments show that the probability of longitudinal crack occurrence decreases from 4.5% to 0.7%, effectively reducing the risk of surface cracks and significantly improving slab surface quality. This work provides reference and guidance for optimizing the continuous casting process and improving the slab surface quality.