Abstract

Crack formation is an issue that significantly undermines the quality and productivity of steel production. In previous studies, a solidification and microstructure model known as InterDendritic Solidification (IDS) has been developed and implemented in various slab casters in Finland. Numerous quality criteria have been derived from the model outputs to identify the general phenomena which increase the risks of defect formation in different steel grades. The aim of this study is to study the feasibility of these criteria in providing input data for predicting quality in a group of defect-prone steel grades with rule-based decision-making and machine learning algorithms. To this end, three steel grades are studied by utilizing measured compositions and comparing the quality criteria with plant data regarding reported defects. The computations are carried out by coupling IDS with a fundamental model for simulating heat transfer (Tempsimu) in continuous casting. The results indicate that for the studied steel grades, phenomenological quality criteria can be applied to predict the formation of cracks and other defects. Trends contributing to increased risks of defect formation are identified for all the studied steel grades, and possibilities for avoiding defects by changes in the compositions of these steel grades are also proposed.