A Statistical and Self-Organizing Maps (SOM) Comparative Study on the Wear and Performance of MgO-C Resin Bonded Refractories Used on the Slag Line of Ladles of a Basic Oxygen Steelmaking Plant

Understanding refractory wear variables on steelmaking ladles is important to promote safe, low-cost, high-performance and quality steel processing. Traditional statistical models are usually applied; however, the difficulties in assuming hypotheses, data multicollinearity and analyzing large amounts of data can lead to prediction errors. In the last decades, new techniques of data analysis have been sought for the construction of deterministic models from large database with computerized procedures. Among these techniques are the artificial neural networks (ANN). The present work is a comparative analysis of wear and performance evaluation of MgO-C refractories from the slag line of steelmaking ladles. It was performed using statistical and the self-organizing maps (SOM) ANN techniques to identify the main variables that cause refractory degradation. The comparative results between classical statistics and SOM analysis showed that the main variables were the desulfurized treatment route, nepheline consumption, calcium silicon additions, ladle furnace use and time, steel permanence time in the ladle, treatment times of secondary refining, type of steel product, total load in the ladles and ladle conditions. Classical statistical and SOM evaluations of the dataset (approximately 6700 heats and 1,457,518 single process data) were able to distinguish the main causes of refractory degradation, confirming the possibility of applying SOM in steelmaking analysis.