Hybrid Models for BOF Steelmaking and Continuous Casting Applications

The steel industry has faced environmental and economic challenges in the past few decades. As a result, cost-effective and robust process control models are in great need to maximize operational efficiency while ensuring product quality. This thesis proposes a hybrid method to model the oxygen steelmaking and continuous casting processes, which combines metallurgical knowledge and data-driven techniques such as machine learning.In the first part of the thesis, a process control model was established for the Basic Oxygen Furnace endpoint carbon, phosphorus, and temperature based on production data collected from Tata Steel India. To begin with, the importance of individual operating parameters was examined with respect to endpoint values by using Pearson’s correlation analysis, mutual information, stepwise regression, random forest, and principal component analysis. Three subsets of optimal features were selected for each endpoint, and an equal number of artificial neural networks were established, tuned, and validated. Secondly, theoretical models based on mass and energy balance were formulated based on operating parameters. Finally, the proposed hybrid model was established via exchanging inputs and outputs among theoretical and neural network models. The hybrid model results suggest that the prediction accuracies are improved with the application of machine learning algorithms. In addition, incorporating theoretical frameworks benefits the hybrid model with enhanced generalization. With the established hybrid model, a graphical user interface software was developed for operators to implement in production. In the second part of the thesis, a time-series deep learning model was established to monitor the nozzle clogging phenomenon in continuous casting based on production data of four steel grades collected from Stelco, Lake Erie Works. Firstly, a clogging index was formulated based on process parameters to indicate the severity of clogging buildup within the nozzle quantitatively. With this index, long short-term memory networks were developed to predict future clogging index and monitor the casting process. The models can predict clogging events and erosion incidents during the continuous casting for all steel grades. With the implementation of this model in production, operators can take corrective actions to reduce the frequency of clogging and improve process efficiency.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.