Statistical and artificial intelligence approaches for energy, raw material, and quality optimization in steel production and hot rolling processes

The steel industry, due to its high-temperature processes and large production volume, is one of the sectors with the highest energy and raw material consumption in the manufacturing industry. Therefore, improving the efficiency of production processes is critically important for achieving both economic benefits and environmental sustainability. In this doctoral study, four different experimental investigations were conducted focusing on scrap-based rebar production processes. First, in the roughing stage of hot rolling, a six-pass rolling parameter optimization was performed using the Taguchi–Grey Relational Analysis method. As a result of this optimization, a 7.3% reduction in energy consumption was achieved, and dimensional consistency was improved. This study contributed to both energy efficiency and dimensional product quality through the systematic determination of process parameters. Subsequently, Multiple Linear Regression (MLR) and Artificial Neural Network (ANN) models were developed and compared for predicting rolling force and material spread. These models were trained using data obtained from FEM analyses, and their prediction accuracies were evaluated. Although the ANN model demonstrated slightly higher accuracy than the MLR model in predicting rolling force and spread, the difference between the two was limited. Therefore, considering its simplicity and ease of interpretation, the MLR model can also be regarded as a viable alternative. Moreover, both models provided a solid foundation for the optimization studies conducted in the next phase. In the third investigation, rolling force values calculated via MLR were optimized using the Artificial Bee Colony (ABC) algorithm, a metaheuristic optimization technique. This approach resulted in a 10.4% reduction in energy consumption and a significant decrease in unit weight variation. Finally, a Deep Neural Network (DNN) model was trained using over 39,000 data points from the production line to predict mechanical properties based on liquid steel composition and process parameters with high accuracy. Through this model, manganese usage was reduced by 12%, preventing approximately 24,000 tons of CO₂ emissions annually and contributing significantly to environmental sustainability. This integrated approach offers valuable contributions to digitalization, quality control, raw material optimization, and reduction of environmental impact in integrated steel production, while also establishing a strong foundation for future studies.