Advanced Monitoring, Simulation and Optimization Approaches for Improving Resource Management in Electric Steelmaking and achieving Sustainable and Circular Production

The steel industry is among the most energy and resource-intensive industrial sectors, as it is characterised by high energy consumptions, extensive use of raw materials, and significant generation of by-products. These factors contribute to its relevant environmental footprint and pose major challenges in the context of global sustainability goals. Enhancing the environmental performance of steel production processes, while maintaining the required quality standards of final products, is therefore a critical priority. Consequently, the steel sector is undergoing a transformative shift aimed at reducing carbon emissions and improving resource efficiency. Achieving these objectives requires the implementation of advanced technologies, alternative energy sources, and integrated process innovations that support a more sustainable and less carbon-intensive production model. Two key elements in this context are represented by scrap and slag. Scrap, which is composed of ferrous and non-ferrous materials, plays a vital role as a secondary raw material in both Electric Arc Furnace (EAF) and Basic Oxygen Furnace (BOF) processes. Its utilisation not only reduces the demand for virgin iron ore and energy but also significantly contributes to lowering greenhouse gas emissions. On the other side, slag is a by-product generated during the steelmaking process. Depending on the process, various types of slag are produced, each with unique chemical and physical characteristics and the possibility of being reused/recycled inside or outside the production route. Therefore, the improvement of low-quality scrap material, the valorisation and optimal management of slag are crucial to achieving sustainability and circularity in the steel sector. The present thesis works in this direction. Two case studies are presented: the first one linked to slag management and valorisation through the estimate of slag characteristics, and the second one centred on identification of copper parts in scrap to reduce the presence of this tramp element by improving the scrap quality. First, process know-how acquisition was carried out considering production units, scrap features and by-product productions. Then, advanced modelling techniques (flowsheet-based modelling and Machine Learning approaches) were developed to predict the characteristics of the slag produced, monitor the steel chemical composition and the potential environmental impacts. In addition, the development of a Decision Support System (DSS) is presented to allow the selection of optimal paths for slag recovery based on economic impacts and CO2 emissions. For the scrap topic, two Deep Learning models based on Faster RCNN + ResNet101 and YOLOv11 architectures were improved and evaluated to detect and identify the presence of copper in scrap automatically. Specifically, during the project, several data collection campaigns were conducted and, after labelling the images, several datasets were created on which the models were trained, validated and tested. Finally, the performance results (mAP and FPS) of the models were compared.