Physical, Mathematical, and Machine Learning Modeling of Iron and Steel Processes

点击翻译按钮获取中文摘要

The mining and metals industry is facing major challenges and the iron and steel industry is no exception. All iron and steel makers are aiming for energy efficient and environmentally friendly operations and at the same time improving and optimizing the associated metallurgical processes to meet the stringent product quality demands at reduced cost. Having all these constraints in mind, iron and steel companies have heavily invested in research and development and one of the major areas has been physical and mathematical modeling of steelmaking and casting processes. However, the iron and steel makers have access to huge amount of process data which have been stored for a long time, and now it is essential to include data driven modeling and machine learning in solving process related problems. Today, the rapid development of modern industry and industry 4.0 is accelerating the discovery of  next-generation hybrid models which combine both fundamental and data driven concepts, and the building of digital twins of each unit operation. Hence, developing digital twins for iron and steelmaking processes is critical to strengthening Canada's competitive position in today's metals industry. Having expertise in process metallurgy, physical and mathematical modeling, and machine learning, the applicant's group at the University of Toronto aims to use quantitative experimental techniques in physical models and generate controlled process data to develop preliminary digital twins of iron and steel processes, and integrate them with industrial data to develop real digital twins of unit operations. With this long-term vision, physical and digital twins for basic oxygen furnaces (BOF), Continuous Caster (CC), Ladle Metallurgy (LMF), and  a Water Atomizer (WA) for the production of metal powders, will be developed. Researchers in our group will extensively use physical and mathematical modeling to understand the complicated underlying physics behind each process, and also generate controlled experimental data. Researchers will also use machine learning based predictions of different outputs for the above mentioned processes. Some of the key questions to be answered are: (i) Where does fundamental physical and mathematical models fail to make accurate predictions? (ii) Are the data driven predictions interpretable? (iii) Are digital twins reliable? (iv) Can we develop hybrid techniques using the power of both fundamental models based on metallurgical principles and data driven models? The short term benefits will be in-depth knowledge of the underlying physics of various unit operations and the applicability of machine learning techniques for process optimization. The long-term benefits will be the development of digital twins and hybrid models which can serve as real time optimization tools and benefit the iron and steel industry immensely. Finally, all the knowledge and discovery can be  cross pollinated to other mining and metals sectors in Canada.