Article ID: ISIJINT-2024-359
In hot rolling of steels, both microstructural evolution and work roll/steel interfacial state are critical for high quality products. Unfortunately, they are taking place in forms of black boxes because they cannot be readily detected during productions. Therefore, accurate modelling evolutionary behaviours of microstructures and surface scales during hot rolling and changes of as-rolled mechanical properties has become significantly important. In this paper, typical semi-empirical models developed since the 1970s and data-to-data models by artificial neural networks are briefly reviewed for their advantages and disadvantages. Then, physical metallurgy guided machine learning is discussed for its superiority in logicalities and accuracies. For the newest development, industrial foundation models (IFM) are proposed to integrate different processes in hot rolling and accelerated cooling, by which recrystallizations of austenite grains, strain induced precipitations, mechanical loading, and changes of interfacial friction coefficients during hot rolling can be simultaneously worked out, and dynamic continuous cooling transformation diagrams are instantly generated to account for variations of mechanical properties based on deep learning and heterogeneous data. Finally, typical applications to hot strip/plate lines for high efficiency production and stable control of mechanical properties are described.
