Machine Learning Model for Thickness Evolution of Oxide Scale During Hot Strip Rolling of Steels

Oxide scales have significant influences on the surface qualities of hot-rolled steels. In the present work, an oxidation kinetic model for anisothermal conditions was employed with the aim of accurately controlling the evolution of scales during hot rolling, which could describe the evolution of scale thickness throughout hot rolling processes based on the scale growth behaviors at different stages, including reheating, rolling, and descaling. To improve the accuracy of the modeling, machine learning with the backpropagation artificial neural network (BPNN) and support vector machine (SVM) was used to calculate the values of oxidation rates and scale deformation ratios, and genetic algorithm (GA) was employed to optimize the model parameters. The modeling demonstrated high accuracy in predicting the evolution of scale thickness in hot rolling by comparing with the measured scale thicknesses. Based on the accurate predictions, the effects of steel compositions, rolling, and descaling processing parameters on scale growth behavior were analyzed, which presents the solution to working out the optimal processing windows for scale control.