Hot-Rolled Steel Plate Thickness Prediction and Bayesian Optimization-Based Rolling Pattern Derivation

In the steel manufacturing process, the hot rolling process involves the thin rolling of materials at a high temperature. Producing hot rolled products with the optimal thickness is one of the most important tasks to meet the customers’ needs. To control the thickness, accurate measuring is essential. Because the thickness gauge is located at the exit of the rolling mill, the head part of hot rolled products cannot be controlled after measurement. Consequently, most thickness defects occur at the head part. In this study, we attempt to predict the thickness of the head part before finishing rolling process by using various machine learning methods. Further, explainable artificial intelligent methods are used to identify the factors that significantly affect to the thickness. Having identified these significant factors, we use Bayesian optimization to discover the optimal rolling pattern in finishing rolling process for the target thickness. It can be seen that the thickness deviation during rolling can be reduced by 29.6% using the optimized rolling pattern.