A Novel Approach for Predicting Rolling Force in Cold Tandem Rolling Process based on Physics-Informed Neural Network

Abstract: In cold tandem rolling process, rolling force is a crucial parameter in the rolling process. Precise prediction of rolling force aids in enhancing the rolling process stability and the precision of the strip product dimensions. However, the rolling force mechanism prediction model and the data prediction model have problems of insufficient prediction accuracy and prediction instability, respectively. Therefore, in this paper, comprehensively evaluating the strengths and weaknesses of mechanism and data model of rolling force, a data and mechanism cooperative rolling force prediction model is proposed. Through the Physics-Informed Neural Network (PINN), the rolling force mechanism constraints are integrated into the neural network model. Then, by collecting rolling process data from actual production, HILL mechanism model, Artificial Neural Network (ANN) data model, and PINN model are applied to rolling force prediction of six different stands in tandem cold rolling mill, and a comparison is made of the prediction effect generated by various models. The findings indicate that the PINN model exbibits superior prediction precision and robustness, leading to enhanced strip quality and the rolling process stability. In addition, the PINN approach is of great significance for engineering fields that want to improve model accuracy and stability through the cooperation of data and mechanism.