Rolling theory-guided prediction of PQF mill rolling force based on DE-GWO-BP

Abstract

When producing high-precision seamless steel pipes, due to their large diameter to wall ratio and poor rolling stability, they mainly rely on subjective judgement and empirical control by operators, which can easily lead to quality defects, especially the problem of uneven wall thickness along the entire length. To alleviate the quality defects of seamless steel pipes, we propose a rolling force prediction modeling method for seamless steel pipe rolling mills that uses a hybrid differential evolution (DE) and grey wolf optimisation (GWO) algorithm optimised BP neural network (DE-GWO-BP), thereby improving the accuracy of rolling force prediction and the accuracy of seamless steel pipe wall thickness control. An offline model was established by collecting historical production data from a seamless steel pipe factory. Under the same data conditions, experimental comparisons were conducted with commonly used BP and GA-BP neural network algorithms in the rolling field. The experimental results indicate that the DE-GWO-BP model has the highest prediction accuracy, with higher prediction accuracy and performance than BP and GA-BP neural network models. Industrial field applications have shown that the DE-GWO-BP model effectively reduces the phenomenon of steel stacking caused by rolling force deviation, which plays an important role in reducing the quality problems of seamless steel pipes and improving the accuracy of seamless steel pipe wall thickness control.