Predicting Post-rolling Flatness by Statistical Analysis

Abstract

A concept to improve post-rolling flatness and offer flat products to the end customer would decrease substantially run-around scrap. This would mean lower energy consumption and lower environmental load per rolled strip. Part of the concept is advanced prediction tools. This paper reports current work in post-rolling flatness prediction of cold-rolled metal strip. The work was tested in an aluminium mill in Sweden where 8-series aluminium is produced. On-line measurements are made in a cold rolling mill and post-rolling measurements in a tension levelling line, using the same measurement technique in both processing lines. This allows measurements to be easily compared. There are too many thermal and mechanical parameters to make a reliable analytical model of the post-rolling flatness. Instead, two statistical methods to predict the post-rolling flatness are evaluated: multiple linear regression and artificial neural networks. Results show that both techniques are suitable for the purpose, but multiple linear regression is preferable.