Bayesian model mixing for cold rolling mills: Test results

Abstract

The contribution presents the results of a collaborative R&D effort of two private companies and two national research institutions, joined at the European level. It was aimed to develop an enhanced on-line predictor of the strip thickness in the rolling gap. The issue dealt with is the absence of a reliable delay-free measurement of the outgoing strip thickness or the gap size, making the thickness control a challenging task. Although several satisfactory solutions have been used for decades, and modern control theory has been exploited as well, the pervasive competition in the field of metal strip processing emphasizes the need of a novel, more precious measuring method. The solution developed within the completed project is based on a parallel run of several adaptive Bayesian predictors whose outputs are continuously mixed to provide the best available rolling gap size prediction. The system was already tested in open loop in a real industrial environment for two reversing cold rolling mills processing steel and copper alloys strips, respectively.