Neural networks modelling of in-bar width in hot strip mill

The steel industry is restructuring to seek higher quality products, rather than a simple increase of the amount of steel produced. Artificial Intelligence provides the necessary tools to improve the current steel processing technology. This thesis deals with the rolling process in a Hot Strip Mill. Width control is considered as a major quality variable, hence this thesis concentrates on Roughing Mill Width Control. Currently, a statistical model is used for width control, which is providing a limited quality performance. Therefore, an accurate model is required in place of any current statistical model. This thesis presents the research work towards an applicational in-bar width model at a roughing mill. This work is under the Australia Research Council (ARC) collaborative project with BHP steel. Flat Products Division, Hot Strip Mill. Neural Networks have been proven to be a very good tool for modelling and prediction, for its nonlinear feature, good interpolation ability and adaptabihty to novelty situation. Hence, a Neural Network was is selected as the tool to perform modelling of the width deformation process. Statistics analysis was carried out to assist improving model accuracy and reduce network redundancy.