InImpact: The Journal of Innovation Impact

Publisher Future Technology Press
Vol. 7 No. 2 KES Transactions on SDM I - Sustainable Design and Manufacturing 2014
Volume Editors KES International
Journal ISSN 2051-6002
Article TitleSuitability of multi-layer perceptron neural network model for the prediction of roll forces and motor powers in industrial hot rolling of high strength steels
Primary AuthorDavid James, Swansea University
Other Author(s) Alex Gater; Nick Lavery; Johann Sienz
Pages 744 - 756
Article ID sdm14-109
Publication Date 01-May-16
AbstractProcess data from a hot rolling mill is analysed to determine the variation of the load requirements of the mill stands with temperature for 13 high strength steel grades. The relationship is shown to be highly non-linear making prediction of rolling loads very difficult. Feed forward multi-layer perceptron (MLP) neural networks are suggested as a means of prediction. A methodology is presented for the generation of a system of MLP neural networks for offline prediction of required rolling forces and also motor powers during finishing rolling of a range of high strength steels. The networks are trained using approximately 12000 coils worth of process data, and prediction errors are less than 10% for load and power in over 90% of the data for most of the stands measured.

 View Paper