The role of neural networks in the optimisation of rolling processes

Applications of neural networks in the rolling of steel are reviewed. The first papers on the topic were published in 1991 and since then the number of publications has steadily increased. In most applications today, so-called back propagation networks are used. After briefly reviewing the various neural network types, the results of two case studies at Rautaruukki cold strip mill are presented. In the first case an efficiency model for tandem cold rolling was developed. By using the model it is possible to study whether a new product with a given width, strength or thickness can be produced, and the optimised mill settings can then be determined. A 1.8% improvement in efficiency was obtained with the model. The second case concerns the prediction of the mechanical properties of steel strips and temper rolling force by using neural network modelling and measured process data. The location of the coils in annealing stacks and their vanadium content were found to explain the deviation in mechanical properties. The temper rolling force could be predicted with good accuracy, which can be exploited in determining mill pre-settings. (C) 1998 Published by Elsevier Science S.A. All rights reserved.