Application of neural-network for improving accuracy of roll-force model in hot-rolling mill

Abstract In this paper, a long-term learning method using neural-network is proposed to improve the accuracy of rolling-force prediction in hot-rolling mill. The statistical analysis shows that the overall thickness accuracy at the first-coil of the lot was very low compared to that of non-first coils. Frequent lot changes from the various product ranges make the conventional short-term learning insufficient to compensate the thickness error at this point. Thus, to solve this problem, a corrective neural-network is trained to predict the rolling force for the first coil and the conventional learning output is used for the remaining coils of the lot. By doing so, the thickness error at the head-end part of the strip is considerably reduced.