Application of neural networks to predict the width variation in a plate mill

The width variation of steel plate during broadside and finish rolling passes in a plate mill has been investigated using neural networks. It was found that the width variation after broadside rolling and finish rolling is affected by the edging ratio, the broadside rolling ratio, the longitudinal rolling ratio, the width deviation after the broadside pass, the temperature, the width-to-thickness ratio, and so on. Neural network modeling of a back propagation learning algorithm with one hidden layer has been conducted on the width variation prediction during the plate rolling. The prediction for the width variation according to rolling sequence was classified into two rolling processes; broadside rolling and finishing rolling. A performance test showed that the well-trained neural network model can interpolate the width variation very effectively. The prediction accuracy was improved by adjusting the adaptive learning algorithm. Based on these prediction models, the width variation of the final plate is much more decreased.