The research on an adaptive rolling load prediction model based on neural networks

In order to improve the predicting precision of rolling load and avoid the waste of several preceding slats due to depending on self-learning excessively, a new approach is proposed in which deformation resistance and friction coefficient should be first predicted based on measured data by neural network, then rolling load could be predicted with an adaptive model. The numerical optimization technique Levenberg-Marquardt is used to train the neural network. The convergence is fast because the parameter /spl mu/ can be modified adaptively. The experiment has proved that the new model can predict the rolling load on temper mill with a high precision.