Self-learning method for rolling model based on continuous surface

Abstract

The traditional self-learning method of rolling model based on division of layer,resulting in the problems of model self-learning coefficients of adjacent layers are jumping greatly,discontinuous and other issues,the construction mechanism of rolling model based on "mechanism model+feature points+quasi-interpolation+self-adaptive" is proposed,which replacing the original layer concept with multidimensional space continuous surface and upgrading the structure of rolling model. Constructing continuous surfaces characterized by feature points and the continuous function is used to interpolate the self-learning coefficients of each feature point in space to obtain the equation of smooth surface. The multidimensional space are continuous and differentiable adjacent layers,so the model self-learning coefficients can correct to any point in the space,which make a great breakthrough with the precision of the rolling model. The model has been successfully applied to on-line calculation of deformation resistance in a large hot strip mill in China. The practical application shows that the prediction accuracy of deformation resistance and rolling force is improved remarkably after the new method on line,the pre-blockade capacity of strip steel due to oversize thickness reduced by 44% to meet the hot strip rolling strip production requirements.