A fast rigid-plastic finite element method for online application in strip rolling

In this paper, a fast rigid plastic finite element method (FEM) is developed for online application in strip rolling. Firstly, a refined neural network model is established to generate an initial guess for the start of the Newton-Raphson iteration. Secondly, the Hessian matrix is divided into sub-domains for parallel computing. Thirdly, the Brent’s method is adopted to speed up the line search for the relaxation factor. Fourthly, the energy functional is partitioned according to the numbering of elements for parallel computing. Fifthly, an energy method with high numerical stability is proposed for predicting the rolling force. Finally, the numerical examples for the strip rolling show that the fast rigid-plastic FEM can meet the requirements both on the computational time and the accuracy.