Prediction of influence parameters on the hot rolling process using finite element method and neural network

Abstract

In the present investigation, a hot rolling process of AA5083 aluminum alloy is simulated using the finite element method. The temperature distribution in the roll and the slab, the stress, strain and strain rate fields, are extracted throughout a steady-state analysis of the process. The main hypotheses adopted in the formulation are: the thermo-viscoplastic behavior of the material expressed by Perzyna constitutive equation and rolling under plane-deformation conditions. The main variables that characterize the rolling process, such as the geometry of the slab, load, rolling speed, percentage of thickness reduction, initial thickness of the slab and friction coefficient, have been expressed in a parametric form giving a good flexibility to the model. The convergence of the results has been evaluated using experimental and theoretical data available in the literature. Since the FE simulation of the process is a time-consuming procedure, an artificial neural network (ANN) has been designed based on the back propagation method. The outputs of the FE simulation of the problem are used for training the network and then, the network is employed for prediction of the behavior of the slab during the hot rolling process.