Finite element model simulation and back propagation neural network modeling of void closure for an extra-thick plate during gradient temperature rolling

The void closure behavior in a central extra-thick plate during the gradient temperature rolling was simulated and a back propagation (BP) neural network model was established. The thermal–mechanical finite element model of the gradient temperature rolling process was first developed and validated. The prediction error of the model for the rolling force is less than 2.51%, which has provided the feasibility of imbedding a defect in it. Based on the relevant data obtained from the simulation, the BP neural network was used to establish a prediction model for the compression degree of a void defect. After statistical analysis, 80% of the data had a hit rate higher than 95%, and the hit rate of all data was higher than 90%, which indicates that the BP neural network can accurately predict the compression degree. Meanwhile, the comparisons between the results with the gradient temperature rolling and uniform temperature rolling, and between the results with the single-pass rolling and multi-pass rolling were discussed, which provides a theoretical reference for developing process parameters in actual production.