Application of neural networks to minimize front end bending of material in plate rolling process

This study proposes an approach that combines a trained neural network with a bisection algorithm to minimize the front end bending of material that occurs during plate rolling. With finite element analysis of plate rolling, front end bending data set was generated under conditions where the three rolling parameters (percentage reduction, entry material thicknesses, and percentage difference in peripheral speed between the top and bottom work rolls) varied at regular intervals. The finite element model was validated by comparing the computed roll forces, with the ones measured from a pilot plate rolling test. The pilot hot plate rolling test, wherein the rotational speeds/rates of two work rolls were independently controlled, was also performed, to validate the proposed approach. The proposed approach predicted the percentage difference in peripheral speed that minimized front end bending of the rolled material within 1 s. When the percentage difference in peripheral speed determined for the selected reduction and entry material thicknesses were input, the measured front end bending was only up to about 5 mm, which is negligible value because the ratio of the front end bending to roll diameter in the pilot plate rolling mill is only 0.0071 (5/700 mm), which is much lower than the ratio (0.02) in an actual plate rolling mill.