Estimation of number of roll passes in roll forming using neural network

Cold rolls, especially the number of passes, to roll-form a flat sheet to a specified cross-section have exclusively been designed by those who have had enough expertise. The potential problem with it lies in that the expertise is difficult to be inherited and to be trained due to absence of documented materials. To help solve the problem, we report a method of estimating the number of roll passes in roll forming using neural network. Based on that the data have many relevant parameters but that an enough number of learning data is not available, we optimize the learning process in the number of iterations, and we contrive a few shape factors. The method is applied to the actual data of 27 symmetrical cross-sections, to obtain the result that the number can be estimated within accuracy of ±1 pass for 85% of the cases.