Studies on the prediction of springback in air vee bending of metallic sheets using an artificial neural network

Springback in air vee bending process is large in the absence of bottoming. Inconsistency in springback might arise due to inconsistent sheet thickness and material properties. Among the various intelligent methods for controlling springback, an artificial neural network (ANN) may be used for real time control by virtue of their robustness and speed. The present work describes the development of an ANN based on backpropagation (BP) of error. The architecture, established using an analytical model for training consisted of 5 input, 10 hidden and two output nodes (punch displacement and springback angle). The five inputs were angle of bend, punch radius/thickness ratio, die gap, die entry radius, yield strength to Young's modulus ratio and the strain hardening exponent, n. The effect of network parameters on the mean square error (MSE) of prediction was studied. The ANN was subsequently trained with experimental data generated from over 400 plane strain bending experiments using combinations of two punch radii, three die radii and three die gaps and five different materials. Updating of the learning rate and the momentum term was found to be beneficial. Testing of the ANN was carried out using experimental data not used during training. It was found that accuracy of predictions depended more on the number of training patterns used than on the ANN architecture. A comparison between batch and pattern modes of training showed that the pattern mode of learning was slower but more accurate. (C) 2000 Published by Elsevier Science B.V.