Prediction of wrinkling in thin-walled tube push-bending process using artificial neural network and finite element method

Abstract

The tube push bending process is a method that is mainly used to bend thin-walled tubes with a small bending radius. In this paper, artificial neural networks (ANNs) are used to predict the creation of wrinkling in a tube during the bending process. Since training the neural network involves many datasets and all of these data are difficult to generate using experiments, a credible finite element (FE) model is developed. The results obtained from the FE model are validated by conducting experimental tests. The results of the FE simulations are used to train, test, and validate the ANN models. Backpropagation neural networks based on the Levenberg–Marquardt algorithm are constructed using five design parameters including: relative bending radius; relative tube diameter; friction between die and tube; friction between tube and mandrel; and pressure as the network inputs and the maximum wrinkling height (MWH) as the single output. Two different ANN models are trained for two types of tube materials: brass and stainless steel 304. The obtained results show that by using the hybrid method that combines the ANN and FE it is possible to predict the MWH created in the push bending method with a high degree of accuracy.