Prediction of Wrinkling in Thin-Walled Tube Hydraulic Push-Pull Bending Process Based on Neural Network

Abstract

The hydraulic push-pull bending process is a method that is suitable to bend thin-walled tubes with a large bending radius. Wrinkling of tubes is difficult to predict due to the comprehensive influence of various factors such as geometric shape, process parameters, and boundary conditions. In this paper, artificial neural networks (ANNs) are used to predict wrinkling morphology in a tube during the bending process. Since neural network training involves many datasets, which are difficult to generate through experiments. A finite element (FE) model of hydraulic push-pull bending process is established, and the accuracy of simulation results is validated by experimental tests. The backpropagation neural network is established with five input parameters, including: relative bending radius (R/D), relative wall thickness (t/D), die clearance (c) and internal pressure (p). And three parameters of wrinkling number (N), maximum wrinkling height (Δh), and maximum wall thickness thickening rate (η) are used as the output. The results of FE simulations are used to train, test, and validate the ANN models. A simple and effective ANN model is established for the prediction of tube bending wrinkling of DC06 material. The results show that the hybrid method combining ANN and FE can predict wrinkling morphology produced by the hydraulic push-pull bending method with a high degree of accuracy.