Cross-sectional performance prediction of metal tubes bending with tangential variable boosting based on parameters-weight-adaptive CNN

Abstract

Metal bending tubes have been widely used in various fields owing to their excellent performance; however, bending defects significantly affect their quality. This study proposes a tangential variable boosting (TVB) scheme to reduce the severity of multiple defects generated during the rotary draw bending (RDB). The boosting pressure die is divided into several parts. Different process parameters were applied to the divided semi-parts based on the tangential stress distribution trend. To explore the effects of the TVB, a cross-sectional defect prediction method is proposed based on a parameter-weight-adaptive convolutional neural network (PWA-CNN). To validate the proposed prediction method, a dataset for training the prediction model was constructed using a series of numerical simulations of aluminum-alloy tubes. The prediction results of the PWA-CNN were then compared with those of the other CNNs and different weight assignment methods. The comparison results show that the PWA-CNN exhibits the best performance in predicting the tube cross-sectional distortion. Finally, the effects of the TVB on the forming quality of the bending tubes were obtained. A reasonable TVB is beneficial for reducing the wall-thinning and contour distortion ratios of the tubes and has a negligible effect on the short-axis variation rate.