A neural networks approach to investigating the geometrical influence on wrinkling in sheet metal forming

Abstract

In sheet metal industries, the ability to predict and avoid surface failures, such as wrinkling, is of great importance. Wrinkling behaviour can be affected by various factors. However, little research has focused on generalisation of the role of geometry on the occurrence of wrinkling. In this paper, the geometrical features of various formed parts are generalised and correlated with surface behaviour (wrinkled or wrinkle-free). A model based on an artificial neural network is introduced to reveal the critical geometrical criteria for wrinkling. This model is a feed forward back propagation neural network with a set of geometrical variables as its inputs and the possibility of the occurrence of wrinkling as its output. After using experimental data to train and test it, the model was applied to new data for prediction of the occurrence of wrinkling. The results are promising. The network can correctly forecast a large portion of the occurrence of wrinkling based on the given information on the geometrical features of points. Besides, by examining the contribution factors of the neural network in its prediction performance, we may also obtain certain information regarding the relative importance of each geometrical variable investigated. It is believed that the neural model can be a useful tool for the product design engineer as well as the workshop practitioner.