Research On The Prediction Of The Bending Springback Based On Grey Theory And Neural Network Model

Springback is an inevitable physical phenomenon in the process of bending. Because of the presence of the springback, the shape and size of the forming parts are largely changed and it is difficult to meet the design requirements. Therefore, how to improve the prediction accuracy of bending springback has been a hot and difficult point in industry.Recently, the surrogate model method is increasingly applied in metal forming process, especially for the prediction of the springback. However, due to the springback affected by a great number of factors, different shape parts have different springback laws. And the prediction accuracy of the surrogate model method applied in the bending springback is not ideal. So the springback remains to be further studied. In addition, at industrial practice, controlling process parameters is the general way to reduce the springback. Based on the several difficulties in affecting the accuracy of the bending springback prediction, which included how to adjust the process parameters to reduce the springback, and how to choose the springback prediction method, as well as how to establish high-precision springback approximate model, the method of combination forecasting creating surrogate model is applied to carry out the research in the bending springback prediction.Firstly, the influence degree of various process parameters in the springback is determined by grey relation analysis method. Then, the greater influence degree of the process parameters are choosen as the design variables.A new grey neural network model, which combines the advantages of grey model and BP neural network model, is developed to predict the bending springback. The model has several improvements in the prediction of bending springback, including employ multi-factor input grey prediction model. Grey model is used to do a coarse prediction task as the main model firstly, and then the auxiliary model-BP neural network is applied to correct errors. At last, a best grey model is obtained through finding the best weight values to optimize the background value corresponding to the differential in the grey model, in order to improve the springback prediction accuracy of grey neural network model.Secondly, the U-shaped piece, which is the typical bending model form NUMISHEET’93, is taken as an example. And the analysis of the impact parameters of the U-shaped piece in springback process is conducted in order to obtain the original data. Then the main influence factors, which are the blank holder force, the friction coefficient between the die and the sheet metal and the die clearance, are determined by grey relational method. At last, an approximate model of grey neural network is established with these three factors as design variables. Experiments are conducted using the model to predict springback after forming and the result show s that the average relative error is only2.26%. Finally, the prediction effect of this method is compared to the other surrogate models of other literature. The result shows that the surrogate model method based on the combination of grey relation and improved grey neural network model, which has less design variables, not only greatly improves springback prediction accuracy, but also significantly enhances the efficiency of the controlling process parameters to reduce the springback.