Research On The Optimization Of Sheet Metal Forming Process Based On RBF Neural Network

Sheet metal stamping process is widely used in various industrial fields.It is the main forming method of thin-wall metal parts.However,because of the complexity of sheet metal forming,the main forming defects in some complex parts and new material parts are wrinkling and cracking.In order to satisfy the requirements of sheet metal forming such as reducing the design cycle of die and product,decreasing the design costs,improving product quality and so on,based on the computer numerical simulation technology,more and more researches on the optimization of sheet metal stamping process parameters are conducted by the combination of experimental design method,approximate model and optimization algorithm.Although the approximation model and the optimization algorithm can effectively solve the problem of sheet forming process parameter optimization,the accuracy of the approximation model directly affects the accuracy of the optimization results.In order to improve the prediction accuracy of the approximate model,two improved RBF neural networks are proposed in this paper.The SA algorithm is used in SA-RBF neural network model to optimize the number and overlapping coefficients of the hidden layer nodes of the K-means clustering RBF neural network;the second RBF neural network is trained by the shared niche technology to solve the problem that the nodes are non-global optimal and the nodes converge slowly for the existing K-means clustering training method.It is proved that the nonlinear approximation performance and the prediction accuracy of the two improved RBF neural networks are significantly improved compared with the K-means clustering RBF neural network,which is applied to the fitting and prediction in the nonlinear function.In order to improve the performance of multi-objective intelligent evolutionary algorithms,the NSGA-II algorithm is studied.The immune operator is used to select the elite individuals of each non-dominated individuals to improve the search performance of NSGA-II algorithm.The application in multi-objective function proves that the improved NSGA-II algorithm has a better performance of multi-objective optimization.The SA-RBF neural network is applied to study the drawbead setting method of sheet metal forming.The NUMISHEET 02 fender is used as the research object,of which six equivalent drawbead forces are used as input variables.Based on the Spearman correlation analysis and the Latin hypercube method,the training samples are generated and the numerical simulations are carried out.The non-linear mapping relationship is established by SA-RBF neural network using the forming quality evaluation function established by wrinkle defects and crack defects.The improved NSGA-Ⅱ algorithm and the grey relational analysis theory are used to determine the optimal equivalent drawbead force and the simulation analysis is carried out.The comparison result shows that this method can obtain better equivalent drawbead force to improve the forming quality.The RBF neural network with shared niche technology is applied to study the loading path of variable blank holder force(VBHF)in sheet metal forming.The double C part is used as the research object,in which 5 different constant BHF are set as input variables.Based on the Spearman correlation analysis and the Latin hypercube method,the training samples are generated and the numerical simulations are carried out.With the maximum thickness and maximum reduction rate as the optimization target,the non-linear mapping relationship is established by the RBF neural network with shared niche technology.The improved NSGA-II algorithm and the grey relational analysis theory are used to determine the optimal loading path of variable blank holder force.The simulation results show that the loading path of variable blank holder force obtained by this method can effectively reduce the forming defects and improve the forming quality.