基于GR神经网络的汽车U形纵梁多工步冲压成形回弹预测分析

Springback Prediction on Multi-steps Stamping Process for U-shaped Auto Longeron Based on GR Neural Network

针对纵梁件单工步成形回弹预测误差较大的缺陷,提出对汽车纵梁件进行多工步成形回弹控制分析;在数值模拟的基础上,建立了基于广义回归神经网络(GRNN)回弹预测模型,并运用该模型对不同凹模圆角与压边力等重要成形工艺参数下的回弹值进行了模拟预测。结果表明,采用的广义回归神经网络模型的预测值与模拟试验值有较好的吻合度,说明广义回归神经网络模型能够准确地预测纵梁多工序后回弹分布。

According to the drawback that the prediction error was larger for the forming springback of longeron part after single operation, the control analysis for multi-steps forming springback of automobile longeron part was put forward; the generalized regression neural network （GRNN） springback prediction model was established based on numerical simulation, and the model was used to do some simulated prediction for springback value under some important technological parameters, such as different fillet of dies and blank-holder force. The results show that the prediction values of generalized regression neural network model and simulated test value have good inosculation, which proves that the generalized regression neural network model can accurately predict the springback distribution of longeron after multistep.