基于BP神经网络的车身U形类冲压件成形回弹预测

Prediction of Springback Value for U Shaped Parts in the Stamping Process of Auto Body Panel Based on BP Neural Network

基于MATLAB平台,将BP神经网络和数值模拟技术应用于冲压回弹预测中。采用三层BP神经网络建立基于变压边力的回弹预测数学模型,由正交实验法安排模拟实验组合,采用有限元软件进行冲压过程的数值模拟,并把端点处的Z向回弹量作为模型目标值。将模拟结果作为神经网络的输入样本对训练网络并建立网络知识源,得到了输入为工艺参数、输出为冲压回弹量的神经网络模型,并通过检验样本检验了ANN模型的准确性。实验表明:将神经网络与正交实验、数值模拟三者结合用于板料冲压参数优化可以明显缩短优化工艺参数的时间,提高工艺设计效率,同时在数值模拟实验次数一定的条件下,能获得比单纯使用正交实验和数值模拟方法更为精确的结果。

BP artificial neural network and FEM simulation were applied to optimize the design for the prediction of springback value in the stamping process based on MATLAB.A three-layer neural network was used to set up mathematical model for springback prediction based on variable pressure-pad-forces.Orthogonal test was arranged for numerical simulation to get Z-displacement at the endpoint,which was used as the target value of the model.The neural network was trained by the above Z-displacement values to form knowledge source,and the general optimal solution was attained through genetic algorithm.Nonlinear relationship between stamping process parameters and quantity of springback was obtained through the neural network,whose accuracy was testified by the test samples.The work performed in this paper shows that the combination of network,orthogonal test and numerical simulation may obviously reduce the time of optimizing process parameters and improve the process design efficiency.At the same time,on condition that the time of numerical simulation is available,the more precise conclusion can be obtained.