论文部分内容阅读
以汽车踏板横梁为研究对象,结合数值模拟技术与GRNN神经网络对零件翻边过程中的回弹情况进行预测。首先采用Autoform对踏板横梁翻边过程进行模拟,并与相同参数下实际零件回弹角进行对比,验证模拟结果的准确性和可替代性。再通过设计正交试验获取不同参数组合下各检测点的回弹角数据作为样本数据,并在MATLAB中对GRNN神经网络进行训练。为保证预测精度,设置多组光滑因子进行训练,发现光滑因子为0.1时,网络具有最优的逼近性能和预测性能,并作为最终网络模型进行检验。通过预测结果与真实结果进行对比,发现预测误差最大为4.3%,满足生产要求。研究表明,GRNN神经网络对板料翻边回弹预测既具有较高效率,又具有较高的精度。
Taking the automobile pedal cross beam as the research object, combining with the numerical simulation technology and the GRNN neural network, the springback condition in the part flanging process is predicted. Firstly, Autoform is used to simulate the flanging process of pedal crossbeam, and compared with the actual part rebound angle under the same parameters to verify the accuracy and substitutability of simulation results. Then, by designing orthogonal test, the springback angle data of each detection point under different parameters are obtained as sample data, and the GRNN neural network is trained in MATLAB. In order to ensure the accuracy of prediction, several sets of smoothing factors were set up for training. When the smoothness factor was found to be 0.1, the network had the best approximation and prediction performance and was tested as the final network model. By comparing the predicted results with the real ones, it is found that the maximum prediction error is 4.3%, which meets the production requirements. The research shows that the GRNN neural network predicts the flanging rebound of sheet metal with high efficiency and high precision.