基于径向基神经网络的中厚板热弯成形收缩特性分析  

Analysis of Shrinkage Characteristics of Medium Plate After Hot Bending Based on Radial Basis Function Neural Network

作　　者：张渝[1] 安治国[1] 

机构地区：[1]重庆交通大学机电与汽车工程学院,重庆400074

出　　处：《热加工工艺》2012年第3期22-24,共3页Hot Working Technology

基　　金：重庆市教委科学技术研究项目(KJ100414)

摘　　要：针对中厚钢板热弯成形冷却收缩变形问题,对影响变形的主要因素进行了分析。采用拉丁超立方抽样与有限元数值模拟相结合的方式得到试验样本数据,并在此基础上利用径向基函数神经网络建立收缩变形量的预测模型。预测值与试验值相比,最大误差小于1.5%,表明该预测模型的有效性。最后根据RBF神经网络预测的收缩变形量,以冷却过程中有限元节点的位移矢量关系为依据,对模具型面进行了补偿修正。结果表明,采用修正后的模具生产出的产品尺寸误差小于0.05%,满足形状精度的要求。According to the problem of shrinkage and deformation of medium plate in cooling process after hot bending,the main factors influencing deformation were analyzed.The sample data were obtained using the method of combination with Latin Hypercube Sampling and finite element simulation.The predicting model of shrinkage deformation was set up based on the above analysis.The predicted value of RBF model was compared with test value,the error is smaller than 1.5% and the RBF predicting model has much higher precision.At last,the compensation of deformation on the die surface was made based on the displacement vector of node during the cooling process and the predicting deformation by using RBF model.The results show that the geometric error of the part produced by the modified die is lower than 0.05% and the accuracy can meet the production requirement.

关 键 词：热弯曲 收缩变形 RBF神经网络 数值模拟 

分 类 号：TG316[金属学及工艺—金属压力加工]