在数值模拟基础上利用神经网络进行缺陷预测

Defects prediction by artificial neural network based on FEM simulation

利用体积可压缩的刚塑性有限元法，对多孔体材料镦粗成形过程进行了数值模拟计算，并结合Ｌｅｅ－Ｋｕｈｎ提出的极限压缩应变准则得到了坯料鼓形表面出现裂纹的一系列极限参数。以有限元数值模拟结果作为学习样本，利用人工神经网络建立了缺陷预测系统。利用该系统的推广能力，获得了多孔体材料在镦粗成形过程中出现裂纹的极限工艺参数。结果表明，ＢＰ网络在缺陷预测方面具有重要的应用价值。

In this paper,the compressible rigid plastic finite element method(FEM) is used to simulate the deformation process of a cylindrical upsetting of porous material A successful analysis of forging process results in the prediction of stress,strain and density fields,which can be used together with the ductile fracture criterion obtained by Lee and Kuhn to check for possible defects Then it describes the application of artificial neural networks(ANN) used for defect prediction A defects prediction system based on the ANN and FEM is established,and the learning samples are gained from the results combining the rigid plastic finite element method(FEM) with the Lee Kuhn's criterion Then limit process parameters are obtained through this system The result shows that the method is effective for defects prediction