摘要
采用BP神经网络算法预测断裂参数J和A2。将三点弯曲试件有限元数值实验结果作为神经网络的训练样本,经过训练得到拓扑结构为4-25-2的BP网络模型。建立了J-积分、约束参数A2这二者与裂纹尺寸、裂纹尖端附近三个应变值之间的非线性映射。结果表明:应用BP网络时,只要选取适当的传递函数、训练函数、隐含层数目、神经元个数、学习速率就可以得到较好的预测,满足应用要求。
J-A 2 theory can well describe the elasticity-plasticity stress and strain fields at crack-tip.The structure integrality assessment needs to evaluate the J-integral and the constraint parameter A 2.Here the BP neural network was introduced to predict these fracture parameters.The FEA numerical simulant results of some bend specimens were used as the training specimens of the neural network.Once training,the BP neural network was established,where the topology structure was 4-25-2.The non-linear relationship was simulated among the crack length,three strain values along the ligament near the crack-tip,J-integral and constraint parameter A 2.The result indicates that BP network model predicts J-integral and constraint parameter A 2 accurately.
出处
《应用力学学报》
CAS
CSCD
北大核心
2009年第2期379-382,共4页
Chinese Journal of Applied Mechanics
关键词
裂纹
BP神经网络
有限元
约束效应
J-积分
crack,BP neural network,finite element,constraint effect,J-integral.
