神经网络预测裂纹尖端约束效应

Crack-tip Constraint Effect Predicted by Neural Network

裂纹尖端约束效应的评估在结构完整性分析中十分重要。基于J-A2双参数弹塑性理论,用有限元对裂纹尖端应力、应变场进行数值模拟。研究用BP神经网络预测裂纹尖端的约束效应,采用单边缺口弯曲(SENB)试件韧带上三个点的应变值作为网络的输入数据,J-积分和约束参数A2作为输出,建立神经网络。实例数值结果表明,神经网络可以很好地模拟韧带上应变值和J-积分及约束参数A2之间的非线性关系,它可用于预测带裂纹构件裂纹尖端的约束效应。

Based on the J-A2 two parameters elastic-plastic theory, the stress and strain fields on the crack-tip were simulated by finite element method. The BP neural network was established, in which the strain data of three points along ligament were input, J-integrals and constraint parameters A2 were output. It is shown that the predicted data agree very well with the numerical experimental data. The neural network can simulate the non-linear relationship of the strains, J-integral and constraint parameter A2, and the constraint level on crack-tip can be predicted.