基于GABP的压力容器表面裂纹断裂研究

Study on Fracture of Pressure Vessel with Surface Crack Based on GABP

压力容器在长期运行过程中表面裂纹问题难以避免,进行基于断裂分析的安全评估对压力容器的稳定运行具有较强的现实意义.针对二维J-积分理论难以应用于表面半椭圆裂纹,数值模拟耗时冗长的问题,论文提出一种采用三维J-积分量化压力容器表面裂纹尖端应力强度,再结合神经网络进行预测的安全评估方法.通过有限元方法计算了1200例不同几何尺寸、裂纹尺寸和内压载荷的含表面裂纹的压力容器问题,分析了半椭圆裂纹尖端三维J-积分结果,构建修正系数F表征材料性能、裂纹尖端奇异性以及容器几何特征对三维J-积分的影响.基于生成的机器学习数据集,搭建反向传播神经网络(BPNN)模型,采用遗传算法优化,形成GABPNN预测模型.结果表明:BPNN和GABPNN模型预测精度高达96%以上,在未知数据上亦可以取得较为准确的结果,可以高效地预测裂纹尖端三维J-积分,对于实现计算机辅助压力容器安全性现场快速评定提供新的思路和方法.

Surface cracks usually exist in pressure vessels of long service.The safety assessment based on fracture analysis is of great practical significance for the stable running of pressure vessels.The conventional method for evaluating the safety of pressure vessels with surface cracks is using the two-dimensional(2D)J-integral,which,however,has two obvious problems in practice.One is the inapplicability of 2D J-integral for evaluating semi-elliptical surface cracks;the other is that the numerical simulation is timeconsuming for elastoplastic vessels.Aiming to solve these two problems,an artificial neural network safety evaluation method based on the three-dimensional(3D)J-integral is proposed in this paper.The 3DJintegral is applied to quantify the stress intensity at the surface crack tip in the pressure vessel,and the trained neural network is constructed to predict the 3DJ-integral.By means of the finite element method(FEM),1200cases of elastoplastic pressure vessels with surface cracks of different geometric sizes,crack sizes and internal pressures are calculated.The 3DJ-integral results of semi-elliptical crack tip are analyzed.A correction factor Fis constructed to characterize the material properties,the singularity of crack tip,and the influence of vessel’s geometry on the 3DJ-integral.Based on the generated machine-learning data set from the FEM calculation,the back propagation neural network(BPNN)model is built,and the GABPNN prediction model is formed by genetic algorithm optimization.The data set is randomly divided into the training set(90%)and the validation set(10%).The training process shows that genetic algorithm optimization can accelerate the convergence speed of neural network and improve the stability of training.The results show that the prediction accuracy of BPNN and GABPNN models is higher than 96%,and relatively accurate results can be obtained from the unknown data.The 3DJ-integral of crack tip can be predicted efficiently,which provides a new idea and method for the realization of computer-aided on-site safety assessment of pressure vessels with surface cracks.