摘要
在工业领域中,压力容器发挥着关键性作用。随着对其要求的不断提高,压力容器朝着精密化、高压化、工作环境复杂化发展,这对压力容器的安全性评估提出了新的要求,表面裂纹是压力容器常见缺陷中最危险的一种,为保障压力容器在使用期间的安全,对含表面裂纹的压力容器安全性做出更为精确的评估具有重要工程意义。通过对含表面裂纹压力容器断裂韧性值研究,给出了不同特征参数下裂纹尖端三维J-积分值,结合神经网络方法,搭建反向传播神经网络(BPNN)模型学习多个参数与三维J-积分之间的关系,研究不同架构对模型性能的影响。采用遗传算法优化BPNN,形成GABPNN结构,探讨了不同种群规模下遗传算法的效果。训练结果最佳的GABPNN模型在验证集上的准确度达到96%以上,在未知数据上亦取得较为准确的结果。为神经网络方法在压力容器安全评定中的应用,进行了有价值的探索。
In the industrial field,pressure vessels play a key role.With the continuous improvement of its requirements,pressure vessels are developing towards precision,high pressure and complex working environment,which puts forward new requirements for the safety assessment of pressure vessels.Surface cracks are the most dangerous of common defects in pressure vessels.In order to ensure the safety of the pressure vessel during use,it is of great engineering significance to make a more accurate evaluation of the safety of the pressure vessel with surface cracks.By studying the fracture toughness values of pressure vessels with surface cracks,three-dimensional J-integral values of crack tip under different characteristic parameters are given.Combined with neural network method,backpropagation neural network(BPNN)model is built to learn the relationship between multiple parameters and three-dimensional J-integral,and the influence of different architectures on model performance is studied.Genetic algorithm was used to optimize BPNN and form GABPNN structure.The effect of genetic algorithm under different population size was discussed.The GABPNN model with the best training results has an accuracy of more than 96%on the verification set and relatively accurate results on the unknown data.A valuable exploration is made for the application of neural network method in the safety assessment of pressure vessels.
作者
李小丽
王磊
Li Xiaoli;Wang Lei(School of Information Engineering and Artificial Intelligence,Zhengzhou Vocational College of Information and Technology,Zhengzhou,He'nan 450046,China;School of Computer Science and Technology,He'nan Polytechnic University,Jiaozuo,He'nan 454003,China)
出处
《化工设备与管道》
CAS
北大核心
2023年第5期22-30,共9页
Process Equipment & Piping
基金
河南省科技攻关项目,井孔多分量阵列式瞬变电磁远探测正反演研究,项目编号:232102320318。
关键词
压力容器
断裂
三维J-积分
神经网络
pressure vessel
severance
three-dimensional J-integration
neural network
