摘要
为快速识别结构内部未知缺陷,提出了一种基于图像四叉树比例边界有限元法和深度学习的缺陷反演模型.采用平衡四叉树递归分解原理对结构域进行网格剖分,并自动加密缺陷边缘区域网格;所得到的平衡四叉树网格模型采用比例边界有限元法进行数值分析,求解精度不受悬挂节点影响,且模型中仅含6种单元模式,自动化程度高,极大降低了缺陷反演模型训练数据集的成本;基于极限学习机作为学习规则,搭建了孔洞缺陷反演的深度学习人工神经网络模型,有效避免了模型训练过程中目标函数迭代陷入局部最优解的问题.采用统计学方法分析反演模型精度,并探究孔洞缺陷尺度及训练集样本数量对反演结果的影响.结果表明:所构建的反演模型得到的缺陷参数与预留的洁净数据相吻合,模型能够精准量化缺陷位置和尺寸.
To quickly identify unknown internal structural flaws,a flaw detection model is proposed based on the image quadtree scaled boundary finite element method(SBFEM)and deep learning.The structural domain is meshed using the balanced quadtree algorithm with the recursive decomposition principle,and the mesh refinement processes are conducted automatically in the flaw boundary regions.The balanced quadtree mesh model is numerically analyzed using SBFEM,which ensures the solution accuracy and does not affect by hanging nodes.The model contains only six types of elements,resulting in a high degree of automation and significantly reducing the cost of the training dataset.A deep learning artificial neural network model for flaw detection is built using the extreme learning machine(ELM)as the learning rule,effectively avoiding the problem of the objective function getting trapped in local optima during model training.Statistical methods are used to analyze the accuracy of the inversion model,and the effects of flaw size and the number of training samples on the results are investigated.The results show that the predicted flaw parameters closely match the reserved clean data,and the model accurately quantifies the location and size of the flaws.
作者
方涛
郑辉
赵文虎
唐森
FANG Tao;ZHENG Hui;ZHAO Wenhu;TANG Sen(School of Infrastructure Engineering,Nanchang University,Nanchang 330031,China;Jiangxi Provincial Key Laboratory of Intelligent Systems and Human-Machine Interaction,Nanchang 330031,China)
出处
《三峡大学学报(自然科学版)》
北大核心
2025年第4期45-52,共8页
Journal of China Three Gorges University:Natural Sciences
基金
国家自然科学基金项目(52109152)
江西省自然科学基金项目(20242BAB25023,20232BAB214086)
江西省大学生创新创业训练计划项目(S202310403051)。
关键词
缺陷反演
深度学习
图像四叉树
比例边界有限元法
flaw detection
deep learning
image quadtree
scaled boundary finite element method
