华南理工大学学报(自然科学版) ›› 2024, Vol. 52 ›› Issue (11): 43-54.doi: 10.12141/j.issn.1000-565X.230685

• 土木建筑工程 • 上一篇    下一篇

局部随机点蚀钢构件的卷积神经网络损伤智能识别

强旭红1(), 田伟潇2, 姜旭1(), 赵波森3   

  1. 1.同济大学 土木工程学院,上海 200092
    2.中国路桥工程有限责任公司,北京 100011
    3.上海市建筑科学研究院有限公司,上海 200232
  • 收稿日期:2023-11-01 出版日期:2024-11-25 发布日期:2024-06-14
  • 通信作者: 姜旭(1982—),男,副教授,博士生导师,主要从事结构工程和智能建造研究。 E-mail:jiangxu@tongji.edu.cn
  • 作者简介:强旭红(1984—),女,副教授,博士生导师,主要从事结构工程和智能建造研究。E-mail:qiangxuhong@tongji.edu.cn
  • 基金资助:
    国家重点研发计划重点专项(2020YFD1100403);上海市科技计划项目(20DZ2253000);同济大学中央高校基本科研业务费专项资金资助项目(22120210577)

Intelligent Method for Identifying Damage of Steel Members with Localized Random Pitting Based on Convolutional Neural Network

QIANG Xuhong1(), TIAN Weixiao2, JIANG Xu1(), ZHAO Bosen3   

  1. 1.College of Civil Engineering,Tongji University,Shanghai 200092,China
    2.China Road and Bridge Corporation,Beijing 100011,China
    3.Shanghai Research Institute of Building Sciences Co. ,Ltd. ,Shanghai 200232,China
  • Received:2023-11-01 Online:2024-11-25 Published:2024-06-14
  • Contact: 姜旭(1982—),男,副教授,博士生导师,主要从事结构工程和智能建造研究。 E-mail:jiangxu@tongji.edu.cn
  • About author:强旭红(1984—),女,副教授,博士生导师,主要从事结构工程和智能建造研究。E-mail:qiangxuhong@tongji.edu.cn
  • Supported by:
    the Key Program of the National Key Research and Development Program of China(2020YFD1100403);the Project of Shanghai Science and Technology Plan(20DZ2253000)

摘要:

海洋等结构服役环境引起的点蚀会对钢结构的安全产生影响,而点蚀形式具有较强的多尺度多参数随机性。为在实际工程中对点蚀进行有效检测与损伤识别,基于卷积神经网络,结合试验研究、数值模拟、理论分析,对钢构件的局部随机点蚀进行系统研究。选用多参数局部随机点蚀数值模型,在遵循点蚀坑深度的分布模型、点蚀坑的直径时变模型的前提下,对点蚀坑的位置分布进行边界限制和交叉限制,利用Python实现点蚀坑在尺寸、位置和数量等方面的随机性,使Abaqus能够批量生成锈蚀位置和锈蚀率各不相同的钢板有限元模型,进行运算分析,得到各有限元模型的振型样本。之后,以有限元模型作为试验原型,将数值试验得到的大量前6阶振型样本作为数据集,用于建立、训练一种适用于损伤位置识别的卷积神经网络模型,并使用有限元数据集对模型的精度进行验证。最后,采用足尺试验的振型结果进一步验证卷积神经网络模型的精度。研究表明,该模型充分考虑了点蚀在形状参数和位置坐标等方面的随机性,参数合理,接近现实中的实际点蚀情况,识别准确率较高,在数值试验中点蚀损伤识别到真实区域及其相邻区域的准确率高达95.9%,在足尺试验中的准确率达到81.2%,能满足钢构件智能损伤识别实际应用的精度需求。

关键词: 钢结构, 局部点蚀损伤, 损伤识别, 卷积神经网络

Abstract:

Pitting induced by the marine environment has a significant impact on the safety of steel structures and its form exhibits a strong multi-scale and multi-parameter randomness. In order to effectively detect and identify damage in actual engineering, this paper systematically investigates local random pitting of steel members via experimental study, numerical simulation, and theoretical analysis based on convolutional neural networks. Firstly, under the premise of following the distribution model of pitting corrosion pit depth and the time-varying model of pitting corrosion pit diameter, the boundary and cross restrictions were imposed on the position distribution of corrosion pits using multi-parameter local random pitting numerical model. Python was utilized to generate randomness in the size, location, and number of pits, allowing Abaqus to generate a large number of finite element models of steel plates with varying rust locations and rust rates, and the mode shape samples of each finite element model were obtained. Then, the finite element model was used as a test prototype, and a large number of samples of the first six-order vibration patterns obtained from numerical tests were used to train a convolutional neural network model for identifying damage location. The accuracy of the model was verified using the finite element data set. Finally, the vibration results of the ruler test were used to further verify the accuracy of the convolutional neural network model. The study shows that the model fully considers the randomness of pitting corrosion in aspects such as shape parameters and position coordinates. The parameters are reasonable, close to the actual pitting corrosion situation in reality, and the recognition accuracy is relatively high. In numerical tests, the model achieved 95.9% accuracy in identifying pitting damage to the real area and its adjacent areas, and 81.2% accuracy in full-scale tests, meeting the requirements for the practical intelligent application of identifying steel component damage.

Key words: steel structure, local pitting damage, damage identification, convolutional neural network

中图分类号: