基于机器学习算法的轨道交通弹簧隔振器损伤识别方法  

E-mail:[e-mail]。Machine learning-based method for damage identification of spring isolators in rail transit system

作　　者：狄会霖 付伟庆 王瑞萍[1] 赵亮 王建 DI Huilin;FU Weiqing;WANG Ruiping;ZHAO Liang;WANG Jian(School of Civil Engineering,Qingdao University of Technology,Qingdao 266525,China;01DYNA(Qingdao)Test&Data Co.,Ltd.,Qingdao 266114,China)

机构地区：[1]青岛理工大学土木工程学院,青岛266525 [2]青岛零一动测数据科技有限公司,青岛266114

出　　处：《青岛理工大学学报》2025年第2期1-8,48,共9页Journal of Qingdao University of Technology

基　　金：国家自然科学基金(52178488)。

摘　　要：为保证轨道交通运行的平稳性和安全性,减少列车运行振动对周围环境的影响,一般采用在浮置板下设弹簧隔振器作为减振措施。随着地铁运行时间的增加,弹簧隔振器不可避免地会产生损伤或吊空,作为隐蔽结构其不易被日常巡检所发现,传统的检测方法效率较低,不能及时发现损伤并维修。利用ABAQUS有限元软件构建车辆-轨道耦合模型,采用均匀设计法对隔振器不同损伤工况进行设计后,通过有限元模型模拟计算,得到弹簧隔振器损伤后浮置板的振动响应。编写支持向量机、决策树和随机森林3种机器学习算法,利用模拟计算数据进行识别网络训练,建立损伤工况和振动响应的映射关系。研究结果表明,支持向量机和随机森林算法对损伤位置的识别效果较好;随机森林算法对损伤数量和损伤程度的识别效果较好。To ensure the smoothness and safety of rail transit operation and reduce the impact of train vibration on the surrounding environment,spring isolators are commonly used under floating slabs for vibration reduction.With the increase of subway operation time,spring isolators inevitably suffer damage or become suspended.As a concealed structure,it is not easy to be detected by daily inspections.Traditional detection methods are inefficient and cannot timely detect damage for maintenance.In this study,a vehicle-track coupling model was constructed using ABAQUS finite element software.Different damage conditions of isolators were designed using uniform design method,and the vibration response of the floating slab after spring isolator damage was obtained through finite element simulation calculations.Three machine learning algorithms of support vector machine,decision tree and random forest were edited,and simulation calculation data were used to train network identification and establish the mapping relationship between damage conditions and vibration responses.The results showed that the support vector machine and random forest algorithms performed better in identifying the location of damage.The random forest algorithm performed better in identifying the number and severity of damage.

关 键 词：弹簧隔振器 损伤识别 机器学习 均匀设计法 

分 类 号：TU352.1[建筑科学—结构工程]